Datasets:

---

BioLaySumm

/

BioLaySumm2025-eLife

like 0

Follow

BioLaySumm Shared Task at ACL 3

Hallmarks of cancer , including rapid growth and aneuploidy , can result in non-oncogene addiction to the proteostasis network that can be exploited clinically . The defining example is the exquisite sensitivity of multiple myeloma ( MM ) to 20S proteasome inhibitors , such as carfilzomib . However , MM patients invariably acquire resistance to these drugs . Using a next-generation shRNA platform , we found that proteostasis factors , including chaperones and stress-response regulators , controlled the response to carfilzomib . Paradoxically , 19S proteasome regulator knockdown induced resistance to carfilzomib in MM and non-MM cells . 19S subunit knockdown did not affect the activity of the 20S subunits targeted by carfilzomib nor their inhibition by the drug , suggesting an alternative mechanism , such as the selective accumulation of protective factors . In MM patients , lower 19S levels predicted a diminished response to carfilzomib-based therapies . Together , our findings suggest that an understanding of network rewiring can inform development of new combination therapies to overcome drug resistance . Protein degradation by the ubiquitin-proteasome system ( UPS ) fulfills essential roles in eukaryotic cells in maintaining proteome homeostasis ( proteostasis ) , signaling , and cell cycle progression . The 26S proteasome is a large macromolecular machine composed of the 20S catalytic core and the 19S regulator , each comprised of more than a dozen different protein subunits . The 19S proteasome regulator binds polyubiquitinated proteins and catalyzes their deubiquitination and delivery to the 20S proteasome core for proteolysis . The 20S core proteasome can also associate with alternative proteasome regulators , such as the 11S complex ( Nathan et al . , 2013; Schmidt and Finley , 2014 ) . As expected because of its essential cellular role , pharmacologic inhibition of the proteasome is inherently toxic . Consequences of proteasome inhibition leading to toxicity include the accumulation of proteasome substrates and the failure to recycle amino acids ( Suraweera et al . , 2012 ) . Intriguingly , multiple myeloma ( MM ) cells are hypersensitive to proteasome inhibition , and two inhibitors of the proteolytic activity of the 20S core , bortezomib and carfilzomib , have been approved for the treatment of MM patients ( Shah and Orlowski , 2009; Buac et al . , 2013; Röllig et al . , 2014 ) . The basis for the hypersensitivity of MM cells to proteasome inhibitors is unclear . One hypothesis poses that the high protein biosynthetic rate coupled the high secretory activity of the plasma cell-like MM cells results in an increased need for clearance of misfolded proteins by the proteasome ( Meister et al . , 2007; Bianchi et al . , 2009; Cenci et al . , 2012 ) . This would render these cells heavily dependent on the proteasome and the proteostasis network at large , and would account for the therapeutic window of proteasome inhibitors in the clinic . Most MM patients initially respond to treatment with proteasome inhibitors , but the tumors eventually develop resistance ( Buac et al . , 2013 ) . To uncover the genetic mechanisms underlying resistance to proteasome inhibitors , and to identify strategies to overcome resistance , we used our next-generation shRNA library ( Kampmann et al . , 2015 ) to screen for genes controlling the sensitivity and adaptation of MM cells to the proteasome inhibitor carfilzomib . Paradoxically , we found that knockdown of 19S regulator components desensitized cells to proteasome inhibition . Previous RNAi screens had not reported this effect ( Chen et al . , 2010; Zhu et al . , 2011 ) , however a haploid mutagenesis screen carried out independently and in parallel to this study also found the protective effect of 19S depletion ( S . Lindquist , personal communication ) . Lower 19S levels were also predictive of diminished response of MM patients to proteasome inhibitor-based therapy . To identify genetic nodes that would delineate specific dependencies of MM cells , as well as those controlling the response of MM cells to proteasome inhibitors , we conducted an RNAi screen using our next-generation shRNA library ( Kampmann et al . , 2015 ) . This library targets each mRNA by ∼25 independent shRNAs and contains thousands of negative control shRNAs to enable robust detection of hit genes . We introduced sublibraries targeting 7712 genes involved in proteostasis , cancer , apoptosis , kinases , phosphatases and drug targets into U-266 MM cells . We then split this population into two subpopulations , each of which was grown either in the absence of drug or exposed to 1-hr pulses of carfilzomib followed by recovery ( Figure 1A ) . This strategy allows for the identification of inherent vulnerabilities ( i . e . genes affecting cell growth ) , as well as genes controlling sensitivity to proteasome inhibition . We identified several hundred genes that modified the response ( either sensitizing or desensitizing ) towards carfilzomib , as well as several hundred genes whose loss impacted cell growth ( Supplementary files 1 and 2 ) . Gene Ontology ( GO ) term enrichment analysis of the hit genes from this primary screen identified the UPS , cell cycle , and translation as major functional categories controlling the cells' response towards proteasome inhibition ( Figure 1B ) . 10 . 7554/eLife . 08153 . 003Figure 1 . Screen for genes controlling the sensitivity of multiple myeloma cells to carfilzomib . ( A ) Screening strategy . ( B ) Gene Ontology ( GO ) categories enriched among the top 50 genes whose depletion results in sensitization carfilzomib and the top 50 genes whose depletion results in desensitization carfilzomib . DOI: http://dx . doi . org/10 . 7554/eLife . 08153 . 003 As expected , the genetic depletion of the multi-drug resistance ABC transporters ( ABCB1 , black circle in Figure 2A ) sensitized cells to carfilzomib . In addition , several nodes of the cytosolic proteostasis network modulated sensitivity to proteasome inhibition , including molecular chaperones ( HSPA4 , HSPA8 , HSPA90AB1; pink circles in Figure 2A ) , and stress response transcription factors ( HSF1 , NFE2L1; purple circles in Figure 2A ) . Conversely , knockdown of several genes directly participating in protein synthesis conferred protection ( green circles in Figure 2A ) , most notable including components of the EIF4F translation initiation complex ( EIF5A , EIF4A1 , EIF4E , EIF4G1 , EIF4G2 , EIF3A , EIF3F ) , as well as the elongation factor EEF2 , ribosomal RNA polymerase ( POLR1D ) , ribosomal proteins ( RPS3A , RPS6 , RPS25 ) , and MTOR , the master regulator of protein synthesis , even though knockdown of these factors in the absence of carfilzomib was detrimental to cell growth ( Figure 2B ) . This finding is consistent with the notion that decreased protein synthesis alleviates the load on proteasome ( Chen et al . , 2010; Cenci et al . , 2012 ) . 10 . 7554/eLife . 08153 . 004Figure 2 . Nodes within the proteostasis network control the response of myeloma cells to carfilzomib . ( A ) Volcano plot showing knockdown effects ( sensitization or desensitization to carfilzomib ) and statistical significance of human genes ( orange dots ) and quasi-genes generated from negative control shRNAs ( grey dots ) . Drug resistance / sensitization phenotypes were previously defined as ρ ( Kampmann et al . , 2013 ) ; a value of −1 corresponds to a twofold sensitization to the drug . Hit genes belonging to functional categories of interest are color-coded as labeled in the panels . ( B ) Volcano plot as in A , except showing effect on growth . Growth phenotypes were previously defined as γ ( Kampmann et al . , 2013 ) ; a value of −1 corresponds to a twofold reduction in growth rate . ( C ) Volcano plot as in A , highlighting the opposing effects of 20S or 19S proteasome knockdown on the sensitivity of cells towards carfilzomib . Note the protective effect is not restricted to the 19S regulator alone , but is shared with the 11S regulator . ( D ) Volcano plot as in C , except showing effect on growth . DOI: http://dx . doi . org/10 . 7554/eLife . 08153 . 00410 . 7554/eLife . 08153 . 005Figure 2—figure supplement 1 . Comparison of growth phenotypes and carfilzomib resistance phenotypes for each targeted gene . Hit genes belonging to functional categories of interest are color-coded . DOI: http://dx . doi . org/10 . 7554/eLife . 08153 . 005 Some of the relevant nodes of the proteostasis network that we identified can be targeted pharmacologically . Based on the protective effect of MTOR knockdown , we hypothesized that its inhibition by rapamycin would desensitize cells to carfilzomib . Indeed , we observed the expected protective effect of rapamycin ( Figure 3 ) . Since MTOR inhibition can also induce autophagy ( Reviewed in Sarkar , 2013 ) , we tested whether the MTOR-independent induction of autophagy by trehalose ( Sarkar et al . , 2007 ) would confer similar protection . Our results support the protective role of autophagy during proteasome inhibition ( Figure 3—figure supplement 1 ) , indicating that MTOR inhibition may desensitize to carfilzomib both through inhibition of translation and induction of autophagy . These experiments illustrate the potential of our functional genomics approach to predict drug–drug interactions on the cellular level . 10 . 7554/eLife . 08153 . 006Figure 3 . Rapamycin desensitizes cells to carfilzomib . Dose-response curves of multiple myeloma ( MM ) cells ( A , B ) and a leukemia cell line ( C ) exposed to carfilzomib after a 24 hr pretreatment with 200 nM rapamycin . FC: fold change of EC50 . Data points are means of two experimental replicates , error bars denote SD . DOI: http://dx . doi . org/10 . 7554/eLife . 08153 . 00610 . 7554/eLife . 08153 . 007Figure 3—figure supplement 1 . Induction of autophagy desensitizes cells to carfilzomib . ( A–C ) Dose-response curves of MM cells ( A , B ) and a leukemia cell line ( C ) exposed to carfilzomib after a 24 hr pretreatment with 100 mM trehalose . FC: fold change of EC50 . Data points are means of four experimental replicates , error bars denote SD . ( D ) The percent viable cells ( compared to untreated ) at the calculated EC50 for carfilzomib are interpolated from the dose response curves in ( A–C ) and Figure 3 . p values are derived from a one-tailed Student's t-test for unpaired samples . ( E ) Measurement of autophagy induction in U-266 cells transduced with a mCherry-EGFP-LC3B autophagic flux reporter and treated with 200 nM rapamycin or 100 mM trehalose for 24 hr . Reduction in GFP signal indicates an increase in autophagic flux . Chloroquine ( 75 μM , 24hr , co-incubation ) was added as a control to show inhibition of autophagic flux . DOI: http://dx . doi . org/10 . 7554/eLife . 08153 . 007 Knockdown of several subunits of the 20S proteasome core itself ( PSMB1 , PSMB4 , PSMB5 , PSMB6 , PSMA2 , PSMA3 , PSMA7 , red circles in Figure 2C ) , as well as genetic depletion of KIAA0368/ECM29 ( brown circle in Figure 2C ) , an adaptor/scaffold protein that associates with the 20S core ( Gorbea et al . , 2004 ) , and NFE2L1 ( purple circle in Figure 2A ) , a transcription factor controlling proteasome biogenesis ( Radhakrishnan et al . , 2010; Steffen et al . , 2010 ) , provided strong sensitization to proteasome inhibition . This finding is consistent with previous studies in which depletion of a protein or pathway targeted by a drug sensitizes cells to the drug used around its EC50 ( Giaever et al . , 1999; Matheny et al . , 2013 ) . Unexpectedly , the genetic depletion of the vast majority of subunits of the 19S proteasomal regulator conferred marked resistance to proteasome inhibition ( blue circles in Figure 2C ) . Notably , 19S subunits were among the strongest protective hits in our screen ( Supplementary file 2 ) . This paradoxical effect also occurred with the depletion of PSME1 and PSME2 , components of the 11S regulator ( light blue circles in Figure 2C ) . While knockdown of 20S and 19S subunits in the presence of carfilzomib had opposing phenotypes , knockdown of either 19S or 20S subunits in the absence of carfilzomib negatively impacted cell growth in all cases ( Figure 2D ) . However , the protective effect of 19S knockdown was not simply mediated by slowing cell growth , since knockdown of many other genes had a similar or more dramatic impact on cell growth without increasing resistance to carfilzomib ( Figure 2C , D , Figure 2—figure supplement 1 , Supplementary files 1 and 2 ) . The opposing protective and sensitizing effects of 19S and 20S subunit knockdown were observed for the vast majority of shRNAs targeting these genes ( Figure 4A , B ) . To validate the generality of our findings , we determined bortezomib sensitivity in a batch retest of shRNAs introduced into JJN-3 , U-266 , and RPMI-8226 MM cells , and K-562 leukemia cells . As we observed in U-266 cells treated with carfilzomib , knockdown of 20S subunits sensitized to proteasome inhibitors , whereas knockdown of 19S subunits desensitized to proteasome inhibition across this cell line panel ( Figure 4C ) . In addition , we introduced single shRNAs into MM cells targeting single subunits of either the 20S core or 19S regulator . In these experiments , expression of the individual shRNAs resulted in about twofold shifts of the dose–response curves to proteasome inhibition ( Figure 4D ) , lending further strong support to the results from the genetic screen . 10 . 7554/eLife . 08153 . 008Figure 4 . Opposing effects of 19S and 20S proteasomal subunit knockdown on carfilzomib sensitivity . ( A , B ) Scatter plots of the frequencies cells expressing different shRNAs targeting a 20S core subunit ( A ) or a 19S regulator subunit ( B ) in untreated or carfilzomib-treated cells . The grey dots represent cells expressing negative control shRNAs . Colored bars indicate the quantitative resistance phenotype ( ρ ) of each shRNA . ( C ) Heatmap showing the protective or sensitizing effect of knocking down subunits of the 19S or 20S proteasomes , respectively , in multiple cell lines . ( D ) Dose-response of U266 cells that constitutively expresses an shRNA targeting the PSMD12 subunit of the 19S proteasome , the PSMB5 subunit of the 20S proteasome , or a negative control shRNA . Data points are means of two experimental replicates , error bars denote SD . DOI: http://dx . doi . org/10 . 7554/eLife . 08153 . 008 To gain insights into the mechanistic basis of the protective effect of 19S knockdown , we first tested whether depletion of 19S subunits changed 20S levels or protects the 20S core from proteasome inhibitors . To this end , we compared the chymotrypsin-like protease activity of the β5 subunits of the 20S core ( which is the direct target of carfilzomib and bortezomib ) in lysates of cells expressing a negative control shRNA or an shRNA targeting the 19S subunit PSMD12 ( characterized in Figure 5—figure supplement 1 ) . We found that PSMD12 knockdown did not substantially increase 20S chymotrypsin-like activity , and did not affect its susceptibility to carfilzomib ( Figure 5A ) . We confirmed these findings by an orthogonal approach measuring the amounts of accessible 20S subunits targeted by carfilzomib ( β5 and LMP7 ) in a proteasome constitutive/immunoproteasome subunit enzyme-linked immunosorbent assay , ProCISE ( Parlati et al . , 2009 ) ( Figure 5B , C ) . Together , these results suggest that the loss of 19S subunits does not lead to a decreased effectiveness of proteasome inhibitors in targeting and inhibiting the 20S core . 10 . 7554/eLife . 08153 . 009Figure 5 . Proteasome activity in U266 cells expressing a negative control shRNA or an shRNA targeting the PSMD12 subunit of the 19S proteasome , and its susceptibility to inhibition by carfilzomib after a 1 hr treatment . ( A ) Fluorometric measurement of the chymotrypsin-like protease activity of the 20S proteasome . ( B , C ) Enzyme-linked immunoabsorbent assay for accessibility of the ( B ) β5 subunit and the ( C ) LMP7 subunits of the 20S proteasome . Data points are means of two experimental replicates , error bars denote SD . DOI: http://dx . doi . org/10 . 7554/eLife . 08153 . 00910 . 7554/eLife . 08153 . 010Figure 5—figure supplement 1 . Characterization of samples used to measure proteasome activity in Figure 5 . ( A ) Immunoblot analysis of PSMD12 and PSMB5 levels . Numbers below the blots correspond to the normalized relative amount ( compared to total protein in each lane ) . ( B ) PSMD12 transcript levels quantified by qPCR and normalized to GAPDH and to the untreated negative control cells . Data points are means of three primer pairs targeting the transcript , error bars denote SD . ( C ) Dose response to carfilzomib . Data points are means of two experimental replicates , error bars denote SD . DOI: http://dx . doi . org/10 . 7554/eLife . 08153 . 010 Our results support a scenario wherein the catalytic activity of the 20S core and its sensitivity to proteasome inhibitors remains unaltered in the face of 19S depletion . Therefore , we reasoned that knockdown of 19S subunits causes profound changes to cellular physiology that desensitize cells to the effects of proteasome inhibition . Because the 19S regulator delivers substrates to the 20S catalytic core ( Liu and Jacobson , 2013 ) , we hypothesized that a loss in 19S function may lead to the selective accumulation of certain proteins , some of which may mitigate the effects of 20S core inhibition . To determine the global effects of 19S knockdown on the proteome , we conducted an exploratory proteomics experiment that suggested a possible accumulation of select substrates ( Figure 6—figure supplement 1 , Supplementary files 3 and 4 ) . Among the proteins accumulating upon 19S knockdown were protein degradation factors , whose accumulation we verified by quantitative immunoblot ( Figure 6 ) . These included SQSTM1/p62 , a cargo receptor protein that delivers polyubiquitylated proteins to aggresome-like bodies degraded by the autophagosome ( Pankiv et al . , 2007 ) , and two subunits of a heterotrimeric complex functioning in endoplasmic reticulum ( ER ) -associated protein degradation: UFD1L and the triple AAA ATPase VCP/p97 ( Wolf and Stolz , 2012 ) . Notably , these factors accumulated upon PSMD12 knockdown both in the presence and absence of carfilzomib , but not ( or to a lesser degree ) upon carfilzomib treatment ( Figure 6 ) . 10 . 7554/eLife . 08153 . 011Figure 6 . Depletion of the 19S protease regulator causes the accumulation of specific substrates . Immunoblot analysis of protein levels in U-266 cells expressing a negative control shRNA or an shRNA targeting 19S subunit PSMD12 , untreated or exposed to low ( 200 nM ) or moderate doses ( 2 μM ) of carfilzomib for 4 hr . Numbers below the blots correspond to the normalized relative amount ( compared to total protein in each lane ) . Numbers on the left margin of each panel indicate molecular weights ( kDa ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08153 . 01110 . 7554/eLife . 08153 . 012Figure 6—figure supplement 1 . Determination of the global effects of 19S proteasome depletion on the proteome . ( A ) Schematic of proteomics experiments in K562 cells expressing a negative control shRNA ( N . C . ) or an shRNA targeting the PSMD6 subunit of the 19S proteasome , either untreated or treated with bortezomib . ( B ) Log2 enrichment of proteins derived from proteome-wide SILAC in cells treated with bortezomib or in cells where the 19S proteasome subunit PSMD6 was knocked down . Overall Pearson correlation between the proteomic changes induced by the two treatments is 0 . 3 . Blue and green lines correspond to the top 50 enriched proteins for each experiment . GO terms enriched within these sets were Protein Degradation ( for PSMD6 knockdown , FDR-corrected p < 0 . 001 ) and Cell cycle ( for bortezomib treatment , FDR-corrected p < 0 . 13 ) ( C ) Increased levels of factors involved in autophagy and the ubiquitin-proteasome system ( UPS ) specifically by PSMD6 knockdown . PSMD6 KD ( bortezomib ) refers to the effect of PSMD6 KD in the context of bortezomib treated samples ( i . e . log2 ratio of PSMD6 KD + bortezomib over control KD + bortezomib ) , whereas Bortezomib ( PSMD6 KD ) refers to the effect of bortezomib treatment in the context of PSMD6 KD cells ( i . e . the log2 ratio of PSMD6 KD + bortezomib over PSMD6 KD + DMSO ) . ( D ) Fold enrichment of ubiquitylated peptides caused by PSMD6 knockdown . Prominent hits represent the UPS and autophagy as protein degradation pathways . DOI: http://dx . doi . org/10 . 7554/eLife . 08153 . 012 By contrast , the abundance of the anti-apoptotic Bcl-2 family member MCL1 , a protein rapidly turned over by the proteasome ( Schwickart et al . , 2010 ) , was sharply increased by carfilzomib treatment , but much less affected by 19S knockdown ( Figure 6 ) . Similarly , 19S knockdown did not lead to an overall enhancement of NFkB signaling , a pathway critical for survival and proliferation of lymphoid malignancies ( Demchenko and Kuehl , 2010 ) ( Figure 6 ) . A caveat of this comparison is that the PSMD12 knockdown was constitutive , whereas the carfilzomib treatment was acute . Notwithstanding , these observations are consistent with the hypothesis that the protective phenotype we observed upon 19S knockdown may arise from the selective upregulation of protein turnover pathways and not from the upregulation of pro-survival pathways . Proteasome inhibitors bortezomib and carfilzomib are used clinically to treat MM patients . Most patients respond to proteasome inhibitors to varying degrees , but eventually develop resistance . We sought to test whether the role of 19S levels in controlling sensitivity to proteasome inhibitors , which we identified in cell lines , would also be relevant in MM patients . To this end , we isolated CD138-positive cells , which include MM cells and plasma cells , from the bone marrow cells of pre-treatment MM patients . Patients then underwent carfilzomib-based combination therapy . Using previously defined criteria ( Durie et al . , 2006 ) , we classified patients based on their response to therapy as complete responders ( including complete response [CR] , and stringent complete response , [sCR] ) or partial responders ( including partial response [PR] , and very good partial response [VGPR] ) . Using flow cytometry , we quantified levels of 19S regulator subunit S7 ( PSMC2 ) , 20S core subunit beta-4 ( PSMB2 ) and aggresomes in pre-treatment CD138-positive cells . We found that 19S proteasome levels were significantly higher in the group of patients who achieved CR after treatment compared to partial responders ( p < 0 . 0007 , Mann–Whitney test; Figure 7A ) . By contrast , levels of the 20S core proteasomes were not significantly different between complete and partial responders ( Figure 7B ) . Similarly , aggresome levels were not predictive of clinical outcomes ( Figure 7C ) . The combination therapy used in the clinical study included lenalidomide , and while we cannot exclude an impact of 19S levels on lenalidomide response , the rapidity and depth of the response in this study ( Korde et al . , 2015 ) suggest that anti-tumor activity is mostly due to carfilzomib , since lenalidomide has consistently been reported to act more slowly and have a much lower CR rate than carfilzomib ( Mateos et al . , 2013 ) . 10 . 7554/eLife . 08153 . 013Figure 7 . 19S proteasomal subunit levels predict the response to carfilzomib-based therapy in patients . Levels of ( A ) 19S subunit PSMC2 , ( B ) 20S subunit PSMB2 and ( C ) aggresomes quantified by flow cytometry in CD138 + bone marrow cells ( including plasma cells and MM cells ) of patients prior to therapy in clinical trial for carfilzomib-based combination therapy . Values are shown for separately for complete responders and partial responders . DOI: http://dx . doi . org/10 . 7554/eLife . 08153 . 013 Taken together , these findings suggest that the desensitization to proteasome inhibition caused by decreased 19S levels is also clinically relevant in MM patients , and that 19S levels are a predictive biomarker of response to proteasome inhibitor-based therapy . The exquisite sensitivity of MM cells to proteasome inhibitors provides a paradigm for non-oncogene addiction in cancer . Using our next-generation RNAi platform to identify genetic determinants of sensitivity to proteasome inhibitors , we found that incapacitating the ‘executioner’ ( the 20S core ) promotes sensitization to the pharmacological insult , while disarming the ‘decision maker’ ( the 19S regulator ) paradoxically results in resistance . In fact , knockdown of almost any 19S subunit conferred carfilzomib resistance , and together they were the group of genes with the strongest protective effect . These opposing phenotypes are consistent with our previous observations in budding yeast , where components of the 20S and 19S proteasome subcomplexes formed separate functional clusters in a genetic interaction map ( Breslow et al . , 2008 ) . The phenotype arising from genetic depletion of the 20S core subunits is readily understandable and in line with previous observations: diminishing the amount of the direct target of an inhibitory drug sensitizes cells to the drug ( Giaever et al . , 1999; Matheny et al . , 2013 ) . Two non-mutually exclusive scenarios account for this observation: silencing the 20S core subunits can lead to ( i ) a crippled proteasome that is easier to inhibit , or to ( ii ) substoichiometric amounts of the silenced subunits which in turn compromises the assembly of functional proteasomes . The protective effect of 19S regulator knockdown was unexpected since 19S and 20S work in concert to degrade ubiquitylated protein substrates . We validated our finding in a range of MM and non-MM cells with different genetic backgrounds , as well as in MM patients , suggesting this is a general , unifying mechanism underlying the response and adaptation to proteasome inhibition . Our data supports that 19S depletion alters the spectrum of proteasome substrates , leading to selective accumulation of factors involved in protein degradation . This mechanism may represent a homeostatic feedback loop that is relevant in normal cellular contexts . In MM cells , upregulation of protein turnover pathways may reduce cellular dependence on the proteasome . Previously reported mechanisms of resistance to proteasome inhibitors in yeast and mammalian cells include mutations or overexpression of the direct drug target , PSMB5 , in the catalytic core of the proteasome ( Oerlemans et al . , 2008; Huber et al . , 2015 ) , but to our knowledge , they have never been identified in patients . Our results support an entirely different mechanism of drug resistance , brought about by rewiring the proteostasis network , which reduces the dependence on the proteasome . A detailed understanding of the role of the proteostasis network in disease remains a major challenge . This is due to the size of the network its dynamic nature , genetic redundancies , and context dependence ( Balch et al . , 2008 ) . Our functional genomics approach proved to be especially well-suited to reveal functionally relevant nodes contributing to the rewiring of the proteostasis network in the context of disease . The notion that alternative protein degradation pathways can desensitize cells to proteasome inhibitors provides a strong motivation for the simultaneous targeting of parallel pathways in combination therapy . Indeed , other groups have investigated the role of autophagy during proteasome inhibition ( Kawaguchi et al . , 2011; Santo et al . , 2012; Komatsu et al . , 2013; Moriya et al . , 2013; Mishima et al . , 2015 ) , and combinations of bortezomib with autophagy inhibitors , including hydroxychloroquine ( Vogl et al . , 2014 ) and the HDAC6 inhibitor ACY-1215 ( ClinicalTrail . gov identifier NCT01323751 ) , are currently in clinical trials for MM . While increasing protein degradation desensitizes cells to proteasome inhibition , our results also suggest and are consistent with an alternative network-level mechanism of resistance: reducing protein synthesis ( Chen et al . , 2010; Cenci et al . , 2012 ) . Both mTOR and the EIF4F complex are master regulators of protein synthesis , and their depletion leads to increased resistance towards proteasome inhibition . By diminishing protein synthesis , the cell could remodel its response to proteasome inhibitors , in this case not through the compensatory upregulation of alternative protein degradation pathways , but by offsetting the reduced proteasome capacity through lowered protein load . Our functional genomics approach also pointed to other nodes in the proteostasis network that emerge as synergistic vulnerabilities with proteasome inhibition . Specifically , knockdown of the master regulator of cytosolic proteostasis , HSF1 , and of individual cytosolic Hsp70 and Hsp90 chaperones sensitized cells to carfilzomib ( Figure 2A ) . These nodes are therefore additional candidate targets for combination therapy with proteasome inhibitors . Hsp90 inhibitors ( Ishii et al . , 2012; Suzuki et al . , 2015 ) and Hsp70 inhibitors ( Braunstein et al . , 2011 ) have previously been reported to act synergistically with bortezomib against MM cells . Notably , we did not find a role for ER proteostasis factors in controlling sensitivity to carfilzomib in U-266 cells . The high secretory activity of MM cells had previously been hypothesized to challenge the folding capacity of the ER , leading to an increased burden of misfolded protein that would explain the increased dependence on the proteasome ( Meister et al . , 2007; Bianchi et al . , 2009 ) . A possible explanation of our findings is that during proteasome inhibition in MM cells , misfolded proteins are still extracted from the ER , as has been reported for 3-Hydroxy-3-methylglutaryl-coenzyme A reductase in the cholesterol biosynthesis pathway ( Morris et al . , 2014 ) , and that their removal from the ER alleviates the proteotoxic load . Our study illustrates the power of our functional genomics approach to uncover biomarkers predictive of drug responses in patients and to guide the rational design of combination therapies that show promise to overcome the urgent clinical problem of drug resistance in cancer . From a clinical perspective , an assay that can predict response to a given therapy would significantly help improving outcomes and reduce toxicities for individual patients with MM . Currently , however , therapeutic decisions are largely based on a clinical trial-and-error basis . Our results support the development of clinical assays predictive of proteasome inhibitor sensitivity , with the potential to become an essential test in every-day treatment decisions for physicians treating patients with MM . The combination of observational genomics in patients and functional genomics in model systems can pave the way for precision medicine , while providing fundamental insights into biology . RPMI-8226 , U-266 and JJN-3 cells were obtained from the German Collection of Microorganisms and Cell Culture ( DSMZ , Braunschweig , Germany ) . K562 cells were a kind gift of Neil Shah ( UCSF ) . RPMI-8226 , U-266 and K562 cells were grown in RPMI medium supplemented with 10% fetal bovine serum ( FBS ) , 4 mM L-glutamine ( L-glut ) and antibiotics ( pen/strep ) . JJN-3 cells were grown in a 50:50 mixture of Iscove's Modified Dulbecco's media ( IMDM ) and high-glucose Dulbecco's Modified Eagle's Medium ( DMEM ) supplemented with 20% FBS and 4 mM L-glut and pen/strep . For proteasome inhibition studies , cells were treated ( 1 ) overnight ( 24 hr ) with increasing doses of either carfilzomib ( Onyx , South San Francisco , CA ) or bortezomib ( Selleck Chemicals , Houston , TX ) to establish dose–response curves , or ( 2 ) for 1 hr with a concentration of the proteasome inhibitor equal to its EC50 , measured 24 hr after the pulse exposure . Following the pulse exposure cells were washed twice with media and replated at a density of 0 . 5 million per milliliter . Exposure times and drug concentrations are indicated in the figures . To induce autophagy , cells were treated with 200 nM rapamycin ( EMD Millipore , Billerica , MA ) or 100 mM D- ( + ) -trehalose dihydrate ( Sigm-Aldrich , St . Louis , MO ) overnight ( 24 hr ) . After exposure to the drug , the cells were pelleted and resuspended in media without rapamycin or trehalose and supplemented with increasing concentrations of carfilzomib and incubated for an additional 24 hr to establish dose–response curves . The retroviral expression construct pBABE-puro mCherry-EGFP-LC3B was a gift from Jayanta Debnath ( Addgene , Cambridge , MA] , plasmid # 22 , 418 ) ( N'Diaye et al . , 2009 ) . Retroviral transductions were carried out by transfection of 16 μg of retroviral vector and 2 μg of a plasmid encoding the envelope protein VSV-G ( Clontech Laboratories , Inc . , Mountain View , CA ) into the packaging cell line GP2-293 ( Clontech ) using lipofectamine 2000 ( Life Technologies , Grand Island , NY ) . 24 hr after transfection , the cells were switched to virus collection medium ( DMEM supplemented with 4% FBS , 15 mM HEPES and 2 mM L-glut ) . 24 hr later ( 48 hr after transfection ) , high-titer retroviral supernatant was collected , filtered through a 0 . 45 μm PVDF membrane and supplemented with 8 μg/ml polybrene . The supernatant was used immediately for infection of 2 . 5 million target cells at a density of 1 million cells per milliliter . Transduction was accomplished by spinoculating the cells at 2000 rpm . After Spinoculation , the cells were recovered , spun down and resuspended at ∼500 , 000 cells per milliliter in the appropriate media . Lentiviral transductions were carried out as previously described ( Kampmann et al . , 2014 ) . Sublibraries of our next-generation shRNA library ( Kampmann et al . , 2015 ) targeting 7712 genes involved in proteostasis , cancer , apoptosis , kinases , phosphatases and drug targets were introduced into U-266 MM cells at an MOI of ∼0 . 3 ( ensuring that most cells express at most one shRNA ) . Cells transduced with the shRNA libraries were selected using puromycin . This population was split into two subpopulations . One was grown in the absence of drug , whereas the other was exposed to 1 hr pulses of carfilzomib at a concentration around the LD50 measured at 24 hr ( 150–200 nM ) , followed by recovery . After 4 rounds of treatment and recovery , treated and untreated cells were harvested , genomic DNA was isolated and the shRNA-encoding cassette was PCR-amplified and subjected to next-generation sequencing as previously described ( Kampmann et al . , 2014 ) . Growth and carfilzomib resistance phenotypes were quantified and p values for hit genes were calculated using gimap software ( Kampmann et al . , 2013 , 2014 ) ( http://gimap . ucsf . edu ) . Gene-based phenotypes were calculated by averaging the phenotypes of the 3 most extreme shRNAs targeting this gene , excluding shRNAs for which there were less than 50 sequencing reads in both the treated and the untreated population . This ad hoc metric is defined arbitrarily as a compromise between averaging fewer shRNAs ( which may be too sensitive to outliers ) and averaging too many shRNAs ( which would include inactive shRNAs and thereby underestimate the effect size ) . An average of 3 shRNAs targeting selected hit genes were selected based on their activity in the primary screen and individually cloned to build a focused custom shRNA library . This library was introduced into U-266 , JJN-3 , RPMI-8226 and K562 cells . Cells were grown untreated , or treated with pulses of bortezomib , following a similar selection strategy as for the primary screen . Growth and bortezomib resistance phenotypes were quantified and phenotypes were averaged for shRNAs targeting the same gene using gimap software ( Kampmann et al . , 2013 , 2014 ) ( http://gimap . ucsf . edu ) . For individual validation experiments , a negative control shRNA ( 5′-CGTTCTTAGGGTGAGTAAGAAATAGTGAAGCCACAGATGTATTTCTTACTCACCCTAAGAACT-3′ ) , an shRNA targeting PSDM12 ( 5′- CAGCCTTTCTCTCAAATCTAGTTAGTGAAGCCACAGATGTAACTAGATTTGAGAGAAAGGCTT-3′ ) or an shRNA targeting PSMB5 ( 5′-CGACGGTGAAGAAGGTGATAGATAGTGAAGCCACAGATGTATCTATCACCTTCTTCACCGTCT-3′ ) were expressed in the lentiviral vector pMK1200 ( Kampmann et al . , 2013 ) or pMK1224 ( details provided on request ) . For GO term enrichment analysis of hit genes from the primary screen , top hits were defined as follows: The 50 genes with the most sensitizing and most desensitizing gene phenotypes and a minimal p value of 10−4 were defined as top sensitizing and desensitizing hits . GO term enrichment p values were calculated using Database for Annotation , Visualization and Integrated Discovery ( DAVID ) ( Huang et al . , 2009a , 2009b ) , using the set of 7712 genes targeted by the sublibraries used in the primary screen as the background . Where different GO-terms encompassing the same subset of hit genes were found , only one is displayed . Similarly , where a GO term described a subset of hit genes of those described by another GO term that had a more significant p value , only the more significant GO term describing a larger set of hit genes is displayed . For GO term enrichment analysis of proteins enriched based on proteome Stable Isotope Labeling by Amino Acids in Cell Culture ( SILAC ) experiments , the top 50 enriched proteins for each pair-wise comparison of samples was analyzed . Cell viability assays were carried out using the CellTiter GLO kit ( Promega Corp . , Madison , WI ) following manufacturer's recommendations . Raw luminescence signals were collected using a SpectraMax M5 plate reader ( Molecular Devices , Sunnyvale , CA ) and an integration time of 500 milliseconds . The raw counts were normalized as the percent of signal relative to untreated cells , or the percent maximum signal when comparing treatments with more than one drug . Sigmoidal dose–response curve fitting for EC50 calculation was performed using the Prism V5 package ( GraphPad Software Inc . , San Diego , CA ) . Total cell lysates were collected in SDS sample buffer ( 62 . 5 mM Tris pH 6 . 8 , 10% glycerol , 2% SDS , 0 . 004% bromophenol blue ) . Lysates were sonicated for ∼15 s to shear the genomic DNA . 2-mercaptoethanol ( 2 ME ) was added to a final concentration of 5% to the lysates just prior to boiling and loading on SDS-PAGE gels . Gel loading normalization was accomplished by one of two methods: ( 1 ) same number of live cells per condition , or ( 2 ) densitometry after SDS-PAGE electrophoresis , transfer onto nitrocellulose membranes and staining with Ponceau stain . Normalized total cell lysate amounts were loaded onto precast Tris-glycine SDS-PAGE gels ( BioRad ) , electrophoresed , and transferred onto 2 . 0 μm pore nitrocellulose membranes . Equal loading per lane was verified by staining the membrane with Ponceau stain . Destained membranes were blocked for 1–2 hr in Tris-borate saline supplemented with 0 . 1% Tween 20 ( TBST ) and 5% non-fat milk . Blocked membranes were incubated overnight with primary antibodies diluted in TBST supplemented with 5% bovine serum albumin . Antibodies and dilutions were as follows: anti-SQSTM1/p62 ( D5E2 ) rabbit monoclonal antibody ( Cell Signaling Technology [Danvers , MA] #8025 , 1:1000 ) , anti-GAPDH rabbit polyclonal antibody ( Abcam [Cambridge , MA] ab9485 , 1:2000 ) , anti-PSMD12 rabbit polyclonal antibody ( Bethyl Laboratories Inc . [Montgomery , TX] #A303-830A , 1:2000 ) , anti-PSMB5 rabbit polyclonal antibody ( Bethyl Laboratories , Inc . , Montgomery , TX , #A303-847A , 1:5000 ) , anti-VCP/p97 rabbit polyclonal antibody ( Cell Signaling Technology #2648 , 1:1000 ) , anti-MCL1 ( D35A5 ) rabbit monoclonal antibody ( Cell Signaling Technology #5453 , 1:1000 ) , anti-UFD1L rabbit polyclonal antibody ( Bethyl Laboratories Inc . #A301-875A 1:500 ) , NFkB pathway antibodies ( NFkB sample kit , Cell Signaling Technology #9936 , all at 1:1000 ) . Membranes were incubated with HRP-conjugated secondary antibodies diluted in TBST supplemented with 5% non-fat milk at a 1:5000 dilution ( Amersham , GE Healthcare Life Sciences [Pittsburgh , PA] NA931 , NA934 ) for 1 hr at room temperature . Blots were developed using luminol-based enhanced chemiluminescence substrates ( SuperSignal West Dura Extended Duration Substrate , or SuperSignal West Femto Maximum Sensitivity Substrate , Life Technologies ) and exposed to radiographic film or imaged directly in a digital gel imager ( Chemidoc XRS+ , BioRad ) . Digital images were automatically adjusted for contrast using the photo editor Adobe Photoshop ( Adobe Systems , San Jose , CA ) . Densitometric quatification of immunoblots was performed using the software package ImageJ ( National Institutes of Health , Bethesda , MD ) ( Schneider et al . , 2012 ) . The background subtracted area under the curve for each band was quantified and normalized to either total protein or loading controls . U-266 cells were transduced with a negative control shRNA or an shRNA targeting PSMD12 ( sequences as described above ) . Cell populations were then split and treated with different concentrations of carfilzomib for 1 hr or left untreated . Proteasome activity was assayed in lysates of these cell populations using the ProCISE assay ( Parlati et al . , 2009 ) or the LLVY-AMC for chymotryptic activity ( Boston Biochem , Cambridge , MA , #S-280 ) following the manufacturer's instructions , except that a buffer consisting of 20 mM Tris , pH 8 . 0 , 0 . 5 mM EDTA without SDS was used . K562 cells expressing negative control shRNA or an inducible shRNA targeting PSMD6 ( as used for proteasome activity assays ) were grown for 7 days in medium containing either standard lysine and arginine ( light isotopes ) , lysine ( 4 , 4 , 5 , 5-D4 ) and arginine ( U-13C6 ) ( medium isotopes ) or lysine ( U-13C6 , U-15N2 ) and arginine ( U-13C6 , U-15N4 ) ( heavy isotopes ) obtained from Cambridge Isotope laboratories ( Cambridge , MA ) . shRNA expression was then induced for 48 hr , after which some cell populations were treated with bortezomib as described for the proteasome activity assays . Cells were combined in two triple-SILAC experiments as follows: Combination A: SILAC light: untreated negative-control; SILAC medium: bortezomib-treated ; SILAC heavy: PSMD6 knockdown and bortezomib-treated . Combination B: SILAC light: untreated negative-control; SILAC medium: PSMD6-knockdown no inhibitor; SILAC heavy: PSMD6-knockdown and bortezomib-treated . SILAC labeled cells were lysed with modified RIPA buffer ( 50 mM Tris–HCl pH 7 . 5 , 150 mM NaCl , 1% Nonidet P-40 , 0 . 1% sodium-deoxycholate , 1 mM EDTA ) containing protease inhibitors ( Complete protease inhibitor mixture tablets , Roche Diagnostics , Indianapolis , IN ) and N-ethylmaleimide ( 5 mM ) . The lysates were incubated for 10 min on ice and subsequently cleared by centrifugation at 16 , 000 × g . An equal amount of protein was mixed from different SILAC states and proteins were precipitated by adding chilled acetone ( final concentration 80% ) and storing overnight at −20°C . Proteins were redissolved in denaturing buffer ( 6 M urea , 2 M thiourea , 10 mM HEPES pH 8 . 0 ) and subsequently reduced with dithiothreitol ( 1 mM ) and alkylated with chloroacetamide ( 5 . 5 mM ) . Proteins were proteolysed with Lysyl endoproteinase C ( Lys-C ) for 6 hr and after fourfold dilution in water with trypsin overnight . The digestion was stopped by addition of trifluoroacetic acid , incubated at 4°C for 2 hr and resulting precipitates removed by centrifugation for 15 min at 4000 × g . Cleared peptides were purified by reversed-phase Sep-Pak C18 cartridges ( Waters Corporation , Milford , MA ) . Di-Gly-lysine containing peptides were enriched using the Ubiquitin Remnant Motif Kit ( Cell Signaling Technology ) , according to the manufacturer's protocol . In brief , peptides were eluted from the Sep-Pak C18 cartridges and incubated with 40 μl of anti-di-Gly-lysine antibody resin in 1X immunoaffinity purification ( IAP ) buffer on a rotational wheel for 4 hr at 4°C ( Wagner et al . , 2011 ) . After centrifugation , the supernatant ( containing unbound peptides ) was removed and used for proteome analysis ( see below ) . The immunoenriched peptides were washed three times with 1X IAP buffer and two times with water . Peptides were eluted with 0 . 15% trifluoroacetic acid in water . Eluted peptides were fractionated into 6 fractions by micro-column-based strong-cation exchange chromatography ( SCX ) and desalted by reversed phase C18 Stage-tips . Similarly , for analysis of protein levels , unbound peptides from the anti-di-Gly-lysine enrichment ( see above ) were fractioned by micro-SCX and purified by reversed phase C18 Stage-tips . Peptide fractions were analyzed on a quadrupole Orbitrap ( Q Exactive , Thermo Scientific , Waltham , MA ) mass spectrometer equipped with a nanoflow HPLC system ( Thermo Scientific ) . Peptides were loaded onto C18 reversed-phase columns and eluted with a linear gradient from 8% to 40% acetonitrile containing 0 . 5% acetic acid . The mass spectrometer was operated in a data-dependent mode , automatically switching between MS and MS/MS . Survey full scan MS spectra ( m/z 300–1200 ) were acquired in the Orbitrap mass analyzer . The 10 most intense ions were sequentially isolated and fragmented by higher-energy C-trap dissociation ( HCD ) . Peptides with an unassigned charge state , as well as peptides with a charge state less than +2 for proteome samples , and +3 for di-glycine-lysine enriched samples , were excluded from fragmentation . Fragment spectra were acquired in the Orbitrap mass analyzer . Raw MS data files were analyzed by the MaxQuant software version 1 . 4 . 1 . 1 ( Cox and Mann , 2008 ) . Parent ion and MS/MS spectra were searched against protein sequences from the UniProt knowledge database using the Andromeda search engine . Spectra were searched with a mass tolerance of 6 ppm in the MS mode , 20 ppm for the MS/MS mode , strict trypsin specificity allowing up to two missed cleavage sites . Cysteine carbamidomethylation was searched as a fixed modification . Amino-terminal protein acetylation , methionine oxidation and N-ethylmaleimide modification of cysteines , and di-Gly-lysine were searched as variable modifications , and di-Gly-lysines were required to be located internally in the peptide sequence . Site localization probabilities were determined by MaxQuant using a post-translational modification scoring algorithm as described previously ( Cox and Mann , 2008 ) . Using the target-decoy search strategy ( Elias and Gygi , 2007 ) and a posterior error probability filter , a false discovery rate of less than one percent was achieved . 23 patients on a clinical research protocol using carfilzomib , lenalidomide , and dexamethasone ( CRd ) treatment in newly diagnosed MM patients were evaluated at the National Institute of Health ( NIH , Bethesda , MD ) . Patients were tested for the expression of 19S and 20S proteosome subunits and aggresome levels in bone marrow plasma cells before start of carfilzomib therapy . Bone marrow aspirates were collected and immunostained using antibodies against CD45 , CD38 , CD138 ( Becton Dickinson , San Jose , CA ) , Proteasome 19S S7 ( 19S ) and Proteasome Subunit Beta Type 4 ( beta4 ) ( Abcam , Cambridge , UK ) . The antibody used for immunodetection of the S7 subunit of the 19S proteasome was previously validated ( Rousseau et al . , 2009 ) . In parallel , cells were labeled with ProteoStat Aggresome Detection Reagent ( Enzo Life Sciences , Farmingdale , NY ) . Multicolor acquisition and analysis was performed using BD FACS CANTO II and DIVA software . Data was expressed as mean fluorescence intensity ( MFI ) ratio using isotype-matched controls . Statistical analysis was performed using Excel ( Microsoft Corporation , Redmond , WA ) and DataPrism ( Seattle , WA ) software . U-266 cells carrying a constitutive non-targeting shRNA or an shRNA targeting the 19S subunit PSMD12 were collected by centrifugation , washed twice with ice-cold PBS and lysed in TRIzol ( Life Techologies , Grand Island , NY ) . RNA was extracted following manufacturer's recommendations . 500 ng of total RNA were reverse transcribed using the SuperScript VILO cDNA synthesis kit ( Life Technologies ) following manufacturer's recommendations . The resulting cDNA reactions were diluted 10-fold with 10 mM Tris pH 8 . 0 and 1% of this dilution was used for each quantitative real-time PCR ( qPCR ) reaction . qPCR reactions were set-up using IQ SYBR Geen Super Mix ( BioRad , Hercules , CA ) in 20 μl reactions . The reactions were ran on a BioRad CFX96 Real Time system ( BioRad ) and analyzed using the CFX Manager Software V3 . 0 ( BioRad ) . All reactions were normalized to an internal loading control ( GAPDH ) and the fold-changes reflecting the extent of knockdown were then normalized to the no drug , negative control shRNA condition . Average values for three different oligonucleotide pairs targeting the PSMD12 trasncript were taken for the calculations . The following oligonucleotides taregting human transcripts were used: Hs_GAPDH_Fwd: 5′-AGCCACATCGCTCAGACAC-3′ , Hs_GAPDH_Rev: 5′-TGGAAGATGGTGATGGGATT-3′ , Hs_PSMD12_exon9-3′UTR_Fwd: 5′- AATGAAAAGGATGGCACAGC-3′ , Hs_PSMD12_exon9-3′UTR_Rev: 5′- TTTGGATCCTTGGGTCTCTG-3′ , Hs_PSMD12_RefSeq_Var1_Fwd: 5′- CGTCAAGATGGAGGTGGACT-3′ , Hs_PSMD12_RefSeq_Var1_Rev: 5′- TCCAGAGAGAGAAGGGTTTCA-3′ , Hs_PSMD12_exon1-3_Fwd: 5′- CGTCAAGATGGAGGTGGACT-3′ , Hs_PSMD12_exon1-3_Rev: 5′- AGATACGGGATGTCGATACCA-3′

Cells have several mechanisms for removing proteins that have been damaged or are no longer needed . One of these mechanisms is carried out by a large protein complex called the proteasome . Drugs that block the proteasome are toxic to all cells , and a type of blood cancer called multiple myeloma is particularly sensitive to these ‘proteasome inhibitors’ . However , tumors in patients with multiple myeloma can also become resistant to these drugs . Using a genetic approach , Acosta-Alvear et al . identified the factors that control the sensitivity of cells to proteasome inhibitors . In particular , reducing the levels of other factors that contribute to protein balance made the cells more sensitive . Using a combination of proteasome inhibitors and drugs that target these other factors could prove to be useful in the fight against multiple myeloma . The proteasome complex contains two types of subunits: regulatory subunits that recognize the proteins that need to be degraded , and catalytic subunits that degrade the proteins . The results of Acosta-Alvear et al . revealed how varying the levels of these two subunits influenced the sensitivity of cells to inhibitors . While decreasing the levels of catalytic subunits made the cells more sensitive , as expected , decreasing the level of regulatory subunits surprisingly made the cells resistant to the inhibitors . A possible explanation for this paradoxical result is that certain proteins are less effectively degraded by the proteasome in these cells , and that the buildup of these proteins protects the cells against the drugs . Acosta-Alvear et al . also found that lower levels of regulatory subunits desensitized multiple myeloma patients to therapy based on proteasome inhibition , suggesting that results from the genetic screen carried out in cells can predict clinical resistance mechanisms and guide the development of future therapies to increase patient survival .

2015

Paradoxical resistance of multiple myeloma to proteasome inhibitors by decreased levels of 19S proteasomal subunits

The diverse transcriptional mechanisms governing cellular differentiation and development of mammalian tissue remains poorly understood . Here we report that TAF7L , a paralogue of TFIID subunit TAF7 , is enriched in adipocytes and white fat tissue ( WAT ) in mouse . Depletion of TAF7L reduced adipocyte-specific gene expression , compromised adipocyte differentiation , and WAT development as well . Ectopic expression of TAF7L in myoblasts reprograms these muscle precursors into adipocytes upon induction . Genome-wide mRNA-seq expression profiling and ChIP-seq binding studies confirmed that TAF7L is required for activating adipocyte-specific genes via a dual mechanism wherein it interacts with PPARγ at enhancers and TBP/Pol II at core promoters . In vitro binding studies confirmed that TAF7L forms complexes with both TBP and PPARγ . These findings suggest that TAF7L plays an integral role in adipocyte gene expression by targeting enhancers as a cofactor for PPARγ and promoters as a component of the core transcriptional machinery . Adipose tissue plays a central role in energy homeostasis by acting as a major lipid storage site as well as an important endocrine tissue . Excess food intake vs energy expenditure invariably leads to obesity , an increasingly prevalent condition in modern societies . Obesity is , in turn , tightly associated with an elevated risk of type-2 diabetes , hypertension , cardiovascular disease , and certain cancers , posing serious public health issues and rapidly escalating costs of health care ( Rosen and MacDougald , 2006; Rosen and Spiegelman , 2006 ) . Accumulation of adipose tissue mass results from increase in both adipocyte size and number . The size of adipocytes depends on the amount of stored lipids , while an increase in adipocyte number ( adipogenesis ) generally results from the expansion of adult precursor cells and subsequent differentiation , an underlying cause of obesity ( Rosen and MacDougald , 2006 ) . Consequently , there is great interest in dissecting the molecular mechanisms regulating adipogenesis and adipose biology . During the past 20 years , numerous studies have focused on the formation of adipocytes using well-established in vitro cell models ( Hollenberg et al . , 1997; Tang et al . , 2003; Zhang et al . , 2004a; Tang et al . , 2005; Tontonoz and Spiegelman , 2008 ) . The utilization of these model cells identified key adipogenic transcriptional activators such as C/EBPα and PPARγ ( Hu et al . , 1995; Brun et al . , 1996; Tang et al . , 2003; Zhang et al . , 2004b ) . Other transcription factors and cofactors , such as KLF15 , KLF9 , MED1 , MED14 , MED15 , MED23 and TAF8 , have also been reported to be involved in either adipocyte commitment or differentiation in model cell lines ( Guermah et al . , 2003; Mori et al . , 2005; Matsumoto et al . , 2007; Wang et al . , 2009; Grøntved et al . , 2010; Pei et al . , 2011 ) . More recently , genome-wide studies using a variety of techniques ( microarray , FAIRE-seq , Quanttrx and mRNA-seq ) have been utilized to screen for additional putative pro-adipogenic factors based on changes in either mRNA levels or chromatin states during adipogenesis ( Gupta et al . , 2011; Waki et al . , 2011 ) . Specifically , adipocyte-specific expression signatures and genome-wide binding maps of pro-adipogenic activators such as PPARγ , C/EBPα , and RXRα have been primarily determined from 3T3-L1 derived adipocytes . Other factors , including ZFP423 , NF1 family proteins , and IRFs have been also implicated in adipogenesis ( Eguchi et al . , 2008; Mikkelsen et al . , 2010; Schmidt et al . , 2011; Waki et al . , 2011; Boergesen et al . , 2012 ) . However , it has become evident that the full complement of key transcriptional regulators that orchestrate adipogenesis remains to be illucidated . The diversity of factors and the mechanisms driving adipocyte formation and cellular differentiation in general remain a challenge . For example , several tissue-specific components of the core transcriptional machinery were recently found to play essential roles in directing cell type-specific programs of transcription and lineage-specific differentiation . The examples of these include TAF4b which was found to be a key component in the development of the mouse ovary and spermatogenesis ( Freiman et al . , 2001; Voronina et al . , 2007 ) , and TAF3 which was found to be required for mouse germ-layer differentiation ( Liu et al . , 2011 ) . Additionally , a TRF3/TAF3 complex was found to be important for mouse myogenesis and zebrafsh hematopoiesis ( Deato et al . , 2008; Hart et al . , 2009 ) . As increasing numbers of cell type- and tissue-specific components of the core machinery become better characterized , a somewhat different notion of how traditional sequence–specific enhancer binding factors cooperate with selective tissue-specific core factors to drive lineage-specific transcription programs has emerged ( D'Alessio et al . , 2009 ) . With respect to adipogenesis , the only hint of such a mechanism came from previous studies of TAF8 , which was reported as a component of TFIID implicated in adipocyte formation using 3T3-L1 cells ( Guermah et al . , 2003 ) . Whether TAF8 operates exclusively as a subunit of TFIID or participates in other molecular transactions , in addition to its in vivo function , remain unknown . Here we found TAF7L , a paralogue of TATA binding protein associated factor 7 , is highly enriched in differentiated C3H10T1/2 adipocytes and bona fide mouse WAT . We have utilized shRNA knockdown and gene knockout strategies to determine the role of TAF7L in adipocyte formation both in vitro and in vivo . In addition , we have explored the consequences of ectopically expressing TAF7L in C2C12 myoblasts to probe its reprogramming capabilities . Further , we carried out genome-wide mRNA-seq and ChIP-seq analysis to survey its functions in adipocyte differentiation . By using a combination of cellular , biochemical , genetic , and genomic approaches , our findings suggest that TAF7L plays an integral role in adipocyte gene expression by targeting enhancers as a cofactor for PPARγ and promoters as a component of the core transcriptional machinery , therefore providing new molecular insights into fat development that may prove useful for developing therapeutic strategies to treat obesity and its associated diseases . To explore the regulatory mechanisms directing adipocyte formation and function , we asked whether there were significant changes to the core promoter recognition components during adipogenesis similar to what had been observed during myogenesis ( Deato and Tjian , 2007 ) . In particular , we set out to determine whether and which TAF subunits of the prototypic core promoter recognition complex TFIID increase or decrease in expression during adipocyte differentiation . Consistent with the previous observations from 3T3-L1 adipogenesis studies ( Guermah et al . , 2003 ) , our analysis of both protein and mRNA levels revealed that TBP and most of the canonical subunits of TFIID are down-regulated during C3H10T1/2 differentiation ( Figure 1A , B ) . Surprisingly however , one subunit ( TAF7L ) previously reported to be a component of TFIID primarily in testis ( Pointud et al . , 2003; Cheng et al . , 2007 ) was found to be significantly up-regulated in differentiated C3H10T1/2 adipocytes ( Figure 1A , B and Figure 1—figure supplement 2A ) and 3T3-L1 adipocytes ( Figure 1—figure supplement 1A , C ) . Importantly , this enrichment appears specific for the adipogenesis process since the mRNA abundance of Taf7l is downregulated to levels comparable to those of other TAF subunits during myogenesis ( Figure 1C ) . To exclude the possibility that Taf7l enrichment reflects a cell culture artifact of C3H10T1/2 adipogenesis , we compared Taf7l mRNA and protein levels in bona fide mouse tissue . In concordance with previous studies , Taf7l is most highly expressed in testis ( Pointud et al . , 2003 ) ( Figure 1D , E ) . Importantly , Taf7l also shows significant expression in WAT and detectable expression in liver , spleen , brown adipose tissue ( BAT ) and kidney , but not in muscle or brain tissue ( Figure 1D , E ) . By contrast , the expression of canonical TFIID subunits such as TAF4 is low in both WAT and muscle as expected ( Figure 1E ) . Taken together , these data indicate that TAF7L is indeed enriched in differentiated C3H10T1/2 and 3T3-L1 adipocytes and bona fide WAT . 10 . 7554/eLife . 00170 . 003Figure 1 . TAF7L is enriched in terminally differentiated adipocytes and bona fide WAT . ( A ) and ( B ) Expression of TAF7L and TFIID subunits prior to and 5 days ( 5D ) post adipogenic induction of C3H10T1/2 cells as shown by RT-qPCR analysis ( A ) and by Western blot ( B ) . ( C ) mRNA levels of TFIID subunits in C2C12 cells and myotubes . ( D ) Taf7l mRNA levels in different mouse tissues detected by RT-qPCR relative to muscle , whose expression level was assigned to 1 as the tissue displaying the lowest Taf7l mRNA levels . ( E ) Western blot analysis of mouse tissues with TAF4 and TAF7L antibodies . mRNA levels in ( A ) and ( C ) was assigned to 1 in C3H10T1/2 and C2C12 cells , mRNA levels in adipocytes and myotubes were compared with C3H10T1/2 and C2C12 cells respectively . *p<0 . 05 , data is mean and s . e . m is from triplicates . RT-qPCR was normalized to the amount of total mRNA and Western blotting analysis was normalized to the amount of total protein . D , days; 10T1/2 , C3H10T1/2 cells; ES , embryonic stem cell; BAT , brown adipose tissue; WAT , white adipose tissue . DOI: http://dx . doi . org/10 . 7554/eLife . 00170 . 00310 . 7554/eLife . 00170 . 004Figure 1—figure supplement 1 . TAF7L is enriched in 3T3-L1 differentiated adipocytes . ( A ) Expression of Taf7l and TFIID subunits prior to and 7 days ( 7D ) post adipogenic induction of 3T3-L1 cells as shown by RT-qPCR analysis ( A ) and by Western blot ( C ) . ( B ) Gene expression of adipocyte marker genes Adipsin , Adipoq and Fabp4 of 3T3-L1 adipocytes prior to and 7 days post adipogenic induction . mRNA levels in 3T3-L1 cells were assigned to 1 , mRNA levels of each gene in 3T3-L1 adipocytes were compared to 3T3-L1 cells , data is mean from triplicates . DOI: http://dx . doi . org/10 . 7554/eLife . 00170 . 00410 . 7554/eLife . 00170 . 005Figure 1—figure supplement 2 . Gene expression analysis of C3H10T1/2 cells during adipogenesis . ( A ) – ( F ) Time course analysis by RT-qPCR analysis of Taf7l and Taf7 ( A ) , C/ebpα ( B ) , Dlk1 and Cyclophilin ( C ) , Fabp4 ( D ) , Pparγ ( E ) and Adipoq ( F ) in C3H10T1/2 cells at 0D , 1D , 2D , 3D , 4D and 5D post adipogenic induction . D , days , mRNA levels in C3H10T1/2 cells at 0D were assigned to 1 , mRNA levels of each gene at 0D , 1D , 2D , 3D , 4D , and 5D in C3H10T1/2 cells during adipogenesis were compared to 0D respectively , and data is mean from triplicates . DOI: http://dx . doi . org/10 . 7554/eLife . 00170 . 005 These findings were surprising for several reasons . First , Taf7l had only been well documented to be critical for directing spermatogenesis in mice , and Taf7l-deficient mice show an impaired male fertility phenotype but no other defects were previously reported ( Akinloye et al . , 2007; Cheng et al . , 2007 ) . Second , although earlier studies of terminal differentiation implicated specific ‘atypical TAFs’ in , for example , skeletal muscle , ovary and testis formation ( Freiman et al . , 2001; Deato et al . , 2008; D'Alessio et al . , 2009 ) , we did not anticipate Taf7l as a potential key player in adipogenesis . Instead , based on previous work , we expected that Taf8 would emerge as the ‘cell-type specific’ TAF involved in adipogenesis ( Guermah et al . , 2003 ) . However , we have found Taf7l to be up-regulated while Taf8 mRNA is down-regulated upon induction of C3H10T1/2 or 3T3-L1 cells to form adipocytes ( Figure 1A and Figure 1—figure supplement 1A ) . To explore this new finding , we set out to investigate the hitherto unrecognized potential role of Taf7l in adipogenesis . To assess whether Taf7l is required for adipogenesis , we first knocked down TAF7L expression in C3H10T1/2 cells and then induced adipogenesis . shTAF7L and control shGFP sequences were transfected into C3H10T1/2 cells to generate puromycin resistant stable TAF7L knockdown or shGFP control cell lines ( Figure 2 ) . As shown by Western blot , shTAF7L significantly reduced TAF7L protein levels both pre- and more dramatically post-adipogenesis while levels of canonical TFIID subunits remained largely unaltered in control and TAF7L knockdown pre-adipogenesis cultures ( Figure 2A , pre- ) . Consistent with our previous observation in terminally differentiated cells , the protein levels of the canonical TFIID subunits become largely decreased in control post-adipogenesis cultures while cells that have been depleted of TAF7L and therefore blocked from differentiation show high levels of TFIID subunits . As expected , PPARγ levels increased in control shGFP cells post induction but showed markedly reduced levels in shTAF7L-treated cells suggesting that TAF7L may directly or indirectly regulate this key adipogenic transcription factor . Our results also suggest that shTAF7L efficiently reduced endogenous TAF7L levels and blocked adipogenesis without significantly affecting TFIID complex integrity ( Figure 2A , post- ) . Next , C3H10T1/2 cells stably treated with shTAF7L and control shGFP were subjected to Oil red O staining to determine the efficiency of adipogenesis . Very few lipid-laden adipocytes formed in TAF7L depleted C3H10T1/2 cells , whereas over 98% of shGFP-treated cells differentiated into mature adipocytes . The few adipocytes that formed in shTAF7L-treated cells appeared smaller and exhibited abnormal morphology compared to untreated or shGFP-treated C3H10T1/2 adipocytes ( Figure 2B , E ) . These results suggest that TAF7L knockdown largely compromised the adipogenic potential of C3H10T1/2 cells , thus functionally implicating TAF7L in adipocyte differentiation . 10 . 7554/eLife . 00170 . 006Figure 2 . TAF7L is required for adipogenesis in vitro . ( A ) Western blot of TAF7L , TAF4 , TAF7 , TBP , GAPDH , and PPARγ protein levels in C3H10T1/2 cells expressing shRNA sequence against GFP as control ( shGFP ) or specifically against TAF7L ( shTAF7L ) pre- ( left panel ) and post-differentiation ( right panel ) . GAPDH protein levels serve as a loading control . ( B ) Oil red O staining in 5 days differentiated C3H10T1/2 cells stably expressing either shGFP or shTAF7L . ( C ) mRNA levels of adipocyte-specific genes by RT-qPCR on differentiated shGFP or shTAF7L C3H10T1/2 cells from ( B ) , mRNA levels in shGFP cells were assigned to 1 , mRNA levels of each gene in shTAF7L cells were compared to shGFP cells , *p<0 . 05 , data is mean and s . e . m is from triplicates . RT-qPCR was normalized to the amount of total mRNA . ( D ) Western blot with FLAG and TBP antibodies showing the expression of FLAG-TAF7LmA and FLAG-TAF7 in shTAF7L stably transfected C3H10T1/2 cells , TBP protein levels serve as a loading control . ( E ) Oil red O staining on shGFP or shTAF7L cells ectopically expressing FLAG , FLAG-TAF7 or FLAG-TAF7LmA 5 days post adipogenesis . ( F ) mRNA levels of adipocyte-specific genes by RT-qPCR in differentiated cells from ( E ) , mRNA levels in shTAF7L + vector cells were assigned to 1 , mRNA levels of each gene in shTAF7L + TAF7 , shTAF7L + TAF7LmA and shGFP cells were compared to shTAF7L + vector cells , data is mean from triplicates . DOI: http://dx . doi . org/10 . 7554/eLife . 00170 . 00610 . 7554/eLife . 00170 . 007Figure 2—figure supplement 1 . Gene expression analysis after TAF7L knockdown in C3H10T1/2 cells . ( A ) – ( F ) Time course of gene expression by RT-qPCR analysis of Taf7l ( A ) , Pparγ ( B ) , Adipoq ( C ) , Glut4 ( D ) , Fabp4 ( E ) , and Klf15 ( F ) in C3H10T1/2 cells stably treated with shGFP or shTAF7L sequences at 0D , 1D , 3D , and 5D post adipogenic induction . D , days; shGFP , control cells; shTAF7L , TAF7L knockdown cells . mRNA levels in shTAF7L-treated C3H10T1/2 cells at 0D were assigned to 1 , mRNA levels of each gene at 0D , 1D , 3D , and 5D in both shGFP and shTAF7L-treated C3H10T1/2 cells during adipogenesis were compared to mRNA levels in shTAF7L-treated C3H10T1/2 cells at 0D respectively , data is mean from triplicates . DOI: http://dx . doi . org/10 . 7554/eLife . 00170 . 007 To address the possibility that the observed adipogenic defects result from off-target effects of shTAF7L treatment , we carried out ‘rescue’ experiments by introducing either an empty vector , a shRNA-resistant vector TAF7LmA ( Ding et al . , 2008 ) , or its control paralogue TAF7 into shTAF7L cells . The TAF7LmA expression vector contains two silent mutations in Taf7l cDNA which renders resistance to RNA-mediated silencing by shTAF7L . As revealed by Western blot , TAF7LmA is efficiently expressed in shTAF7L cells , similar to the TAF7 construct ( Figure 2D ) . Next , we induced adipogenesis followed by Oil red O staining 5 days post induction . Our results indicate that introduction of TAF7LmA restored adipocyte formation ( Figure 2E ) and elevated expression of adipocyte-specific genes compared to vector control in shTAF7L cells ( Figure 2F ) . In contrast , overexpression of TAF7 failed to restore the adipogenic defects caused by the loss of TAF7L ( Figure 2E ) . Collectively , these results support the notion that TAF7L is likely an important player in adipogenesis , at least in the C3H10T1/2 cell differentiation model . To identify the full range of genes regulated by TAF7L in adipocyte differentiation , we performed mRNA-seq to profile global gene expression patterns in C3H10T1/2 cells prior to ( 10T1/2-pre ) and after adipogenesis ( 10T1/2-post ) ( Figure 3A ) . Next , we verified our mRNA-seq results by single gene RT-qPCR assays for a handful of well-characterized adipocyte-specific genes such as Fabp4 , Glut4 , Adipsin , Lpl , as well as control genes such as Mef2c and Frzb ( data not shown ) ; these results confirmed high concordance between the RT-qPCR assays and the genome-wide mRNA-seq data , although RT-qPCR generally gave 3- to 4-fold higher sensitivity compared to mRNA-seq . We also surveyed a set of 2360 genes upregulated by 10-fold or more in C3H10T1/2 cells post- vs pre-differentiation ( Figure 3 ) ; this analysis identified nearly all the well characterized adipocyte-specific genes including a large proportion of genes involved in adipocyte development and function ( Figure 3D ) . 10 . 7554/eLife . 00170 . 008Figure 3 . TAF7L is required for the expression of adipocyte-specific genes . ( A ) and ( B ) , mRNA-seq data on gene expression of C3H10T1/2 cells pre- ( horizontal axis ) and post-adipogenesis ( vertical axis ) ( A ) ; mRNA-seq data on gene expression in C3H10T1/2 cells pre-adipogenesis ( horizontal axis ) and C3H10T1/2 treated with shTAF7L post-adipogenesis ( vertical axis ) ( B ) . Orange dots in ( A ) mark genes upregulated during adipogenesis; blue dots in ( A ) mark genes unchanged or downregulated during adipogenesis . Circled genes were tested individually in RT-qPCR analysis . R indicates the correlation of the expression programs between two compared cells ( 10T1/2-post vs 10T1/2-pre in ( A ) , 10T1/2-shTAF7L-post vs 10T1/2-pre in ( B ) ) . ( C ) TAF7L knockdown blocks the upregulation of the adipocyte-specific genes which occurs during normal adipogenesis , pink circle represents 2360 genes upregulated in 10T1/2-post by 10-fold ( 10× ) from ( A ) ; orange circle represents 2226 genes unchanged in 10T1/2-shTAF7L-post ( B ) compared to ( A ) , 2083 genes in the overlapping intersect region account for 88% of total upregulated 10× genes in ( A ) . ( D ) List of gene ontology analysis hits showing a few typical adipocyte genes involved in fat cell differentiation and metabolic processes . DOI: http://dx . doi . org/10 . 7554/eLife . 00170 . 008 Next , we measured gene expression programs in differentiated C3H10T1/2 cells after depletion of TAF7L ( 10T1/2-shTAF7L-post ) . Importantly , genes normally up-regulated following adipogenesis ( Figure 3A , shown in orange ) showed similarly low expression levels in induced shTAF7L cells as in pre-adipocytes ( Figure 3B , orange ) . Strikingly , 2083 out of 2360 genes ( 88% ) that are highly upregulated during adipocyte differentiation failed to be induced upon adipogenic induction in the absence of TAF7L ( Figure 3B , C ) . RT-qPCR analysis of several representative marker genes confirmed that TAF7L knockdown dramatically reduced their mRNA levels compared to control shGFP in post-adipogenesis cells ( Figure 2C and Figure 2—figure supplement 1 ) . These data suggest that the induction of adipocyte-specific genes is markedly compromised in the absence of TAF7L . Moreover , the overall transcriptional profile of differentiated shTAF7L-treated C3H10T1/2 cells ( shTAF7L-post ) largely matched the expression levels of pre-differentiated C3H10T1/2 cells ( 10T1/2-pre ) ( R = 0 . 98 ) ( Figure 3B ) ; while mature adipocytes ( 10T1/2-post ) is distinct ( R = 0 . 84 ) ( Figure 3A ) . Thus , loss of TAF7L in C3H10T1/2 cells severely impaired its adipogenic potential rendering shTAF7L-treated cells in an undifferentiated state ( Figures 2B and 3A , B ) . These findings were also confirmed using a different shTAF7L construct and RT-qPCR analysis ( data not shown ) . Taken together , these results strongly implicate TAF7L in potentiating efficient adipogenesis of C3H10T1/2 cells by serving as an important regulator of adipocyte-specific gene expression . To assess Taf7l function in adipogenesis in vivo , we first isolated primary adipocyte fibroblasts from WAT of Taf7l knockout ( KO ) mice and littermate controls ( WT ) and tested their ability to undergo adipogenesis . As revealed by Oil red O staining , Taf7l KO fibroblasts produced very few , if any , lipid-filled adipocytes compared to WT cells in response to adipogenic induction ( Figure 4A ) . RT-qPCR analysis confirmed that ablation of Taf7l also suppresses the upregulation of adipocyte-specific genes during differentiation ( Figure 4B ) . These results indicate a requirement for Taf7l in adipocyte differentiation of primary adipocyte fibroblasts , consistent with our results from C3H10T1/2 cells . As expected , loss of Taf7l caused no obvious change in mRNA or protein levels of other TFIID subunits ( data not shown ) , suggesting that deletion of Taf7l is unlikely to affect TFIID integrity or function in vivo , in agreement with previous observations that Taf7l-deficient mice appear normal except for germ cell developmental defects ( Cheng et al . , 2007 ) . 10 . 7554/eLife . 00170 . 009Figure 4 . Taf7l is required for WAT development in vivo . ( A ) Oil red O staining to detect mature adipocytes from 5 day differentiated primary fibroblasts derived from adipose tissue of wild-type ( WT ) and Taf7l-deficient mice ( KO ) . ( B ) mRNA levels of adipocyte-specific genes by RT-qPCR on WT and Taf7l KO primary fibroblasts post differentiation from ( A ) , mRNA levels in WT cells were assigned to 1 , mRNA levels of each gene in Taf7l KO cells were compared to WT cells , *p<0 . 05 , data is mean and s . e . m is from triplicates . RT-qPCR was normalized to the amount of total mRNA . ( C ) Average food intake of WT and KO mice from week 4 to week 9 after birth . n = 9 . ( D ) Average body weights of WT and KO littermates from week 4 to week 9 after birth , n = 9 . ( E ) H&E and FABP4 antibody stain subcutaneous fat cells in E18 . 5 WT and Taf7l KO embryos , left panel magnification , ×5; right panel magnification , ×20; red arrows indicate fat cells stained by FABP4 . ( F ) Taf7l KO mice exhibits less fat tissue than WT littermate . Shown are representative photographs of 1-month-old mice with skin removed from both front and back views . ( G ) Taf7l KO mouse exhibits less fat formation than WT littermate . Shown is a representative photograph of 4-month-old mouse with skin removal . DOI: http://dx . doi . org/10 . 7554/eLife . 00170 . 009 To determine the influence of Taf7l KO in early fat development in vivo , we next examined the adipose formation in E18 . 5 embryos of WT control and Taf7l KO littermates . We utilized haematoxylin and eosin ( H&E ) staining on transversal sections of the interscapular region to visualize the overall structure of skin and underlying tissue including subcutaneous fat , connective tissue and muscle . In particular , we used FABP4 antibody staining to localize developing subcutaneous adipose tissue . We observed that FABP4+ lipid-laden cells are significantly diminished in the subcutaneous layer of Taf7l KO mice compared to WT littermate controls with a concomittant increase in layers of connective-like tissue under the skin of Taf7l KO mice ( Figure 4E ) . These results suggest that loss of Taf7l impairs WAT development in the later stages of embryogenesis ( Gupta et al . , 2010 ) . It also appears that in Taf7l-deficient mice , there may be an imbalance in mesodermal-derived lineages as revealed by the appearance of a thicker layer of subcutaneous connective-like tissue . We next examined the formation of WAT in 1-month and 4-month-old Taf7l KO and control WT littermates . The results revealed that 5 out of 12 Taf7l KO mice exhibited a noticeable reduction in both their subcutaneous and abdominal white fat pads compared to control WT animals ( Figure 4F , G ) , while both groups consumed similar amounts of food and have similar growth curves ( Figure 4C , D ) . Taken together , our observations with isolated primary adipose fibroblasts , E18 . 5 embryos , and fat pads from young mice are consistent with the notion that Taf7l likely functions in potentiating mouse WAT development . A complementary strategy to probe the capacity of transcription factors to influence specific differentiation pathways involves the ‘reprogramming of cell fate’ . For instance , transdifferentiation of C2C12 myoblasts into adipocytes by ectopic expression of PPARγ and/or C/EBPa under adipogenic permissive conditions helped establish these sequence-specific enhancer binding factors as key regulators of adipogenesis ( Tontonoz et al . , 1994a , 1994b; Hu et al . , 1995; Kajimura et al . , 2009 ) . Therefore , we tested the adipogenic function of TAF7L by an analogous ‘gain of function’ approach with forced introduction of TAF7L or control vector into C2C12 myoblasts . First , we generated TAF7L-expressing ( C2C12 . TAF7L ) or control C2C12 ( C2C12 . CNTL ) stable cell lines by transfecting either FLAG-TAF7L or empty vector followed by neomycin selection . As detected by FLAG antibody through Western blot analysis , C2C12 . TAF7L stable cells achieved modestly elevated expression levels of FLAG-TAF7L protein ( Figure 5B ) . Similarly , C2C12 . TAF7L cells express roughly eightfold higher Taf7l mRNA levels than C2C12 . CNTL cells ( Figure 5—figure supplement 1A ) . Next , we treated both C2C12 . TAF7L and C2C12 . CNTL stable cell lines with the four standard adipogenic inducers for 5 days and then applied Oil red O staining . A large proportion of C2C12 . TAF7L cells developed into lipid-laden cells while no detectable C2C12 . CNTL cells produced lipid droplets ( Figure 5A ) . Gene expression analysis by RT-qPCR confirmed that C2C12 . TAF7L cells have markedly increased mRNA levels of a subset of adipocyte-specific genes including Adipsin , Resistin , Pparγ , C/EBPα , Adipoq and Fabp4 relative to C2C12 . CNTL cells post differentiation ( Figure 5C and Figure 5—figure supplement 1B , C , E , F ) . By contrast , myoblast-gene Myf5 is downregulated in C2C12 . TAF7L cells prior to and post adipogenic induction ( Figure 5—figure supplement 1D ) . Furthermore , we performed mRNA-seq on differentiated C2C12 . TAF7L and C2C12 . CNTL cells and these genome-wide expression studies revealed that indeed , a number of adipocyte-specific genes become highly upregulated in C2C12 . TAF7L cells compared to C2C12 . CNTL cells post adipogenesis ( Figure 5D ) . Gene ontology analysis of genes upregulated fivefold or more in differentiated C2C12 . TAF7L cells vs C2C12 . CNTL cells indicated that ∼32% of these genes are involved in either adipocyte differentiation or function . Notably , in accordance with the role of Taf7l in spermatogenesis that was reported previously , ∼10% of these up-regulated genes are involved in spermatogenesis and sexual reproduction ( Figure 5E ) . Taken together , these results indicate that ectopic expression of TAF7L in C2C12 myoblasts , even at modestly elevated levels , is capable of inducing upregulation of Taf7l itself and other important adipogenic transcription factors including Pparγ and C/ebpα ( Figure 5—figure supplement 1A–C ) thereby reprogramming a significant portion of C2C12 cells into adipocytes upon induction ( Figure 5—figure supplement 1E , F ) , providing further evidence for TAF7L as a pro-adipogenic regulator . 10 . 7554/eLife . 00170 . 010Figure 5 . Ectopic expression of TAF7L transdifferentiates C2C12 myoblasts into adipocytes under adipogenic induction . ( A ) C2C12 myoblasts expressing empty vector ( C2C12 . CNTL ) or TAF7L ( C2C12 . TAF7L ) were stained with Oil red O 5 days after inducing adipogenesis . ( B ) Western blot analysis on ectopic expression levels of FLAG-TAF7L in C2C12 . 7L and C2C12 . CNTL cells , β-actin protein level is served as a loading control . CNTL , C2C12 . CNTL; TAF7L , C2C12 . TAF7L . ( C ) mRNA levels of adipocyte marker genes are measured by RT-qPCR in C2C12 . TAF7L cells compared with C2C12 . CNTL cells 5 days post adipogenesis , mRNA levels of genes in C2C12 . CNTL cells were assigned to 1 . *p<0 . 05 , data is mean and s . e . m is from triplicates . RT-qPCR was normalized to the amount of total mRNA . ( D ) mRNA-seq analyzes genes activated by TAF7L in C2C12 . TAF7L compared to C2C12 . CNTL post adipogenesis . Red dots represent genes upregulated in C2C12 . TAF7L-post cells; blue dots represent genes unaltered or downregulated in C2C12 . TAF7L-post cells compared to C2C12 . CNTL-post cells after adipogenic induction . ( E ) Major gene functional groups from genes activated above fivefold by TAF7L in C2C12 cells post adipogenic induction through gene ontology analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 00170 . 01010 . 7554/eLife . 00170 . 011Figure 5—figure supplement 1 . Gene expression analysis of TAF7L-expressing C2C12 cells . ( A ) – ( F ) Time course of gene expression by RT-qPCR analysis of Taf7l ( A ) , Pparγ ( B ) , C/ebpα ( C ) , Myf5 ( D ) , Adipoq ( E ) , and Fabp4 ( F ) in C2C12 . CNTL and C2C12 . TAF7L cells at 0D , 1D , 2D , 3D , 4D and 5D post adipogenic induction . D , days; CNTL , C2C12 . CNTL; TAF7L , C2C12 . TAF7L . mRNA levels in C2C12 . CNTL cells at 0D were assigned to 1 , mRNA levels of each gene at 0D , 1D , 2D , 3D , 4D , and 5D in both C2C12 . CNTL and C2C12 . TAF7L cells during adipogenesis were compared to mRNA levels in C2C12 . CNTL cells at 0D respectively , data is mean from triplicates . DOI: http://dx . doi . org/10 . 7554/eLife . 00170 . 011 To explore the potential mechanism by which TAF7L functions during adipogenesis , we took advantage of chromatin immunoprecipitation combined with deep sequencing ( ChIP-seq ) to map TAF7L , TBP , and Pol II binding profiles genome-wide prior to and after adiopgenesis . Mapped ChIP tags were analyzed by intersecting MACS and Grizzly Peak algorithms ( Zhang et al . , 2008; Feng et al . , 2011; Harrison et al . , 2011 ) to identify binding regions for each factor . In agreement with the low concentration of TAF7L shown in Western blots prior to differentiation ( Figure 1A ) , only one significant peak was identified in C3H10T1/2 cells compared to 18 , 672 significant TAF7L binding peaks detected after adipocyte formation . At the same time , we found comparably large numbers of peaks for TBP ( 12 , 883 and 14 , 587 ) and Pol II ( 14 , 502 and 11 , 424 ) in pre- and post-differentiation C3H10T1/2 cells . As an example of our TAF7L ChIP-seq data , the profiles of two typical adipocyte-specific genes Adipoq and Klf15 , a general highly expressed gene Rfc4 , and a nonactive gene Ccdc37 are shown in Figure 6A , B . The expression level of each gene before and after differentiation can be deduced from the enrichment levels of Pol II . 10 . 7554/eLife . 00170 . 012Figure 6 . TAF7L binds strongly on the majority of genes upregulated during adipogenesis . ( A ) Read accumulation for eight ChIP-seq datasets including TAF7L , PPARγ , TBP and Pol II before ( _pre ) and after ( _post ) adipocyte differentiation at the Rfc4 and Adipoq gene loci . ( B ) The same as in ( A ) at the Ccdc37 and Klf15 gene loci . Vertical axis is 0–500 reads for all factors , co-localized peaks were marked with boxes , black boxes indicate promoters and red boxes indicate enhancers , solid lines denote active genes and dashed lines denote inactive gene . ( C ) Frequency ( vertical axis ) of TAF7L occupancy on gene expression groups ( horizontal axis ) including unchanged ( low , med , high ) ( three blue dots regions from left-bottom to right-top in Figure 3A ) , downregulated ( blue dots in left-bottom region in Figure 3A ) , and upregulated ( >5× , >50× , two orange dots regions from lower to higher in Figure 3A ) . ( D ) Average TAF7L binding signal strength ( vertical axis ) on the core promoters ( 500 bp from TSS ) and proximal enhancers ( 500 bp to 5 kb from TSS ) of three major gene expression groups as in ( C ) . ( Regular t-test for ( C ) and ( D ) , NS is no significant , *p<0 . 05 , ***p<0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00170 . 012 We split the genome into three groups representing unchanged , downregulated , and upregulated ( >5× , >50× ) genes based on their expression pattern and levels before and after differentiation based on mRNA-seq data . Analysis of our ChIP-seq and mRNA-seq data suggests that among all active genes in adipocytes , TAF7L binds to >65% of adipocyte-specific genes that are highly upregulated ( >50× ) during adipogenesis while binding much less frequently at both core promoters ( <500 bp from TSS ) and proximal enhancer regions ( 500 bp to 5 kb from TSS ) near genes unaltered or downregulated during differentiation ( Figure 6C ) . Next we compared the expression levels of genes in adipocytes to the binding intensity of TAF7L and found a strong positive correlation in the three expression groups . For instance , examination of 3 , 468 TAF7L peaks , representing 20% of the total peaks located at proximal enhancers revealed that the average TAF7L binding strength on genes upregulated >50× is twice as strong as on genes induced between 5–50× in adipocytes relative to C3H10T1/2 cells , and eight times stronger than on genes down-regulated following adipogenesis ( Figure 6D ) , suggesting that TAF7L binding frequency and strength is highly correlated with upregulated genes during adipogenesis . By mapping the genomic binding sites , we found that both TBP and Pol II display greater than 63% occupancy at transcriptional start sites ( TSS ) of highly expressed genes in both undifferentiated C3H10T1/2 cells and adipocytes , consistent with their roles in mediating global and general transcription functions . Interestingly , comparing the binding regions of TAF7L , TBP and Pol II in adipocytes revealed that TAF7L only partially ( 30% ) colocalizes with TBP and Pol II at a subset of promoters , while a greater proportion ( 45% ) of TAF7L peaks localizes to enhancer regions where TBP and Pol II are generally not found ( Figure 7B , C ) . This surprising finding suggests that TAF7L may function via additional mechanisms other than as a subunit of canonical TFIID in regulating adipocyte differentiation . 10 . 7554/eLife . 00170 . 013Figure 7 . TAF7L colocalizes and associates with PPARγ and TBP . ( A ) Two top motifs ( motif 1 and motif 2 ) were found in TAF7L binding sites . Motif1 p<2e-20 ) matches with PPARγ binding motif and motif 2 p<3e-10 ) matches with C/EBPα binding motif . ( B ) Overlap of PPARγ peaks with TAF7L peaks in adipocytes , each circle represents the total peaks from ChIP-seq for a factor and the overlapped region represents the common binding peaks of the factors . ( C ) Similar as in ( B ) ; Pol II , TBP and TAF7L peaks from ChIP-seq overlap with each other in adipocytes . ( D ) Table showed the total peak numbers of each factor in adipocytes from ChIP-seq and the percentage of genome-wide peak overlapping between TAF7L and PPARγ , Pol II , TBP , IgG control . ( E ) FLAG tagged TAF7L , HA tagged PPARγ were overexpressed in 293T cells , immunoprecipitations were performed on both FLAG and HA antibodies and followed by Western blotting with FLAG and HA antibodies . ( F ) The same procedures were performed on FLAG tagged PPARγ and HA tagged TBP . ( G ) The same procedures were performed on FLAG tagged TAF7L and HA tagged TBP as in ( E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00170 . 013 To determine mechanisms by which TAF7L operates via its occupancy at enhancers , we probed its potential association with adipocyte-specific enhancer binding transcription factors . First , we applied an unbiased genome-wide approach to identify sequence-specific transcription factors that could interact with TAF7L . We analyzed the sequences surrounding TAF7L binding sites and identified several DNA consensus sequence motifs enriched in TAF7L peaks . Next , we compared these motifs with all known sequence-specific recognition elements of transcription factors , which led to the identification of several binding motifs of well-known adipogenic transcription factors including PPARγ and C/EBPγ ( Ge et al . , 2008 ) ( Figure 7A ) . We then used ChIP-seq to directly map the genome-wide binding profiles of PPARγ and detected 4 , 121 and 12 , 893 significant peaks for PPARγ in C3H10T1/2 cells pre- and post-differentiation , respectively . We then validated our ChIP-seq data by comparing our new data with published ChIP-seq data for PPARγ , C/EBPα , and RXRα in 3T3-L1 derived adipocytes 6 days post-differentiation ( Mikkelsen et al . , 2010; Schmidt et al . , 2011; Waki et al . , 2011 ) . This extensive analysis revealed that the majority of PPARγ binding sites we mapped overlap with those previously reported although some differences in specific binding sites were observed , likely due to inherent differences between 3T3-L1 and C3H10T1/2 cells . A direct comparison of TAF7L and PPARγ genome-wide binding profiles revealed that 26% of TAF7L peaks were co-occupied by PPARγ and reciprocally , 37% of PPARγ peaks were co-bound by TAF7L ( Figure 7B , D ) . Similarly , TAF7L binds to 20% of RXRα bound loci genome-wide . Moreover , TAF7L also co-localizes with 25% of C/EBPα binding and vice versa . Collectively , these findings provide indirect evidence for a functional association between TAF7L and adipogenic activators PPARγ , RXRα and C/EBPα in adipocytes . Indeed , the relationship between TAF7L , PPARγ , C/EBPα and TBP/Pol II at a genome-wide scale suggests that TAF7L might also functions as an enhancer-associated co-activator that connects adipogenic activators with the core promoter recognition machinery to potentiate adipocyte differentiation . Prompted by the extensive co-localization between TAF7L , TBP and PPARγ , we set out to examine the potential physical association between these factors . First , we constructed TAF7L , TBP and PPARγ with either FLAG or HA Tags for expression in 293T cells in pair-wise combinations . Next , we performed co-immunoprecipitations ( co-IP ) between PPARγ , TAF7L and TBP with either FLAG or HA antibodies , followed by Western blot analysis to detect the FLAG- or HA-tagged proteins . Intriguingly , these co-IP assays showed that PPARγ can efficiently pull down TAF7L and vice versa with ( data not shown ) or without the addition of PPARγ ligand rosiglitozone ( Figure 7E ) ; by contrast , PPARγ was unable to co-IP TBP ( Figure 7F ) . As expected , TAF7L and TBP can pull down each other reciprocally ( Figure 7G ) , which is consistent with previous observations ( Pointud et al . , 2003 ) . To confirm that the association between TAF7L and PPARγ or TBP is direct and not mediated via DNA/chromatin interactions , we included benzonase treatment in our co-IP assays . Eliminating DNA in these co-IP experiments did not alter the binding interactions we observed between TAF7L and PPARγ or TBP ( data now shown ) . As expected , over-expression of TAF7L in C3H10T1/2 cells and adipocytes allows co-IP of other endogenous TFIID subunits including TBP and TAF4 ( data not shown ) , suggesting that some tagged-TAF7L can integrate into native TFIID complexes . Taken together , these protein:protein binding assays suggest that TAF7L can physically associate with both PPARγ and TBP/TFIID either directly or indirectly via presently unidentified protein factors . These studies provide a potential mechanism by which TAF7L may serve as a cofactor linking specific adipogenic activators , proximal enhancers and the core transcription apparatus . It is well-documented that the adult human body contains cells residing in the adipose tissue , referred to as Adipose-derived Stem Cells ( ASCs ) ( DeLany et al . , 2005; Feve , 2005; Gonzalez , 2005; Gerlach et al . , 2012 ) . ASCs resemble mesenchymal stem cells ( MSCs ) in terms of their ability to differentiate into multiple lineages including adipocytes , myotubes , osteocytes , and cartilage under appropriate developmental cues ( Gornostaeva et al . , 2006 ) . Given that increased numbers of adipocytes , a major underlying cause of obesity , are primarily derived from MSCs and/or ASCs ( Bowers and Lane , 2008 ) , we chose C3H10T1/2 MSCs as our cell culture model system for studying adipogenesis in large measure because MSCs efficiently recapitulate aspects of adipocyte differentiation and in vivo fat development . Using this MSC culture model as well as Taf7l KO mouse model for our in vitro and in vivo studies , we unexpectedly identified Taf7l as a key regulator of adipogenesis; adding a new piece of the molecular puzzle to the critically important regulators of fat development in mammalian organisms . We found the effect of Taf7l in adipogenesis to be quite robust wherein its loss led to extensive down-regulation of genome-wide adipocyte-specific gene expression in cell culture and defects in WAT development in vivo . We are particularly intrigued by the manner in which TAF7L seems to operate–serving both as an integral component of TFIID at the core promoter and as a key co-activator interacting directly with PPARγ or other adipocyte-specific transcriptional factors ( ATFs ) at proximal enhancers of adipocyte-specific genes on genome-wide scale ( Figure 8 ) . Thus , a hitherto unrecognized cell-type selective core regulator with an apparent dual mechanism of action has been identified that influences the pro-adipogenic transcriptional control network . It is conceivable that TAF7L and associated regulatory factors in this newly discovered pathway may reveal potentially useful therapeutic drug targets to combat obesity and its related diseases . 10 . 7554/eLife . 00170 . 014Figure 8 . Dual functions of TAF7L in adipocyte differentiation . TAF7L expression is enriched during C3H10T1/2 MSCs adipocyte differentiation while other TFIID subunits ( TAFs ) decrease in expression . TAF7L regulates adipogenesis by associating with TBP as a component of adipocyte TFIID complex at promoters and with PPARγ or other adipocyte transcriptional factors ( ATFs ) as a cofactor at enhancers on adipocyte-specific genes , providing the mechanisms of its dual roles during differentiation . General highly-expressed genes are those with high expression before and after adipocyte differentiation include a portion of housekeeping genes; adipocyte-specific genes are those required for adipocyte differentiation and highly upregulated during adipocyte differentiation . TAFs , TBP-associated factors; ATFs , adipocyte transcriptional factors; BEs , binding elements . DOI: http://dx . doi . org/10 . 7554/eLife . 00170 . 014 It came as a surprise to find Taf7l playing such an essential role in adipogenesis because previous studies had primarily reported a specific role of Taf7l in spermatogenesis ( Vernon et al . , 1986; Armstrong et al . , 1994; Hashmi et al . , 2011 ) . Indeed , even after being clued into the potential contribution of Taf7l in fat tissue development , our analysis of Taf7l KO mice mainly revealed defects in WAT formation at certain stages of development and there were no gross , easily-observable morphological abnormalities in young animals except when the underlying fat pads were dissected for direct inspection to reveal the partial penetrance of the lean phenotype . Also , previous studies had no reason to examine the expression of Taf7l in adipose tissue because Taf7l mRNA and protein levels are relatively low in adipose tissue compared with testis . We also note that although TAF7L protein levels become highly elevated when MSCs differentiate into adipocytes , its mRNA levels increase only modestly ( ∼2× ) compared to typical adipocyte marker genes such as Fabp4 and Glut4 during adipogensis ( Kajimura et al . , 2008; Seale et al . , 2008 ) . It is therefore not surprising that a role for Taf7l in adipocytes could have been overlooked in previous studies . This cautionary tale also suggests that other key transcriptional regulators could likewise go undetected; suggesting that more studies will likely be required to take a fuller accounting of the multiple factor combinations that have evolved to orchestrate the diversity of gene regulatory pathways during differentiation and development of metazoans . Although spermatogenesis and adipogenesis represent two biologically distinct differentiation processes , their common dependence on Taf7l may imply some underlying and perhaps hidden commonality . For example , one can speculate that energy storage and metabolic homeostasis are tightly related/linked to reproductive abilities since both are highly sensitive to and dependent on nutrient availability ( Vernon et al . , 1986; Hashmi et al . , 2011 ) . Another possible link involved steroid hormones ( i . e . testosterone ) , which are essential for spermatogenesis , are derived from cholesterol ( Chowdhury and Mukherjee , 1976; Mahmoud et al . , 1985 ) . In any case , our genome-wide mRNA-seq expression analysis unambiguously identified Taf7l-mediated enrichment of genes involved in both adipogenesis and reproductive processes . However , further investigation will be required to assess whether there are more direct links between adipogenesis and spermatogenesis and whether mutant Taf7l involved in male infertility also disturbs normal fat development and metabolism . Our data clearly showed that TAF7L binds to and regulates the expression of a subset of white adipose genes , interestingly we also observed downregulation of some brown adipocyte genes ( Cidea , Ucp1 and Elovl3 ) caused by TAF7L knockdown ( Figure 2C ) ( Kajimura et al . , 2009; Kajimura et al . , 2010 ) . However , our incipient attempts failed to pinpoint significant defects in interscapular BAT formation in Taf7l KO mice . Thus , future studies will be required to determine whether Taf7l is also involved in BAT development . Further , our preliminary attempts to measure metabolic functions suggest that Taf7l-deficient mice showed changed serum glucose levels after 24 hr fasting compared with control WT littermates , it will be interesting to explore whether and how Taf7l KO alters energy metabolism . Therefore , further detailed investigations will be required to more rigorously delineate the potential function of Taf7l in energy homeostasis . Mouse Taf7l full length cDNA was isolated from mouse testis tissue , amplified by PCR and then sequenced . Full length Taf7l , Taf7 , and Pparγ cDNAs were inserted into pCMV 3×FLAG-10 vector to construct pCMV 3×FLAG-TAF7L/TAF7/PPARγ . Taf7l , Tbp , and Pparγ full length cDNAs were cloned into the pCS2+ vector with either HA or FLAG tag at their N-terminus . pLKO . 1 shGFP and pLKO . 1 shTAF7L vectors were purchased from Open Biosystems . shTAF7L-resistant pCMV 3×Flag-TAF7LmA was made by site-directed mutagenesis to introduce two silent mutations in shTAF7L targeted TAF7L cDNA sequence ( Ding et al . , 2008 ) . A fragment of the Taf7l cDNA corresponding to residues 400–600 was cloned into the pGEX-4T-1 vector carrying a GST tag . GST-TAF7L ( a . a . 400–600 ) was expressed and purified from E . coli and injected into rabbits by Covance ( Covance Research Products Inc . , Denver , PA ) . Bleeds were collected after three boosts and TAF7L antibodies were tested and confirmed by in vitro transcribed and translated TAF7L protein and whole protein extracts from mouse testis of WT and Taf7l KO mice ( data not shown ) . The antisera obtained were affinity-purified using antigen immobilized on Affigel 10/15 resin ( Bio-rad , Hercules , CA ) . For Pol II antibody , monoclonal anti-Pol II ( 8WG16 ) was concentrated from hybridoma supernatant with Protein A Sepharose Beads ( GE Healthcare , Piscataway , NJ ) . Antibody information: anti-TAF4 ( BD 612054 ) , anti-TBP ( abcam 62126 ) , anti-FLAG ( Sigma , F3165 ) , anti-HA ( abcam 9110 ) , anti-β-actin ( Sigma , A2228 ) , anti-TAF7 ( abnova H00006879-M01 ) , anti-FABP4 ( abcam 66682 ) , Pol II ( monoclonal 8GW16 , protein-A purified ) , PPARγ ( sc-7196 ) , mouse and rabbit IgG ( prepared in-house and concentrated with Protein A Sepharose Beads ) . C3H10T1/2 , 3T3-L1 , C2C12 , HeLa , and 293T cells were cultured in high glucose DMEM with 10% fetal bovine serum at 10% CO2 . C3H10T1/2 shGFP and shTAF7L lentiviral shRNA knockdown stable cell lines were established by transfecting pLKO . 1 shGFP or pLKO . 1 shTAF7L into C3H10T1/2 cells and then subjecting to puromycin selection for 3 weeks . C3H10T1/2 shTAF7L cells were transfected with pCMV 3×Flag vector , pCMV 3×Flag-TAF7 or pCMV 3×Flag-TAF7LmA and then subjected to G418 and puromycin double selection for 3 weeks to establish shTAF7L+vector , shTAF7L+TAF7 or shTAF7L+TAF7LmA cell lines used in ‘rescue’ experiments in Figure 2E , F . C2C12 . CNTL and C2C12 . TAF7L were established by transfecting pCMV 3×Flag vector or pCMV 3×Flag-TAF7L vector into C2C12 cells and then underwent G418 selection for 3 weeks . For adipogenesis , 3T3-L1 and C3H10T1/2 cells were grown in high glucose DMEM supplemented with 10% fetal bovine serum . At confluence , cells were exposed to induction medium containing dexamethasone ( 1 μM ) , isobutylmethylxanthine ( IBMX , 0 . 1 mM ) , insulin ( 5 μg/ml ) , rosiglitazone ( 1 μM ) , and 10% FBS . 3 days later , cells were further cultured in high glucose DMEM containing insulin ( 5 μg/ml ) and rosiglitazone ( 1 μM ) until they were ready for harvest . C3H10T1/2 cells form mature adipocytes 5 days post induction; 3T3-L1 cells require 7–8 days to form adipocytes . For Oil red O staining , pre- and post-differentiated C3H10T1/2 and 3T3-L1 cells , WT and Taf7l KO adipose-derived primary fibroblasts , shGFP and shTAF7L-treated C3H10T1/2 cells , C2C12 . CNTL and C2C12 . TAF7L cells were washed once in PBS and fixed with freshly prepared 4% formaldehyde in 1×PBS for 30 min , followed by standard Oil red O staining method described previously ( Tang et al . , 2004 ) . For C2C12 myogenesis , C2C12 cells are cultured in maintenance media until confluence was reached . 2 days post confluence , cells were switched to differentiation media comprised of low glucose DMEM , 2% horse serum , and 5 μg/ml insulin , 3 days later , change fresh differentiation media and culture cells for additional 2 days , differentiated myotubes were harvested and purified by collecting the suspended cells after splitting and reseeding the cells for 1 hr . Total RNA from cultured cells or mouse tissues was isolated using QIAGEN RNeasy Plus mini columns according to the manufacturer's instructions ( Qiagen Inc . , Germantown , MD ) . For RT-qPCR analysis , 1 μg total RNA was reverse transcribed using cDNA reverse transcription kit ( Invitrogen , Carlsbad , , CA ) . SYBR green reactions using the SYBR Green PCR Master Mix ( Applied Biosystems , Warrington , UK ) were performed according to the manufacturer's instruction using an ABI 7300 real time PCR machine ( Applied Biosystems , Foster City , CA ) . Relative expression of mRNA was determined after normalization to total RNA amount . Student's t-test was used to evaluate statistical significance . Whole cell extracts were prepared from cells by homogenization in lysis buffer containing 50 mM Tris–Cl , pH 8 . 0 , 500 mM NaCl , and 0 . 1% Triton X-100 , 10% glycerol and 1 mM EDTA , supplemented with protease inhibitor cocktail ( Roche , Indianapolis , IN ) and phenylmethylsulphonyl fluoride ( PMSF ) . Fifteen micrograms ( μg ) of whole-cell lysates were separated by SDS-PAGE and transferred to nitrocellulose membrane . For immunoblotting , membranes were blocked in 10% milk , 0 . 1% Tween-20 in TBS for 30 min , and then incubated with TAF7L , TAF4 , TAF7 , FLAG , β-actin , PPARγ and TBP antibodies for 2 hr at room temperature; detailed Western blotting procedure was performed as previously described ( Zhou et al . , 2006 ) . 500 μg whole-cell extracts from 293T cells transfected with FLAG-TAF7L and HA-PPARγ , FLAG-PPARγ and HA-TBP , or FLAG-TAF7L and HA-TBP were immunoprecipitated with FLAG or HA antibodies at 4°C for overnight under the conditions of 0 . 3 M NaCl and 0 . 2% NP-40 , 30 μl protein A/G beads were added and incubated for additional 2 hr at 4°C , after extensive washing with buffer containing 0 . 15 M NaCl and 0 . 1% NP-40 , remaining beads were subjected to 10% SDS-PAGE and followed by western blotting analysis with FLAG and HA antibodies to detect tagged-proteins in the inputs and IPs as previously described ( Ding et al . , 2008 ) . The derivation of Taf7l-knockout mice has been previously described ( Cheng et al . , 2007 ) . All animal experiments were performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health . All of the animals were handled according to approved animal use protocols ( #R007 ) by Animal Care and Use Committee ( ACUC ) of the University of California , Berkeley . Mice were maintained on a standard rodent chow diet with 12 hr light and dark cycles . Taf7l KO mouse line was maintained on a C57/Bl6 background . Genotyping was performed by PCR as previously described ( Cheng et al . , 2007 ) . Fresh inguinal adipose tissues were removed from 3 week old euthanized WT and Taf7l KO mice and finely minced , digested with 0 . 25% trypsin for 30 min at 37°C , and centrifuged for 5 min at 2 , 000g . The pellet was resuspended in culture media before plated on gelatin coated plates . Cells were cultured at 37°C in high glucose DMEM supplemented with 20% FBS . Adipocyte differentiation and staining were followed the same procedure as C3H10T1/2 cells . For histological analysis on interscapular tissue of E18 . 5 embryos from WT and Taf7l KO mice , freshly-harvested mouse embryos were genotyped and then interscapular regions of embryos were transversally dissected and then fixed in 10% formaldehyde for 24 hr at 4°C; tissue was embedded in paraffin using the microwave method and then sectioned into 8–10 μm sections to mount on slides . This method and the following immunohistochemistry by haematoxylin and eosin ( H&E ) staining were performed using the method described previously by Steven Ruzin ( Schichnes et al . , 1998 ) , and FABP4 immunostaining method was modified from the one described previously ( Kajimura et al . , 2008 ) ( Gupta et al . , 2010 ) . Fix C3H10T1/2 cells and differentiated adipocytes with 1% formaldehyde for 10 min at room temperature then use 0 . 125 M glycine to stop the crosslinking for an additional 5 min . Collect cells and extract nuclei with extraction buffer . The chromatin obtained from C3H10T1/2 cells and adipocytes was fragmented to sizes ranging from 175 to 225 bp using a Covaris-S2 sonicator ( Covaris , Inc . , Woburn , MA ) for a total processing time of 40 min ( 20 s on , 20 s off ) . 900 µg of the sonicated chromatin was used in each immunoprecipitation reaction as previously described ( Liu et al . , 2011 ) with the Pol II , TBP , PPARγ , and TAF7L antibodies , mouse and rabbit IgG were used as negative controls respectively . Preparation of the sequencing libraries on the DNA samples of the immunoprecipitation from antibodies and IgGs precisely followed the instructions from Illumina ( Illumina Inc . , San Diego , CA ) , qualities of the libraries were assessed by 2100 Bioanalyzer ( Functional Genomics Laboratory , Berkeley , CA ) and then subjected to ultra-high throughput sequencing on an Illumina HiSeq 2000 sequencer ( GSL core facility , Berkeley , CA ) as previously described ( Fong et al . , 2011; Liu et al . , 2011 ) , each sample yielded ∼50–200 million reads . Sequenced reads were mapped to the July 2007 assembly of the mouse genome ( UCSC version mm9 , NCBI37 ) using Bowtie ( Langmead et al . , 2009 ) with the command-line options ‘-c -q -n 2 -l 48 -m 1’ , thereby keeping for further analyses only reads that mapped uniquely to the genome with at most two mismatches at the first 48 bases . To accurately identify significant binding positions across the genome , we incorporated two peak calling methods , as described previously ( May et al . , 2012 ) . First , we used MACS ( Zhang et al . , 2008 ) ( version 1 . 4 , with default settings except: --nomodel --shiftsize = 110 --pvalue=1e-2 --mfold = 10 , 10 , 000 --slocal = 2000 --llocal = 20 , 000 ) , with an FDR significant threshold of 0 . 05 ( using IgG ChIP as a control ) . Second , we applied the Grizzly peak fitting algorithm ( Harrison et al . , 2011 ) , which uses a model-based iterative approach to accurately identify multiple binding loci at every enriched region . Peaks were then associated with the nearest TSS ( using gene annotation from UCSC , version mm9 ) , and classified as promoter peaks ( up to 500 bp from start site ) , proximal enhancer peaks ( larger than 500 bp and less than 5 kb ) , distal enhancer peaks ( less than 50 kb from TSS ) , or none ( further away than any start site ) . The DNA sequence associated with each of the peaks is a 250 bp fragment that is centered on the highest point peak within the TAF7L binding region as defined by Grizzly Peak . Weeder ( version 1 . 4 . 2 ) ( Pavesi et al . , 2004 ) searches the inputted sequences for enriched motifs . The Weeder output is then passed through STAMP ( Mahony and Benos , 2007 ) to compare the identified motif's frequency matrix against the JASPAR v2010 database and identify which transcription factors have that DNA binding motif . Total RNAs were extracted from C3H10T1/2 cells and adipocytes , differentiated shTAF7L-C3H10T1/2 cells , differentiated C2C12 . CNTL and C2C12 . TAF7L cells by RNeasy Plus Mini Kit ( Qiagen ) , 8 μg of each sample was used to purify mRNA and subsequently converted into to mRNA-seq library using mRNA-Seq Sample Prep Kit ( Illumina ) and sequenced on an Illumina HiSeq 2000 sequencer . 50 bp paired-end were used for the C3H10T1/2 samples and 100 bp single-end were used for the C2C12 samples , both resulting in over 175 million reads yield per sample . The reads were than mapped to the mouse transcriptome ( created using UCSC table browser , version mm9 , on February 2012 ) , using TopHat ( Trapnell et al . , 2009 ) , version v1 . 4 . 0 . , using default parameters . We then applied cufflinks ( Trapnell et al . , 2010 ) , version v1 . 3 . 0 , using the default parameters except: --max-mle-iterations 1 , to estimate the digital expression levels at each transcript . Due to the high number of sequence reads , these steps were done in seven batches of 25 million reads each , per sample . Raw and mapped sequencing reads are available from the National Center for Biotechnology Information's GEO database ( http://www . ncbi . nlm . nih . gov/geo/ ) under accession number GSE41937 . All supplemental data , including gene-by-gene ChIP-seq and gene expression mRNA-seq data , are available at http://eisenlab . org/data/TAF7L . A genome browser with mapped mRNA and ChIP profiles and other related data discussed in the manuscript can be accessed at http://eisenlab . org/data/TAF7L/browse or directly at http://genome . ucsc . edu/cgi-bin/hgTracks ? hgS_otherUserName=tomkap&hgS_otherUserSessionName=TAF7L

The development of a single fertilized egg into a highly complex animal is determined by its genome , with a process called differential gene regulation exerting exquisite control over gene expression to ensure that various specialized cells are generated and that many types of tissue are produced . However , the mechanisms responsible for controlling gene expression and , therefore mammalian development , are poorly understood . Researchers have developed a number of in vitro cell culture models to elucidate the details of differential gene regulation , and this approach has been used to characterize adipocytes—cells that store energy in the form of fat—for close to two decades . The formation of adipocytes , a process known as adipogenesis , has been extensively studied , but there remain major gaps in our knowledge: for example , the identities of many of the transcriptional regulators that are responsible for the differentiation of mesenchymal stem cells into adipocytes remain a mystery . This task is complicated by the fact that some of these regulators are involved in the differentiation of multiple cell lines , and that some of them also have multiple roles in the generation of a single cell type . In addition to being of fundamental interest , improving our knowledge of the properties and behavior of adipocytes is essential for tackling the increasing prevalence of obesity in the developed world . Zhou et al . now report that TAF7L—a gene that was previously thought to be involved only in the production of sperm cells—has two roles in the differentiation of stem cells to form adipocytes . Using a combination of cellular , biochemical , genetic and genomic techniques , they show that TAF7L interacts with PPARγ , an important adipocyte transcriptional regulator at enhancer sites on the genome to increase the transcription of genes that are involved in adipogenesis . They also show that TAF7L interacts with a general transcription factor called TBP ( short for TATA-binding protein ) at promoter sequences , again to increase the expression of genes involved in adipogenesis . Moreover , they show that the expression of TAF7L in myoblasts—precursor cells that usually become muscle cells—can induce the formation of fat cells rather than muscle cells . Furthermore , mice lacking TAF7L are lean compared to their normal littermates . A clearer understanding of the underlying causes of fat cell formation could lead to the development of new approaches for the treatment of obesity and associated diseases .

2013

Dual functions of TAF7L in adipocyte differentiation

The histone acetyltransferase ( HAT ) Mof is essential for mouse embryonic stem cell ( mESC ) pluripotency and early development . Mof is the enzymatic subunit of two different HAT complexes , MSL and NSL . The individual contribution of MSL and NSL to transcription regulation in mESCs is not well understood . Our genome-wide analysis show that i ) MSL and NSL bind to specific and common sets of expressed genes , ii ) NSL binds exclusively at promoters , iii ) while MSL binds in gene bodies . Nsl1 regulates proliferation and cellular homeostasis of mESCs . MSL is the main HAT acetylating H4K16 in mESCs , is enriched at many mESC-specific and bivalent genes . MSL is important to keep a subset of bivalent genes silent in mESCs , while developmental genes require MSL for expression during differentiation . Thus , NSL and MSL HAT complexes differentially regulate specific sets of expressed genes in mESCs and during differentiation . Pluripotent mouse embryonic stem cells ( mESCs ) have the ability to self-renew or to differentiate into all cell types . Specific transcription factors like Oct4 , Sox2 , and Nanog form a core transcriptional network , which is required for the maintenance of mESC pluripotency ( Orkin et al . , 2008 ) . Chromatin-modifying enzymes further regulate transcriptional mESCs networks and cellular differentiation processes and can be associated with activation or repression of genes ( Orkin and Hochedlinger , 2011 ) . Histone acetylation is important for mESC pluripotency and is regulated by the concerted action of histone acetyltransferases ( HATs ) and histone deacetylases ( HDACs ) ( Meshorer and Misteli , 2006 ) . Acetylation of histone proteins leads to an open and dynamic chromatin conformation allowing an active transcription state , which is also a signature of mESC pluripotency ( Meshorer , 2007; Niwa , 2007; Efroni et al . , 2008 ) . During differentiation of mESCs , the overall transcription rates decrease , whereas the chromatin structure becomes more compact with a global reduction of histone H3 and H4 acetylation . In line with the requirement of histone acetylation in mESC maintenance and differentiation , genetic deletion or knockdown of several HATs affects mESC pluripotency ( Lin et al . , 2007; Fazzio et al . , 2008; Gupta et al . , 2008; Thomas et al . , 2008; Zhong and Jin , 2009; Li et al . , 2012 ) . HATs can be classified into two predominant families: the GCN5-related N-acetyltransferase ( GNAT ) family ( i . e . Gcn5 and p300 ) and the Moz-Ybf2/Sas3-Sas2-Tip60 ( MYST ) family ( i . e . , Tip60 and Mof [male absent on the first] ) ( reviewed in Kimura et al . , 2005 ) . These enzymes often function as part of multi-protein co-activator complexes ( reviewed in Lee and Workman , 2007 ) . Mof ( also known as Kat8 or Myst1 ) , is a MYST-type HAT specific for histone H4 lysine 16 acetylation ( H4K16ac ) ( Hilfiker et al . , 1997; Smith et al . , 2001 , 2005; Taipale et al . , 2005 ) and has been shown to be the catalytic subunit of two distinct protein complexes in Drosophila ( d ) and mammals: the male-specific lethal ( MSL ) and the non-specific lethal ( NSL ) complexes ( Smith et al . , 2005; Mendjan et al . , 2006; Cai et al . , 2010; Raja et al . , 2010 ) . In Drosophila , the dMSL complex is targeted to transcribed regions of male X-chromosomal genes , where it mediates dosage compensation ( reviewed in Straub and Becker , 2007; Gelbart and Kuroda , 2009; Conrad and Akhtar , 2011 ) . In contrast , dNSL is present at gene promoters of male and female chromosomes , where it regulates transcription of housekeeping genes ( Prestel et al . , 2010; Raja et al . , 2010; Feller et al . , 2012; Lam et al . , 2012 ) . Mof itself and subunits of the Mof-containing dMSL and dNSL HAT complexes are required for the binding of the two Drosophila Mof-containing complexes at promoters and gene bodies , which leads to H4K16 acetylation and gene expression ( Raja et al . , 2010; Kadlec et al . , 2011 ) . Inactivation of Mof in mice ( m ) leads to early embryonic lethality as Mof−/− embryos fail to develop beyond the expanded blastocyst stage and die at implantation ( Gupta et al . , 2008; Thomas et al . , 2008 ) . Mof deletion correlated with cell cycle defects and cell death . Moreover , mESCs could not be derived from Mof−/− mouse embryos . In agreement , it was shown that Mof plays an essential role in the maintenance of mESC pluripotency ( Li et al . , 2012 ) . H4K16 acetylation levels were undetectable in Mof−/− embryos , whereas the acetylation of other histone lysine residues was unaffected ( Thomas et al . , 2008 ) . Surprisingly , loss of H4K16 acetylation upon neuronal differentiation of mESCs did not alter higher-order chromatin compaction ( Taylor et al . , 2013 ) . Moreover H4K16ac and Mof were reported to be present at the transcription start sites ( TSSs ) of expressed genes in mESCs ( Li et al . , 2012; Taylor et al . , 2013 ) . Details on the function of the mammalian Mof-containing MSL and NSL complexes have only recently started to emerge and revealed that Mof fulfils different functions within the two HAT complexes . Human ( h ) MSL complex is composed of the subunits MSL1 , MSL2 , MSL3 , and MOF ( Smith et al . , 2005; Mendjan et al . , 2006 ) , while the hNSL complex is composed of nine subunits: NSL1 , NSL2 , NSL3 , MCRS1 , WDR5 , PHF20 , HCF1 , OGT1 , and MOF ( Mendjan et al . , 2006; Cai et al . , 2010 ) . Mammalian NSL complex appears to have broader substrate specificity than the MSL complex , as it is also able to acetylate non-histone targets ( Li et al . , 2009 ) . However , the function of MSL and NSL complexes in mammalian cells and especially their role in establishing mESC pluripotency in not well understood . To better understand the role of MSL and NSL in gene regulation and their individual contribution in epigenetic changes in mESCs , we have analysed these two Mof-containing complexes by chromatin immunoprecipitation coupled with high throughput sequencing ( ChIP-seq ) and by shRNA knockdown ( KD ) experiments in mESCs . The obtained genome-wide binding maps show that MSL and NSL locate to a large number of expressed genes and each complex has a distinct binding profile at promoters or gene bodies . Our combined ChIP-seq and KD data indicate that MSL and NSL have a combinatorial effect on a given set of genes , whereas some specific loci are only MSL- or only NSL-dependent . Our data indicate that NSL binds exclusively at promoters , while MSL binds more in gene bodies . We show that in mESCs NSL regulates cell growth whereas MSL is the main HAT complex acetylating histone H4K16 . MSL is present at mESC-specific genes . Moreover , MSL binds to and regulates developmental genes in mESCs and during differentiation . Altogether our data demonstrate that MSL and NSL complexes are present at expressed genes in mESCs , but that MSL is essential for regulation of key mESC-specific and bivalent developmental genes . To understand the global role of the two Mof-containing complexes in chromatin remodelling and how this regulates genes linked to self-renewal , proliferation , and/or differentiation , we set out to analyse the genome-wide binding of MSL and NSL in mESCs . To this end , we raised antibodies targeting Msl1 or Nsl1 , which are specific subunits of the MSL or NSL complexes , respectively , and are known to play a role in the assembly and the regulation of these complexes ( Raja et al . , 2010; Kadlec et al . , 2011 ) . The specificity of the purified antibodies was demonstrated by western blot assays and immunoprecipitations followed by mass spectrometry using the multidimensional protein identification technology ( MudPIT ) ( Figure 1 ) . Western blot assays indicated that both of the generated antibodies are specific ( Figure 1A , B ) . In addition , both antibodies immunoprecipitated ( IP-ed ) the endogenous MSL and NSL complexes with the previously described polypeptide composition ( Cai et al . , 2010; Figure 1C ) . Importantly , Mof was identified in both IP-ed MSL or NSL complexes , in the same range of abundance than Msl1 , or Nsl1 ( Figure 1C , Figure 1—source data 1 for all identified proteins by MudPIT ) . Gel filtration followed by western blot analyses further indicated that Msl1 and Nsl1 are only present in Mof-containing complexes as they have eluted from the Superose 6 column in the same molecular weight containing fractions as their respective entire endogenous MSL ( about 240 kDa ) , or NSL ( about 800 kDa ) complexes ( Figure 1D ) . Of note the enzymatic subunit Mof was detected in the respective MSL and NSL complexes , but in addition as a potentially free form in the 50 kDa range fractions ( Figure 1D ) . These results together demonstrate the incorporation of all nuclear Msl1 , or Nsl1 , together with Mof , in their respective endogenous complexes and a fraction of ‘free’ Mof that is not present in either MSL or NSL . 10 . 7554/eLife . 02104 . 003Figure 1 . Msl1 and Nsl1 incorporate into endogenous complexes in mESCs . ( A and B ) Western Blot analysis using the raised anti-Msl1 ( 3208 ) or anti-Nsl1 ( 3130 ) antibodies on nuclear extracts . Preimmune sera ( pre ) were used as negative controls . ( C ) Anti-Msl1 ( 3208 ) or anti-Nsl1 ( 3130 ) antibodies were used to immunoprecipitate protein complexes from mESC nuclear extracts . The IP-ed complexes were then analysed by multidimensional protein identification technology ( MudPIT ) . The identified MSL- or NSL-containing complex proteins and their relative protein abundance in the samples are represented by normalized spectral abundance factor ( NSAF ) ( Zybailov et al . , 2006 ) . NSAF allows the comparison of abundance of individual proteins in multiple independent samples and in multiprotein complexes ( Florens et al . , 2006; Paoletti et al . , 2006 ) . The colour intensity reflects of the NSAF values multiplied by 1000 ( as indicated ) . ( D ) Gel filtration of mESC nuclear extracts . Every second fraction eluted from a Superose 6 column was analysed for the presence of Nsl1 , Msl1 , and Mof by Western Blot . Molecular weight markers for the corresponding fractions are indicated on the top of the panel . DOI: http://dx . doi . org/10 . 7554/eLife . 02104 . 00310 . 7554/eLife . 02104 . 004Figure 1—source data 1 . List of identified proteins of MudPIT analyses . SAF and NSAF values of all significantly enriched proteins in the Msl1 3208 or Nsl1 3130 IP in NE of mESCs . DOI: http://dx . doi . org/10 . 7554/eLife . 02104 . 004 To characterize the genome-wide role of MSL and NSL complexes , we carried out ChIP-seq analysis in mESCs using the above-characterized anti-Msl1 and anti-Nsl1 antibodies . The obtained binding maps of Msl1 and Nsl1 in mESCs were then compared at the UCSC genome browser to publicly available ChIP-seq data for Mof , H4K16 acetylation ( H4K16ac ) , RNA Polymerase II ( Pol II ) , and DNAse hypersensitive sites ( DHS ) . At a representative genomic locus , Nsl1 peaks were detected at the TSSs of four expressed genes , where they co-localized with Mof , Pol II and DHSs , whereas Msl1 binding peaks were usually broader and together with H4K16ac downstream of Pol II peaks ( Figure 2A ) . As previously reported , Mof is present at promoters , gene bodies ( GBs ) and intergenic regions ( Li et al . , 2012 ) . 10 . 7554/eLife . 02104 . 005Figure 2 . Distinct binding profiles of Msl1 and Nsl1 at active genes . ( A ) UCSC genome browser tracks representing Msl1 , Nsl1 , H4K16ac , Mof and Pol II ChIP-seq data . The Input serves as control . ( B ) Scatter Plot showing the Pearson correlation between Msl1 and Nsl1 densities at all identified MACS14 peak regions . Densities were normalized to the control ( Input ) and represented as log2 values . ( C ) Mapping of Msl1 and Nsl1 identified MACS14 peaks to different genomic regions ( promoter-TSS , genebody [GB] or intergenic regions ) using HOMER ( Heinz et al . , 2010 ) . Identified peaks are listed in Figure 2—figure supplement 1A and Figure 2—source data 1 . Validation of identified peaks is shown in Figure 2—figure supplement 1B , C . ( D ) Heatmap showing k-means clustering of Msl1 , Nsl1 , Mof and H4K16ac using the TSSs of all ENSEMBL transcript IDs as reference coordinates . Densities are presented ±2 kb around reference coordinates . Input serves as negative control . ( E and F ) Average binding profiles of Msl1 , Nsl1 and Mof ( E ) at a region of +1 kb around the annotated TSSs and ( F ) 1 kb upstream of the TSS , in the GB and 1 kb downstream of the TTS . Only Nsl1 or Msl1 positive genes were taken into consideration . The Input serves as control and tag densities were normalized to the input . See Figure 2—figure supplement 2 for validation of ChIP-seq data . DOI: http://dx . doi . org/10 . 7554/eLife . 02104 . 00510 . 7554/eLife . 02104 . 006Figure 2—source data 1 . List of Msl1 and Nsl1 MACS14 peaks . DOI: http://dx . doi . org/10 . 7554/eLife . 02104 . 00610 . 7554/eLife . 02104 . 007Figure 2—source data 2 . List of Msl1 and Nsl1 positive genes . DOI: http://dx . doi . org/10 . 7554/eLife . 02104 . 00710 . 7554/eLife . 02104 . 008Figure 2—figure supplement 1 . Identification and validation of Msl1 and Nsl1 binding sites . ( A ) Identified binding sites of Msl1 and Nsl1 using MACS14 ( p-value<10−5 ) ( Zhang et al . , 2008 ) , their presence at the promoter ( TSS ) or genebody ( GB ) of ENSEMBL transcript IDs and the total number of positive ENSEMBL genes . The annotation was conducted with HOMER ( Heinz et al . , 2010 ) . ( B and C ) ChIP-qPCR in mESCs using the anti-Msl1 ( 3208 ) or anti-Nsl1 ( 3130 ) antibodies as indicated . Primers were designed at randomly selected MACS14 peaks with increasing tag densities ( t ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02104 . 00810 . 7554/eLife . 02104 . 009Figure 2—figure supplement 2 . Knockdown ( KD ) of Msl1 or Nsl1 through lentiviral shRNA vectors . ( A and C ) ESCs were treated with sh control , sh Msl1 or sh Nsl1 expressing lentiviral vectors , 5 days after lentiviral infection total RNA was isolated and RT-qPCR was carried out . Samples were normalized to actin and the sh control was set to 100% . Error bars represent standard deviation of three independent experiments . ( B and D ) Validation of the Msl1 or Nsl1 downregulation in sh control or shRNA conditions by Western Blot . 20 μg of proteins were loaded per lane and normalized by ponceau staining . ( E and F ) anti-Msl1 ( E ) or anti-Nsl1 ( F ) ChIP was carried out on sh control ( E ) or sh Nsl1 ( F ) KD-treated cells . The results of ChIP-qPCR amplifications at selected genes ( TSS or genebody [gb] regions ) are shown . Fold enrichment higher than five is defined as binding . DOI: http://dx . doi . org/10 . 7554/eLife . 02104 . 009 Using MACS14 algorithm ( Zhang et al . , 2008 ) we determined high-confidence binding sites ( peaks ) for Msl1 or Nsl1 ( Figure 2—figure supplement 1A , Figure 2—source data 1 ) and selected peaks with various tag densities for ChIP-qPCR validation . The Msl1 and Nsl1 enrichments at five different loci as detected by ChIP-qPCR faithfully reflected the tag densities measured by ChIP-seq ( Figure 2—figure supplement 1B , C ) . To further verify the specificity of the Msl1 and Nsl1 ChIP-seq results , we used lentiviral small hairpin ( sh ) RNA vectors to knockdown ( KD ) Msl1 or Nsl1 in mESCs ( Figure 2—figure supplement 2A–D ) and tested by ChIP-qPCR the decrease of Msl1 or Nsl1 binding at the TSSs and in the GBs of two genes that were co-bound by these factors ( Figure 2—figure supplement 2E , D ) . The predominant binding of Msl1 to GBs was lost upon Msl1 KD , whereas Nsl1 binding to TSS was reduced following Nsl1 depletion . These results confirm our ChIP-seq analyses . Next , we asked whether the two complexes bind to common or different loci . A pairwise comparison of the MSL or NSL enrichment at all high confidence binding loci revealed that the binding of both complexes show two populations and have a Pearson correlation coefficient of 0 . 23 ( p-value=1 . 88 × 10−160 ) ( Figure 2B ) . This indicates a significant overlap between Msl1 and Nsl1 binding populations , but suggests also a differential genome-wide binding of MSL and NSL . To know at which genomic regions the identified peaks localize , each peak was annotated either to promoter , GB ( containing introns , exons , untranslated regions and transcription termination sites together ) or intergenic regions . 74% of all Msl1 peaks are detected at GBs ( Figure 2C ) . In contrast , the majority of identified Nsl1 peaks are present at promoter regions ( 67% ) ( Figure 2C ) . Moreover , only about 10% of all Msl1- or Nsl1-binding sites map to intergenic regions ( as defined above , excluding introns ) . The majority of the 9890 Msl1- , or 6251 Nsl1-specific binding sites are at promoter and/or GB regions ( Figure 2C ) , and after removal of redundant genes , we defined 5844 Msl1- and 4755 Nsl1-bound genes ( Figure 2—figure supplement 1A , Figure 2—source data 2 ) . As only 10% of the binding sites were detected at intergenic regions , we focused our further analyses on the role of MSL and NSL complexes in gene regulation at the promoter and/or GB regions . To understand the genome-wide binding of MSL and NSL , we compared by k-means clustering either Msl1 , or Nsl1 binding profiles with that of Mof and the presence of H4K16ac at 30 , 300 ENSEMBL transcription start sites ( TSSs ) . In good agreement with our results showing that in mESCs Msl1 and Nsl1 incorporate in the endogenous MSL or NSL complexes , respectively , genome-wide Mof binding overlaps with that of Msl1 and Nsl1 around most of the TSSs , which are also H4K16ac positive ( Figure 2D ) . Interestingly , at promoters , MSL and NSL complexes have distinct binding profiles . Nsl1 and Mof show a sharp binding peak centred at the TSSs , while the average Msl1 binding profile is similar to H4K16ac ( see below ) and extends downstream from the TSSs in the GB regions ( Figure 2E ) . Moreover , the Msl1 and Mof signals are enriched downstream of promoters at GBs , whereas the control and the Nsl1 signals are not ( Figure 2F ) . Altogether our results demonstrate that MSL and NSL bind mostly to distinct sites in mESCs . NSL binds directly to the TSS region of genes , while the genome-wide location of MSL is both at TSSs and downstream of the TSSs of bound genes . To assess the relationship between Msl1 and Nsl1 binding and gene expression in mESCs , we took advantage of available RNA-seq data ( Tippmann et al . , 2012 ) and compared the average expression of Msl1- , or Nsl1-bound genes ( in median log2 FPKM values ) to that of all ENSEMBL genes ( Figure 3A ) . The median expression values for Msl1- , or Nsl1-positive genes were significantly higher as compared to all ENSEMBL genes , demonstrating that Msl1 and Nsl1 are mostly present at expressed genes in mESCs . 10 . 7554/eLife . 02104 . 010Figure 3 . Msl1 and Nsl1 bind to active genes , but participate differentially to gene expression . ( A ) Boxplots showing the log2 of RNA FPKM expression values from mESCs of all analysed , Msl1- , or Nsl1-bound ENSEMBL genes . ( B , C and D ) RNA expression values are ranked into five groups , where group 5 represents the highest RNA expression level and group 1 the lowest ( see bottom of panels B–D ) . Boxplots show the tag density of the nearest peak to the TSS for ( B ) Pol II , ( C ) Msl1 and ( D ) Nsl1 tag densities around the TSSs at the five groups . Only density values higher than zero were taken into consideration . The median is different between groups , if the notches of the boxplots do not overlap . DOI: http://dx . doi . org/10 . 7554/eLife . 02104 . 010 To determine whether the binding strength of Msl1 or Nsl1 correlates with gene expression , we compared Msl1 and Nsl1 enrichment around TSSs with gene expression data from the corresponding bound genes . Msl1- or Nsl1-positive genes were divided into five categories according to their expression levels ( Figure 3B–D ) . As a control , in the same five categories densities of Pol II peaks at promoters correlated with gene expression with decreasing Pol II densities from highly to poorly expressed genes ( Figure 3A; Barski et al . , 2007 ) . Importantly , the boxplot representation revealed a similar correlation as Pol II between Msl1 binding and gene expression , indicating that the stronger the gene is expressed the higher Msl1 and Pol II are enriched at the binding sites ( Figure 3B , C ) . In contrast , there is no significant difference of the Nsl1 median values between the five groups , indicating that Nsl1-binding to promoters is not proportional with the level of expression ( Figure 3D ) . Our results thus demonstrate that both Msl1 and Nsl1 bind to active genes , but that only the binding strength of Msl1 , and not that of Nsl1 , correlates with mRNA levels , suggesting a different dynamic and/or functional behaviour of the two complexes at the regulated loci . As H4K16 is a known target of Mof in Drosophila and mammals ( Hilfiker et al . , 1997; Smith et al . , 2001 , 2005; Taipale et al . , 2005 ) , we compared Msl1 or Nsl1 binding sites with the presence of H4K16ac . Our scatter plot analyses indicate that there is a general overlap of Msl1 or Nsl1 with H4K16ac , whereas the correlation between Msl1 binding sites and H4K16ac is better ( Pearson correlation coefficient 0 . 57 ) than between Nsl1 and H4K16ac ( Pearson correlation coefficient 0 . 32 ) , which is also reflected in the corresponding p-values ( Figure 4A , B ) . The comparison of the distribution patterns of Msl1 , Nsl1 , Pol II and H4K16ac around the TSSs ( ±2 kb ) of all Msl1- and Nsl1-bound genes further indicates that the Msl1 binding profile is more similar to the genome-wide presence of H4K16ac , than that of Nsl1 ( Figure 4C ) . H4K16ac levels are enriched downstream of the TSSs overlapping with the binding profile of Msl1 ( Figure 4C ) . In contrast , the centre of the Nsl1 binding profile centred at the TSS region does not overlap with that of the H4K16ac peak ( Figure 4C ) . These binding profiles suggest a link between H4K16 acetylation and the MSL HAT complex ( Figure 4 ) . 10 . 7554/eLife . 02104 . 011Figure 4 . MSL affects H4K16 acetylation in mESCs . ( A and B ) Scatter Plots indicating the Pearson correlation and Pearson p-values between H4K16ac and Msl1 ( A ) or Nsl1 ( B ) densities at Msl1 peaks or Nsl1 peaks . Log2 represented tag densities were calculated at peak regions and normalized to the control ( Input ) data set . ( C ) Average binding profiles of Msl1 , Nsl1 , Pol II and H4K16ac at a region of +2 kb around all ENSEMBL promoters . Only Nsl1 and Msl1 positive genes are taken into consideration . The input serves as control and tag densities are normalized to the input . ( D ) mESCs were treated for 5 days with lentiviral vectors expressing sh control , sh Msl1 , or sh Nsl1 interfering RNAs . Total histones were isolated by acidic extraction and H4K16ac , H4K5ac , and H4K8ac levels were analysed by western blot . Histones were normalized using an antibody against non-modified histone 3 ( H3 ) . KD efficiencies were tested in Figure 2—figure supplement 2A–D . DOI: http://dx . doi . org/10 . 7554/eLife . 02104 . 011 Although it was demonstrated that Mof depletion in embryos results in a loss of H4K16ac ( Gupta et al . , 2008; Thomas et al . , 2008 ) , the exact contribution of the two Mof-containing HAT complexes to H4K16 acetylation remains to be determined . To address this question , we analysed the global acetylation of H4K16 after Msl1 or Nsl1 KD and also quantified acetylation of H4K5 and H4K8 , two other proposed substrates for hNSL in differentiated human cells ( Cai et al . , 2010 ) . Western blot analyses of total histone proteins from mESCs expressing shRNAs targeting Msl1 or Nsl1 revealed a dramatic reduction of H4K16ac upon Msl1 depletion , whereas Nsl1 KD did not affect H4K16ac levels ( Figure 4D ) . This is in good agreement with the differential Msl1 and Nsl1 ChIP-seq profiles ( Figure 4C ) . Moreover , H4K5ac and H4K8ac levels did not change in cells expressing either Msl1 or Nsl1 shRNA ( Figure 4D ) . Altogether , the above results indicate that in mESCs ( i ) the enzymatic activity of the MSL complex is responsible for H4K16 acetylation downstream of the TSS , ( ii ) MSL is the main acetylase for H4K16 and ( iii ) the global H4K16 acetylating function of MSL cannot be compensated by other HAT complexes . To understand the role of the two Mof-containing complexes for gene regulation and regulatory pathways in mESC , we further characterized genes bound either individually or together by Msl1 and/or Nsl1 . Out of 10 , 600 Msl1- and Nsl1-bound genes about one quarter are co-bound by both complexes , while 3274 are only bound by Msl1 and 2185 only by Nsl1 ( Figure 5A , Figure 2—source data 2 ) . Our statistical analyses showed that these numbers are significant ( Figure 5—figure supplement 1 ) . To identify genes regulated specifically by MSL and/or by NSL , we analysed genes bound by only Msl1 , by only Nsl1 , or together by Msl1 and Nsl1 for gene ontology ( GO ) . All three categories are enriched for GO terms such as metabolic process , gene expression , cell proliferation , and cell cycle . These GO terms represent housekeeping functions of every cell type , but can also be related to the cellular homeostasis of ESCs . Interestingly , genes bound by only Msl1 are enriched for GO terms such as embryo development , stem cell differentiation and maintenance ( Figure 5B ) . Importantly , almost 50% of all reference genes associated with stem cell maintenance are Msl1 positive ( Figure 5B ) . 10 . 7554/eLife . 02104 . 012Figure 5 . MSL and NSL bind to shared and specific gene sets . ( A ) Venn Diagram showing the overlap between Msl1 and Nsl1 binding sites at ENSEMBL genes . For statistical analyses see Figure 5—figure supplement 1A–C . Binding at TSSs and gene bodies was considered together . Genes are listed in Figure 2—source data 2 . ( B ) Gene ontology analysis using Manteia ( Tassy and Pourquie , 2013 ) of only Msl1 binding sites , only Nsl1 binding sites , or common binding sites . Significant GO terms for only Msl1 binding sites are highlighted by a box . ( C ) Differentially expressed genes were identified with the DEseq analysis ( Anders and Huber , 2010 ) . The table represents genes expressed only in mESCs ( when compared to NPCs ) . The overlap with all Msl1 positive , all Nsl1 positive or only complex specific genes was calculated . Statistical analyses in Figure 5—figure supplement 2A–D indicates significant enrichment of Msl1 at mESC-specific genes . ( D and E ) Bound genes were divided into three categories: Common ( category 1 ) , specific for Nsl1 ( category 2 ) and specific for Msl1 ( category 3 ) . From these categories genes were chosen for ChIP-qPCR using the anti-Msl1 ( D ) or anti-Nsl1 ( E ) antibodies . Fold enrichment higher than five was defined as specific binding and Msl1 and Nsl1 presence was analysed at the indicated genes in each category . Bar charts represent the mean and standard deviation of 2–3 independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 02104 . 01210 . 7554/eLife . 02104 . 013Figure 5—figure supplement 1 . MSL and NSL significantly bind to shared and specific gene sets . ( A–C ) Bootstrap statistical analyses ( ‘Materials and methods’ ) were carried out with a random selection of 10 , 600 genes ( IDs ) out of a total pool of 26 , 460 ENSEMBL IDs . Histograms represent the average numbers of observed IDs in the three random sets . The average numbers and SDs are: only Msl1 1303 ± 27 ( A ) , Msl1 and Nsl1 1027 ± 23 ( B ) and for only Nsl1 870 ± 22 ( C ) . These averages are significantly far from the experimentally determined numbers ( p-value<1 . 0e−90 ) shown with an arrow in bold . The z-cores are 75 ( only Msl1 ) , 66 ( Msl1 and Nsl1 ) and 60 ( Nsl1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02104 . 01310 . 7554/eLife . 02104 . 014Figure 5—figure supplement 2 . MSL , but not NSL locates to mESC-specific genes . ( A–D ) Bootstrap statistical analyses ( ‘Materials and methods’ ) were carried out with a random selection of 282 genes ( IDs ) out of a total pool of 26 , 460 ENSEMBL IDs . Histograms represent the average numbers of observed IDs in the four random sets . The average numbers and SDs are: 62 ± 7 ( all MSL ) ( A ) , 35 ± 6 ( only MSL ) ( B ) , 51 ± 7 ( all NSL ) ( C ) and 23 ± 5 ( only NSL ) ( D ) . The averages obtained by random selection are significantly far from those obtained experimentally ( shown with an arrow in bold ) for MSL ( A and B ) : p-value=3 . 27e−18 ( all MSL ) and 1 . 11e−31 ( only MSL ) . However , the averages are not significantly different ( shown with an arrow in bold ) for NSL ( C and D ) : p-value=0 . 107 ( all NSL ) and p-value=0 . 1177 ( only NSL ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02104 . 014 Therefore , we investigated the presence of Msl1 ( and Nsl1 ) at mESC-specific genes . Available RNA-seq data of mESCs ( Tippmann et al . , 2012 ) allowed us to define 282 genes expressed only in pluripotent mESCs . Out of these 282 mESC-specific genes , 123 ( 44% ) are bound by Msl1 and only 40 ( 14% ) are Nsl1 positive . Furthermore , about 100 mESC-specific genes are bound exclusively by Msl1 , while 16 genes are bound only by Nsl1 ( Figure 5C ) . Our statistical analyses indicate that only MSL binding at mESC-specific genes is higher than random ( Figure 5—figure supplement 2 ) . To validate these bioinformatics analyses , all Msl1- and/or Nsl1-bound genes were divided into three categories ( Figures 2B and 5A ) : Msl1- and Nsl1-bound genes ( category 1 ) , genes bound only by Nsl1 ( category 2 ) and genes bound only by Msl1 ( category 3 ) . Msl1 and Nsl1 binding to the three gene categories were validated by ChIP-qPCR on a few selected genes . In agreement with all our above analyses , we observed that Msl1 binds to TSSs and/or GB regions of most genes from category 1 and 3 ( Figure 5D ) , whereas Nsl1 is detected mostly at the TSSs of genes from category 1 and 2 ( Figure 5E ) . Our results also show that Msl1 positive genes contain several mESC specific genes , including genes related to the core pluripotency network ( e . g . , Oct4 , Nanog and Sox2 ) ( Figure 5E ) . In summary , we demonstrate that the two Mof-containing complexes bind to shared , MSL- , or NSL-specific gene sets , but only the MSL complex is present at genes regulating the ESC pluripotency network and developmental processes . In mouse embryos ablation of Mof results in lethality at embryonic day 3 . 5 and Mof also affects mESCs pluripotency ( Gupta et al . , 2008; Thomas et al . , 2008; Li et al . , 2012 ) . To further analyse the cellular roles of MSL and NSL in mESCs , Msl1 , or Nsl1 were individually depleted by shRNA KD ( see Figure 2—figure supplement 2A–D ) . To exclude compensation between MSL and NSL complexes , a double KD of Msl1 and Nsl1 ( shMsl1/shNsl1 ) was also carried out ( Figure 6—figure supplement 1A ) . Next , total cell numbers were counted over 6 days . These analyses indicated that KD of Msl1 reduces slightly cell proliferation , while the KD of Nsl1 , or the double KD of Msl1 and Nsl1 lead to a much slower cell growth ( Figure 6A ) . When analysing cell morphology under these KD conditions , we did not observe any change in mESC shape . ( Figure 6B ) As the reduction of cell numbers under the KD conditions was not due to apoptosis ( Figure 6—figure supplement 1B ) , we next carried out cell cycle analyses . These FACS measurements demonstrated that mESCs treated with shMSL1 , shNsl1 and shMsl1/shNsl1 accumulate in the G1-phase of the cell cycle , with shNsl1 and shMsl1/shNsl1 being more severe than shMsl1 ( Figure 6C ) . These results together suggest that Nsl1 might be more required for regulating housekeeping genes involved in cellular homeostasis of mESCs , as reflected in higher number of G1-phase cells and decreased cell proliferation of shNsl1 and shMsl1/Nsl1 mESCs . 10 . 7554/eLife . 02104 . 015Figure 6 . NSL influences cell growth and cell cycle of mESCs . ( A ) Cell proliferation analyses by cell counting over 6 days of control , or indicated KD mESCs . Error bars represent the standard deviation of three independent experiments . See Figure 6—figure supplement 1A for validation of KD efficiency of sh Msl1/Nsl1 double KD mESCs . ( B ) Morphology of control and Msl1 and Nsl1 single or double KD mESCs at 6 days after lentiviral infection using a reverse-phase microscope with a 10x magnification . ( C ) Cell cycle analyses of control and KD mESCs by propidium iodide staining followed by FACS analyses . Cell numbers of G1- , S- or G2-phases are represented in percentages after analyses with CellQuest Pro software . See Figure 6—figure supplement 1B for apoptosis analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 02104 . 01510 . 7554/eLife . 02104 . 016Figure 6—figure supplement 1 . shMsl1 , shNsl1 and double KD mESCs do not undergo apoptosis . ( A ) mESCs were treated with sh control or a mix of sh Msl1 and sh Nsl1 ( sh Msl1/Nsl1 ) expressing lentiviral vectors . 5 days after lentiviral infection Msl1 and Nsl1 downregulation in sh control or sh Msl1/Nsl1 mESCs was tested by Western Blot . 20 μg of proteins were loaded per lane and normalized by ponceau staining and the blots were revealed with the indicated antibodies . ( B ) Cell death analyses of sh control , sh Msl1 , sh Nsl1 and sh Ms1/Nsl1 mESCs 6 days after lentiviral infection . Sh control mESCs treated with 10 mM of H2O2 were used as positive control . Phosphatidylserine appearance at the outer cell membrane of apoptotic cells was analysed by colorimetric measurement at 550 nm using the APOpercentage assay . DOI: http://dx . doi . org/10 . 7554/eLife . 02104 . 016 To better understand the function of genes bound by MSL and NSL in mESCs , genome-wide expression changes were analysed by microarrays . To this end total RNA was isolated from control mESCs , or mESCs depleted for either Msl1 , or Nsl1 ( Figure 2—figure supplement 2A–D ) . In shMsl1 KD cells , 275 genes were found to be downregulated ( with Msl1 itself is in the downregulated list ( Figure 7—source data 1 ) ) , and 500 genes upregulated , as compared to control KDs . By comparing Msl1-bound and Msl1-regulated genes ( Figure 7A ) , we found that Msl1 is present at about 30% ( 105 genes ) of all downregulated and at 20% ( 107 genes ) of all upregulated genes . In Nsl1 KD conditions , 1158 genes are downregulated ( including Nsl1 itself ) and 429 are upregulated , as compared to KD controls ( Figure 7—source data 1 ) . By comparing the genome-wide binding ( ChIP-seq ) and expression data changes following KD , we show that Nsl1 is present at 43% ( 441 genes ) of all downregulated and at 5% ( 30 genes ) of all upregulated genes ( Figure 7B ) . The Msl1- , or Nsl1-KD affected genes determined by the microarray analyses were then confirmed by RT-qPCR under Msl1- , or Nsl1-KD conditions ( Figure 7C , D ) . Interestingly , we noticed genes known to be involved in differentiation , like Nestin and Ntrk1 , in the upregulated genes of shMsl1 mESCs ( Figure 7C ) . 10 . 7554/eLife . 02104 . 017Figure 7 . Analysis of Msl1 and Nsl1 regulated genes in mESCs . ( A and B ) Venn Diagram of Msl1 ( A ) or Nsl1 ( B ) MACS14 binding sites at ENSEMBL genes and down- or upregulated genes in Msl1 or Nsl1 KD cells . Down- and upregulated genes are listed in Figure 7—source data 1 . For validation of KD efficiencies see Figure 2—figure supplement 2A–D . ( C and D ) RT-qPCR validation of down- and up-regulated genes in sh Msl1 ( C ) or sh Nsl1 ( D ) mESCs . ( E ) GO analyses using Manteia ( Tassy and Pourquie , 2013 ) of all downregulated genes in Msl1 KD , Nsl1 KD and Mof KO mESCs . See Figure 7—figure supplement 1 for further GO analyses . ( F ) Gene expression changes in Mof KO vs wild-type mESCs according to Li et al . ( 2012 ) of genes involved in stem cell maintenance , represented as fold change . DOI: http://dx . doi . org/10 . 7554/eLife . 02104 . 01710 . 7554/eLife . 02104 . 018Figure 7—source data 1 . Down- and upregulated genes in shMsl1 and shNsl1 mESCs . DOI: http://dx . doi . org/10 . 7554/eLife . 02104 . 01810 . 7554/eLife . 02104 . 019Figure 7—figure supplement 1 . Msl1 , Nsl1 and Mof regulate mESC-unspecific genes . ( A ) GO analysis ( Tassy and Pourquie , 2013 ) of downregulated genes in sh Msl1 , sh Nsl1 and Mof KO mESCs . ( B ) Oct4 expression in cell extracts prepared from sh control , sh Msl1 , sh Nsl1 and sh Msl1/Nsl1 mESCs was analysed by Western Blot . Extracts from mouse embryonic fibroblasts ( mEFs ) were used as negative control . 20 μg of proteins were loaded per lane and normalized by ponceau staining , and the blots were developed with an anti-Oct4 antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 02104 . 01910 . 7554/eLife . 02104 . 020Figure 7—figure supplement 2 . Msl1 and Mof binding at pluripotency genes . ( A and B ) Msl1 , Nsl1 and Mof binding together with Pol II and H4K16ac profiles at the ( A ) Pou5f1 ( Oct4 ) and ( B ) Sox2 locus at the UCSC genome browser . The Input serves as control . DOI: http://dx . doi . org/10 . 7554/eLife . 02104 . 020 Altogether , our results show that the bound genes of which the expression is affected following either Msl1 or Nsl1 KD are those genes , which absolutely require either MSL or NSL for their correct regulation . Note however , that the relatively weak overlap between Msl1- and/or Nsl1-bound genes on one side and Msl1- and/or Nsl1-regulated genes on the other , may reflect that either the KDs were not sufficiently efficient , and/or that the global gene expression analysis detected only changes in the steady state levels of the mature mRNAs and not the changes in the neosynthesized pre-mRNAs . Along these lines , in our experimental system Msl1 , or Nsl1 single , or double KD mESCs do not loose Oct4 expression ( Figure 7—figure supplement 1A ) , a common marker of mESC pluripotency . Since MSL and NSL are transcriptional co-activators , we were interested in the biological function of downregulated genes upon KD of either Msl1 or Nsl1 . We also included available expression data from Mof knock-out ( KO ) mESCs ( Li et al . , 2012 ) to overcome the above-described limitations . Analysing the biological function of Msl1 , Nsl1 , or Mof downregulated genes , we observed GO terms like metabolic processes , gene expression , cell death , or cell cycle control ( Figure 7—figure supplement 1B ) . However , only downregulated genes in Mof KO mESCs are significantly enriched for GO terms like stem cell differentiation or maintenance ( Figure 7E ) . Several of these genes are amongst the Msl1 positive mESC-specific genes , such as Nanog , Sox2 , Oct4 ( Li et al . , 2012 ) ( Figure 7F ) . Moreover , these genes are bound by Msl1 and Mof ( Figure 7—figure supplement 2 ) . Thus , the Mof-regulation and the exclusive binding of Msl1 and Mof to these key pluripotency genes suggest that the MSL complex is a regulator of the pluripotency network in mESCs . Our above analyses have shown that Msl1 binds not only to mESC-specific genes , but that it locates also to silent , or very weakly expressed , genes that become expressed to control mESC differentiation ( Figure 8A , Figure 8—figure supplement 1A ) . Importantly , KD of Msl1 leads to the upregulation of developmental genes , such as Nestin and Ntrk1 ( Figure 7C ) . These genes often contain both positive ( H3K4me3 ) and negative ( H3K27me3 ) epigenetic modifications ( Figure 8A , Figure 8—figure supplement 1A ) . It is well established that H3K4me3 and H3K27me3 histone modifications co-localize at bivalent domains , which are poised for a quick activation during distinct differentiation processes ( Azuara et al . , 2006; Bernstein et al . , 2006 ) . The Ezh2 subunit of the polycomb repressive complex 2 ( PRC2 ) , which catalyzes histone H3K27 tri-methylation , is also a good marker of bivalent domains ( Bernstein et al . , 2006; Ku et al . , 2008 ) . To determine whether Msl1 , or Nsl1 , would bind to bivalent domains genome-wide , we compared the combined list of all Msl1 and Nsl1 binding sites with Pol II and Ezh2 profiles , together with H3K4me3 and H3K27me3 marks ( Figure 8B ) . The heatmap indicates that about 343 Msl1 binding sites significantly co-localize with Ezh2 , H3K27me3 , and H3K4me3 , which define the bivalent domains ( see Cluster C in Figure 8B , Figure 8—figure supplement 1B for statistical analyses ) . Importantly , these 343 bivalent domain sites are negative for Nsl1 binding . The presence of Msl1 ( our study ) and Mof ( Li et al . , 2012 ) at bivalent genes in mESCs suggests that the MSL complex is involved in keeping these developmental genes silenced , but poised for activation in mESCs . 10 . 7554/eLife . 02104 . 021Figure 8 . MSL regulates mESC differentiation . ( A ) Msl1 binding together with Nsl1 , H3K4me3 , H3K27me3 and Pol II at the Nestin locus in mESCs . For the Hes5 gene locus see Figure 8—figure supplement 1A . ( B ) Heatmap showing k-means clustering of Msl1 , Nsl1 , Pol II , H3K4me3 , Ezh2 and H3K27me3 using all Msl1 and Nsl1 binding sites as reference coordinates . Densities are presented −/+2 kb around reference coordinates . Based on the density profiles of all data sets , the heatmap is divided into different categories ( as indicated ) . For statistical analyses of Msl1 positive bivalent genes in Cluster C see Figure 8—figure supplement 1B . ( C and D ) mRNA expression measurements by RT-qPCR of bivalent genes , which are also key markers for NPC differentiation , under sh control ( dark grey ) and sh Msl1 conditions ( light grey ) in pluripotent mESCs ( C ) or in mNPCs ( D ) NPC formation and KD efficiency of Msl1 in NPCs was validated in Figure 8—figure supplement 1C , D . DOI: http://dx . doi . org/10 . 7554/eLife . 02104 . 02110 . 7554/eLife . 02104 . 022Figure 8—figure supplement 1 . MSL binds to bivalent genes . ( A ) H3K27me3 and Ezh2 profiles together with Nsl1 , Msl1 , H3K4me3 and Pol II at the bivalent Hes5 gene at the UCSC genome browser . The input serves as negative control . ( B ) Bootstrap statistical analyses ( ‘Materials and methods’ ) were carried out with a random selection of 13 , 505 genes ( IDs ) out of a total pool of 26 , 460 ENSEMBL IDs . The histograms represent the average numbers of observed IDs in the random sets . The average number and SD is: 165 ± 10 using a random selection ( p-value=4 . 48e−66 ) . The experimentally obtained 343 Msl1 positive bivalent genes in Cluster C ( indicated by a bold arrow ) is significantly far from the average of randomly selected gene IDs . ( C and D ) mESCs were treated with sh control or sh Msl1 expressing lentiviral vectors , 5 days after lentiviral infection mESCs were cultured without LIF . RA was added for additional 4 days to induce NPC formation . ( C ) Msl1 downregulation in sh control , or sh Msl1 conditions were tested by Western Blot as indicated . 20 μg of proteins were loaded per lane and normalized by tubulin and ponceau staining . ( D ) Morphology was analysed with a reverse-phase microscope using a 10x magnification . DOI: http://dx . doi . org/10 . 7554/eLife . 02104 . 022 As MSL , but not NSL , was found to be required for keeping bivalent genes silent or low expressed in pluripotent mESCs , we asked whether MSL could regulate bivalent gene expression during mESCs differentiation . For this , mESCs were differentiated into neuronal progenitor cells ( NPCs ) under control and Msl1 KD conditions ( Figure 8—figure supplement 1C ) . Bivalent genes such as Pax6 , Hes5 , Mapt2 and Nestin , which are also considered as key markers of NPC differentiation , were upregulated in pluripotent mESC under Msl1 KD conditions ( Figure 8C; and see above ) . In contrary , but in agreement of the regulatory role of the MSL complex at these genes , these key developmental marker genes were downregulated in NPCs in which Msl1 was silenced by shRNA expression during NPC differentiation ( Figure 8D ) . Note however , that Msl1 KD cells morphologically are still able to form NPC-like cells ( Figure 8—figure supplement 1D ) . These results together indicate the important regulatory requirement of the MSL complex first for keeping the subset of bivalent genes poised for activation in mESCs and then for turning them on during mESC differentiation . In this study , we analysed two Mof-containing complexes , MSL and NSL to understand their transcription regulation function in mESCs . The proteomic characterization of the mMsl1- , or mNsl1-containing complexes indicated that the subunit composition of the mMSL and the mNSL complexes is identical to human complexes ( Figure 1C and see Mendjan et al . , 2006; Cai et al . , 2010 ) . Importantly , the comparison of the abundance of Msl1 or Nsl1 with Mof in the respective complexes ( Figure 1C ) and our gel filtration analyses ( Figure 1D ) demonstrated the incorporation of Msl1 and Nsl1 together with Mof in their respective endogenous complexes . Moreover , the gel filtration experiment indicated the potential existence of ‘free’ Mof that may not be present in either MSL or NSL . Together , the proteomic analyses suggest that there is no free Msl1 or Nsl1 in the nuclei of mESCs and that Msl1 and Nsl1 are specific to the MSL and NSL complexes , respectively . This observation is important for our study as it indicates that the ChIP binding profiles obtained with either anti-Msl1 or anti-Nsl1 antibodies represent the behaviour of the corresponding endogenous Mof-containing MSL , or NSL HAT complexes . Furthermore , our findings are consistent with previous observations , which suggested that Msl1 and Nsl1 directly interact with Mof in their respective complexes , to stabilize the assembly of these complexes and to regulate their HAT activity ( Raja et al . , 2010; Kadlec et al . , 2011 ) . Similarly to Drosophila ( Prestel et al . , 2010; Raja et al . , 2010; Kadlec et al . , 2011; Feller et al . , 2012; Lam et al . , 2012 ) , our results demonstrate that mouse MSL and NSL have mainly distinct binding profiles at transcribed genes in mESCs . NSL binding overlaps with Pol II binding and DHSs at TSSs , while MSL locates more downstream of promoters towards the GB ( Figure 2 ) . These evolutionary conserved binding profiles further suggest that the function of the MSL and NSL complexes in transcriptional regulation are also conserved between Drosophila and vertebrate cells . Surprisingly , in mESCs the KD of Msl1 , but not that of Nsl1 , leads to a global loss of H4K16ac , without reducing H4K5ac and H4K8ac levels ( Figure 4D ) . In differentiated human cells Mof , and subunits of MSL ( i . e . Msl1 , Msl3 ) or the Nsl1 subunit of NSL have been reported to be crucial for global H4K16 acetylation by either MSL or NSL , respectively ( Li et al . , 2009; Zhao et al . , 2013 ) . Thus , while the subunit composition of the two Mof-containing MSL and NSL complexes is conserved between mESCs and differentiated human cells ( Figure 1C; Cai et al . , 2010 ) , the function of the NSL complex seems to be differently regulated in pluripotent mESCs than in differentiated cells . Our observation that the KD of the Nsl1 subunit of NSL does not abolish global H4K16ac levels in mESCs suggests that in these pluripotent cells NSL may have a very localized HAT activity around TSSs of bound genes and the acetylation at these loci cannot be detected in total histone preparations , in contrary to the Msl1 KD . This difference may also be due to the more dynamic recruitment of NSL by mESC-specific factors . In contrast to MSL , NSL binding to promoters does not correlate with RNA expression levels ( Figure 3B–D ) . This further suggests that the two Mof-containing complexes have different mechanisms of action in transcriptional regulation in mESCs . Moreover , the depletion of MSL function is supposed to recapitulate those chromatin perturbations and related cellular changes that are caused by the Mof KO and are linked to H4K16ac loss . Strikingly , the gene expression of only a small number of MSL or NSL bound genes was directly affected when either Msl1 or Nsl1 was depleted ( Figure 7A , B ) . This might be due to either inefficient KDs , or the measurement of steady-state mature mRNA levels under our experimental conditions . As contrary to Msl1 and Nsl1 , ‘free’ Mof was detected in mESC nuclear extracts ( Figure 1D ) , we cannot exclude the possibility that to certain extent Mof alone could compensate for the function of the MSL or NSL complex under Nsl1 and Msl1 KD conditions . However , the abolished global H4K16ac levels in shMsl1 mESCs would rather propose that other transcriptional co-activators , modifying other histone residues than H4K16 , could compensate the role of H4K16ac in transcriptional activation . In summary , our data demonstrating that MSL is the main HAT complex responsible for global H4K16ac in mESCs ( Figure 4D ) , together with the finding that genome-wide binding profiles of Msl1 and Mof overlap with H4K16ac , ( Figure 2D ) suggest that the co-activator role of MSL is linked to its H4K16 acetylation function at the bound genes . The transitions between distinct chromatin states , from the open acetylated chromatin of the pluripotent mESCs to the more compact deacetylated chromatin of the differentiated cells , suggest the requirement for a tightly regulated chromatin acetylating/deactylating balance that participates in defining pluripotency on one hand and the consequent commitments for distinct differentiation pathways on the other hand . The HAT Mof is important for mESC pluripotency . Mof-deficient embryos have slight cell cycle defects and undergo cell death ( Thomas et al . , 2008 ) . Under our experimental conditions , KD of Msl1 and Nsl1 alone or together did not affect expression of key transcription factors of the pluripotency network ( Figures 6A , Figure 7 ) . Our observation that in mESCs Nsl1 KD leads to decreased cell numbers during proliferation and an increase in cells in the G1-phase of the cell cycle shows that NSL might influence mESC proliferation ( Figure 6B–D ) . This further indicates that NSL might be required for the homeostasis of mESC , either by directly regulating transcription or through acetylation of non-histone targets . H4K16ac was shown to promote chromatin fibre decompaction in vitro ( Shogren-Knaak et al . , 2006; Robinson et al . , 2008; Allahverdi et al . , 2011 ) . Our data showing that Msl1 KD abolishes global H4K16ac levels , together with the aberrant chromatin compaction observed in the Mof-deficient embryos ( Thomas et al . , 2008 ) , suggest that the MSL complex is an important factor in establishing high acetylation levels required for more open chromatin conformation and consequent mESC pluripotency . In addition to its role as a general regulator of H4K16ac in mESCs , the MSL complex seems to be recruited to ESC-specific loci to regulate different steps in the transcription process , such as ( i ) chromatin accessibility , ( ii ) pre-initiation complex formation and/or ( iii ) Pol II transcription elongation rates . Importantly , the Msl1-bound mESC-specific genes are regulated by Mof . Note however , these Mof- and Msl1-bound and Mof-regulated genes were not affected by the KD of Msl1 ( Figure 5B–D , Figure 7E , F ) . As above explained this may be due to the different experimental systems used here and the Mof KO study . Nevertheless , we assume that these genes are regulated by the whole MSL complex . Altogether , the exclusive co-binding of Msl1 and Mof to pluripotency genes suggests that the MSL complex is a regulator of the pluripotency network in mESCs . Bivalent genes , which are either repressed or expressed at very low levels in mESCs can be directly upregulated or completely silenced upon differentiation ( Azuara et al . , 2006; Bernstein et al . , 2006 ) . So far , little is known about the function of HATs at bivalent genes . Interestingly , we show that a subset of bivalent genes ( about 350 genes ) is bound by MSL in ESCs ( Figure 8A , B , Figure 8—figure supplement 1A ) and consequently that the KD of Msl1 results in the upregulation of a subset of bivalent genes in mESCs ( Figure 8C ) . Consistent with our study , Mof has also been shown to be present at bivalent genes ( Li et al . , 2012 ) . It seems that the MSL complex , probably in concert with HDACs and/or other chromatin remodelling factors , can have a silencing function at these bivalent genes . In contrast , the same genes require MSL for expression during differentiation ( Figure 8D ) . Even though the morphology of NPCs was not obviously influenced under Msl1 KD conditions , expression of key developmental NPC genes , such as Pax6 and Hes5 , were downregulated during NPC differentiation ( Figure 8D ) . Thus , our findings together with the observation that Mof is also binding to bivalent genes in mESCs , strongly suggests that the presence of MSL at bivalent loci is important for keeping these bivalent genes poised in pluripotent mESCs , allowing a quick transcriptional upregulation of the same genes during mESC differentiation . In summary , MSL and NSL are key transcriptional co-activators at a large number of expressed genes in mESCs , whereas each complex has a distinct binding profile either at promoters ( NSL ) or gene bodies ( MSL ) . MSL and NSL have overlapping and distinct roles in transcriptional regulation in mESCs . NSL binds mostly to genes with housekeeping functions and mediates mESC proliferation suggesting that NSL is important for the cellular homeostasis of mESCs . MSL is the main acetyltransferase complex acetylating H4K16 . Moreover , MSL binds to mESC-specific genes , which are de-regulated in Mof ablated mESCs . Moreover MSL is present at bivalent domains in mESCs , where it may poise genes for activation during mESC differentiation . Importantly , expression of those genes is directly regulated by MSL in differentiated NPCs . In the future , it will be interesting to investigate how the genome-wide function of MSL and NSL changes during distinct mESC differentiation pathways . Wild-type male mESCs ( E14 . wt ) were cultivated on 0 . 1% gelatine ( Sigma , France ) and CD1 feeder cells ( 37°C , 5% CO2 ) in DMEM ( 4 . 5 g/l glucose ) w-Glutamax-I , 15% foetal calf serum ESC-tested , leukemia inhibiting factor ( 5 μg ) ( Sigma ) , 50 mM ß-Mercaptoethanol ( Invitrogen , France ) , penicillin/streptomycin ( Invitrogen ) , 200 mM L-glutamine ( Invitrogen ) , and non-essential amino acids ( GIBCO , France ) . To work under feeder-free conditions cells were treated with 1 mg/ml Collagenase ( GIBCO ) and 2 mg/ml Dispase ( GIBCO ) and cultivated for one passage without feeder cells on 0 . 1% gelatine ( Sigma ) coated plates . Experiments were conducted at passage 26–29 . Mouse embryonic fibroblasts ( 3T3 ATCC ) were cultivated in DMEM ( 4 . 5 g/l glucose ) , 10% newborn calf serum and gentamycin ( Invitrogen ) . For NPC generation , we followed the protocol of Bibel et al . ( 2007 ) . Briefly , 6 × 106 mESC were cultured in DMEM ( 4 . 5 g/l glucose ) w-Glutamax-I , 10% foetal calf serum ESC-tested , 50 mM ß-Mercaptoethanol ( Invitrogen ) , penicillin/streptomycin ( Invitrogen ) , 200 mM L-glutamine ( Invitrogen ) , and non-essential amino acids on bacteriological Petri dishes ( 37°C , 5% CO2 ) to start differentiation . After 4 days retinoic acid ( 5 μm ) ( Sigma ) was added to induce NPC formation . Experiments were conducted 8 days after differentiation . Polyclonal anti-Msl1 ( 3208 ) and anti-Nsl1 ( 3130 ) antibodies were generated by immunization of rabbits with the N-terminal ( 3-210 amino acids ) region of mouse Msl1 or C-terminal region ( 762-1037 amino acids ) of mouse Nsl1 . The fragment was amplified and cloned in pET28b ( Novagen , France ) vector to express proteins in E . coli ( BL21 ) . For primer sequences see Supplementary file 1 . Polyclonal antibodies were purified through Affi-Gel columns ( Bio-Rad ) . For WB analysis anti-Msl1 ( 3208 ) or anti-Nsl1 ( 3130 ) antibodies were diluted 1:2000 . Nuclear extracts were prepared from 30 P15 plates of mESCs with 80% confluency as described in Demeny et al . ( 2007 ) . Proteins of 3 mg ( Msl1 ) or 1 mg ( Nsl1 ) mESC nuclear extracts were immunoprecipitated ( IP ) with 100 μl protein A Sepharose beads and 20 μl of the anti-Msl1 ( 3208 ) or 20 μl of the anti-Nsl1 ( 3130 ) antibody . Antibody-protein A Sepharose containing the bound proteins were washed three times with IP buffer ( 25 mM Tris-HCl pH 7 . 9 , 10% glycerol , 0 . 1% NP40 , 0 . 5 mM DTT , 5 mM MgCl2 ) and 100 mM KCl and afterwards with IP buffer containing 250 mM KCl . Proteins were eluted from protein A Sepharose beads 150 μl of 0 . 1 M Glycine pH 2 . 6 . Elutions were neutralized by adding 50 μl of 2 M Tris pH 8 . 5 . MudPIT analyses were performed as previously described ( Washburn et al . , 2001; Florens et al . , 2006 ) . In summary , protein mixtures were TCA precipitated , urea-denaturated , reduced , alkylated , and digested with endoproteinase Lys-C ( Roche ) followed by modified trypsin digestion ( Promega ) . Peptide mixtures were loaded onto a triphasic 100 μm diameter fused silica microcapillary column described as follows ( McDonald and Yates , 2002 ) . Loaded microcapillary columns were placed in-line with a Quaternary Dionex Ultimate 3000 HPLC pump and a LTQ Velos linear ion trap mass spectrometer equipped with a nano-LC electrospray ionization source ( ThermoFischerScientific ) . A fully automated 12-steps MudPIT run was performed as previously described ( Florens et al . , 2006 ) during which each full MS scan ( from 300 to 1700 m/z range ) was followed by 20 MS/MS events using data-dependent acquisition . Proteins were identified by database searching using SEQUEST ( Eng et al . , 1994 ) within ThermoProteome Discoverer 1 . 3 and 1 . 4 ( ThermoFischerScientific ) . Tandem mass spectra were searched against a Mus musculus protein sequence database containing 16 , 604 entries ( from the Swissprot 2013-04-03 release ) . In all searches , cysteine residues were considered to be fully carboxyamidomethylated ( +57 Da statically added ) and methionine considered to be oxidized ( +16 Da dynamically added ) . Proteins were considered as specific in a given IP data set if they were absent or 10-fold minimum enriched as compared to a MOCK IP , performed on the same protein input by using a non-specific antibody targeting yeast TAF90 . Relative protein abundance for each protein in either the anti-Msl1 , or the anti-Nsl1 IPs was estimated by the calculation of a Normalized Spectral Abundance Factor ( NSAF ) ( Zybailov et al . , 2006 ) . NSAF values were calculated from the spectral counts of each identified protein . To account for the fact that following enzymatic digestion larger proteins result in more peptides/spectras than small proteins , each given spectral count was divided by the corresponding protein length to provide a spectral abundance factor ( SAF ) . To obtain NSAF , SAF values were normalized against the sum of all SAF values in the corresponding run . Thus , NSAF values obtained from a given protein mixture , such as immunoprecipitated protein complexes , allow the comparison of the abundance of a given protein/subunit to another in the same mixture/complex . For gel filtration a Superose 6 ( 10/300 ) column pre-equilibrated in 25 mM Tris pH 7 . 9 , 1 mM DTT , 5 mM MgCl2 , 150 mM KCl , and 5% Glycerol was used . 250 μl calibration mix containing Dextran Blue ( 2 MDa ) and Biorad calibration kit ( ref 151-1901 ) with marker sizes of 670 kDa , 158 kDa , 44 kDa , 17 kDa , and 1 . 35 kDa were injected at 0 . 3 μl/min . 500 μl of mESC nuclear extract containing 1 mg protein was injected and run at 0 . 3 μl/min . 40 fractions were collected and analysed by western blot . For western blot the anti-Mof ( A300-992a; Bethyl ) antibody was used . ChIP was carried out as described previously with slight modifications ( Krebs et al . , 2011 ) . At 80% confluency mESCs were cross-linked with 1% formaldehyde for 10 min at room temperature , lysed and shared mechanically using the Covaris E210 to obtain a chromatin fragment size of 200–500 bp . IP were carried out using 500 μg of chromatin . For the IP 3 μg of purified Msl1 3208 or Nsl1 3130 antibodies were used . The input was obtained from 50 μg of chromatin , pre-cleared , and directly reverse crosslinked . DNA was purified using a Qiaquick ( Qiagen , France ) column . Quantitative real-time PCR ( qPCR ) was performed with SYBR Green ( Roche ) . Primer sequences are summarized in the Supplementary file 1 . 10 ng of precipitated DNA obtained from ChIP was used for Solexa sequencing . To create a genomic library , we followed the instructions of NEXTFlex v12 . 03 ( BIO Scientific ) for Msl1 and the NEBNext protocol ( E6240; Biolabs ) for Nsl1 . Libraries were validated with the Agilent Bioanalyzer . Single reads run sequencing was conducted with the HiSeq 2000 . Image analysis and base calling were done with the Illumina pipeline ( 1 . 8 . 2 ) . The July 2007 Mus musculus genome assembly ( NCBI37/mm9 ) from NCBI was used for the sequence alignment by the software Bowtie ( 0 . 12 . 7 ) ( Langmead , 2010 ) . All analyses were conducted with unique reads . Bed files were used to create read density ( wig ) files by extending reads to 200 bp length and creating 25 bp bins . We further included following sequencing datasets , which were obtained from Gene Expression Omnibus ( www . ncbi . nlm . nih . gov/geo/ ) in our analysis: Input ( GSM798320 ) ( Karmodiya et al . , 2012 ) , RNA Polymerase II ( GSM307623 ) , H3K4me3 ( GSM307618 ) , H3K27me3 ( GSM307619 ) , Ezh2 ( GSM327668 ) ( Mikkelsen et al . , 2007 ) , H4K16ac ( GSM1156617 ) ( Taylor et al . , 2013 ) , and Mof ( GSM915227 ) ( Li et al . , 2012 ) . Fastq files were generated from SRA lite format and aligned to the NCBI37/mm9 assembly using Bowtie ( 0 . 12 . 7 ) ( Langmead , 2010 ) . DHS were obtained from Encode/UW ( GSM1014154 ) . To detect Msl1 and Nsl1 peaks , the algorithm MACS14 ( Zhang et al . , 2008 ) was applied using default parameters with slight modifications . For Msl1 peak detection the p-value cutoff was set to 10−5 , no shifting model was built and the shift size was defined as 200 . The annotation was based on the ENSEMBL 67 database ( mm9 ) . Peaks were annotated to genomic features ( TSS , TTS , CDS Exons , 5'UTR , 3′UTR , Introns and intergenic ) using the software HOMER ( 4 . 2 ) ( Heinz et al . , 2010 ) with default parameters . To calculate the Msl1 , Nsl1 , or Pol II enrichment at a given gene either the peak tag density of the nearest peak to the TSS ( in a region of +2 kb ) , was obtained through MACS14 ( Zhang et al . , 2008 ) or the total tag density around the TSS ( +2 kb ) was taken . Further analysis and graphical representation were conducted using the software R . Density profiles around the TSS and GB were obtained through seqMINER ( Ye et al . , 2011 ) . For the comparison and analysis of genomic features between data sets , the software BEDTools ( 2 . 17 . 0 ) ( Quinlan and Hall , 2010 ) was used . Scatter plots and Pearson correlations with Pearson p-values were obtained by calculating the log2 values of read densities normalized to the control at the given peaks or around ENSEMBL transcription start sites . K-means linear clustering was conducted and represented with seqMINER . Venn diagrams were generated with Biovenn ( Hulsen et al . , 2008 ) . Manteia ( v . 2 ) was used for GO analysis of batch gene entries to understand the biological function ( Tassy and Pourquie , 2013 ) . Only GO levels between 1 and 10 were taken into consideration and compared between groups . To verify the statistical significance of the obtained Msl1- or Nsl1-bound gene groups in Figure 5A , C and Figure 8B , we performed bootstrap statistical analyzes for Figure 5A , C and Figure 8B . In all these analyses , we used the total pool of 26 , 460 ENSEMBL genes . Next out of these pools , we randomly selected the same number of total events ( genes or binding sites ) than those determined non-randomly in the corresponding figures ( i . e . , 10600 in Figure 5A; 282 in Figure 5C and 13 , 505 in Figure 8 ) . This random selection was then compared with the different given interest gene lists ( i . e . 3274 , 2570 , and 2185 for Figure 5A ) and the number of genes ( IDs ) belonging to the non-random experimental group was determined . We repeated this process of random selection and gene list crossings 10 , 000 times and represented the number of IDs and their observed frequencies as histograms ( see corresponding figure supplements ) . For each gene list , we computed an average ( mean ) and a standard deviation ( sd ) of the number of random matches . A z-score is computed as: z = ( mean-expect ) /sd , where 'expect' is the number of expected interest genes . p-values associated to these scores are indicated in the corresponding figure legends . On each histogram we indicated in bold the number of IDs found in the non-random experimental group . The p-value represents the significance of the difference between the randomly found average and the experimental ID numbers . Gene expression levels are based on the ENSEMBL 67 database ( mm9 ) . Raw data of mESCs and NPCs were taken from Gene Expression Omnibus ( GSE34473 ) and processed using the software tools TopHat ( Trapnell et al . , 2009 ) and HTSeq with default parameters . FPKM ( fragments per kilobase of exon per million fragments mapped ) values were calculated with Cufflinks ( Roberts et al . , 2011 ) . Differentially expressed genes ( DE ) in mESCs and NPCs were identified with the bioconductor package DESeq ( 1 . 14 . 0 ) ( Anders and Huber , 2010 ) using default parameters . shRNA approaches were conducted with pLKO . 1 puro shRNA vectors ( Sigma–Aldrich , France ) of the TRC2 library . For Ns1 KD the TRCN0000241466 shRNA clone and for Msl1 KD the TRCN0000241378 shRNA clone was used . Double KD of Msl1 and Nsl1 was conducted with equal amount of the TRCN0000241466 and TRCN0000241378 shRNA clones . For control the shRNA non-target control ( Product No . SHC002 ) was applied . Production of lentiviral particles as well as infection of mESCs was conducted according to the manufacturer's protocol . 3 days after viral transfection of 2 × 106 mESCs selection with puromycin ( 2 μg /ml ) ( InvivoGen ) was started . Experiments were conducted 5 days after viral transfection . KD efficiency was tested at RNA levels through reverse transcriptase ( RT ) -qPCR ( see Supplementary File 1 ) and at protein levels through western blot of whole protein extracts . Moreover , mRNA expression of selected genes was analysed by ( RT ) -qPCR , whereas primer sequences are summarized in Supplementary file 1 . Total RNA , which was used for gene expression profiles and cDNA synthesis , was isolated with TRIzol reagent ( Invitrogen ) and treated with DNAse . cDNA was synthesized with Transcriptor reverse transcriptase ( Roche ) using random hexamers according to the manufacturer's protocol . For normalization of protein amount by WB analyses the anti-Tubulin ( T6557; Sigma-Aldrich ) antibody and ponceau solution ( Sigma-Aldrich ) was applied . To analyse the pluripotency state of KD mESCs the anti-Oct4 ( 611202; BD Labs ) antibody was used . To analyse cell morphology images were taken with the digital inverted EVOS XL core ( Fisher Scientific , France ) microscope using a 10X objective . Histones were prepared from mESCs by lysing cells in 10 mM HEPES , pH 7 . 5 , 1 . 5 mM MgCl2 , 10 mM KCl , 0 . 5 mM DTT , 100 mM Natrium Butyrate , and 0 . 2 M HCl for 30 min on ice , centrifuged and dialysed first against 0 . 1 M acidic acid and then against water . Samples were analysed by western blot for histone modifications using the anti-H4K16ac ( 07-329; Millipore , France ) , anti-H4K5ac ( 51997; Abcam , UK ) , anti-H4K8ac ( 15823; Abcam ) , and anti-H3 ( 1791; Abcam ) antibody . Cell growth analyses was started 6 days after lentiviral infection by plating 1 × 105 mESCs on 0 . 1% gelatine coated per 6-well plates in triplicates . mESCs were counted in triplicates using Neubauer cell counting chambers at indicated time points . mESCs were split every second day to 1 × 105 mESCs/6-well . 7 days after lentiviral infection 5 × 105 mESCs were dissolved in 1 ml PBS ( 0 . 1% NaCitrate and 0 . 1% TritonX 100 ) . Propidium iodide ( 50 μg/ml ) was added and after 4 hr incubation on ice cells were analysed by the FACS calibur . Data were analysed using the CellQuest software . Cell death was examined using the APOPercentage apoptosis assay ( A1000/DC79; Biocolor , France ) following the manufacturer's instructions . As a positive control apoptosis was induced with 10 mM hydrogen peroxide for 8 hr in sh control cells . Absorbance was read at 550 nm and normalized to the blank control ( without cells ) . Experiments were designed with three independent biological replicas . Biotinylated cDNA targets were prepared , starting from 150 ng of total RNA , using the Ambion WT Expression Kit ( Cat 4411974 ) , and the Affymetrix GeneChip WT Terminal Labelling Kit ( Cat 900671 ) according to Affymetrix recommendations . Following fragmentation and end-labeling , 3 μg of cDNAs were hybridized on GeneChip Mouse Gene 2 . 0 ST arrays ( Affymetrix , UK ) for whole-transcript expression profiles . Washed and stained chips were scanned with the GeneChip Scanner 3000 7 G ( Affymetrix ) at a resolution of 0 . 7 μm . Obtained raw data ( . CEL intensity files ) were processed with Affymetrix Expression Console software version 1 . 1 to calculate probe set signal intensities using Robust Multi-array Average ( RMA ) algorithms with default settings . To select the DE genes , we used the fold change rank ordering statistics ( FCROS ) method ( Dembele and Kastner , 2014 ) . In the FCROS method , k pairs of test/control samples are used to compute fold changes ( FC ) . For each pair of test/control samples , obtained FCs for all genes are ranked in increasing order . Ranks that result are associated to genes . Then , the k-ranks of each gene are used to calculate a statistic , and resulting probability ( f-value ) is used to identify the DE genes with an error level of 5% . Msl1 and Nsl1 ChIP-seq data sets as well as gene expression profiles of sh control , sh Msl1 and sh Nsl1 mESCs are deposited at Gene Expression Omnibus ( www . ncbi . nlm . nih . gov/geo/ ) under the accession numbers: GSE53797 and GSE56646 .

Embryonic stem cells are special cells that have the ability to become many different types of cells , such as skin , muscle , or neuronal cells . This process is called differentiation . They can also undergo a process called self-renewal to produce more embryonic stem cells . These processes are controlled by a complex network of enzymes , and the production of these enzymes depends on various genes within the organism being expressed as proteins . The DNA that holds the genetic information inside cells spends most of its time wrapped around proteins called histones: this allows the DNA molecules—which can be up to several metres long in some species—to fit inside the cell nucleus; it also protects the DNA molecules , which are quite fragile , from damage . Enzymes that attach chemical groups called acetyl groups to histones have a central role in controlling the self-renewal and differentiation of embryonic stem cells . Mof is an enzyme that attaches an acetyl group to a specific position in a particular histone . It is a subunit within two larger protein complexes that were originally identified in flies: the male-specific lethal ( MSL ) complex , which is only found in male flies , and the non-specific lethal ( NSL ) complex , which is found in both male and female flies . These complexes have been widely studied in flies , and the role of the Mof enzyme is also reasonably well understood in mammals . However , the roles of the MSL and NSL protein complexes in mammals are not fully understood . Ravens et al . have now used a combination of a technique called ChIP-seq ( which can identify binding sites anywhere in the genome ) and genetic ‘knock down’ experiments to explore the roles of these two complexes in mouse embryonic stem cells and neuronal progenitor cells . There is some overlap between the genes that the complexes act on . However , NSL acts on some genes than MSL does not act on , and vice versa . NSL mostly acts on genes that have ‘housekeeping’ functions and are expressed in many different cell types . MSL binds more to genes that are specific to embryonic stem cells , and acts on genes required for the development of neuronal progenitor cells . This means that NSL regulates the growth of embryonic stem cells , whereas MSL controls their development and differentiation .

2014

Mof-associated complexes have overlapping and unique roles in regulating pluripotency in embryonic stem cells and during differentiation

How phenotypically distinct states in isogenic cell populations appear and stably co-exist remains unresolved . We find that within a mature , clonal yeast colony developing in low glucose , cells arrange into metabolically disparate cell groups . Using this system , we model and experimentally identify metabolic constraints sufficient to drive such self-assembly . Beginning in a uniformly gluconeogenic state , cells exhibiting a contrary , high pentose phosphate pathway activity state , spontaneously appear and proliferate , in a spatially constrained manner . Gluconeogenic cells in the colony produce and provide a resource , which we identify as trehalose . Above threshold concentrations of external trehalose , cells switch to the new metabolic state and proliferate . A self-organized system establishes , where cells in this new state are sustained by trehalose consumption , which thereby restrains other cells in the trehalose producing , gluconeogenic state . Our work suggests simple physico-chemical principles that determine how isogenic cells spontaneously self-organize into structured assemblies in complimentary , specialized states . During the course of development , groups of isogenic cells often form spatially organized , interdependent communities . The emergence of such phenotypically heterogeneous , spatially constrained sub-populations of cells is considered a requisite first step towards multicellularity . Here , clonal cells proliferate and differentiate into phenotypically distinct cells that stably coexist , and organize spatially with distinct patterns and shapes ( Newman , 2016; Niklas , 2014 ) . Through such collective behavior , groups of cells can maintain orientation , stay together , and specialize in different tasks through the division of labor , while remaining organized with intricate spatial arrangements ( Ackermann , 2015; Newman , 2016 ) . In both eukaryotic and prokaryotic microbes , such organization into structured , isogenic but phenotypically heterogeneous communities , is widely prevalent , and also reversible ( Ackermann , 2015 ) . Such phenotypic heterogeneity within groups of clonal cells enables several microbes to persist in fluctuating environments , thereby providing an adaptive benefit for the cell community ( Wolf et al . , 2005; Thattai and van Oudenaarden , 2004 ) . A well studied example of spatially organized , phenotypically heterogeneous groups of cells comes from the Dictyostelid social amoeba , which upon starvation transition from individual protists to collective cellular aggregates that go on to form slime-molds , or fruiting bodies ( Bonner , 1949; Du et al . , 2015; Kaiser , 1986 ) . Indeed , most microbes show some such complex , heterogeneous cell behavior , for example in the extensive spatial organization within clonal bacterial biofilms and swarms ( Kearns et al . , 2004; Kolter , 2007 ) , or in the individuality exhibited in Escherichia coli populations ( Spudich and Koshland , 1976 ) . Despite its popular perception as a unicellular microbe , natural isolates of the budding yeast , Saccharomyces cerevisiae , also form phenotypically heterogeneous , multicellular communities ( Cáp et al . , 2012; Koschwanez et al . , 2011; Palková and Váchová , 2016; Ratcliff et al . , 2012; Váchová and Palková , 2018; Veelders et al . , 2010; Wloch-Salamon et al . , 2017 ) . However , despite striking descriptions on the nature and development of phenotypically heterogeneous states within groups of cells , the rules governing the emergence and maintenance of new phenotypic states within isogenic cell populations remain unclear . Current studies emphasize genetic and epigenetic changes that are required to maintain phenotypic heterogeneity within a cell population ( Ackermann , 2015; Sneppen et al . , 2015 ) . In particular , many studies emphasize stochastic gene expression changes that can drive phenotypic heterogeneity ( Süel et al . , 2007; Ackermann , 2015; Balázsi et al . , 2011; Blake et al . , 2003 ) . Further , groups of cells can produce adhesion molecules to bring themselves together ( Halfmann et al . , 2012; Halme et al . , 2004; Octavio et al . , 2009; Váchová and Palková , 2018 ) , or support possible co-dependencies ( such as commensal or mutual dependencies on shared resources ) within the populations ( Ackermann , 2015 ) . Such studies now provide insight into why such heterogeneous cell groups might exist , and what the evolutionary benefits might be . However , an underlying biochemical logic to explain how distinct , specialized cell states can emerge and persist in the first place is largely absent . This is particularly so for isogenic ( and therefore putatively identical ) groups of cells in seemingly uniform environments . In essence , are there simple chemical or physical constraints , derived from existing biochemical rules and limitations , that explain the emergence and maintenance of heterogeneous phenotypic states of groups of clonal cells in space and over time ? Contrastingly , a common theme occurs in nearly all described examples of phenotypically heterogeneous , isogenic groups of cells . This is a requirement of some ‘metabolic stress’ or nutrient limitation that is necessary for the emergence of phenotypic heterogeneity and spatial organization , typically in the form of metabolically inter-dependent cells ( Ackermann , 2015; Campbell et al . , 2016; Cáp et al . , 2012; Johnson et al . , 2012; Liu et al . , 2015 ) . This idea has been explored experimentally , where approaches that systems-engineer metabolic dependencies between non-isogenic cells can result in interdependent populations that constitute mixed communities ( Campbell et al . , 2016; Campbell et al . , 2015; Embree et al . , 2015; Wintermute and Silver , 2010 ) . These findings suggest that biochemical constraints derived from metabolism may determine the nature of phenotypic heterogeneity , and the spatial organization of cells in distinct states within the population . Therefore , if we can understand what these biochemical constraints are , and discern how metabolic states can be altered through these constraints , this may address how genetically identical cells can self-organize into distinct states . In this study , using clonal yeast cells , we experimentally and theoretically show how metabolic constraints imposed on a population of isogenic cells can determine the production , accumulation , and utilization of a specific , shared resource . The selective utilization of this resource enables the spontaneous emergence and persistence of cells exhibiting a counter-intuitive metabolic state , with spatial organization . These metabolic constraints create inherent threshold effects , enabling some cells to switch to new metabolic states , while restraining other cells to the original state which produces the resource . This thereby drives the overall self-organization of cells into specialized , spatially ordered communities . Finally , this group of spatially organized , metabolically distinct cells confer a collective growth advantage to the community of cells , rationalizing why such spatial self-organization of cells into distinct metabolic states benefits the cell community . Using a well-studied S . cerevisiae isolate as a model ( Reynolds and Fink , 2001 ) , we established a simple system to study the formation of a clonal colony with irregular morphology . On 2% agar plates containing a complex rich medium with low glucose concentrations , S . cerevisiae forms rugose colonies with distinct architecture , after ~5–6 days ( Figure 1A ) . Such colonies do not form in the typical , high ( 1–2% ) glucose medium used for yeast growth ( Figure 1A ) . Thus , as previously well established ( Granek and Magwene , 2010; Reynolds and Fink , 2001 ) , glucose limitation ( with other nutrients such as amino acids being non-limiting ) drives this complex colony architecture formation . Currently , the description of such colonies is limited to this external rugose morphology , and does not describe the phenotypic states of cells and/or any spatial organization in the colony . With only such a description , as observed in Figure 1A , the mature colony surface has an internal circle and some radial streaks near the periphery . We carried out a more detailed observation of entire colonies under a microscopic bright-field ( using a 4x lens ) . Here , we unexpectedly noticed what appeared to be distinct internal patterning , and apparent spatial organization of cells within the colony ( Figure 1B ) . As categorized purely based on these observed differences in visual optical density ( ‘dark’ or ‘light’ ) , regions between the colony center and periphery had optically dense ( dark ) networks spanning the circumference of the colony , interspersed with optically rare regions . In contrast , the periphery of the mature colony appeared entirely light ( Figure 1B ) . Based simply on these optical traits alone , we categorized cells present in these regions of the colony as dark cells and light cells ( Figure 1B ) . At this point , our description is visual and qualitative , and does not imply any other difference in the cells in either region . However , this visual description is both robust and simple , and hence we use this nomenclature for the remainder of this manuscript . Since these structured colonies form only in glucose-limited conditions , we hypothesized that dissecting the expected metabolic requirements during glucose limitation might reveal drivers of this internal organization within the colony . The expected metabolic requirements of cells growing in glucose limited conditions are as follows: first , all cells would be expected to have constitutively high expression of the high-affinity hexokinase ( Hxk1p ) ( Lobo and Maitra , 1977; Rodríguez et al . , 2001 ) . Further , during glucose-limited growth , all cells are expected to carry out extensive gluconeogenesis , as the default metabolic state ( Broach , 2012; Haarasilta and Oura , 1975; Yin et al . , 2000 ) . Indeed , we confirmed this second expectation by measuring the amounts of the gluconeogenic enzymes Pck1 ( phosphoenolpyruvate carboxykinase ) and Fbp1 ( fructose-1 , 6-bisphosphatase ) , in short-term ( 4–5 hr ) liquid cultures of log-phase cells growing in either high glucose medium , or in the same glucose-limited medium we used for colony growth . Expectedly , we observed very high amounts of these gluconeogenic enzymes in cells growing in glucose-limited medium ( Figure 1—figure supplement 1A ) , reiterating that even in well-mixed glucose-limited , cells are in a strongly gluconeogenic state . In order to now examine the mature colony and dissecting expected metabolic requirements in these conditions , we first designed visual indicators for these metabolic hallmarks of yeast cell growth in low glucose . We engineered two different fluorescent reporters , one dependent on HXK1 expression ( mCherry under the HXK1 gene promoter ) , and the second on PCK1 expression as an indicator of gluconeogenic activity ( mCherry under the PCK1 gene promoter ) ( Figure 1—figure supplement 1B ) . Cells carrying these reporters were seeded to develop into colonies , and the expression levels of these reporters were monitored in the mature , rugose colony ( 5–6 days ) . Expectedly , the HXK1-promoter dependent reporter showed constitutive , high expression in all cells across the entire colony ( Figure 1C ) . Contrastingly , only the dark cells exhibited high gluconeogenesis reporter activity ( Figure 1C ) . Notably , the light cells entirely lacked detectable gluconeogenic reporter activity ( Figure 1C ) . To better quantify this phenomenon , cells were dissected out from dark or light regions respectively ( under the light microscope , using a fine needle ) , and the percentage of fluorescent cells in each region was measured using flow cytometry . Based on flow cytometric readouts , ~80% of the isolated dark cells showed strong fluorescence for the gluconeogenic reporter , while ~97% of the light cells were non-fluorescent for gluconeogenic activity ( Figure 1C , Figure 1—figure supplement 1C ) . This spatial distribution of gluconeogenic activity is shown as a quantitative heat-map histogram overlaid on the entire colony , in Figure 1—figure supplement 2A . Since this observation was based solely on reporter activity , in order to more directly examine this observation , we estimated native protein amounts of enzymes associated with gluconeogenesis ( Pck1 , Fbp1 , and Icl1- Isocitrate lyase from the glyoxylate shunt ) in isolated light cells and dark cells . Only the dark cells showed expression of the gluconeogenic enzymes ( Figure 1D , Figure 1—figure supplement 2B ) . Finally , we measured steady-state amounts of trehalose and glycogen within dark and light cells , using these metabolites as unambiguous biochemical readouts of the end-point biochemical outputs of gluconeogenesis ( François et al . , 1991 ) . We observed that the dark cells had substantially higher amounts of both trehalose and glycogen ( Figure 1E ) , indicating greater gluconeogenic activity in these cells . Collectively , these results strikingly reveal that intracellular gluconeogenic activity is spatially restricted to specific regions , resulting in a distinct pattern of metabolically specialized zones within the colony . In the given nutrient conditions of low glucose , gluconeogenesis is an expected , constitutive metabolic process , essential for cells . This can therefore be considered as a necessary , permitted metabolic state in this condition . Paradoxically , in these mature colonies , gluconeogenic activity was spatially restricted to only within the dark cell region , with no discernible gluconeogenic activity in the cells located in the light region . This absence of gluconeogenic activity in these light cells , while concomitant with a constitutively high level of hexokinase activity , therefore poses a biochemical paradox . What might the metabolic state of these light cells be ? To quickly address this using a crude but useful readout , we compared the ability of freshly isolated light and dark cells to proliferate in both gluconeogenic ( low glucose ) , and non-gluconeogenic ( high glucose ) growth conditions . For simplicity , isolated light cells and dark cells were inoculated either into a medium where gluconeogenesis is essential ( 2% ethanol +glycerol as a sole carbon source ) , or in high ( 2% ) glucose medium where cells rely on high glycolytic and pentose phosphate pathway ( PPP ) activity , and initial cell proliferation was monitored . Here , cells that had been growing in high glucose were used as a control . Expectedly , the dark cells grew robustly and reached significantly higher cell numbers ( 0D600 ) compared to the light cells in the gluconeogenic condition ( Figure 2A ) . Conversely , light cells grew robustly when transferred to the high glucose medium , as compared to the dark cells ( Figure 2A ) . While this was an overly simple , and not definitive experiment , counter-intuitively , this result suggested that despite being in a low-glucose environment , the light cells were well suited for growth in high glucose , and therefore might be in a metabolic state suited for growth in glucose . We therefore decided to more systematically investigate this phenomenon . In the presence of glucose , yeast cells typically show high glycolytic and PPP activities , as part of the Crabtree ( analogous to the Warburg ) effect ( Crabtree , 1929; De Deken , 1966; Figure 2B ) . Therefore , if the light cells in the colony were indeed behaving as though present in more glucose-replete conditions , they should exhibit high PPP activity . To test this , we first designed a fluorescent PPP-activity reporter ( mCherry under the control of the transketolase 1 ( TKL1 ) ( Walfridsson et al . , 1995 ) gene promoter , Figure 1—figure supplement 1B ) , and monitored reporter activity across the mature colony . Indeed , only the light cells exhibited high PPP-reporter activity ( Figure 2C , Figure 2—figure supplement 1A ) . This spatial restriction of high PPP activity across the colony is also shown as an overlaid quantitative heat-map histogram in Figure 2—figure supplement 1A . Next , we directly addressed the possibility of these light cells exhibiting relatively high PPP activity . For unambiguously testing this , we utilized a stable-isotope based metabolic flux experiment to assess the flux towards PPP in light and dark cells . Light and dark cells isolated from colonies were pulsed with 13C-labeled glucose ( for ~5 min ) , metabolites extracted , and the incorporation of this carbon label into the late PPP intermediates ribulose-5-phosphate ( R-5-P ) and sedoheptulose-7-phosphate ( S-7-P ) was measured by liquid chromatography/mass spectrometry ( LC/MS/MS ) . The relative amounts of these all-carbon labeled PPP intermediates were compared between the two cell types ( light or dark ) . Notably , light cells incorporated significantly higher levels of 13C labeled glucose into PPP metabolites compared to the dark cells ( Figure 2D , and Figure 2—figure supplement 1B , and see Table 1 for MS parameters ) , showing that the light cells are in a high PPP activity state . Finally , we assessed if other biochemical end-point outputs requiring high PPP activity/flux were also high in the light cells . High nucleotide synthesis is a canonical consequence of enhanced PPP activity ( Nelson and Cox , 2013 ) . The carbon backbone ( ribose-5-phosphate ) of newly synthesized nucleotides is derived from the PPP , while the nitrogen backbone comes from amino acids ( Nelson and Cox , 2013; Figure 2B , and see Table 1 for MS parameters ) . We devised another metabolic flux-based experiment to assess de novo nucleotide biosynthesis in light and dark cells , as an end-point collective readout of high PPP activity coupled with amino acid utilization . Light and dark cells , isolated from colonies were pulsed with a 15N-label ( ammonium sulfate +aspartate ) , and incorporation of this label into nucleotides was measured by liquid chromatography/mass spectrometry ( LC/MS/MS ) . Light cells had higher flux into nucleotide biosynthesis , compared to the dark cells ( Figure 2E , and see Table 1 for MS parameters ) . Taken together , we surprisingly find that light cells exhibit multiple metabolic hallmarks of cells growing in glucose-replete conditions , including increased PPP activity , and increased nucleotide biosynthesis . Thus , in the spatially organized colony , the light cells and dark cells have contrary metabolic states . This is despite the expectation that the gluconeogenic state , exhibited by the dark cells , is the plausible metabolic state in the given growth conditions . Notably , the light cells or dark cells , when isolated and reseeded as a new colony , both develop into indistinguishable , complex colonies ( Figure 2F ) . This reiterates that these phenotypic differences between the light and dark cells are fully reversible , and do not require genetic changes . Collectively , these data reveal that cells within the colony organize into spatially separated , metabolically specialized regions . Within these regions , cells exhibit complimentary metabolic states . One of these states , where cells have high PPP activity , is counter-intuitive and cannot obviously be sustained given the external nutrient environment . What determines the emergence and spatial organization of a group of cells , in these contrary metabolic states ? Particularly , what can explain the emergence and proliferation of the light cells , which exhibit this counter-intuitive metabolic state , while the colony maintains a large subset of cells in the dark state ? To address this , we built a coarse-grained mathematical model . This model incorporates simple processes derived from our current experimental data , to simulate the formation of a colony of ‘light’ and ‘dark’ cells . The model was intentionally coarse-grained , since its purpose was only to find a minimal , biologically consistent combination of processes that is sufficient to produce the overall spatial structure and composition of cell states observed in the colonies . The intention behind the model was not to decipher all possible molecular details that explain this phenomenon . The model should only sufficiently account for both the emergence of light cells , as well as their spatial organization with dark cells . Such a model could therefore suggest constraints that determine the emergence of light cells , and the organization of the colony with the observed organization , which can then be experimentally tested . While building this model , we included a range of processes that must be considered , based on our experimental data thus far ( Figure 3A ) . This includes ( i ) the dark cells switching to a light state , ( ii ) the production of some resource by dark cells , which may be shared/utilized by the cells , ( iii ) diffusion parameters for this resource , ( iv ) consumption of this resource , and ( v ) rates of cell division are included ( Figure 3A ) . Next , we constructed a two-dimensional square grid of ‘locations’ for groups of cells within the colony ( Figure 3B ) . Here , each location is either empty or occupied by a group of ~100 cells ( also see Materials and methods for full details ) . Note: we intentionally coarse-grain the grid ( for computational simplicity , in order to simulate colony sizes comparable to real colonies ) by approximating that the locations either consist of all light or all dark cells . This is a major simplification that was necessary . At each time step ( 12 min of real time ) , all the processes shown in Figure 3A are executed across the spatial grid using the outlined algorithm ( Figure 3C ) . In such an implemented algorithm , ( i ) all cells consume all available nutrients ( present in saturating amounts ) , while free glucose concentrations are negligible , ( ii ) dark cells grow and divide in the given conditions , ( iii ) dark cells produce a resource/resources as a consequence of their existing metabolic ( gluconeogenic ) state , ( iv ) this resource diffuses around the grid and is freely available , ( v ) dark cells switch to the light state if sufficient resource is present at their location , and lastly , ( vi ) the resource when consumed can sustain the light state cells , which can expand if there is an empty location in the neighborhood . If the resource is not present in that location the light cells will switch back to dark cells . All processes occur at specified rates , allowing for stochasticity . Finally , this existence of a shared resource is surmised because , logically the emergence of light cells from dark can happen only if the local nutrient environment enables a switch to the new metabolic state . In each simulation , empty grids are seeded with 1257 occupied locations , with 95–99% of the cells in the dark state . After ~750 time steps ( corresponding to ~6 days ) a simulated wild-type colony looks typically as shown in Figure 3D ( also see Video 1 ) . A range of resource amounts , growth rates and diffusion of the resource were included in control simulations ( see Figure 3—figure supplement 1A–C ) . Strikingly , the simulated spatial organization ( Figure 3D , and Video 1 ) recapitulates most obvious features of a real colony ( Figure 3D ) . These are: at the edge of the initial circular inoculation of the colony is a ring of dark cells , the outermost part of the colony is made up of outcrops of light cells , and from this ring of dark cells emanate clusters of dark cells penetrating into the outcrops of light cells . This is despite the simplicity of the rules in the model , including its flattening into 2D . In the simulation , for the first 40–45 time steps , the colony remains small and predominantly dark , while the resource builds up . Then , dark cells start to switch to light . When this happens within the bulk of the colony , these light cells have restricted division due to spatial constraints . Around 100 to 150 time steps later , light cells emerge at the perimeter of the colony , and then rapidly divide and expand ( Figure 3D , Video 1 ) . In order to test if the processes of Figure 3A are all required for this behavior , we examined three comprehensive control scenarios: ( i ) dark cells do not produce a resource ( and therefore in this case for growth light cells depend only on amino acids or other pre-supplied resources in the medium ) , ( ii ) dark cells cannot switch to the light state , or ( iii ) light cells produce a resource that is needed by dark cells to grow ( a straw-man scenario , since initially in the actual colony all cells were in a dark state , as shown earlier ) . None of these cases produces the wild-type spatial organization , over a wide range of parameter values ( Figure 3E , as well as a range of parameters explored in Figure 3—figure supplement 2A–D , and simulations in Videos 2–4 ) . Summarizing , this simple model successfully recaptures the general features of the spatial patterning and organization of real colonies . This includes the overall general architecture , and spatial organization of light and dark cells . Two simple take-home points emerge from this model , for such spatial distribution of cells in these two metabolic states , across the developing colony . First , the model requires that dark ( gluconeogenic ) cells will produce a resource that is needed by dark cells to switch to the light state . Second , a resource produced by the dark cells is required to sustain the light state . Collectively , in our model , these metabolic constraints are sufficient to determine the overall spatial organization of metabolically distinct , specialized cells . Does any gluconeogenic metabolite ( s ) determine the organization of these cells , consistent with these requirements suggested by experimental and modeled data ? Such a metabolite must logically satisfy the following three criteria . First , this resource should be available in the extracellular environment ( i . e . released by cells ) , second , cells must selectively be able to take up this resource , and third , the resource should be metabolized within cells to produce glucose/a glucose-like product capable of fueling a glycolytic and PPP-active state . Further , if this were indeed a ‘controlling resource’ that determined the emergence of light cells , preventing the uptake and utilization of this resource should prevent the emergence and proliferation of only the light cells , but leave the dark cells unaffected . In order to identify such a candidate metabolite , we considered all possible outputs of gluconeogenesis: the storage carbohydrates/sugars glycogen and trehalose , the polysaccharides of the cell wall ( chitin , mannans , glycans ) , and glycoproteins ( Figure 4A ) ( Jules et al . , 2008; Kayikci and Nielsen , 2015 ) . The large molecular size of glycogen , chitins , and complex glycosylated proteins , the lack of known cellular machinery for their uptake , and the difficulty in efficiently breaking them down make them all unlikely candidates to be the resource controlling the emergence of light cells . Contrastingly , trehalose has unique properties making it a plausible candidate . It is a small , non-reducing disaccharide composed of two glucose molecules . Trehalose has been observed in the extracellular environment in yeast ( Parrou et al . , 2005 ) , and yeast can uptake trehalose through disaccharide transporters ( Jules et al . , 2008; Stambuk , 1998 ) . Further , trehalose can be rapidly and specifically hydrolyzed to two glucose molecules , which can fuel glycolysis and re-entry into the cell division cycle ( Laporte et al . , 2011; Shi et al . , 2010; Shi and Tu , 2013 ) . These diverse data therefore presented trehalose as an excellent putative candidate metabolite that controlled the emergence of cells in the light state . To test this possibility , we first measured extracellular trehalose in colonies . Free trehalose was readily detectable in the extracellular environment of these colonies ( Figure 4B ) . To test if trehalose could be differentially transported into either light or dark cells , we first estimated amounts of a primary trehalose transporter , Mal11 ( Stambuk , 1998 ) in these cells . Mal11 protein amounts were substantially higher in the light cells compared to the dark cells ( Figure 4C , and quantified in Figure 4—figure supplement 1A ) . To unambiguously , directly estimate trehalose uptake , we isolated light and dark cells from a mature colony , and exogenously added 13C-trehalose . We then measured intracellular levels of labeled trehalose present in these cells , by extracting and estimating metabolite amounts ( by LC/MS/MS ) ( see Table 1 for MS parameters ) . Notably , the light cells rapidly accumulated 13C-trehalose ( Figure 4D ) , while the dark cells did not , suggesting robust , preferential uptake of extracellular trehalose . Finally , we estimated the ability of light and dark cells to break-down and utilize trehalose . For this , we first measured the expression of the predominant neutral trehalase in yeast ( Nth1 ) ( Jules et al . , 2008 ) , in the light and dark cells . Light cells had substantially higher Nth1 amounts than the dark cells ( Figure 4E , and quantified data shown in Figure 4—figure supplement 1B ) . We also measured enzymatic activity for Nth1 ( in vitro , using cell lysates ) , and found that the light cells had ~2 fold higher in vitro enzymatic activity , compared to the dark cells ( Figure 4F ) . Collectively , these data suggested that the light cells were uniquely able to preferentially take up more trehalose , break it down to glucose , in order to potentially utilize it to sustain a metabolic state with high PPP activity . Since these data suggested that trehalose uptake and utilization would be preferentially high in the light cells , we directly tested this using a quantitative metabolic flux based approach . For this we used stable-isotope labeled trehalose , and measured trehalose uptake , breakdown and utilization . To the isolated light and dark cells , 13C-labeled trehalose was externally provided , and intracellular metabolites extracted from the respective cells . The intracellular amounts of 13C -labeled glycolytic and PPP intermediates were subsequently measured using LC/MS/MS ( Figure 5A and Figure 5—figure supplement 1A , also see Table 1 for MS parameters ) . 13C –labeled glucose-6-phosphate ( which enters both glycolysis and the PPP ) , the glycolytic intermediates glyceraldehyde-3-phosphate and 3-phosphoglycerate , and the PPP intermediates 6-phosphogluconate , ribulose-5-phosphate and sedoheptulose-7-phosphate all rapidly accumulated exclusively in the light cells ( Figure 5A , Figure 5—figure supplement 1A ) . Since the labeled carbon can come only from trehalose , these data indicate both the breakdown of trehalose to glucose , as well as the subsequent utilization of glucose for these pathways . Indeed , the labeled forms of these metabolites were only above the detection limit in dark cells ( Figure 5A and Figure 5—figure supplement 1A ) . Thus , these data demonstrate that external trehalose is preferentially taken up only by the light cells , and utilized to fuel the complimentary metabolic state of the light cells , with high glycolytic and PPP activity . Finally , we tested if the sharing and differential utilization of trehalose determined both the emergence and the proliferation of light cells . An explicit prediction is made both in our model , and our hypothesis based on these experimental data . This is: preventing uptake and/or utilization of trehalose should prevent cells from switching to the light state . To test this prediction , we generated strains lacking NTH1 ( which cannot break-down trehalose to glucose ) , and MAL11 ( which will have reduced trehalose uptake ) , allowed them to develop into mature colonies , and compared the amounts of light cells in each . Compared to wild-type colonies , cells lacking the major trehalose uptake transporter ( Δmal11 ) formed colonies with very few light cells ( Figure 5B ) . Note: while Mal11 shows a high affinity for trehalose , S . cerevisiae has multiple sugar transporters with reduced affinity for any disaccharide . Therefore , cells lacking MAL11 may take up trehalose with lower efficiency . In these cells , the ability to break-down trehalose remains intact . More importantly , in colonies of cells lacking trehalase ( Δnth1 ) , and which therefore cannot efficiently breakdown internal trehalose , had nearly no detectable light cells , based on brightfield microscope observations ( Figure 5B ) . This result was more quantitatively estimated in colonies of cells with these respective genetic backgrounds , using the expression of the fluorescent PPP reporter . Again , almost no PPP reporter activity was observed in the Δnth1 cell colonies , while very few cells with PPP reporter activity were seen in Δmal11 colonies ( Figure 5C ) . As controls , we ensured that there were no defects in the expression of the PPP reporters in cells from these genetic backgrounds . Correspondingly , we also quantified the percentage of dark , highly gluconeogenic cells ( as determined using the gluconeogenesis reporter ) , in colonies from each of these genetic backgrounds . The percentage of gluconeogenic cells was proportionately higher in the Δmal11 ( ~73% ) , and Δnth1 ( ~80% ) colonies compared to the wild-type colony ( ~65% ) ( Figure 5D ) . Thus , controlling the uptake and utilization of the resource ( trehalose ) directly regulates the emergence of cells in the light state . Collectively , these data demonstrate that trehalose is the shared gluconeogenic resource that determines the emergence , and persistence , of light cells within the structured colony . Our experimental data showing the organization of dark and light cells was obtained from ~5–6 day old , mature colonies . However , in our simulations of the temporal development of the colony , we observed that the dense network of dark ( gluconeogenic ) cells form first , followed by a very late appearance of light cells ( Figure 6A and Video 1 ) . This late appearance of light cells in the simulations comes from an inherent threshold effect included within the model . Here , the external build-up of the shared resource made by the dark cells is required . At a sufficient built-up concentration , this resource will trigger the switching of some cells to light cells . Light cells in turn will consume the resource , reducing the available amounts , thereby preventing other cells from switching to this new state . This threshold-effect therefore predicts a delayed , rapid emergence of light cells , and also enables such a pattern of distinct cell groups to form . If this threshold requirement is removed in the simulation ( for example when replaced by a rate of switching from dark to light that depends linearly on the amount of resource ) , the resultant colony remains small , and the organized pattern of cells in two states does not occur . This is shown in Figure 6A , and Video 5 . This small colony size is largely due to low resource amounts to support the proliferation of the light cells , since there are insufficient dark cells remaining to produce the resource . This is also clearly seen in control simulations with a range of resource amounts , and linear switching , as shown in Figure 6—figure supplement 1A–1D . Contrastingly , in the model that successfully simulates the colony development , the externally available amount of the resource builds-up , reaches the threshold ( where cells switch to the light state ) , and then rapidly decreases , if the light cells also consume the resource ( Figure 6B , upper panel ) . This therefore prompted us to more closely examine the development of actual colonies over time , for these properties . We first estimated the relative amounts of extracellular , free trehalose in the colony over time . Notably , trehalose amounts steadily increased over 4 days , and subsequently rapidly decreased ( Figure 6B lower panel ) . This rapid decrease in trehalose after day four is despite a steady , continuing increase in the total number of cells in the colony ( Figure 6—figure supplement 1E ) . We next monitored the development of colonies over time , to determine when the light cells emerge . Using just the bright-field image reconstruction of the colonies , during this time course , the intensity of dark cells steadily increased , and organized into the mesh-like network over 4 days ( Figure 6C ) . However , the light cells appeared only after ~4 days , and rapidly increased in number ( Figure 6C ) . We more quantitatively estimated this , using strains expressing the gluconeogenic- or the PPP-reporter ( Figure 6D ) . Notably , the increase in total fluorescence intensity due to the gluconeogenic-reporter in the colony ( over time ) was relatively linear over the first four days ( r2 = 0 . 99 ) , increasing with the steady increase in the number of cells ( Figure 6—figure supplement 1E ) . Contrastingly , the increase in the PPP reporter activity over the first five days was clearly non-linear and switch like , with very low signal intensity for the first three days , and then a rapid emergence of signal over days four and five ( Figure 6E ) . This indicated a cooperative , switch like emergence of , and increase in these light cells . A useful biophysical measure of cooperativity ( more commonly used for protein-ligand binding characteristics ) is the Hill coefficient . We adopted the Hill equation , using the amount of PPP reporter fluorescence ( instead of ligand-receptor binding ) , to estimate cooperativity in the system . Over the first five days the increase in PPP-reporter activity showed a Hill coefficient greater than 1 , indicating a positively cooperative switch of cells to the light state ( Figure 6E ) . This nicely correlates with the build-up , and rapid decrease in external trehalose around day 4 ( Figure 6B ) . These data also show that the peripheral location of the light cells cannot simply be due to possible greater access to glucose in the medium , since for the first ~4 days , there are no cells in the periphery with high PPP reporter activity . Their emergence is indeed rapid , and switch-like . In summary , data from model simulations and experiments show that initially the gluconeogenic cells increase in number , leading to release and build-up of the resource ( trehalose ) in the local environment . At this time there are no light cells in the colony . Above a threshold concentration of trehalose , some cells rapidly switch to light state with high PPP activity . The further expansion of these light cells correlates with rapid consumption of the extracellular trehalose that sustains this state . These data suggest a threshold effect , where the controlling resource , trehalose , needs to build up above a certain amount , in order for cells to switch to the contrary , high PPP activity state . Finally , we wondered if such an emergence of light cells with high PPP activity might benefit the community of cells as a whole . To address this , we compared the long term colony expansion of wild-type cells , with colonies comprised of cells lacking the neutral trehalase ( Δnth1 ) . Cells in the Δnth1 colonies cannot utilize trehalose to produce glucose , and as shown earlier , will remain in a gluconeogenic state . Therefore , in these Δnth1 cell colonies light cells will be absent . However , these cells are still capable of normal gluconeogenesis ( and trehalose production ) . Strikingly , we observed that as the respective colonies expanded over time ( ~21 days ) , the wild-type colonies spread over a significantly greater area on the plate , while the Δnth1 colonies were unable to expand as efficiently ( Figure 7A and Figure 7B ) . This shows that the emergence and proliferation of light cells are important for the expansion of the colony . Since the dark cells are required for the emergence and existence of the light cells , collectively , these data suggest how the community uses cells in distinct metabolic states to maximize growth and spatial expansion , possibly to forage for new nutrients . Collectively , we present a simple model proposing how cells in metabolically distinct states spontaneously emerge and spatially self-organize within a yeast colony , as summarized in Figure 7C . In low glucose conditions , cells begin in a uniform gluconeogenic state , which is the expected metabolic state in this nutrient condition . The gluconeogenic cells produce a resource ( trehalose ) , that is now externally available . This resource builds up to above a threshold amount . At this threshold , some cells take up and consume trehalose , breaking it down to glucose . These cells spontaneously switch to the complimentary metabolic state , with high PPP and glycolytic activity ( i . e . the light state ) ( Figure 7C ) . These light cells can remain in this metabolic state only so as long as the resource ( trehalose ) is externally available . However , as trehalose is consumed by these cells , the available amount of external trehalose itself drops below the threshold . The surrounding dark cells therefore remain trapped in a gluconeogenic state , continuing to produce trehalose . Thereby , a predictable fraction of cells , constrained spatially , will remain in each metabolic state , resulting in specialized cell groups and division of metabolic labor . Thus , biochemically heterogeneous cell states can spontaneously emerge and spatially self-organize . An implicit concept emerging from this study is that of threshold amounts of a controlling or sentinel metabolite that regulates a switch to a new metabolic state . By definition , such a metabolite must be produced by cells present in a certain ( original ) metabolic state . But when this metabolite is utilized , it must have the ability to switch cells to an entirely distinct metabolic state . Further , the emergence and expansion of cells in the new state will be rapid and switch-like , resembling a bistable system ( Pomerening , 2008 ) . This idea of metabolites controlling cell states is an emerging area of interest ( Cai and Tu , 2011; Krishna and Laxman , 2018 ) , but has not been studied in the context of groups of cells organizing into distinct groups or metabolic states ( and therefore different phenotypic properties ) . We speculate what the advantages of such spatially organized , phenotypically distinct states within a group of clonal cells might be , in this example . Here , we observe that the organized community with cells in distinct states has clear advantages , in being able to spatially expand better ( Figure 7C ) . For sessile microbes such as yeast , this ability to forage for better nutrients is important for their survival . This might also convey other advantages , and uncovering those are obvious areas of future studies . Since the inherent properties of the cells in the distinct states are different , this raises the deeper possibility that these advantages come from physical and chemical properties of the cells , which arise from their distinct metabolic states . Regardless , our study substantially advances descriptions of yeast ‘multicellularity’ from simple dimorphism , aggregated cells , or three-dimensional colony forms ( Cáp et al . , 2012; Koschwanez et al . , 2011; Palková and Váchová , 2016; Ratcliff et al . , 2012; Váchová and Palková , 2018; Veelders et al . , 2010; Wloch-Salamon et al . , 2017 ) , to self-organized , phenotypically heterogeneous cell states exhibiting division of labor and metabolic interdependence . Strikingly , the nature of spatial patterning allied with division of labor that we observe in yeast is reminiscent of true multicellular systems ( Newman , 2016; Niklas , 2014 ) . Also , the cell states in these yeast colonies can be considered commensal , since trehalose is a necessary output of gluconeogenesis , and therefore a default , biochemically non-limiting output in dark cells . Since trehalose controls the emergence and maintenance of light cells in the complimentary metabolic state , it thus can be considered a resource benefiting the light state . Thus simple , metabolism-derived constraints are sufficient to determine how contrary biochemical states can spontaneously emerge and be supported , in conjunction with spatial structure . Such organization of cells into specialized , labor-divided communities expands on the role of reaction-diffusion systems ( particularly activator-depleted substrate schemes ) in controlling cellular patterning ( Gierer and Meinhardt , 1972; Kondo and Miura , 2010; Newman , 2016 ) , with a metabolic resource threshold being central to the emergence and stabilization of a new phenotype ( Cai and Tu , 2011; Krishna and Laxman , 2018 ) . A deeper dissection of what such constraints can permit will therefore advance our general understanding of how specialized cell states can emerge and be stabilized . Metabolic cross-feeding is best understood currently in multi-species microbial communities , where this has been inferred largely using inter-species genomic comparisons ( Ackermann , 2015; D'Souza et al . , 2018; Goldford et al . , 2018; Tyson et al . , 2004 ) . Further , metabolic sharing has typically been demonstrated using synthetically engineered systems where mutual dependencies are created ( Campbell et al . , 2016; D'Souza et al . , 2018; Mee et al . , 2014; Pande et al . , 2015; Wintermute and Silver , 2010 ) . The spatial organizations of any such populations remain challenging to model . Biochemically identifying metabolites that are conclusively exchanged between cooperating cells remains difficult , and the significance of such putative metabolite exchange challenging to interpret ( Ackermann , 2015; D'Souza et al . , 2018 ) . Finally , such studies have emphasized non-isogenic systems , where genetic changes stabilize different phenotypes , and auxotrophies define the nutrient sharing or cooperation ( Ackermann , 2015 ) . Contrastingly , here we directly identify a produced metabolic resource , and demonstrate how its availability and differential utilization can control the emergence of cells in opposing metabolic states , in a clonal population . We also explain how the spontaneous spatial organization into phenotypically distinct cell groups occurs . Thus , our study also goes beyond stochastic gene expression ( Ackermann , 2015; Balázsi et al . , 2011; Blake et al . , 2003 ) to explain how phenotypic heterogeneity and specialization can emerge in clonal populations . By considering these metabolism-derived rules , and thereby manipulating available metabolic resources , we suggest how it can be viable to program the formation , structure or phenotypic composition of isogenic cell populations . Collectively , such simple physico-chemical constraints can advance our understanding of how isogenic cells can self-organize into specialized , labor-divided groups , as a first step towards multicellularity . The prototrophic sigma 1278b strain ( referred to as wild-type or WT ) was used in all experiments . Strains with gene deletions or chromosomally tagged proteins ( at the C-terminus ) were generated as described ( Longtine et al . , 1998 ) . Strains used in this study are listed in Table 2 . The growth medium used in this study is rich medium ( 1% yeast extract , 2% peptone and 2% glucose or 0 . 1% glucose ) . All strains were grown overnight at 30 °C in either rich medium or minimal medium . 5 microliters of the overnight cultures were spotted on rich medium ( low glucose ) ( 1% yeast extract , 2% peptone , 0 . 1% glucose and 2% agar ) . Plates were incubated at 30 °C for 7 days unless mentioned otherwise . For observing colony morphology , colonies were imaged using SZX-16 stereo microscope ( Olympus ) wherein the light source was above the colony . Bright-field imaging of 7 day old colonies were done using SZX-16 stereo microscope ( Olympus ) wherein the light source was below the colony . Epifluorescence microscopy imaging of 7 day old gluconeogenesis reporter colonies ( pPCK1-mCherry ) , pentose phosphate pathway ( PPP ) reporter colonies ( pTKL1-mCherry ) and HXK1 reporter colonies ( pHXK1-mCherry ) were imaged using the red filter ( excitation of 587 nm , emission of 610 nm ) of SZX-16 stereo microscope ( Olympus ) . Similar protocol was followed for imaging 1 day to 6 day old colonies . Light cells and dark cells isolated from 7 day old wild-type colonies harboring either the gluconeogenesis reporter , PPP reporter or the HXK1 reporter were re-suspended in 1 ml of water . The percentage of fluorescent cells were determined by running the samples through a flow cytometer , and counting the total number of mCherry positive cells in a total of 1 million cells . Light cells and dark cells isolated from wild-type colonies without the fluorescent reporter were used as control . Trehalose and glycogen from yeast samples were quantified as described previously , with minor modifications ( Shi et al . , 2010 ) . 10 OD600 of light cells and dark cells from 7 day old wild-type colonies ( rich medium , 0 . 1% glucose ) were collected . After re-suspension in water , 0 . 5 ml of cell suspension was transferred to four tubes ( two tubes for glycogen assay and the other two tubes for trehalose assay ) . When sample collections were complete , cell samples ( in 0 . 25 M sodium carbonate ) were boiled at 95–98°C for 4 hr , and then 0 . 15 ml of 1 M acetic acid and 0 . 6 ml of 0 . 2 M sodium acetate were added into each sample . Each sample was incubated overnight with 1 U/ml amyloglucosidase ( Sigma-Aldrich ) rotating at 57°C for the glycogen assay , or 0 . 025 U/ml trehalase ( Sigma-Aldrich ) at 37°C for the trehalose assay . Samples were then assayed for glucose using a glucose assay kit ( Sigma-Aldrich ) . Glucose assays were done using a 96-well plate format . Samples were added into each well with appropriate dilution within the dynamic range of the assay ( 20–80 µg/ml glucose ) . The total volume of sample ( with or without dilution ) in each well was 40 microliters . The plate was pre-incubated at 37°C for 5 min , and then 80 µl of the assay reagent from the kit was added into each well to start the colorimetric reaction . After 30 min of incubation at 37°C , 80 microliters of 12 N sulfuric acid was added to stop the reaction . Absorbance at 540 nm was determined to assess the quantity of glucose liberated from either glycogen or trehalose . For measurement of extracellular trehalose measurement , single wild-type colony ( 1 day to 7 day old colony ) was re-suspended in 100 microliters of water and centrifuged at 20000 g for 5 min . Supernatant was collected and buffered to a pH of 5 . 4 ( optimal for trehalase activity ) using sodium acetate buffer ( pH 5 . 0 ) . 0 . 025 U/ml trehalase ( Sigma-Aldrich ) was added and samples were incubated at 37°C overnight . Glucose concentration was estimated as described earlier . Neutral trehalase activity assay was performed as described earlier with the following modifications ( De Virgilio et al . , 1991 ) . Briefly , 1 OD600 of light cells and dark cells isolated from 7 day old wild-type colonies ( rich medium , 0 . 1% glucose ) were washed twice with ice-cold water . For permeabilization , cells were re-suspended in tubes containing equal volume of 1% Triton-X in assay buffer ( 200 mM tricine buffer ( Na+ ) ( pH 7 . 0 ) ) and immediately frozen in liquid nitrogen . After thawing ( 1–4 min at 30°C ) , the cells were centrifuged ( 2 min at 12000 g ) , washed twice with 1 ml of ice-cold assay buffer and immediately used for the trehalase assay . Trehalase assay was performed in 50 mM tricine buffer ( Na+ ) ( pH 7 . 0 ) , 0 . 1 M trehalose , 2 mM manganese chloride ( MnCl2 ) and the Triton X-100 permeabilized cells in a total volume of 400 microliters . After incubation for 30 min at 30°C , the reaction was stopped in a boiling water bath for 3 min . Glucose concentration in the supernatant was determined using the glucose assay kit ( Sigma-Aldrich ) . Approximately ten OD600 cells were collected from respective cultures , pelleted and flash frozen in liquid nitrogen until further use . The cells were re-suspended in 400 microliters of 10% trichloroacetic acid ( TCA ) and lysed by bead-beating three times: 30 s of beating and then 1 min of cooling on ice . The precipitates were collected by centrifugation , re-suspended in 400 microliters of SDS-glycerol buffer ( 7 . 3% SDS , 29 . 1% glycerol and 83 . 3 mM tris base ) and heated at 100°C for 10 min . The supernatant after centrifugation was treated as the crude extract . Protein concentrations from extracts were estimated using bicinchoninic acid assay ( Thermo Scientific ) . Equal amounts of samples were resolved on 4% to 12% bis-tris gels ( Invitrogen ) . Western blots were developed using the antibodies against the respective tags . We used the following primary antibody: 538 monoclonal FLAG M2 ( Sigma-Aldrich ) . Horseradish peroxidase–conjugated secondary antibody ( anti-mouse ) was obtained from Sigma-Aldrich . For Western blotting , standard enhanced chemiluminescence reagents ( GE Healthcare ) were used . Light cells and dark cells isolated from wild-type colonies grown in different media were rapidly harvested and metabolites were extracted as described earlier ( Walvekar et al . , 2018 ) . Metabolites were measured using LC-MS/MS method as described earlier ( Walvekar et al . , 2018 ) . Standards were used for developing multiple reaction monitoring ( MRM ) methods on Sciex QTRAP 6500 . Metabolites were separated using a Synergi 4µ Fusion-RP 80A column ( 100 × 4 . 6 mm , Phenomenex ) on Agilent’s 1290 infinity series UHPLC system coupled to the mass spectrometer . For positive polarity mode , buffers used for separation were- buffer A: 99 . 9% H2O/0 . 1% formic acid and buffer B: 99 . 9% methanol/0 . 1% formic acid ( Column temperature , 40°C; Flow rate , 0 . 4 ml/min; T = 0 min , 0% B; T = 3 min , 5% B; T = 10 min , 60% B; T = 11 min , 95% B; T = 14 min , 95% B; T = 15 min , 5% B; T = 16 min , 0% B; T = 21 min , stop ) . For negative polarity mode , buffers used for separation were- buffer A: 5 mM ammonium acetate in H2O and buffer B: 100% acetonitrile ( Column temperature , 25°C; Flow rate: 0 . 4 ml/min; T = 0 min , 0% B; T = 3 min , 5% B; T = 10 min , 60% B; T = 11 min , 95% B; T = 14 min , 95% B; T = 15 min , 5% B; T = 16 min , 0% B; T = 21 min , stop ) . The area under each peak was calculated using AB SCIEX MultiQuant software 3 . 0 . 1 . For detecting 15N label incorporation in nucleotides , 15N Ammonium sulfate ( Sigma-Aldrich ) and 15N Aspartate ( Cambridge Isotope Laboratories ) with all nitrogens labeled were used . For 13C-labeling experiment , 13C Trehalose with all carbons labeled ( Cambridge Isotope Laboratories ) was used . All the parent/product masses measured are enlisted in Table 1 . For all the nucleotide measurements , release of the nitrogen base was monitored in positive polarity mode . For all sugar phosphates , the phosphate release was monitored in negative polarity mode . The HPLC and MS/MS protocol was similar to those explained above . The model simulation code is available via GitHub ref: https://github . com/vaibhhav/yeastmetabcolony . The model consists of ( i ) a population of light and dark cells , and ( ii ) a shared metabolic resource that is produced by , and is accessible to the cells . Therefore , the dynamic processes involved can be broadly divided into those pertaining to the cells of the colony and those pertaining to the shared resource . The cells and resource occupy a 2-D square grid , which represents the surface of an agar plate . If one takes each grid length to correspond to 50 µm in real space , then , given the average size of a yeast cell at 5 µm , a single grid location can be imagined to contain upto 100 cells , which we term ‘cell blocks’ . We coarse-grain the model such that each location is either empty , occupied by light cell block , or a dark cell block . That is , we ignore the possibility that cell blocks might be mixed . This is simply for computational ease . A more detailed model consisting of smaller grid lengths such that each location could hold at most a single cell would exhibit the same behavior as the coarse-grained one , but would require much larger grid sizes and longer computational times in order to simulate realistic sized colonies . With the coarse-graining , our simulations use a 250 × 250 grid . Each grid location also contains saturating amounts of amino acids , as well as a certain level of the shared metabolic resource . If a location has a cell block , that block also has internal levels of the amino acids and the resource , which may be different from the external level in that location . We start with an approximately circular colony 20 grid lengths in radius ( covering 1257 grid locations ) in the center of the 250 × 250 grid . 95–99% of these 1257 cell blocks are in the dark state , while 1–5% are in the light state , distributed randomly in the colony . The concentration or level of the shared resource is set to zero at every location . However , at all times , we assume the presence , throughout the grid , of saturating amounts of amino acids that are required for the ( slow ) growth of the dark cells . The grid is updated at discrete time steps . Each time step corresponds to 12 min in real time , and all simulations are run for 750 time steps , that is 150 hours of real time ( ~6 days ) . In each time step , we first go over every cell block to implement the following processes: If a block at location ( x , y ) is dark , then: If a block at location ( x , y ) is light , then: ( The above set of rules and parameters is for simulating the wild-type colony . For the variations highlighted in the main text ( Figures 3E and 6A , bottom row ) , see the ‘Variants of the wild-type model’ section below . ) These processes implement growth of cells , as well as production and consumption of amino acids and the shared metabolic resource . Subsequent to this , in each time step , we allow diffusion of the resource levels across the grid ( the “external” level at the location , not the internal levels in cell blocks ) , using a numerical scheme called Forward Time Central Space ( FTCS ) . Say that the value of the resource at time t and location ( x , y ) is given by Ux , yt . The FTCS scheme updates the value simultaneously at all locations using the following formula:Ux , yt+ΔT=Ux , yt+DΔTΔL2 ( Ux-1 , yt+ Ux+1 , yt+ Ux , y-1t+ Ux , y+1t- 4Ux , yt ) where ΔT is the time step and ΔL is the space step , or grid length , and D is the diffusion constant for the resource . Figure 3E , Figure 6A ( bottom row ) and Figure 6—figure supplement 1C–D showcase some of the final colonies generated by the simulations when the rules described above are varied . The following changes were made to the rules/parameters to generate these . 3E ( i ) : No sharing: Set R = 0 . 3E ( ii ) : No switching from dark to light state: Set p=0 . 3E ( iii ) : ‘Reverse’ sharing: Set R = 0 . When a cell block is light it adds R’=0 . 07 to the amino acid grid at the same location . 6A ( bottom row ) : No resource thresholding: Set S = 0 . Figure 6—figure supplement 1C–D: Linear switching: Set S = 0 . The probability of switching from dark to light state , p , is now a linear function of the locally available resource with a maximum value of 1 . 0 . That is , p = max ( m*Ux , yt , 1 . 0 ) , where m is a parameter that sets the slope of this linear relationship .

Under certain conditions , single-celled microbes such as yeast and bacteria form communities of many cells . In some cases , the cells in these communities specialize to perform specific roles . By specializing , these cells may help the whole community to survive in difficult environments . These co-dependent communities have some similarities to how cells specialize and work together in larger living things – like animals or plants – that in some cases can contain trillions of cells . Research has already identified the genes involved in creating communities from a population of identical cells . It is less clear how cells within these communities become specialized to different roles . The budding yeast Saccharomyces cerevisiae can help to reveal how genetic and environmental factors contribute to cell communities . By growing yeast in conditions with a low level of glucose , Varahan et al . were able to form cell communities . The communities contained some specialized cells with a high level of activity in a biochemical system called the pentose phosphate pathway ( PPP ) . This is unusual in low-glucose conditions . Further examination showed that many cells in the community produce a sugar called trehalose and , in parts of the community where trehalose levels are high , cells switch to the high PPP state and gain energy from processing trehalose . These findings suggest that the availability of a specific nutrient ( in this case , trehalose ) , which can be made by the cells themselves , is a sufficient signal to trigger specialization of cells . This shows how simple biochemistry can drive specialization and organization of cells . Certain infections are caused by cell communities called biofilms . These findings could also contribute to new approaches to preventing biofilms . This knowledge could in turn reveal how complex multi-cellular organisms evolved , and it may also be relevant to studies looking into the development of cancer .

2019

Metabolic constraints drive self-organization of specialized cell groups

Lipid droplets are lipid storage organelles found in nearly all cell types from adipocytes to cancer cells . Although increasingly implicated in disease , current methods to study lipid droplets in vertebrate models rely on static imaging or the use of fluorescent dyes , limiting investigation of their rapid in vivo dynamics . To address this , we created a lipid droplet transgenic reporter in whole animals and cell culture by fusing tdTOMATO to Perilipin-2 ( PLIN2 ) , a lipid droplet structural protein . Expression of this transgene in transparent casper zebrafish enabled in vivo imaging of adipose depots responsive to nutrient deprivation and high-fat diet . Simultaneously , we performed a large-scale in vitro chemical screen of 1280 compounds and identified several novel regulators of lipolysis in adipocytes . Using our Tg ( -3 . 5ubb:plin2-tdTomato ) zebrafish line , we validated several of these novel regulators and revealed an unexpected role for nitric oxide in modulating adipocyte lipid droplets . Similarly , we expressed the PLIN2-tdTOMATO transgene in melanoma cells and found that the nitric oxide pathway also regulated lipid droplets in cancer . This model offers a tractable imaging platform to study lipid droplets across cell types and disease contexts using chemical , dietary , or genetic perturbations . Lipid droplets are cellular organelles that act as storage sites for neutral lipids and are key regulators of cellular metabolism ( Wilfling et al . , 2013 ) . Lipid droplets are present in most cell types and are characterized by a monophospholipid membrane surrounding a hydrophobic lipid core ( Tauchi-Sato et al . , 2002; Wilfling et al . , 2013 ) . Cells maintain energetic homeostasis and membrane formation through the regulated incorporation and release of fatty acids and lipid species from the lipid droplet core ( Kurat et al . , 2009; Kuerschner et al . , 2008; Fujimoto et al . , 2007; Zimmermann et al . , 2004 ) . Importantly , lipid droplets can assume various functions during cellular stress through the sequestration of potentially toxic lipids and misfolded proteins ( Bailey et al . , 2015; Listenberger et al . , 2003; Vevea et al . , 2015; Fei et al . , 2009 ) , maintenance of energy and redox homeostasis ( Liu et al . , 2015; Liu et al . , 2017 ) , regulation of fatty acid transfer to the mitochondria for β-oxidation ( Rambold et al . , 2015; Herms et al . , 2015 ) , and the maintenance of endoplasmic reticulum ( ER ) membrane homeostasis ( Chitraju et al . , 2017; Bosma et al . , 2014; Velázquez et al . , 2016; Vevea et al . , 2015 ) . Moreover , a recent study demonstrated that lipid droplets actively participate in the innate immune response ( Bosch et al . , 2020 ) and , conversely , can be hijacked by infectious agents like hepatitis C virus to facilitate viral replication ( Barba et al . , 1997; Miyanari et al . , 2007; Vieyres et al . , 2020 ) . The role of lipid droplets in metabolic homeostasis and cellular stress is critical across multiple cell types and has also been increasingly implicated in cancer . For example , lipid droplets can act as a storage pool in cancer cells after they take up lipids from extracellular sources , including adipocytes ( Kuniyoshi et al . , 2019; Nieman et al . , 2011; Zhang et al . , 2018 ) . Lipid droplets are ubiquitous across most cell types; however , they are essential to the function of adipocytes in regulating organismal energy homeostasis ( Zimmermann et al . , 2004; Bergman et al . , 2001 ) . White adipocytes , which contain a large unilocular lipid droplet , can be readily labeled by lipophilic dyes ( Minchin and Rawls , 2017a; Fam et al . , 2018 ) . However , in vivo imaging of lipid droplets , in adipocytes or other cell types , is currently highly limited . Understanding the dynamics of lipid droplets in vivo , rather than in fixed tissues , is important since the size of the lipid droplet can change very rapidly in response to fluctuating metabolic needs ( Bosch et al . , 2020; Fam et al . , 2018 ) . In mice , much of adipose tissue imaging utilizes tissue fixation and sectioning , which can fail to preserve key aspects of the tissue structure ( Berry et al . , 2014; Xue et al . , 2010 ) . Whole-mount imaging approaches in mice can be combined with adipocyte-specific promoters; however , these methods still require tissue dissection and can be limited by tissue thickness ( Berry and Rodeheffer , 2013; Chi et al . , 2018 ) . Zebrafish offer a tractable model to address these limitations given the ease of high-throughput imaging of live animals . This is especially true with the availability of relatively transparent strains such as casper , which allows for detailed in vivo imaging without the need for fixation of the animal ( White et al . , 2008 ) . Although less well studied than other vertebrates , zebrafish adipose tissue is highly similar to mammalian white adipose tissue , and a detailed work has classified the timing , dynamics , and location of zebrafish adipose tissue development ( Minchin and Rawls , 2017a ) . Current imaging approaches using lipophilic fluorescent dyes or analogs in vivo have advanced our understanding of lipid droplets in adipocytes and other cell types ( Minchin and Rawls , 2017a; Minchin et al . , 2018; Minchin and Rawls , 2017b; Carten et al . , 2011; Farber et al . , 2001; Otis and Farber , 2016 ) ; however , these methods can require extensive and repeated staining , which may restrict the ability to read out dynamic changes over time . Furthermore , fluorescent dyes such as BODIPY and NileRed have limitations in their specificity for the lipid droplet ( Daemen et al . , 2016 ) . Finally , although a probe-free imaging approach to study subcellular lipid populations has been recently described ( Høgset et al . , 2020 ) , this method is restricted to early-stage zebrafish , which would fail to capture post-embryonic cell populations and tissues , including adipocytes . Here , we report the development of an in vivo lipid droplet reporter using a -3 . 5ubb:plin2-tdTomato transgene in the casper strain . To date , transgenic lipid droplet reporters have been restricted to cell culture systems and invertebrate model organisms such as Caenorhabditis elegans ( C . elegans ) and Drosophila ( Targett-Adams et al . , 2003; Beller et al . , 2010; Liu et al . , 2014; Liu et al . , 2015 ) although a similar approach in zebrafish was recently described while this manuscript was in review ( Wilson et al . , 2021 ) . We demonstrate that the -3 . 5ubb:plin2-tdTomato reporter faithfully marks the lipid droplet , which enables robust in vivo imaging . We show that this reporter can be applied to visualize adipocytes and to monitor adipose tissue remodeling in response to dietary and pharmacologic perturbations . Furthermore , we report the discovery of novel pharmacologic regulators of adipocyte lipolysis such as nitric oxide and demonstrate that several of these compounds can modulate adipose tissue area in our in vivo system . To facilitate the study of lipid droplets in novel contexts outside of adipocytes , we also generated a zebrafish melanoma cell line ( ZMEL ) ( Heilmann et al . , 2015 ) expressing -3 . 5ubb:plin2-tdTomato ( ZMEL-LD ) . We confirm that this cell line can be used to monitor changes in lipid droplet production in response to both known and novel regulators of lipolysis . We anticipate that these models will be highly valuable as a high-throughput imaging platform to investigate lipid droplets in both adipose tissue biology and disease contexts such as cancer . To create a specific fluorescent reporter for lipid droplets in zebrafish , we fused tdTomato to the 3’ end of the plin2 cDNA . We chose plin2 because it is a well-known lipid droplet-associated protein that is ubiquitously expressed on lipid droplets across cell types ( Olzmann and Carvalho , 2019 ) . We generated stable transgenic zebrafish expressing -3 . 5ubb:plin2-tdTomato and sought to validate whether the construct faithfully marks lipid droplets ( Figure 1A ) . White adipocytes are fat cells known for their large unilocular lipid droplet ( Fujimoto and Parton , 2011; Heid et al . , 2014 ) , so we expected expression of the PLIN2-tdTOMATO fusion protein on the surface of the adipocyte lipid droplet ( Figure 1A ) . Since the adipocyte lipid droplet occupies the majority of space in the cell ( Fujimoto et al . , 2020 ) , existing methods to visualize zebrafish adipocytes rely on lipophilic dyes and lipid analogs , which incorporate into the lipid droplet ( Zhang et al . , 2018 ) . Thus , in addition to labeling individual lipid droplets , we reasoned that the PLIN2-tdTOMATO fusion protein can also function as a reporter for adipocytes since these cells would have the largest and unilocular lipid droplets . In adult zebrafish , subcutaneous adipocytes are known to reside proximally to the tail fin ( Minchin and Rawls , 2017a ) . When we imaged 6-month-old adult Tg ( -3 . 5ubb:plin2-tdTomato ) zebrafish , we detected PLIN2-tdTOMATO expression in the zebrafish tail fin adipocytes , which colocalized with BODIPY staining ( Figure 1B , C ) . Lipophilic dyes such as BODIPY stain the lipid-rich core of the lipid droplet while lipid droplet resident proteins , such as PLIN2 , localize to the lipid droplet membrane ( Zhang et al . , 2018 ) . As expected , higher magnification images of tail adipocytes revealed that PLIN2-tdTOMATO expression was on the outside of the lipid droplet , whereas the BODIPY staining was on the interior of each droplet in the adipocyte ( Figure 1D ) . Similarly , immunohistochemistry ( IHC ) on the Tg ( -3 . 5ubb:plin2-tdTomato ) zebrafish tail fin showed that adipocytes express tdTOMATO ( Figure 1E ) . Taken together , this data demonstrates that the PLIN2-tdTOMATO fusion protein functions as a fluorescent lipid droplet reporter that can be applied to visualize adipocytes in vivo . Visceral adipose tissue , otherwise known as abdominal fat , plays an important role in metabolism and participates in pathological processes of obesity , aging , and metabolic syndromes ( Tchernof and Després , 2013 ) . Because PLIN2-tdTOMATO labeled subcutaneous adipocytes in the adult zebrafish tail fin , we wondered whether we could use the Tg ( -3 . 5ubb:plin2-tdTomato ) zebrafish to visualize other adipose depots in vivo such as visceral adipocytes . In 21 days post-fertilization ( dpf ) zebrafish , visceral adipose tissue is composed of abdominal and pancreatic visceral adipocytes predominantly located on the right flank near the swim bladder ( Figure 2A; Minchin and Rawls , 2017a ) . To determine whether Tg ( -3 . 5ubb:plin2-tdTomato ) visceral adipocytes express PLIN2-tdTOMATO , we imaged around the swim bladder of zebrafish where we expected development of abdominal visceral adipocytes ( Figure 2B ) . Visceral adipocytes visualized in brightfield demonstrate colocalization of PLIN2-tdTOMATO and BODIPY , as we observed for subcutaneous adipocytes ( Figure 2C ) . IHC confirmed that the abdominal and visceral adipocytes of Tg ( -3 . 5ubb:plin2-Tomato ) express tdTOMATO ( Figure 2G ) . Previous studies have shown that overexpression of perilipin proteins alters lipid accumulation in adipocytes ( Sawada et al . , 2010 ) . To test whether constitutive expression of our PLIN2-tdTOMATO transgenes altered adiposity , we compared the visceral adipose tissue of Tg ( -3 . 5ubb:plin2-tdTomato ) fish to their wild-type siblings . Using BODIPY staining , we did not detect differences in visceral adipose tissue area ( Figure 2D ) . In addition to dpf , standard length can indicate developmental progress in zebrafish ( Parichy et al . , 2009 ) . Similarly , we found no difference in standard lengths between wild-type siblings and Tg ( -3 . 5ubb:plin2-tdTomato ) at 21 dpf ( Figure 2E ) . We normalized visceral adipose tissue area to standard length , similar to a Body Mass Index ( BMI ) in mammals , and did not detect differences between wild-type and Tg ( -3 . 5ubb:plin2-tdTomato ) fish ( Figure 2F ) . To comprehensively assess whether constitutive PLIN2-tdTOMATO expression alters adiposity , we imaged adipose depots of wild-type and Tg ( -3 . 5ubb:plin2-tdTomato ) zebrafish from 5 dpf larvae to adult fish . Using BODIPY staining to visualize adipose tissue , we observed similar development of the major adipose depots between wild-type and Tg ( -3 . 5:ubbplin2-tdTomato ) zebrafish ( Figure 2—figure supplement 1A–B ) . Furthermore , we detected PLIN2-tdTOMATO expression at the corresponding time points and adipose depots within Tg ( -3 . 5ubb:plin2-tdTomato ) ( Figure 2—figure supplement 1A–B ) . Thus , the Tg ( -3 . 5ubb:plin2-tdTomato ) zebrafish faithfully recapitulates normal adipose tissue development . After confirming that we could image visceral adipose tissue in Tg ( -3 . 5ubb:plin2-tdTomato ) , we wanted to test whether this could be a tractable platform to image adipose tissue remodeling . We first verified whether we could use Tg ( -3 . 5ubb:plin2-tdTomato ) to track reduction in visceral adiposity . Fasting is a well-known mechanism for reducing adiposity , since it will induce lipolysis and lead to a reduction in the size of the adipocyte lipid droplet ( Henne et al . , 2018; Longo and Mattson , 2014; Rambold et al . , 2015; Tang et al . , 2017 ) . To test this , wild-type and Tg ( -3 . 5ubb:plin2-tdTomato ) zebrafish were fed or fasted for 7 days and then imaged to measure standard length and adipose tissue area ( Figure 3A ) . As expected , we observed a reduction in the BODIPY-stained visceral adipose tissue in both wild-type and Tg ( -3 . 5ubb:plin2-tdTomato ) fasted zebrafish ( Figure 3B ) . Similarly , we measured a significant reduction in adipose tissue area , standard length , and normalized area to standard length between fed and fasted fish ( Figure 3C , D , E ) . Furthermore , we did not see differences in adiposity or standard length between wild-type and Tg ( -3 . 5ubb:plin2-tdTomato ) fish within the same diet ( Figure 3C , D , E ) . This suggests that the transgenic expression of PLIN2-tdTOMATO reflects adipose tissue remodeling consistent with wild-type fish . Combined with the capacity for high-throughput in vivo imaging in zebrafish , we sought to use Tg ( -3 . 5ubb:plin2-tdTomato ) as a model to study lipid droplet dynamics in visceral adipocytes . One challenge we encountered was the autofluorescence from the zebrafish intestinal loops and gallbladder present in the tdTOMATO and GFP channels ( Figure 3F ) . To remove background fluorescence , we developed an image analysis pipeline in MATLAB to segment the visceral adipocytes in juvenile Tg ( -3 . 5ubb:plin2-Tomato ) zebrafish ( Figure 3F ) . Next , we combined our image analysis pipeline with the ability to do repeated imaging in zebrafish to more granularly quantify adipose tissue remodeling in fed or fasted Tg ( -3 . 5ubb:plin2-tdTomato ) fish ( Figure 3G ) . We saw the expected increase in adipose tissue area , standard length , and normalized area to standard length over 7 days in fed Tg ( -3 . 5ubb:plin2-tdTomato ) fish ( Figure 3H–J ) . The standard lengths for fasted fish were significantly lower than those for fed fish and remained stable over 7 days , which we attribute to food restriction , disrupting zebrafish development ( Figure 3I ) . Notably , we found that visceral adiposity was not reduced until after 2 days of fasting ( Figure 3H , J ) . In addition to fasting as a dietary perturbation , we also pharmacologically reduced adipose tissue . To achieve this , we used Forskolin , a drug that is known to induce lipolysis through cAMP signaling ( Litosch et al . , 1982 ) . We treated juvenile zebrafish for 24 hr with either dimethyl sulfoxide ( DMSO ) or 5 µM Forskolin and imaged the adipocytes ( Figure 3—figure supplement 1A ) . We detected a reduction in both the adipose tissue area and normalized area to standard length in the Forskolin-treated fish , but no differences in standard length ( Figure 3—figure supplement 1B–D ) . Thus , Tg ( -3 . 5ubb:plin2-tdTomato ) can be used as an in vivo model to visualize adipocytes , with the benefits of avoiding staining steps and allowing for repeated imaging with high-throughput image analysis in zebrafish . Having shown that we could use Tg ( -3 . 5ubb:plin2-tdTomato ) to image and measure reduction in adipose tissue , we tested whether we can use our model to detect an increase in adiposity . Zebrafish have been used as a model for diet-induced obesity and share pathophysiological perturbations seen in mammals , but few studies have focused on architectural changes of visceral adipose tissue ( Chu et al . , 2012; Landgraf et al . , 2017; Oka et al . , 2010 ) . We sought to determine if we could detect increases in visceral adiposity from a high-fat diet ( HFD ) . We fed juvenile zebrafish with either control feed ( 12% crude fat ) or HFD ( 23% crude fat ) for 7 days and subsequently imaged the adipose tissue ( Figure 4A , B ) . After a week of HFD feeding , we observed that HFD-fed fish developed significantly increased visceral adiposity compared to the fish fed with control feed ( Figure 4C , E ) . Although less dramatic between 7 and 14 days of HFD , we found that significantly greater visceral adiposity persisted after 2 weeks of HFD ( Figure 4C , E ) . Interestingly , we did not detect differences in the standard lengths of the control and HFD-fed fish , suggesting that this formulation of HFD leads to specific enlargement of visceral adipose tissue ( Figure 4E ) . Our results demonstrate that Tg ( -3 . 5ubb:plin2-tdTomato ) is an effective and unique tool to visualize visceral adipose tissue remodeling induced by HFD , which can be widely applied to study obesity . To meet fluctuating nutritional needs of the cell , lipid droplets are remodeled through lipolysis to regulate lipid mobilization and metabolic homeostasis ( Krahmer et al . , 2013; Olzmann and Carvalho , 2019; Paar et al . , 2012 ) . As a major lipid depot for the body , white adipose tissue is critical to lipid availability and cycles through lipolytic flux in response to energy demands ( Duncan et al . , 2007 ) . In disease contexts such as cancer , adipocytes undergoing lipolysis act as a lipid source for neighboring cancer cells ( Lengyel et al . , 2018 ) . Adipocyte-derived lipids have been directly shown to promote cancer progression in ovarian ( Nieman et al . , 2011 ) , breast ( Balaban et al . , 2017 ) , and melanoma cancer cells ( Zhang et al . , 2018 ) . Due to growing evidence of adipocyte and cancer cell cross-talk as a metabolic adaptation for tumor progression , there is significant interest in disrupting lipid transfer between adipocytes and cancer cells . Leveraging our model to visualize lipid droplets in adipocytes , we became interested in identifying novel compounds that remodel adipocyte lipid droplets through lipolysis . In mammalian systems , the most commonly used cell line to study lipolysis is 3T3-L1 cells , which can be differentiated in vitro to resemble adipocytes ( Zebisch et al . , 2012 ) . We first used the 3T3-L1 system to rapidly identify lipolysis inhibitors at high throughput and then test those hits using our zebrafish lipid droplet reporter . We reasoned that compounds that inhibit lipolysis in vitro would cause an increase in the size of the lipid droplets in vivo . To achieve this , we differentiated mouse 3T3-L1 fibroblast cells into adipocytes and conducted a chemical screen for compounds that inhibit lipolysis ( Figure 5A ) , measured by quantifying glycerol in the media , a gold standard readout of lipolysis in this system ( Hellmér et al . , 1989 ) . As a positive control , we used Atglistatin , an inhibitor of adipose triglyceride lipase ( ATGL ) that is known to be the rate-limiting step of lipolysis and has been shown to inhibit lipolysis in cell lines and mouse models ( Mayer et al . , 2013; Schweiger et al . , 2017 ) . We confirmed that Atglistatin potently inhibits lipolysis in 3T3-L1 adipocytes ( Figure 5B ) . We then screened through a library of 1280 compounds of diverse chemical structures to find novel inhibitors of lipolysis . Overall , we found 29 out of 1280 compounds that led to at least a 40% reduction in lipolysis as measured by glycerol release into the media . Looking more closely at the top 10 hits from this screen , we noted that 2 of the top 10 hits ( Auranofin and JS-K ) modulated nitric oxide ( Figure 5A ) . Nitric oxide can be used for post-translational modification of proteins via S-nitrosylation ( Stamler et al . , 2001 ) . A previous study has shown that increased nitric oxide has a suppressive role on lipolysis , and Auranofin , a thioredoxin reductase inhibitor that promotes S-nitrosylation , can inhibit lipolysis in 3T3-L1 cells ( Yamada et al . , 2015 ) . Similarly , JS-K is a nitric oxide donor purported to promote S-nitrosylation , but it has not been shown to play a role in lipolysis ( Nath et al . , 2010; Shami et al . , 2003 ) . Given that both of these top hits were in the same pathway , we chose these for in vivo validation . Next , we asked whether these drugs could modulate lipid droplet size and lead to increased adiposity in zebrafish . We first established the maximal tolerated doses of Atglistatin , Auranofin , and JS-K in vivo ( Figure 5—figure supplement 1A–C ) , and then tested their effects on adipose tissue in the Tg ( -3 . 5ubb:plin2-tdTomato ) fish ( Figure 5D–F ) . As expected , treatment of juvenile zebrafish for 24 hr with Atglistatin caused a significant increase in adipose tissue area without affecting standard length or fish viability at the maximally tolerated dose . Consistent with our screen results , we found that JS-K also significantly increased adiposity ( Figure 5C , E ) . These effects were specific to the adipose tissue as standard length was not affected ( Figure 5D ) . These data indicate that modulators of nitric oxide can inhibit lipolysis in cell lines and lead to increased adiposity in vivo in zebrafish . Moreover , this approach demonstrates the power of this system to dissect the relationship between novel modulators of lipolysis ( i . e . , nitric oxide ) and adiposity in vivo . Upon uptake of adipocyte-derived lipids , cancer cells can store excess lipids in lipid droplets ( Lengyel et al . , 2018 ) . Accumulation of lipid droplets in melanoma cells has been associated with increased metastatic potential and worse clinical outcomes ( Fujimoto et al . , 2020; Zhang et al . , 2018 ) . The mechanisms regulating subsequent lipolysis from the lipid droplets in cancer cells are not well understood , but we reasoned that some of the same mechanisms ( i . e . , ATGL , nitric oxide ) used in adipocytes might also be used in cancer cells . To test this , we created a stable zebrafish melanoma cell line ( ZMEL ) that expressed the -3 . 5ubb:plin2-tdTomato construct ( Heilmann et al . , 2015 ) to generate the ZMEL-LD ( lipid droplet ) reporter cell line ( Figure 6A ) . Because melanoma cells at baseline only have few small lipid droplets , we induced their formation via extrinsic addition of oleic acid , a key fatty acid that can be transferred from the adipocyte to the melanoma cell ( Zhang et al . , 2018 ) . We found that after a pulse of oleic acid for 24 hr , we could easily detect PLIN2-tdTOMATO expression surrounding lipid droplets marked by the lipid droplet dye Monodansylpentane ( MDH ) ( Figure 6B , Video 1 ) , similar to what we saw in the adipocytes ( Figure 1 ) . Similar to what we did in the whole fish , we wanted to be sure that the PLIN2-tdTOMATO transgene itself did not alter lipid droplets , so we compared lipid droplet formation in ZMEL-LD and parental cell line ( ZMEL-GFP ) via imaging and flow cytometry ( Figure 6—figure supplement 1A , B ) . Using the lipid droplet dye Lipidtox as a proxy for lipid droplet content , we pulsed the cells with increasing amounts of oleic acid and found no difference in Lipidtox median fluorescence intensity ( MFI ) between the two cell lines ( Figure 6—figure supplement 1B ) , indicating that the transgene was not having any unexpected effect . Next , we assessed the sensitivity of the PLIN2-tdTOMATO transgene versus Lipidtox in reporting changes in lipid droplets of ZMEL-LD . While both PLIN2-tdTOMATO and Lipidtox expression increased accordingly with oleic acid , PLIN2-tdTOMATO expression provided a much greater dynamic range ( Figure 6—figure supplement 1C–F ) , highlighting its advantages over standard lipid dyes like Lipidtox . To validate whether the lipolysis-inhibiting compounds we identified above could modulate lipid droplets in ZMEL-LD cells , we utilized the same flow cytometry assay to measure PLIN2-tdTOMATO expression . We treated ZMEL-LD cells for 72 hr with either BSA or oleic acid as controls for low- or high-lipid droplet cell populations ( Figure 6C ) . We then tested the effects of the lipolysis inhibitors Atglistatin , Auranofin , and JS-K . We pulsed the ZMEL-LD cells with oleic acid for 24 hr ( to induce lipid droplets ) and then measured the subsequent decay in signal over the ensuing 48 hr , which is expected to decrease due to gradual lipolysis of the lipid droplets . Compared to cells with oleic acid pulse and DMSO , cells given JS-K did not differ in the percent of lipid droplet-positive cells ( Figure 6D , E ) . In contrast , cells treated with Atglistatin and Auranofin demonstrated significantly higher lipid droplet-positive cells ( Figure 6D , E ) . These data indicate that similar to adipocytes , ATGL is a key regulatory step in lipolysis in melanoma cells . Moreover , we found that nitric oxide , which was identified in our adipocyte screen , is similarly a modulator of lipolysis in the melanoma context and can be utilized for future studies to target adipocyte-melanoma cell cross-talk . We do not yet understand why different nitric oxide donors are more or less potent in adipocytes ( where JS-K is a better inhibitor in vivo ) versus melanoma cells ( where Auranofin is a better inhibitor ) , but this could reflect differences in pharmacokinetics or cell-type-specific lipid droplet regulation . Lipid droplets are cytosolic storage organelles for cellular lipids , which are dynamically regulated in response to metabolic and oxidative perturbations ( Jarc and Petan , 2019 ) . Changes in the lipid droplet content of a cell can occur in response to a variety of factors , including hypoxia , reactive oxygen stress , ER stress , and alterations in nutrient availability ( Bailey et al . , 2015; Bensaad et al . , 2014; Chitraju et al . , 2017; Velázquez et al . , 2016; Vevea et al . , 2015; Cabodevilla et al . , 2013; Nguyen et al . , 2017 ) . However , the regulatory mechanisms driving these processes remain incompletely understood . Furthermore , lipid droplets are highly heterogeneous , and the pathways that regulate lipid droplet dynamics in specific cell types warrant investigation . To address such questions , we report the first lipid droplet reporter in a vertebrate model organism . We show that our plin2-tdtomato reporter faithfully marks the lipid droplet in vivo . As a further validation of our approach , while our manuscript was in review , a similar approach using knockin at the endogenous plin2 promoter was recently described ( Wilson et al . , 2021 ) . The combination of our reporter with the in vivo system of the casper zebrafish enables flexible and robust imaging approaches to examine lipid droplet regulation and function . In particular , the ease of chemical and genetic manipulation of the zebrafish combined with high-throughput imaging approaches enables interrogation of relevant pathways in a cell type-specific manner . Furthermore , the capacity for intravital imaging creates the opportunity to conduct longitudinal analysis of lipid droplet dynamics across developmental time and in disease contexts between single animals . Here , we demonstrate the capabilities of the Tg ( -3 . 5ubb:plin2-tdTomato ) line by taking advantage of the fact that white adipocytes are readily labeled by PLIN2-tdTOMATO expression . This labeling enables the study of individual adipocytes and adipose tissue in adult and juvenile zebrafish . We show that the Tg ( -3 . 5ubb:plin2-tdTomato ) reporter can be used as an alternative to lipophilic fluorescent dyes to study adipose tissue across zebrafish developmental stages while preserving normal adipose tissue development . We establish the utility of this transgenic line to study the regulation of adipose tissue by both diet and pharmacologic perturbations . We focused on visceral adipose tissue due to its role as an endocrine organ and the central regulator of organismal metabolism ( Fox et al . , 2007; Le Jemtel et al . , 2018; Verboven et al . , 2018 ) . We show that our Tg ( -3 . 5ubb:plin2-tdTomato ) line is sensitive enough to capture quantitative changes in visceral adipose tissue after short-term pharmacologic treatments with known regulators of adipocyte lipolysis . Furthermore , by combining top hits from a large-scale in vitro chemical screen in 3T3-L1 adipocytes with our reporter we uncovered a novel role for nitric oxide in modulating adipocyte lipolysis and adipose tissue dynamics . Beyond pharmacologic perturbations , we also demonstrate the power of this line to perform longitudinal analyses of diet-induced perturbations on adipose tissue area across multiple time points . This work yielded compelling insights into the dynamics of adipose tissue response to both prolonged fasting and high-fat diet . Differences in the kinetics of each response suggest a complex relationship between adipose tissue development and the nutritional and energetic requirements of the developing organism , which merits future investigation . Collectively , these data illustrate the potential of our model to yield novel insights into the regulation of visceral adipose tissue , including the context of obesity . While our studies show that this tool can be used to increase our understanding of adipocyte biology , it can also be utilized to study lipid droplets in other contexts as well . Lipid droplets are ubiquitous across almost all cell types . Therefore , this model could be applied to study the regulation of lipid droplets in the development and function of other adipose depots and additional cell types , such as muscle and hepatocytes ( Bosma , 2016; Wang et al . , 2013 ) . In the disease context , we focused on the role of lipid droplets in cancer , since tumor cells can take up lipids from adipocytes and then package them into lipid droplets ( Balaban et al . , 2017; Lengyel et al . , 2018; Nieman et al . , 2011; Zhang et al . , 2018 ) . This transfer of lipids has been linked to disease progression , making the regulation of lipid release from the lipid droplet through subsequent lipolysis in the tumor cell of particular interest . We found that regulation of lipolysis by ATGL and nitric oxide pathways is conserved between adipocytes and melanoma cells although phenotypes downstream of nitric oxide may be cell type specific . Collectively , this underscores the complexity of lipid droplet regulation and emphasizes the importance of studying these processes in multiple cell types . We believe that our model will serve as a powerful tool to study cell type-specific regulation of lipid droplet biogenesis and function while preserving the endogenous structural and metabolic environment of an in vivo system . To clone the plin2 cDNA , tissue from the muscle and heart of adult casper zebrafish was dissected , pooled , and then RNA was isolated using the Zymogen Quick RNA Miniprep Kit ( Zymo Research , Irvine , USA; catalog #R1054 ) according to manufacturer's instructions . The Invitrogen SuperScriptIII First-Strand Synthesis SuperMix Kit ( Thermo Fisher , Waltham , USA; catalog #18080400 ) was used according to manufacturer's instructions to produce cDNA . CloneAmp HiFi PCR Premix ( Takara , Mountain View , USA; catalog #639298 ) was used to PCR amplify the PLIN2 cDNA and gel purified via NucleoSpin Gel and PCR Clean Up ( Takara , Mountain View , USA; catalog #740609 . 50 ) . To generate pME-PLIN2-tdTOMATO , the PLIN2 cDNA was inserted on the 5’ end of pME-tdTOMATO using In-Fusion HD Cloning Plus ( Takara , Mountain View , USA; catalog #638920 ) . Gateway cloning using the Gateway LR Clonase Enzyme mix ( Thermo Fisher , Waltham , USA; catalog #11791019 ) was employed to create the -3 . 5ubb:plin2-tdTomato construct with p5E-ubb , pME-PLIN2-tdTOMATO , and p3E-polyA into pDestTol2pA2-blastocidin ( cells ) ( Heilmann et al . , 2015 ) or pDestTol2CG2 ( zebrafish ) ( Kwan et al . , 2007 ) . All zebrafish experiments were carried out in accordance with institutional animal protocols . All zebrafish were housed in a temperature- ( 28 . 5°C ) and light-controlled ( 14 hr on , 10 hr off ) room . Fish were initially housed at a density of five fish per liter and fed three times per day using rotifers and pelleted zebrafish food . Anesthesia was done using Tricaine ( Western Chemical Incorporated , Ferndale , USA ) with a stock of 4 g/l ( protected for light ) and diluted until the fish was immobilized . All procedures were approved by and adhered to Institutional Animal Care and Use Committee ( IACUC ) protocol #12-05-008 through Memorial Sloan Kettering Cancer Center . The ubb:plin2-tdTomato plasmid was injected into casper embryos with Tol2 mRNA to introduce stable integration of the ubb:plin2-tdTomato cassette . Fish with GFP+ hearts ( due to pDestTol2CG ) were selected and outcrossed with casper fish to produce the F1 generation . F1 zebrafish with GFP+ hearts and validated PLIN2-tdTOMATO-expressing adipocytes were outcrossed to generate F2 and F3 generation zebrafish for experiments . Zebrafish were imaged using an upright Zeiss AxioZoom V16 Fluorescence Stereo Zoom Microscope with a x0 . 5 ( for adult fish ) or x1 . 0 ( for juvenile fish ) adjustable objective lens equipped with a motorized stage , brightfield , and Cy5 , GFP , and tdTomato filter sets . To acquire images , zebrafish were lightly anesthetized with 0 . 2% Tricaine . Images were acquired with the Zeiss Zen Pro v2 and exported as CZI files for visualization using FIJI or analysis using FIJI ( to manually quantify standard length ) and MATLAB ( Mathworks , Natick , USA ) . Our adipocyte segmentation approach utilized the Image Processing Toolbox within MATLAB . Because the zebrafish gut is highly autofluorescent , we chose a threshold for the GFP channel to categorize as background signal and subtracted it from a determined threshold for the tdTOMATO channel . We used a set size to crop images around the tdTOMATO-positive signal and created a mask for the adipose tissue . Within the masked area , we applied a higher tdTOMATO threshold to segment the fluorescent signal from the adipocytes . Finally , we quantified the number of pixels above the threshold to quantify adipose tissue area . MATLAB code is available for download at https://github . com/dlumaquin/PLIN2-tdT-Adipo-Quant . git ( copy archived at swh:1:rev:decae59e912bd79d43f8559d5a4ee32316233153 ) ; Anthony , 2021 . To quantify BODIPY-stained adipose tissue , we utilized FIJI to autothreshold GFP signal . We removed background autofluorescence by subtracting Cy5 autothresholded signal . We used the polygon tool to outline and quantify the resulting segmented adipose area . For visualization purposes , the segmented images were color filtered on Adobe Photoshop from grayscale to gold color scale . Adult zebrafish were placed in tanks with 10 ng/µl BODIPY 493/503 ( Thermo Fisher , Waltham , USA; catalog #D3922 ) for 30 min in the dark . Fish were washed and then placed in new tanks with fresh water for 2 hr . Fish were washed again to remove any residual BODIPY , then anesthetized and imaged as indicated above for whole adipose tissue . Higher resolution images of zebrafish adipocytes were acquired using the Zeiss LSM 880 inverted confocal microscope using a x10 objective . Zebrafish were lightly anesthetized with 0 . 2% Tricaine and mounted on a glass bottom dish ( MatTek , Ashland , USA; catalog #P35G-1 . 5–20 C ) with 0 . 1% low gelling agarose ( Sigma-Aldrich , St . Louis , USA; catalog #A9045-25G ) . For BODIPY staining , 5 , 14 , and 21 dpf fish were incubated with 5 ng/µl BODIPY 493/503 in dishes for 30 min in the dark . Fish were washed with fresh E3 every 45 min for 1 . 5 hr , then anesthetized and imaged as indicated above for whole adipose tissue . 42 dpf and adult zebrafish were stained in p1000 tip boxes for 30 or 15 min , respectively , in the dark . Fish were then placed back on system to wash with fresh water for 1 . 5 hr and then anesthetized and imaged as indicated below for whole adipose tissue . To track adipose depot development across multiple stages , F2 ( -3 . 5ubb:plin2-tdTomato ) zebrafish were outcrossed with caspers to generate F3 fish expressing the transgene and control siblings . Control and Tg ( -3 . 5ubb:plin2-tdTomato ) fish were raised to a standard density of 25 fish per 2 . 8 l tank . At the appropriate time points , 5 , 14 , 21 , and 42 dpf or adult-stage zebrafish were removed from the tank and stained for BODIPY as described above . Images were acquired in three segments per fish along the anterior to posterior axis to capture the entire body of the fish . Adipose depots were classified based on the previously described system ( Minchin and Rawls , 2017a ) . To determine the presence of each adipose depot , images were thresholded in FIJI for GFP ( BODIPY493/503 ) or tdTomato ( -3 . 5ubb:plin2-tdTomato ) using control fish as the reference . A depot was scored as present based on positive signal corresponding to the presence of adipocytes for at least one sub-depot in the appropriate anatomic location . Zebrafish were sacrificed in an ice bath for at least 15 min . For adults , zebrafish tails were dissected . For juvenile zebrafish , the entire fish was used for fixation . Selected zebrafish were fixed in 4% paraformaldehyde for 72 hr at 4°C , washed in 70% ethanol for 24 hr , and then paraffin embedded . Fish were sectioned at 5 µm and placed on Apex Adhesive slides , baked at 60°C , and then stained with antibodies against tdTomato ( 1:500 , Rockland , #600-401-379 ) . Histology was performed and stained by Histowiz . Tg ( -3 . 5ubb:plin2-tdTomato ) F2 fish were outcrossed with caspers to generate the F3 generation . F3 fish were raised at a standard density of 25 fish per 2 . 8 l tank . For BODIPY staining experiments , 21 dpf fish were separated into new tanks that received standard feed or were fasted for 7 days . Prior to imaging , food was withheld for 3–6 hr to clear the gut . Fish were anesthetized with Tricaine and imaged as described above to quantify BODIPY-stained visceral adipose tissue area and standard length . For the time course fast , 21 dpf Tg ( -3 . 5ubb:plin2-tdTomato ) fish were separated into new tanks that received standard feed or fasted . Prior to imaging , food was withheld for 3–6 hr to clear the gut . Fish were anesthetized with Tricaine and imaged on days 0 , 2 , 5 , and 7 to quantify PLIN2-tdTOMATO-positive visceral adipose tissue area and standard length . Tg ( -3 . 5ubb:plin2-tdTomato ) F3 zebrafish were raised at a standard density of 25 fish per 2 . 8 l tank . At 21 dpf , the zebrafish were placed in 0 . 8 l tanks and fed either a high-fat or control diet ( Sparos , Portugal ) for up to 14 days . Fish were then imaged for PLIN2-tdTOMATO expression at days 0 , 7 , and 14 after the start of diet . Prior to imaging , fish were put in a new tank and food withheld for ~16 hr . Zebrafish were at equal density for control and experimental groups , ranging from 15 to 30 fish per tank . Fish were fed 0 . 1 g feed per tank per day split over two feedings . The high-fat and control diets were customized and produced at Sparos Lda ( Olhão , Portugal ) , where powder ingredients were initially mixed according to each target formulation in a double-helix mixer , and thereafter ground twice in a micropulverizer hammer mill ( SH1; Hosokawa-Alpine , Germany ) . The oil fraction of the formulation was subsequently added and diets were humidified and agglomerated through low-shear extrusion ( Dominioni Group , Italy ) . Upon extrusion , diets were dried in a convection oven ( OP 750-UF; LTE Scientifics , United Kingdom ) for 4 hr at 60°C , and subsequently crumbled ( Neuero Farm , Germany ) and sieved to 400 μm . Experimental diets were analyzed for proximal composition . The Sparos control diet contains 30% fishmeal , 33% squid meal , 5% fish gelatin , 5 . 5% wheat gluten , 12% cellulose , 2 . 5% soybean oil , 2 . 5% rapeseed oil , 2% vitamins and minerals , 0 . 1% vitamin E , 0 . 4% antioxidant , 2% monocalcium phosphate , and 2 . 2% calcium silicate . The Sparos HFD contains 30% fishmeal , 33% squid meal , 5% fish gelatin , 5 . 5% wheat gluten , 12% palm oil , 2 . 5% soybean oil , 2 . 5% rapeseed oil , 2% vitamins and minerals , 0 . 1% vitamin E , 0 . 4% antioxidant , 2% monocalcium phosphate , and 2 . 2% calcium silicate . 3T3-L1 cells were acquired from ZenBio and their differentiation protocol was followed . Cells were received at passage 8 and split to a maximum of passage 12 as per the recommendations of the company . 96-well plates were coated with fibronectin ( EMD Millipore , Burlington , USA; catalog #FC010 ) diluted 1:100 in phosphate-buffered saline ( PBS ) for at least 30 min to promote improved adherence of cells to the dish . 3T3-L1 cells were first cultured in PM-1-L1 Preadipocyte Medium and allowed to grow to 100% confluence . PM-1-L1 medium was changed every 48–72 hr . 48 hr after reaching 100% confluence , cells were changed to DM-1-L1 Differentiation Medium for 72 hr and then changed to AM-1-L1 Adipocyte Medium . AM-1-L1 Adipocyte Medium was changed every 48–72 hr . Once in AM-1-L1 , the medium was changed gently with a multichannel pipette , and only 150 µl of the 200 µl was replaced to prevent touching the bottom of the well with the pipette tip . After 2–3 weeks in AM-1-L1 , the 3T3-L1 developed significantly large lipid droplets and were used in the screen . The LOPAC library includes 1280 clinically relevant compounds with annotated targets or pathways . The workflow of the screen involved drug or vehicle control of the 3T3-L1 adipocytes for 24 hr in serum-free media . After 24 hr , 100 µl of the media supernatant was collected to measure secreted glycerol using the Free Glycerol Reagent ( Sigma-Aldrich , St . Louis , USA; catalog F6428 ) following the associated glycerol assay protocol . The medium ( screen media ) used for drug treatment was phenol-free DMEM supplemented with 0 . 2% BSA FFA-free ( Sigma-Aldrich , St . Louis , USA; catalog 9048-46-8 ) . The 1280 compounds were aliquoted as 2 µl at 1 mM into 16 × 96-well plates and stored at −20°C . Upon thawing , 198 µl of screen media was added to the well , bringing the final drug concentration for all compounds in the screen to 10 µM . Control vehicle was 1% DMSO served as a negative control and 1 uM isoproterenol served as a positive control in the screen . This medium containing LOPAC drugs , DMSO , and isoproterenol was transferred to 3T3-L1 cells and incubated for 24 hr . To measure glycerol release as a readout for lipolysis , 100 µl of Free Glycerol Reagent was aliquoted per well of a 96-well plate . 10 µl of supernatant media from 3T3-L1 adipocytes was then added to each well . A standard curve was produced by using Glycerol Standard Solution ( Sigma-Aldrich , St . Louis , USA; catalog G7793 ) . The plate was incubated at 37°C for 5 min and then developed with a plate reader set to detect absorbance at 540 nm . Using the standard curve , a fit equation was developed in Excel to convert the absorbance values into glycerol concentrations . To take into account differences that occur in wells on the edge versus middle of the plate , all well positions across all plates in the screen were averaged to create a normalization factor for any given position on the plate . These normalized values were then used to determine top hits for compounds that block lipolysis . 3T3-L1s were differentiated on a fibronectin-coated 96-well dish . At the start of the lipolysis experiment , 3T3-L1s were changed to serum-free DMEM supplemented with 0 . 2% BSA FFA-free ( Sigma-Aldrich , St . Louis , USA; catalog 9048-46-8 ) . The medium was supplemented with 1% DMSO for negative control or 1 uM isoproterenol to induce lipolysis or ±100 µM Atglistatin ( Sigma-Aldrich , St . Louis , USA; catalog SML1075 ) to block lipolysis and cells were incubated for 24 hr . To measure glycerol release , 100 µl of Free Glycerol Reagent was aliquoted per well of a new 96-well plate . 10 µl of supernatant media from 3T3-L1 adipocytes was then added to each well . A standard curve was produced by using Glycerol Standard Solution ( Sigma-Aldrich , St . Louis , USA; catalog G7793 ) . The plate was incubated at 37°C for 5 min and then developed with a plate reader set to detect absorbance at 540 nm . Using the standard curve , a fit equation was developed in Excel to convert the absorbance values into glycerol concentrations . Tg ( -3 . 5ubb:plin2-tdTomato ) zebrafish were outcrossed with caspers to generate the F2 or F3 generation . F2 or F3 fish were raised at a standard density of 50 fish per 6 . 0 l tank . For drug treatment , fish were removed from the system at 21 dpf and placed at a density of one fish per well in a six-well plate with 10 ml of E3 per well . To evaluate viability , fish were treated for 24 hr and quantified for live and dead larvae . After a 24 hr incubation with the drug , fish were anesthetized with Tricaine and imaged using the described protocol to quantify ( 1 ) standard length and ( 2 ) area of PLIN2-tdTOMATO expression corresponding to visceral adipose tissue area . Fish were treated with Forskolin ( Sigma-Aldrich , St . Louis , USA; catalog #F6886 ) , Auranofin ( Sigma-Aldrich , St . Louis , USA; catalog #A6733 ) , JS-K ( Sigma-Aldrich , St . Louis , USA; catalog #J4137 ) , or Atglistatin ( Sigma-Aldrich , St . Louis , USA; catalog #SML1075[1] ) , which were all dissolved in DMSO . The ZMEL zebrafish melanoma cell line was derived from a tumor of a mitfa:BRAFV600E/p53-/- zebrafish as described previously ( Heilmann et al . , 2015 ) . ZMEL cells constitutively express eGFP driven by the mitfa promoter ( Heilmann et al . , 2015 ) . ZMEL cells were grown at 28°C in a humidified incubator in DMEM ( Gibco , Waltham , USA; catalog #11965 ) supplemented with 10% fetal bovine serum ( FBS ) ( Gemini Bio , #100–500 ) , 1x penicillin/streptomycin/glutamine ( Gibco , Waltham , USA; catalog #10378016 ) , and 1x GlutaMAX ( Gibco , Waltham , USA; catalog #35050061 ) . To generate the ZMEL-LD cells , ZMEL cells were nucleofected with the ubb:plin2-tdtomato plasmid using the Neon Transfection System ( Thermo Fisher , Waltham , USA; catalog #MPK10096 ) , selected for 2 weeks in blasticidine-supplemented media at 4 µg/µl ( Sigma-Aldrich , St . Louis , USA; catalog #15205–25 MG ) , and FACS sorted for GFP and tdTOMATO double-positive cells . ZMEL and ZMEL-LD cells underwent routine Mycoplasma testing , most recently in January 2021 . Eight-well Nunc Lab-Tek Chambered Coverglass was coated with 1:100 dilution of fibronectin in Dulbecco's Phosphate Buffered Saline ( DPBS ) ( Millipore Sigma , Burlington , USA; catalog #FC010-5MG ) for 30 min and then washed with DPBS ( Thermo Scientific , Waltham , USA; catalog #14190–250 ) . ZMEL-GFP or ZMEL-LD cells were seeded at 30 , 000 cells per well and left to adhere for 24 hr . A medium supplemented with oleic acid ( Sigma-Aldrich , St . Louis , USA; catalog #O3008-5ML ) was added for 24 hr . Cells were fixed with 2% paraformaldehyde ( Santa Cruz Biotechnology , Santa Cruz , USA; catalog #sc-281692 ) for 45 min and washed with DPBS . For MDH staining , cells were permeabilized with 0 . 1% Triton-X ( Thermo Fisher , Waltham , USA; catalog #PI85111 ) for 30 min at room temperature , washed , and stained with 1:500 MDH ( Abcepta , San Diego , USA; catalog #SM1000a ) for 15 min . Cells were imaged on the Zeiss LSM 880 inverted confocal microscope with AiryScan using a x63 oil immersion objective . For Lipid staining , cells were stained with 1:500 Lipidtox Deep Red ( Thermo Fisher , Waltham , USA; catalog #H34477 ) and 1:2000 Hoechst 33342 ( Thermo Fisher , Waltham , USA; catalog #H3570 ) for 30 min . Cells were imaged on the Zeiss LSM 880 inverted confocal microscope using a x40 oil immersion objective . Confocal stacks were visualized via FIJI , and 3D reconstruction was created using Imaris ( Bitplane Inc , Concord , USA ) . ZMEL Dark ( no fluorescence ) , ZMEL-GFP , and ZMEL-LD cells were plated on fibronectin-coated six-well plates at a density of 500 , 000 cells in 1 ml of media per well . At 24 hr after plating , cells were given either 150 µM of BSA or oleic acid with 1 µl of DMSO . At 48 and 72 hr after plating , lipid droplet low and high controls were switched to fresh media with 150 µM of BSA or oleic acid with 1 µl of DMSO . Cells pulsed with oleic acid received fresh media with 150 µM of BSA with either 40 µM Atglistatin , 0 . 5 µM Auranofin , or 0 . 5 µM JS-K . At 96 hr after plating , cells were trypsinized , washed with DPBS , and resuspended in DMEM supplemented with 2% FBS , 1x penicillin/streptomycin/glutamine , and 1x GlutaMAX . Cells were stained for viability with 1:1000 DAPI and strained through the Falcon FACS Tube with Cell Strainer Cap ( Thermo Fisher , Waltham , USA; catalog #08-771-23 ) . For Lipidtox comparison , cells were given either BSA or indicated concentrations of oleic acid for 24 hr . Cells were trypsinized , washed with DPBS , stained with 1:250 Lipid Deep Red and 1:1000 DAPI for 10 min , and strained through the Falcon FACS Tube with Cell Strainer Cap . Data was acquired via the Beckman Coulter CytoFLEX Flow Cytometer ( Beckman Coulter , Miami , USA ) and analyzed using FlowJo software ( BD Biosciences , San Jose , USA ) . Schematics and illustrations were generated using Biorender on biorender . com . All statistical analyses were performed using GraphPad Prism 8 ( Graphpad , San Diego , USA ) . Data are presented as mean ± 95% confidence interval ( CI ) or standard error of the mean ( SEM ) . p<0 . 05 was considered statistically significant . Statistical tests used are noted in the figure legend . All experiments were done with at least three independent replicates . For in vivo experiments , N denotes the number of independent experiments while n denotes the number of individual fish . Imaging analyses utilized FIJI , Imaris , and MATLAB software . All zebrafish cell lines and transgenic lines are available upon request . In addition , the Tg ( -3 . 5ubb:plin2-tdTomato ) zebrafish will be deposited at the Zebrafish International Resource Center .

Organisms need fat molecules as a source of energy and as building blocks , but these ‘lipids’ can also damage cells if they are present in large amounts . Cells guard against such toxicity by safely sequestering lipids in specialized droplets that participate in a range of biological processes . For instance , these structures can quickly change size to store or release lipids depending on the energy demands of a cell . It is possible to image lipid droplets – using , for example , dyes that preferentially stain fat – but often these methods can only yield a snapshot: tracking lipid droplet dynamics over time remains difficult . Lumaquin , Johns et al . therefore set out to develop a new method that could label lipid droplets and monitor their behaviour ‘live’ in the cells of small , transparent zebrafish larvae . First , the fish were genetically manipulated so that a key protein found in lipid droplets would carry a fluorescent tag: this made the structures strongly fluorescent and easy to track over time . And indeed , Lumaquin , Johns et al . could monitor changes in the droplets depending on the fish diet , with the structures getting bigger when the animal received rich food , and shrinking when resources were scarce . Finally , experiments were conducted to screen for compounds that could lead to lipids being released in fat cells . The new imaging technique was then used to confirm the effect of these molecules in live cells , revealing an unexpected role for a signalling molecule known as nitric oxide , which also turned out to be regulating lipid droplets in cancerous cells . Further work then showed that drugs affecting nitric oxide could modulate lipid droplet size in both normal and tumor cells . This work has validated a new method to study the real-time behavior of lipid droplets and their responses to different stimuli in living cells . In the future , Lumaquin , Johns et al . hope that the technique will help to shed new light on how lipids are involved in both healthy and abnormal biological processes .

2021

An in vivo reporter for tracking lipid droplet dynamics in transparent zebrafish

Since the highly conserved exosome complex mediates the degradation and processing of multiple classes of RNAs , it almost certainly controls diverse biological processes . How this post-transcriptional RNA-regulatory machine impacts cell fate decisions and differentiation is poorly understood . Previously , we demonstrated that exosome complex subunits confer an erythroid maturation barricade , and the erythroid transcription factor GATA-1 dismantles the barricade by transcriptionally repressing the cognate genes . While dissecting requirements for the maturation barricade in Mus musculus , we discovered that the exosome complex is a vital determinant of a developmental signaling transition that dictates proliferation/amplification versus differentiation . Exosome complex integrity in erythroid precursor cells ensures Kit receptor tyrosine kinase expression and stem cell factor/Kit signaling , while preventing responsiveness to erythropoietin-instigated signals that promote differentiation . Functioning as a gatekeeper of this developmental signaling transition , the exosome complex controls the massive production of erythroid cells that ensures organismal survival in homeostatic and stress contexts . The highly conserved exosome complex , an RNA-degrading and processing machine , is expressed in all eukaryotic cells ( Januszyk and Lima , 2014; Kilchert et al . , 2016 ) . The first subunit was discovered from an analysis of mechanisms controlling yeast rRNA synthesis . Mutations of Exosc2 ( Rrp4 ) disrupt 5 . 8S rRNA 3’-end processing ( Mitchell et al . , 1996 ) . Exosc2 assembles into a complex with components homologous to bacterial 3’ to 5’ exoribonuclease ( PNPase ) ( Mitchell et al . , 1997 ) . Nine exosome complex subunits form a cylindrical core consisting of the RNA binding subunits Exosc1 ( Csl4 ) , Exosc2 ( Rrp4 ) and Exosc3 ( Rrp40 ) , which cap a ring formed by Exosc4 ( Rrp41 ) , Exosc5 ( Rrp46 ) , Exosc6 ( Mtr3 ) , Exosc7 ( Rrp42 ) , Exosc8 ( Rrp43 ) and Exosc9 ( Rrp45 ) ( Liu et al . , 2006; Makino et al . , 2013; Makino et al . , 2015; Wasmuth et al . , 2014 ) ( Figure 1A ) . Despite homology with bacterial PNPases , the vertebrate core subunits lack RNA-degrading activity ( Liu et al . , 2006 ) . Whereas Dis3 ( Rrp44 ) ( nuclear ) and Dis3L ( cytoplasmic ) catalytic subunits bind the same position of the core complex , adjacent to Exosc4 and Exosc7 , the predominantly nuclear catalytic subunit Exosc10 ( Rrp6 ) binds the opposite site ( Dziembowski et al . , 2007; Makino et al . , 2015 ) ( Figure 1A ) . The catalytic subunits , which may function redundantly in certain contexts , mediate RNA degradation and/or processing ( Kilchert et al . , 2016 ) . Unlike Dis3L and Exosc10 , which are strictly exoribonucleases , Dis3 is also an endoribonuclease ( Lebreton et al . , 2008; Tomecki and Dziembowski , 2010 ) . The core subunits , except Exosc1 , are considered to confer structural integrity ( Liu et al . , 2006 ) . 10 . 7554/eLife . 17877 . 003Figure 1 . Exosc8 or Exosc9 downregulation disrupts protein-protein interactions within the exosome complex . ( A ) Crystal structure and model of the human exosome complex ( Liu et al . , 2006 ) . Solid line , direct interactions; Dashed line , indirect interactions . ( B ) Real-time RT-PCR analysis of mRNA expression ( mean ± SE , 3 independent replicates ) in G1E-ER-GATA-1 cells 48 hr post-infection with either Exosc8 or Exosc9 shRNA expressing retrovirus . Values normalized to 18S expression and relative to the control . ( C ) Left: representative image of a semi-quantitative Western blot of Exosc2 co-immunoprecipitated with anti-Exosc3 antibody in G1E-ER-GATA-1 whole cell lysates prepared 48 hr post-Exosc8 or Exosc9 knockdown . Right: densitometric analysis of band intensity relative to the input for each knockdown condition ( mean ± SE , 3 independent replicates ) . Statistical analysis of control and treatment conditions was conducted with the Student’s T-test . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . Source data is available in Figure 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 17877 . 00310 . 7554/eLife . 17877 . 004Figure 1—source data 1 . This Excel spreadsheet contains the values of each independent replicate for data presented as histograms ( mean ± SE ) in Figure 1 . Sheet 1: Figure 1B mRNA expression of Exosc8 and Exosc9 normalized to 18S . Sheet 2: Figure 1C densitometric analysis of Exosc2 immunoblots ( pull down/input ) from an Exosc3 immunoprecipitation 48 hr after Exosc8 or Exosc9 knockdown . DOI: http://dx . doi . org/10 . 7554/eLife . 17877 . 00410 . 7554/eLife . 17877 . 005Figure 1—figure supplement 1 . The RNA binding exosome complex component Exosc3 suppresses erythroid maturation . ( A ) qRT-PCR analysis of Exosc3 mRNA in primary erythroid precursor cells 72 hr post-infection with shRNA-expressing retrovirus ( mean ± SE , 5 biological replicates ) . Values are normalized to 18S expression and relative to the control . ( B ) Erythroid maturation analyzed by flow cytometric quantitation CD71 and Ter119 staining 72 hr post-Exosc3 knockdown in primary erythroid precursor cells . Representative flow cytometry plots , with the R1-R5 gates denoted ( 5 biological replicates ) . ( C ) Percentage of primary erythroid precursor cells in R1-R5 populations 72 hr after Exosc3 knockdown ( mean ± SE , 5 biological replicates ) . Statistical analysis of control and treatment conditions was conducted with the Student’s T-test . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 17877 . 005 The apparent diversity of exosome complex-regulated RNAs ( Schneider et al . , 2012 ) suggests the complex controls a plethora of cellular processes . Exosome complex subunits regulate cell differentiation and are implicated in human pathologies . Exosc8 , Exosc9 and Dis3 suppress erythroid maturation of primary murine erythroid precursor cells ( McIver et al . , 2014 ) . EXOSC7 , EXOSC9 and EXOSC10 maintain human epidermal progenitor function ( Mistry et al . , 2012 ) . In principle , the exosome complex might control differentiation through complex RNA remodeling mechanisms or by targeting factors mediating the balance between self-renewal and lineage commitment , proliferation/amplification or terminal differentiation . DIS3 has been implicated as a tumor suppressor mutated in multiple myeloma ( Chapman et al . , 2011 ) and is overexpressed in colorectal cancer ( de Groen et al . , 2014 ) . EXOSC8 , EXOSC3 or EXOSC2 mutations cause syndromes with complex phenotypes including neurological defects , cerebellar hypoplasia , retinitis pigmentosum , progressive hearing loss and premature aging ( Boczonadi et al . , 2014; Di Donato et al . , 2016; Wan et al . , 2012 ) . As downregulating exosome complex subunits can yield compelling phenotypes , and exosome complex subunit mutations yield pathologies , it is instructive to consider whether the phenotypes reflect complex disruption or subunit-specific activities . By conferring exosome complex integrity , all exosome complex activities might require structural subunits , or sub-complexes might have distinct functions ( Kiss and Andrulis , 2011 ) . The structural subunit requirement for complex stability in vitro ( Liu et al . , 2006 ) , and lethality due to loss of structural components in yeast ( Allmang et al . , 1999a; Allmang et al . , 1999b ) support the importance of the intact complex . However , core subunit downregulation revealed little overlap in the ensembles of regulated RNAs in Drosophila ( Kiss and Andrulis , 2010 ) and differentially influenced RNA processing in humans ( Tomecki et al . , 2010 ) . Moreover , Arabidopsis Exosc1 , Exosc2 and Exosc4 knockouts yielded distinct phenotypes ( Chekanova et al . , 2007 ) . While investigating these models in the context of erythroid maturation , we discovered that Exosc8 or Exosc9 downregulation disrupted protein/protein interactions within the complex and greatly decreased expression of the receptor tyrosine kinase Kit . Loss of Stem Cell Factor ( SCF ) -induced Kit signaling occurred concomitant with precocious acquisition of erythropoietin signaling , which drives erythroid maturation . As Kit stimulates erythroid precursor cell proliferation , our results establish a paradigm in which the exosome complex regulates a receptor tyrosine kinase to orchestrate a vital developmental signaling transition dictating proliferation/amplification versus differentiation . Previously , we demonstrated that downregulating exosome complex subunits ( Exosc8 , Exosc9 or Dis3 ) in murine fetal liver erythroid precursor cells induced erythroid maturation ( McIver et al . , 2014 ) . Analogous to Exosc8 and Exosc9 ( McIver et al . , 2014 ) , experiments in which Exosc3 expression is impaired suggest that this protein also suppresses maturation of primary murine fetal liver lineage-negative hematopoietic precursor cells . In particular , Exosc3 downregulation using two distinct shRNAs increased the R4 ( late basophilic/orthochromatic erythroblasts ) cell population nine fold ( p=0 . 006 and p=0 . 01 for the two shRNAs , respectively ) ( Figure 1—figure supplement 1 ) . However , it remains unclear whether the single subunit perturbations impact exosome complex integrity . To address this , we developed a co-immunoprecipitation assay to test whether individual components mediate complex integrity ( Figure 1A ) . Using the X-ray crystal structure of the exosome complex as a guide ( Liu et al . , 2006 ) , the strategy involved testing whether downregulating endogenous Exosc8 or Exosc9 alter interactions between endogenous Exosc2 and Exosc3 , subunits that do not interact directly in the complex ( Figure 1A ) . As Exosc2 and Exosc3 are only expected to co-immunoprecipitate when residing in the complex or a sub-complex , the extent of co-immunoprecipitation constitutes a metric of complex integrity . We conducted the protein/protein interaction analysis in G1E cells stably expressing a conditionally active GATA-1 allele ( G1E-ER-GATA-1 ) , which mimic a normal erythroid precursor cell , the proerythroblast ( Weiss et al . , 1997 ) . Estradiol activation of ER-GATA-1 induces an erythroid transcriptional program and recapitulates a physiological window of erythroid maturation ( Welch et al . , 2004 ) . G1E-ER-GATA-1 cells were infected with retroviruses expressing control ( luciferase ) shRNA or shRNAs targeting Exosc8 or Exosc9 . Whole cell lysates prepared 48 hr post-infection were immunoprecipitated with anti-Exosc3 or isotype-matched control antibody , and Western blotting was conducted with anti-Exosc2 antibody . Whereas knocking down Exosc8 or Exosc9 by ~75% ( Figure 1B ) did not alter Exosc2 levels in the input , the knockdowns reduced the amount of Exosc2 recovered with the anti-Exosc3 antibody ( Figure 1C ) . Densitometric analysis indicated that Exosc8 and Exosc9 knockdowns reduced the amount of Exosc2 co-immunoprecipitated with Exosc3 by 58 ( p=0 . 007 ) and 87% ( p=1 . 3 × 10–4 ) , respectively ( Figure 1C , right ) . Of relevance to this result , yeast Exosc8 ( Rrp43 ) mutations decrease exosome complex stability and RNA binding ( Lourenco et al . , 2013 ) . As Exosc8 or Exosc9 downregulation disrupted Exosc3-Exosc2 interactions that only occur in the complex , erythroid maturation resulting from downregulating either of these subunits is associated with dismantling or destabilizing intra-complex protein-protein interactions . Since Exosc8 or Exosc9 downregulation disrupts the exosome complex and promotes erythroid maturation , we investigated how integrity of the complex creates an erythroid maturation barricade . Although the parameters dictating the decision of whether an erythroid precursor cell undergoes sustained proliferation or differentiates into an erythrocyte are incompletely understood , cytokine signaling is a key determinant ( Lodish et al . , 2010 ) . It is instructive to consider the relationship between the exosome complex-mediated erythroid maturation barricade and signaling mechanisms that orchestrate proliferation versus differentiation . In principle , the complex might enhance signaling that favors proliferation or oppose signaling mediating differentiation . Whereas SCF/Kit signaling supports hematopoietic stem/progenitor cell ( HSPC ) and erythroid precursor cell proliferation and survival ( Lennartsson and Ronnstrand , 2012 ) , erythropoietin ( Epo ) /Epo receptor signaling uniquely promotes terminal differentiation ( Munugalavadla and Kapur , 2005; Wu et al . , 1995b ) . Epo and SCF/Kit can synergistically promote proliferation and survival ( Joneja et al . , 1997; Muta et al . , 1994; Sui et al . , 1998; Wu et al . , 1997 ) . To test whether the exosome complex and Epo signaling are functionally interconnected , murine fetal liver erythroid precursor cells were infected with control or Exosc8 shRNA-expressing retroviruses . After culturing cells for 48 hr in media containing increasing amounts of Epo ( 0–0 . 5 U/ml ) , we conducted flow cytometric analysis with Annexin V and the membrane-impermeable dye DRAQ7 to quantitate live cells ( DRAQ7-/Annexin V- ) , as well as late ( DRAQ7+/Annexin V+ ) and early ( DRAQ7-/Annexin V+ ) apoptosis ( Figure 2A , B ) . Without exogenous Epo , live cells decreased from 66 to 24% ( control versus knockdown cells ) ( p=2 × 10–5 ) , and late apoptotic cells increased proportionally ( 24 to 67% , p=3 × 10–4 ) . Exosc8 downregulation increased late apoptosis two-fold ( p=0 . 02 ) when cells were cultured in media containing 0 . 001 U/ml Epo . Exosc8 downregulation did not significantly influence the percentage of live and late apoptotic cells when cells were cultured with higher Epo concentrations . Surprisingly , downregulating Exosc8 rendered cell integrity hypersensitive to limiting concentrations of Epo . 10 . 7554/eLife . 17877 . 006Figure 2 . Exosome complex disruption renders primary erythroid cells hypersensitive to limiting erythropoietin concentrations . ( A ) Flow cytometric analysis with Annexin V and the membrane-impermeable dye DRAQ7 to quantitate apoptosis with control and Exosc8-knockdown primary erythroid cells expanded for 48 hr under Epo-limiting conditions . ( B ) Quantification of the percentage of primary erythroid cells in live , late and early apoptotic populations ( mean ± SE , 4 biological replicates ) . Statistical analysis of control and treatment conditions was conducted with the Student’s T-test . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . Source data is available in Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 17877 . 00610 . 7554/eLife . 17877 . 007Figure 2—source data 1 . This Excel spreadsheet contains the values of each biological replicate for data presented as line graphs ( mean ± SE ) in Figure 2 . Sheet 1: Figure 2B percentage of cells from each biological replicate found in the live , early apoptotic , late apoptotic and necrotic flow cytometry gates in control and Exosc8 knockdown cells after 48 hr culture under Epo-limiting conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 17877 . 007 As downregulating Exosc8 promotes erythroid maturation , the commencement and/or progression of maturation when Epo is limiting might be incompatible with cell survival . Alternatively , lowering Epo might impair a mechanism that supports erythroid precursor proliferation , independent of enhanced maturation; thus , precursor cell survival would be compromised . Using flow cytometry to quantitate the erythroid cell surface markers Ter119 and CD71 , we determined the impact of Exosc8 downregulation on erythroid precursor cell maturation under normal and Epo-limiting conditions . In this expansion culture , control cells were largely unaffected by limiting Epo . Whereas Exosc8 downregulation stimulated erythroid maturation ( 1 . 5 fold increase in R2 to R3 transition , p=0 . 003 ) in media containing 0 . 5 U/ml Epo , maturation of Exosc8-knockdown cells did not proceed without exogenous Epo ( Figure 3A , B ) . Without Epo , Exosc8 downregulation also increased the percentage of immature erythroid precursors ( R1 ) . Morphological analysis of DRAQ7- cells using Giemsa stain provided further evidence for the Epo-dependent R2 to R3 transition resulting from Exosc8 knockdown ( Figure 3C ) . The exosome complex-dependent erythroid maturation barricade was also quantitated using an alternative flow cytometric assay based on CD44 staining and cell size . Exosc8 or Exosc9 downregulation induced accumulation of more mature erythroid cells when cultured in 0 . 5 U/ml Epo ( Figure 3—figure supplement 1 ) . Orthochromatic erythroblasts ( gate IV ) increased 8 fold ( p=0 . 006 ) or seven fold ( p=3 . 5 × 10–5 ) post-Exosc8 or -Exosc9 downregulation , respectively . Erythrocytes ( gate VI ) increased two fold ( Exosc8 , p=0 . 005 , Exosc9 , p=0 . 005 ) . Immature erythroid cells ( proerythroblasts and basophilic erythroblasts , gates I and II ) decreased proportionally . Exosc8 downregulation reduced the number of early SCF , IL-3 and Epo-dependent BFU-E ( Burst Forming Unit – Erythroid ) colonies by 10 fold ( p=4 . 7 × 10–10 ) . However , the number of Epo dependent CFU-E ( Colony Forming Unit – Erythroid ) colonies was unaffected ( Figure 3D ) . 10 . 7554/eLife . 17877 . 008Figure 3 . Erythropoietin is required for erythroid differentiation induced by disrupting the exosome complex . ( A ) Flow cytometric quantification of erythroid markers CD71 and Ter119 in live control and Exosc8-knockdown erythroid precursor cells cultured for 48 hr in Epo-limiting media . Representative plots with R1-R5 gates denoted . ( B ) Quantitation of the percentage of live cells in control and Exosc8-knockdown conditions from the R1-R4 and non erythroid gates ( mean ± SE , 4 biological replicates ) . ( C ) Representative images of Wright-Giemsa-stained , DRAQ7-negative erythroid precursor cells , infected with control or ShExosc8 retrovirus . Cells were cultured with or without Epo for 48 hr ( Scale bar , 10 μm ) . ( D ) Representative images ( left ) and quantitation ( right ) of erythroid colony forming unit activity with FACS-sorted R1 cells 24 hr after Exosc8 knockdown ( mean ± SE , 6 biological replicates ) ( Scale bar 200 μm ) . Statistical analysis of control and treatment conditions was conducted with the Student’s T-test . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . Source data is available in Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 17877 . 00810 . 7554/eLife . 17877 . 009Figure 3—source data 1 . This Excel spreadsheet contains the values for each biological replicate for data presented as either line graphs or histograms ( mean ± SE ) in Figure 3 . Sheet 1: Figure 3B the percent live cells found in the R1 , R2 , R3 , R4 and R5 flow cytometry gates in control and Exosc8 knockdown cells after 48 hr culture in Epo-limiting conditions . Sheet 2: Figure 3D the CFU-E and BFU-E counts from colony assays performed after 24 hr infection with shExosc8 . DOI: http://dx . doi . org/10 . 7554/eLife . 17877 . 00910 . 7554/eLife . 17877 . 010Figure 3—figure supplement 1 . Analysis of the exosome complex-mediated erythroid maturation barricade using a distinct flow cytometric assay . ( A ) Erythroid maturation of primary erythroid precursor cells 72 hr post-infection with shExosc8- or shExosc9-expressing retroviruses analyzed by flow cytometric quantification of CD44 and side scatter ( SSC ) . Representative flow cytometry plots with gates I to VI are depicted . ( B ) Percentage of erythroid cells detected in gates I through VI ( 3 biological replicates , mean ± SE ) . ( C ) Representative images of Wright-Giemsa-stained erythroid cells from the sorted , gated ( I-IV ) populations under control conditions ( Scale bar , 10 μm ) . Statistical analysis of control and treatment conditions was conducted with the Student’s T-test . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 17877 . 010 Erythroid maturation instigated by Exosc8 downregulation was Epo-dependent . We tested whether Exosc8 downregulation corrupted a proliferation/amplification mechanism , thereby promoting maturation . SCF signaling supports erythroid precursor proliferation ( Munugalavadla and Kapur , 2005 ) , whereas Epo signaling promotes maturation ( Wu et al . , 1995b ) . SCF and Epo can synergistically promote proliferation ( Joneja et al . , 1997; Sui et al . , 1998; Wu et al . , 1997 ) . Using a phospho-flow cytometry assay ( Figure 4A ) ( Hewitt et al . , 2015 ) , we quantitated the capacity of SCF ( Figure 4B ) or Epo ( Figure 4C ) to instigate cell signaling using the shared downstream substrate Akt . SCF induced maximal Akt phosphorylation in immature erythroblasts ( Ter119-/CD71high ) ( 5 . 5 fold , p=3 × 10–6 ) . As erythroid maturation progressed to Ter119+/CD71high and Ter119+/CD71low stages , the SCF response was diminished ( Figure 4B ) . Exosc8 downregulation abrogated SCF-mediated induction of phospho-Akt ( Figure 4B ) . Epo induced maximal Akt phosphorylation in Ter119+/CD71high erythroblasts , and Exosc8 downregulation accelerated acquisition of this signaling response . Whereas Epo did not affect Akt phosphorylation in control Ter119-/CD71high erythroblasts , Epo increased phospho-Akt 4 fold ( p=0 . 003 ) in Exosc8-knockdown Ter119-/CD71high erythroblasts ( Figure 4C ) . Similar results were obtained using a phospho-flow cytometric assay to quantitate phosphorylation of ERK , an additional shared downstream effector of SCF and Epo signaling , although phospho-ERK was higher in the unstimulated Exosc8 condition in comparison with the control condition ( Figure 4—figure supplement 1 ) . Thus , downregulating an exosome complex subunit that dismantles intra-complex protein-protein interactions abrogates SCF signaling that supports precursor proliferation/amplification , while precociously inducing pro-differentiation Epo signaling . By orchestrating this developmental signaling transition , the exosome complex ensures a balance between proliferation/amplification and differentiation - a balance that shifts physiologically towards differentiation as GATA-1 represses genes encoding exosome complex subunits ( McIver et al . , 2014 ) . Artificially , shRNA-mediated downregulation of exosome complex subunits , which impairs exosome complex integrity , skews the balance . 10 . 7554/eLife . 17877 . 011Figure 4 . Exosome complex sustains proliferation signaling , while suppressing pro-differentiation signaling . ( A ) Experimental scheme: lineage-negative cells were isolated from E14 . 5 fetal livers , and infected with luciferase or Exosc8 shRNAs . Cells were cultured for 48 hr and sorted into Ter119+ and Ter119- populations using beads . After 1 hr of serum-starvation , cells were stimulated for 10 min with 10 ng/ml SCF or 2 U/ml Epo and fixed/permeabilized before staining for CD71 and p-Akt . ( B ) Top: p-Akt staining after stimulation with 10 ng/ml SCF in control and Exosc8-knockdown cells ( 6 biological replicates ) . Bottom: Relative p-Akt MFI after stimulation with 10 ng/ml SCF in control and Exosc8-knockdown cells . MFI expressed relative to unstimulated Ter119-/CD71low control ( mean ± SE , 6 biological replicates ) . ( C ) Top: p-Akt staining after stimulation with 2 U/ml EPO in control and Exosc8-knockdown cells ( 6 biological replicates ) . Bottom: Relative p-Akt MFI after stimulation with 2 U/ml Epo in control and Exosc8-knockdown cells . MFI expressed relative to unstimulated Ter119-/CD71low control ( mean ± SE , 6 biological replicates ) . ANOVA identified any significant variation within the experiment , and a Tukey-Kramer test identified the statistical relationship between each pair of samples . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . Source data is available in Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 17877 . 01110 . 7554/eLife . 17877 . 012Figure 4—source data 1 . This Excel spreadsheet contains the values for each biological replicate for data presented as histograms ( mean ± SE ) in Figure 4 . Sheet 1: Figure 4A and B p-Akt MFI after 10 min stimulation with either SCF or Epo 48 hr post-Exosc8 knockdown . DOI: http://dx . doi . org/10 . 7554/eLife . 17877 . 01210 . 7554/eLife . 17877 . 013Figure 4—figure supplement 1 . Flow cytometric analysis of ERK phosphorylation reveals Exosc8 requirement to confer Kit signaling and to suppress Epo signaling . ( A ) Top: p-ERK staining after 10 min stimulation with 10 ng/ml SCF in control and Exosc8-knockdown cells ( 5 biological replicates ) . Bottom: p-ERK MFI expressed relative to unstimulated Ter119-/CD71low control ( mean ± SE , 5 biological replicates ) . ( B ) Top: p-ERK staining after 10 min stimulation with 2 U/ml EPO in control and Exosc8-knockdown cells ( 5 biological replicates ) . Bottom: p-ERK MFI expressed relative to the Ter119-/CD71 low control ( mean ± SE , 5 biological replicates ) . Initially ANOVA identified any significant variation between experimental groups . A Tukey-Kramer test subsequently identified the statistical relationship between each pair of samples . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001DOI: http://dx . doi . org/10 . 7554/eLife . 17877 . 013 As downregulating Exosc8 abrogated SCF-induced Kit signaling , we tested whether Exosc8 is required for Kit expression in the erythroblast plasma membrane . Using flow cytometry with primary fetal liver erythroblasts , R2 erythroblasts expressed the highest levels of cell surface Kit , and Exosc8 downregulation reduced cell surface Kit to a nearly undetectable level ( Figure 5A ) . Similarly , Exosc9 downregulation reduced cell surface Kit 6 fold in R2 cells ( Figure 5B ) . We tested whether exosome complex disruption reduced Kit transcription , total Kit protein and/or Kit transit to the cell surface . Exosc8 downregulation reduced Kit mRNA and primary transcript levels at all stages of erythroid maturation ( Figure 5C ) . By 24 hr post-infection with shExosc8-expressing retrovirus , total Kit protein declined 10 fold ( p=0 . 046 ) in Ter119- erythroid precursor cells ( Figure 5C ) . 10 . 7554/eLife . 17877 . 014Figure 5 . Exosome complex requirement for Kit expression . ( A ) Left: surface Kit in R1-R3 populations 48 hr post-Exosc8 knockdown . Representative plots . Right: surface Kit MFI relative to control R1 ( mean ± SE , 5 biological replicates ) . ( B ) Left: surface Kit in R1-R3 cells 48 hr post-Exosc9 knockdown . Representative plots . Right: surface Kit MFI relative to control R1 ( mean ± SE , 8 biological replicates ) . ( C ) Left: real-time RT-PCR of Kit mRNA and primary transcripts in sorted R1-R3 populations 72 hr post-infection with shControl or shExosc8 normalized to 18S and relative to control R1 ( mean ± SE , 6 biological replicates ) . Middle: Kit Western blot with Ter119- cells 24 hr post-infection ( mean ± SE , 3 biological replicates ) . ( D ) Left: real-time RT-PCR of erythroid mRNAs in sorted R1-R3 populations 72 hr post-infection with shControl or shExosc8 ( mean ± SE , 6 biological replicates ) . Middle: GATA-2 and GATA-1 Western blot with Ter119- cells 24 hr post-infection . Right: densitometric analysis normalized to tubulin and relative to shControl ( mean ± SE , 3 biological replicates ) . ( E ) qRT-PCR of Exosc8 and Kit mRNA and GATA-1/Exosc8-regulated cell cycle arrest genes in primary erythroid precursor cells 24 hr post-infection . Normalized to 18S and relative to the control ( mean ± SE , 5 biological replicates ) . ( F ) Cell cycle analysis of control and Exosc8-knockdown Ter119- cells 24 ( top ) and 72 hr ( bottom ) post-infection ( mean ± SE , 6 biological replicates ) ( G ) qRT-PCR analysis of Exosc8 and Kit mRNA in G1E cells 48 hr post-infection with shControl or shExosc8 retrovirus , normalized to 18S and expressed relative to the control ( mean ± SE , 3 independent experiments ) ( H ) Cell surface Kit expression in infected ( GFP+ ) and uninfected ( GFP- ) populations of G1E cells 48 hr post-infection with shExosc8 ( mean ± SE , 3 independent experiments ) . Statistical analysis of control and treatment conditions was conducted with the Student’s T-test *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . Source data is available in Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 17877 . 01410 . 7554/eLife . 17877 . 015Figure 5—source data 1 . This Excel spreadsheet contains the values for each biological replicate for data presented as either line graphs or histograms ( mean ± SE ) in Figure 5 . Sheet 1: Figure 5A Kit MFI in the R1 , R2 , R3 , R4 and R5 population 48 hr after Exosc8 knockdown . Sheet 2: Figure 5B Kit MFI in the R1 , R2 , R3 , R4 and R5 population 48 hr post-Exosc9 knockdown . Sheet 3: Figure 5C Kit mRNA and primary transcript expression sorted R1-R3 populations 72 hr post-infection with shControl or shExosc8 normalized to 18S and densitometry analysis of Kit protein in Ter119- cells 24 hr post-knockdown . Sheet 4: Figure 5D mRNA expression of erythroid genes in sorted R1-R3 populations 72 hr post-infection with shControl or shExosc8 and densitometry analysis of GATA-1 and GATA-2 protein Ter119- cells 24 hr post-knockdown . Sheet 5: Figure 5E Expression of Exosc8 , Kit and GATA-1/Exosc8-regulated cell cycle arrest genes in primary erythroid precursor cells 24 hr post-infection , normalized to 18S . Sheet 6: Figure 5F Cell cycle analysis of control and Exosc8-knockdown Ter119− cells 24 and 72 hr post-infection . Sheet 7: Figure 5G Exosc8 and Kit mRNA expression in G1E cells 48 hr post-infection with shControl or shExosc8 retrovirus , normalized to 18S . Sheet 8: Figure 5H Kit MFI in infected ( GFP+ ) and uninfected ( GFP- ) populations of G1E cells 48 hr post-infection with shExosc8 . DOI: http://dx . doi . org/10 . 7554/eLife . 17877 . 015 To further dissect the mechanism underlying exosome complex-dependent expression of Kit and establishment of SCF-induced Kit signaling , we tested whether exosome complex disruption influenced other genes downregulated during erythroid maturation . Since GATA-2 directly activates Kit transcription in immature erythroblasts , and GATA-1 directly represses Kit transcription during maturation ( Jing et al . , 2008; Munugalavadla et al . , 2005 ) , we tested whether exosome complex disruption impacted GATA-2 and GATA-1 levels . Exosc8 downregulation modestly increased Gata2 mRNA in R3 cells , but GATA-2 protein levels were not significantly affected in Ter119- cells ( Figure 5D ) . Previously , we demonstrated that Exosc8 knockdown did not influence Gata1 mRNA levels ( McIver et al . , 2014 ) . GATA-1 increased slightly in Exosc8-knockdown Ter119- cells ( 1 . 4 fold , p=0 . 02 ) ( Figure 5D ) . GATA-2 activates transcription of Samd14 , encoding a facilitator of SCF/Kit signaling ( Hewitt et al . , 2015 ) . In certain contexts , Kit signaling creates a positive autoregulatory loop that increases Kit transcription ( Zhu et al . , 2011 ) . Downregulating Exosc8 had little to no effect on Samd14 expression ( Figure 5D ) . To assess whether Exosc8 downregulation reduced expression of genes resembling Kit in being transcriptionally downregulated upon maturation , we quantitated expression of Vim ( encoding vimentin ) ( DeVilbiss et al . , 2015 ) and the GATA-2 target gene Hdc ( encoding histidine decarboxylase ) ( Katsumura et al . , 2014 ) . Downregulating Exosc8 did not affect the maturation-dependent reduction in Vim or Hdc mRNA levels ( Figure 5D ) . The Exosc8 requirement for Kit transcription and expression of functional , cell surface Kit therefore does not involve a major alteration in GATA factor levels nor a general Exosc8 activity to sustain expression of genes destined for downregulation upon erythroid maturation . Previously , we demonstrated that exosome complex disruption induces erythroid precursor cells to arrest in the G1 phase of the cell cycle and is associated with increased expression of genes promoting ( or implicated in promoting ) cell cycle arrest ( McIver et al . , 2014 ) . As erythroid maturation requires cell cycle progression ( Pop et al . , 2010 ) , we asked whether Kit downregulation precedes , is concomitant with , or a consequence of GATA-1 and Exosc8-mediated regulation of cell cycle arrest genes . At 24 hr post-infection of fetal liver erythroid precursor cells with shRNA-expressing retrovirus , Exosc8 mRNA declined by 70% ( p=0 . 005 ) . Whereas Kit mRNA decreased by 46% ( p=0 . 002 ) , expression of the GATA-1/Exosc8-regulated cell cycle-regulatory genes Cdkn1b ( p27Kip1 ) , Ddit3 , Gadd45a , Gas2l1 and Trp53inp1 was not significantly altered ( Figure 5E ) . Analysis of the cell cycle status of Kit-expressing Ter119- erythroid precursor cells 24 and 72 hr post-infection revealed only a slight increase in the percentage of G1 cells at 24 hr ( 30% in control and 33 . 5% in Exosc8-knockdown cells ( p=2 × 10–6 ) ) . At 72 hr post-infection , 49 . 5% of Exosc8-knockdown Ter119- cells were in G1 , in comparison with 36% of control cells ( p=1 . 6 × 10–6 ) ( Figure 5F ) . Thus , the prominent cell cycle arrest resulting from Exosc8 downregulation occurs after Kit is repressed . To more definitively establish the relationship between Kit downregulation and cell cycle arrest , we tested whether Exosc8 downregulation reduces Kit expression in a system not competent for differentiation . shRNA-mediated Exosc8 downregulation in GATA-1-null G1E proerythroblast-like cells reduced Kit mRNA by 37% ( p=0 . 002 ) ( Figure 5G ) , and Kit cell surface expression by 44% in the GFP+ population ( p=3 . 3 × 10–8 ) . Importantly , cell surface Kit was unaffected in the GFP- population ( Figure 5H ) . In aggregate , these analyses support a model in which exosome complex disruption downregulates Kit expression independent of alterations in cell cycle status and is not a consequence of cell cycle arrest or cellular maturation . Consistent with these results , mining RNA-seq data from normal and Exosc3-/- ES cells ( Pefanis et al . , 2015 ) revealed Kit expression 5 . 9 fold lower in mutant versus control cells ( transcripts per million , false discovery rate < 0 . 05 ) . We tested whether the Exosc8 requirement for Kit primary transcript , mRNA and protein expression involved alterations in the distribution of transcriptionally-competent serine 5-phosphorylated RNA polymerase II ( Pol II ) at Kit . Using quantitative ChIP analysis with control and Exosc8-knockdown Ter119- cells , Exosc8 downregulation reduced phospho-Ser5 Pol II occupancy within the coding region ( +5 kb ) and 3’ UTR , but not at the promoter ( Figure 5E ) . Exosc8 downregulation did not alter phospho-Ser5 Pol II occupancy at the active Rpb1 gene or the inactive Krt5 gene ( Figure 6A ) . These results provide further evidence that exosome complex-mediated Kit expression involves a transcriptional mechanism . As GATA-1 levels increased slightly after Exosc8 downregulation in Ter119- cells , and GATA-1 represses Kit transcription , we tested whether GATA-1 occupancy at Kit was altered . Exosc8 downregulation did not influence GATA-1 occupancy at -114 , +5 and +58 kb sites , relative to the promoter ( Figure 6A ) . The nearly complete Kit repression upon Exosc8 downregulation did not involve detectable changes in GATA-1 occupancy . 10 . 7554/eLife . 17877 . 016Figure 6 . Exosome complex occupies the Kit locus and is required for active RNA Polymerase II occupancy at Kit . ( A ) qChIP of serine 5-phospho Pol II , and GATA-1 occupancy at Kit in control and Exosc8-knockdown Ter119- erythroid precursor cells 24 hr post-infection ( mean ± SE , 6 independent experiments ) . ( B ) qChIP of Exosc9 occupancy at Kit and promoters of other exosome complex-regulated erythroid genes ( Alas2 , Hbb-b1 and Slc4a1 ) in erythroid precursor cells after culturing for 48 hr ( mean ± SE , 3 biological replicates ) . Statistical analysis of control and treatment conditions was conducted with the Student’s T-test . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . Source data is available in Figure 6—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 17877 . 01610 . 7554/eLife . 17877 . 017Figure 6—source data 1 . This Excel spreadsheet contains the values for each biological replicate for data presented in histograms ( mean ± SE ) in Figure 6 . Sheet 1: Figure 6A qChIP of serine 5-phospho Pol II , and GATA-1 occupancy at Kit in control and Exosc8-knockdown Ter119- erythroid precursor cells 24 hr post-infection . Sheet 2: Figure 6B qChIP of Exosc9 occupancy at Kit and promoters of other exosome complex-regulated erythroid genes ( Alas2 , Hbb-b1 and Slc4a1 ) in erythroid precursor cells after culturing for 48 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 17877 . 017 In Drosophila , a genome-wide analysis of Exosc10 ( Rrp6 ) and Exosc3 ( Rrp40 ) chromatin occupancy revealed occupancy predominantly at chromatin insulators and promoters of active genes ( Lim et al . , 2013 ) . However , considering that the exosome complex associates with elongating RNA polymerase II ( Andrulis et al . , 2002 ) and rapidly degrades promoter upstream transcripts ( PROMPTs ) , which are believed to be generated at many genes ( Lubas et al . , 2011; Preker et al . , 2011; Preker et al . , 2008 ) , one might predict that the exosome complex has a broader distribution in and surrounding genes . To investigate the mechanism by which the exosome complex confers Kit transcription and suppresses Alas2 , Hbb-b1 and Slc4a1 transcription ( McIver et al . , 2014 ) , we asked whether the exosome complex occupies these loci . Quantitative ChIP analysis with fetal liver erythroid precursor cells expanded for 48 hr revealed endogenous Exosc9 occupancy at the Kit promoter , the coding region and the 3’-UTR , with occupancy higher at the promoter and coding region sites versus the 3'-UTR ( Figure 6B ) . Exosc9 also occupied the Alas2 , Slc4a1 and Hbb-b1 promoters , with less occupancy at the MyoD promoter ( Figure 6B ) . Exosc9 occupancy implies that the exosome complex directly regulates transcription of these loci . Although Kit downregulation correlated with erythroid maturation induced by exosome complex disruption , whether this reflects causation was unclear . To establish whether Kit downregulation is required for erythroid maturation induced by exosome complex disruption , we asked whether enforced Kit expression in Exosc8-knockdown Ter119- cells opposed differentiation . At 24 hr post-infection , there was little to no difference in the maturation state of control and Exosc8-knockdown cells ( Figure 7A , B ) . By 48 hr , Exosc8 downregulation increased R3 cells from 35 to 63% ( p=4 × 10–7 ) , while reducing R2 cells two-fold ( p=4 × 10–5 ) . Kit expression prevented Exosc8 knockdown-dependent maturation , reducing R3 cells from 63 to 27% ( p=1 × 10–7 ) . Kit expression , concomitant with Exosc8 downregulation , rescued R2 cells ( increased from 21 to 42% , p=8 × 10–4 ) to a level indistinguishable from control cells ( 45% ) . At 48 hr , Kit expression in control cells increased R2 cells from 45 to 66% ( p=6 × 10–5 ) . At 72 hr , Exosc8-knockdown cells matured further ( 18% R4 , 56% R3 and 17% R2 versus 2% R4 ( p=4 × 10–5 ) , 41% R3 ( p=3 × 10–5 ) , and 47% R2 ( p=3 × 10–6 ) for control cells ) . Expressing Kit in Exosc8-knockdown cells reduced R4 from 18 to 7% ( p=4 × 10–5 ) , R3 from 56 to 28% ( p=9 × 10–7 ) , while increasing R2 from 17 to 38% ( p=5 × 10–5 ) . At 72 hr , Kit expression increased R2 cells from 47 to 68% ( p=6 × 10–5 ) . 10 . 7554/eLife . 17877 . 018Figure 7 . Functional link between Kit downregulation and erythroid differentiation induced by disrupting the exosome complex . ( A ) Erythroid maturation analyzed by flow cytometric quantitation of CD71 and Ter119 post-Exosc8 knockdown and/or Kit expression in primary erythroid precursor cells expanded for 72 hr . Representative flow cytometry plots , with the R1-R5 gates denoted . ( B ) Left: relative Kit MFI post-Exosc8 knockdown and/or Kit overexpression ( mean ± SE , 4 biological replicates ) . Right: percentage of primary erythroid precursor cells in the R1-R4 gates ( mean ± SE , 4 biological replicates ) . ( C ) Left: relative Kit MFI 48 hr post-Exosc8 knockdown in cells infected with increasing amounts of a Kit-expressing retrovirus . The arrow depicts Kit downregulation resulting from knocking-down Exosc8 . Right: percentage of erythroid precursor cells in the R3 population 48 hr post-infection with shExosc8 in cells infected with increasing amounts of Kit-expressing retrovirus . The arrow depicts the increased R3 population post-Exosc8 knockdown . ANOVA identified any significant variation between experimental groups then a Tukey-Kramer test identified the statistical relationship between each pair of samples , *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . Source data is available in Figure 7—source data 1DOI: http://dx . doi . org/10 . 7554/eLife . 17877 . 01810 . 7554/eLife . 17877 . 019Figure 7—source data 1 . This Excel spreadsheet contains the values for each biological replicate for data presented in line graphs ( mean ± SE ) in Figure 7 . Sheet 1 2 and 3: Figure 7B Kit MFI and percentage of erythroid precursor cells in the R1 , R2 , R3 R4 and R5 populations 24 , 48 and 72 hr post Exosc8 knockdown and/or Kit overexpression . Sheet 4: Figure 7C Kit MFI and percentage of erythroid precursor cells in the R1 , R2 , R3 , R4 and R5 population 48 hr post-Exosc8 knockdown in cells infected with increasing amounts of a Kit-expressing retrovirus . DOI: http://dx . doi . org/10 . 7554/eLife . 17877 . 019 We evaluated the relationship between the level of Kit required to oppose erythroid maturation , caused by exosome complex disruption , and endogenous Kit expression . Cells were infected with a range of retrovirus concentrations to establish the amount required to restore Kit to the endogenous level after Exosc8 downregulation . Exosc8 downregulation reduced Kit MFI six fold ( p=0 . 001 ) . At 48 hr post-infection , 50 μl of Kit retroviral supernatant ( half of that used in the time-course ) increased Kit cell surface expression in Exosc8-knockdown cells to approximately the endogenous level ( 1 . 2 fold higher than control ) . Under these conditions , infection of control cells increased Kit cell surface expression three fold ( p=0 . 002 ) . Exosc8 downregulation , without ectopic Kit expression , increased the percentage of R3 cells ~two fold ( p=0 . 003 ) 48 hr post-infection . The increased R3 population cells induced by Exosc8 downregulation was prevented by enforced Kit expression at 1 . 2 fold higher than the endogenous level ( control R3 , 19 . 7% versus Exosc8/Kit R3 , 16 . 5% ) . A three fold increase in Kit cell surface expression in the control condition decreased the R3 population cells from 19 . 7% to 6% ( p=0 . 001 ) ( Figure 7C ) . As enforced Kit expression negated erythroid maturation induced by exosome complex disruption , the exosome complex activity to establish and/or maintain Kit expression is a critical step in the exosome complex-dependent maturation process . Regulating the proliferation and differentiation balance of stem and progenitor cells through intrinsic and extrinsic mechanisms ensures normal development and physiology . Whereas exaggerated proliferation of stem and/or progenitor cells can underlie cancer ( He et al . , 2004 ) , a differentiation bias can exhaust the precursor cells . As the intestinal epithelial layer is replaced every 4-5 days , the proliferation versus differentiation balance of intestinal stem cells must be exquisitely regulated ( Clevers et al . , 2014 ) . Wnt signaling promotes intestinal stem cell self-renewal , while Bone Morphogenetic Protein ( BMP ) signaling suppresses self-renewal and promotes differentiation ( Clevers et al . , 2014; He et al . , 2004 ) . In hematopoiesis , SCF signaling promotes erythroid precursor cell proliferation at the expense of differentiation ( Haas et al . , 2015; Muta et al . , 1995 ) . Epo provides a pro-differentiation stimulus ( Wu et al . , 1995b ) . During stress erythropoiesis , Epo , SCF and glucocorticoids synergistically promote erythroid cell expansion ( Kolbus et al . , 2003; Wessely et al . , 1997 ) and Kit downregulation promotes maturation ( Haas et al . , 2015; Munugalavadla et al . , 2005 ) . Epo and Kit can synergize ( Joneja et al . , 1997; Sui et al . , 1998; Wu et al . , 1995a ) or be antagonistic ( Haas et al . , 2015; Joneja et al . , 1997; Kosmider et al . , 2009 ) . Kit signaling induces Epo receptor tyrosine phosphorylation ( Wu et al . , 1995a ) , although synergistic enhancement of proliferation may involve convergence of SCF and Epo signals on ERK1/2 via distinct pathways ( Sui et al . , 1998 ) . Constitutively active Kit inhibits Epo-mediated Akt phosphorylation , increasing apoptosis in mature erythroid cells ( Haas et al . , 2015 ) . GATA-1 and the cooperating transcription factor Scl/TAL1 repress Kit transcription as erythroblasts acquire Epo-dependence ( Munugalavadla et al . , 2005; Vitelli et al . , 2000 ) . While multiple pathways are implicated in controlling the proliferation and differentiation balance of stem and progenitor cells , many questions remain regarding how pathways intersect or function in parallel . Herein , we demonstrated that the exosome complex is a critical determinant of the proliferation versus differentiation balance of erythroid precursor cells , and many unanswered questions existed regarding how this balance is controlled ( Figure 8 ) . The exosome complex establishes the balance by ensuring expression of a receptor tyrosine kinase , Kit , which is essential for proliferation , under conditions in which progenitors are not competent to transduce Epo pro-differentiation signals . Disrupting exosome complex integrity downregulated Kit and SCF-mediated proliferation signaling , while inducing Epo signaling ( Figure 8 ) . Physiologically , GATA-1 represses expression of exosome complex subunits , instigating the developmental signaling transition . Reduced SCF signaling upon exosome complex disruption would be predicted to underlie or contribute to activation of Epo signaling in erythroid precursor cells ( Figures 4 , 5 ) and promote erythroid differentiation ( Figures 3 , 7 ) . Our results establish a paradigm in which an RNA-regulatory machine orchestrates the balance between opposing developmental signaling pathways . Considering the apparent ubiquitous expression of the exosome complex , it is attractive to propose that the paradigm may operate to control additional progenitor cell transitions . 10 . 7554/eLife . 17877 . 020Figure 8 . Exosome complex function to orchestrate developmental signaling pathways that control proliferation versus differentiation . The master regulator of erythropoiesis GATA-1 represses Kit transcription and upregulates EpoR transcription , thus establishing the developmental signaling circuitry for erythroid maturation . GATA-1 represses genes encoding exosome complex subunits , which promotes erythroid maturation . The exosome complex confers Kit expression and establishes competence for SCF-induced Kit signaling . Disruption of this mechanism abrogates Kit signaling and instigates Epo signaling , which favors erythroid precursor maturation versus self-renewal . DOI: http://dx . doi . org/10 . 7554/eLife . 17877 . 020 Mice homozygous for the W mutation ( white spotting ) in Kit die perinatally from severe macrocytic anemia ( Bernstein et al . , 1990; Waskow et al . , 2004 ) . The anemia likely reflects the Kit requirement for HSPC genesis , maintenance and function ( Bernstein et al . , 1990; Ding et al . , 2012 ) . Our analyses with primary fetal liver hematopoietic precursors cultured under conditions that support erythroid precursor growth and differentiation demonstrated that Exosc8 downregulation reduces Kit expression and signaling in erythroid precursors . These findings would not have been predicted from prior work with W mutant multipotent hematopoietic precursors in vivo , given the impact of Kit signaling on multiple cell types . Exosome complex functions have expanded considerably since the discovery of its role in rRNA maturation ( Mitchell et al . , 1996 ) to include coding and non-coding RNA degradation and processing ( Kilchert et al . , 2016; Schneider et al . , 2012 ) . The exosome complex is enriched at actively transcribed genes ( Andrulis et al . , 2002; Lim et al . , 2013 ) , regulates transcription start site usage ( Jenks et al . , 2008; Kuehner and Brow , 2008 ) , influences heterochromatin and post-transcriptional gene silencing ( Eberle et al . , 2015 ) and regulates superenhancer activity ( Pefanis et al . , 2015 ) . Gene promoters are bidirectional in nature ( Andersson et al . , 2015 ) , and the exosome complex rapidly degrades PROMPTs ( Lubas et al . , 2011; Preker et al . , 2011; Preker et al . , 2008 ) . Depletion of exosome complex components stabilizes PROMPTs and additional cryptic transcripts present at gene loci , including PSARs ( promoter-associated small RNAs ) , TSARs ( terminator-associated small RNAs ) , PALRs ( promoter-associated long RNAs ) , TSSa RNAs ( transcription start site-associated RNAs ) and eRNAs ( enhancer RNAs ) ( Jensen et al . , 2013; Lubas et al . , 2015; Pefanis et al . , 2015 ) . The exosome complex is also implicated in transcriptional regulation , and since regulatory RNAs can be integral components of transcriptional mechanisms ( Belostotsky , 2009; Jensen et al . , 2013 ) , this can be explained , in part , through RNA-regulatory activity . While knowledge of molecular mechanisms underlying exosome complex function are sophisticated , many questions remain unanswered regarding its functions in distinct cellular contexts during development , physiology and pathologies . Genes encoding exosome complex components are linked to pathologies including cancer ( DIS3 ) ( Chapman et al . , 2011 ) , immune disorders ( EXOSC9 and EXOSC10 ) ( Allmang et al . , 1999b ) and congenital neurological disorders ( EXOSC3 and EXOSC8 ) ( Boczonadi et al . , 2014; Rudnik-Schoneborn et al . , 2013; Wan et al . , 2012 ) . Despite these human disease links , the apparent diversity of RNA targets and multiple functions of the exosome complex , little is known about its role in cell fate decisions . Downregulating EXOSC10 , EXOSC9 or EXOSC7 expression caused differentiation of human epididymal progenitor cells . Mechanistically , the exosome complex represses GRHL3 , a key transcription factor promoting epidermal differentiation ( Mistry et al . , 2012 ) . In summary , we have described the fundamental importance of the exosome complex to orchestrate a critical developmental signaling transition that determines whether erythroid precursors differentiate . Given the importance of ascribing biological functions for regulatory RNAs , the machinery mediating their biosynthesis and exosome complex functions to process diverse RNAs , it will be instructive to identify exosome complex-regulated RNA ensembles mediating cell fate transitions . Furthermore , we anticipate that the results described herein can be leveraged to improve strategies for the industrial-level generation of erythroid cells to achieve therapeutic goals . Finally , considering the common occurrence of Kit-activating mutations in cancers ( Lennartsson and Ronnstrand , 2006 ) , it will be of considerable interest to analyze the paradigm described herein in physiological states in vivo , to determine if it can be extrapolated to cancer , and to establish whether the exosome complex constitutes a promising therapeutic target for Kit-driven pathologies . Protein structure coordinate files for the human exosome complex ( Liu et al . , 2006 ) were downloaded from the Research Collaboratory for Structural Bioinformatics Protein Data Bank ( www . RCSB . org , accession number 2NN6 ) . Images were generated using PyMOL ( www . PyMOL . org , Schrödinger , New York , NY ) . Primary erythroid precursors were isolated from E14 . 5 mouse C57BL/6 fetal livers using EasySep negative selection Mouse Hematopoietic Progenitor Cell Enrichment Kit ( StemCell Technologies , Vancouver , Canada ) as described ( McIver et al . , 2014 ) . G1E-ER-GATA-1 ( RRID:CVCL_D047 ) cells were cultured in Iscove’s Modified Dulbecco’s Medium ( IMDM ) ( ThermoFisher , Waltham , MA ) containing 15% FBS ( Gemini , West Sacramento , CA ) , 1% antibiotic/antimycotic ( Corning , Tewksbury , MA ) , 2 U/ml erythropoietin ( Amgen , Thousand Oaks , CA ) , 120 nM monothioglycerol ( Sigma , St Louis , MO ) , 0 . 6% conditioned medium from an SCF producing CHO cell line , and 1 μg/ml puromycin ( Gemini ) ( Fujiwara et al . , 2009 ) . G1E ( RRID:CVCL_D046 cells were cultured without puromycin . G1E cells were derived from GATA-1-null murine ES cells . ES cells were cultured under conditions that promoted the development of definitive erythroid cells ( Weiss et al . , 1997 ) . G1E-ER-GATA-1 cells are G1E cells stably expressing GATA-1 fused to the ligand-binding domain of human estrogen receptor ( Gregory et al . , 1999 ) . G1E and G1E-ER-GATA-1 cells were a kind gift from Dr . Mitchell J . Weiss ( St . Judes ) . Fetal liver erythroid precursors cells were cultured in StemPro-34 ( ThermoFisher ) with 1x nutrient supplement ( ThermoFisher ) , 2 mM glutamax ( ThermoFisher ) , 1% penicillin-streptomycin ( ThermoFisher ) , 100 μM monothioglycerol ( Sigma ) , 1 μM dexamethasone ( Sigma ) , 0 . 5 U/ml of erythropoietin , and 1% conditioned medium from a kit ligand producing CHO cell line . Cells were cultured in a humidified incubator at 37°C ( 5% carbon dioxide ) . The vectors expressing MiR-30 context luciferase , murine Exosc8 ( Rrp43 ) , Exosc9 ( Rrp45 ) shRNAs were described ( McIver et al . , 2014 ) . MiR-30 context shExosc3 1 sequence: TGCTGTTGACAGTGAGCGAACTGGCAGAGAGTTGACATATTAGTGAAGCCACAGATGTAATATGTCAACTCTCTGCCAGTCTGCCTACTGCCTCGGA . MiR-30 context shExosc3 2 sequence: TGCTGTTGACAGTGAGCGCAAGACCATTCAGCAGACGTTATAGTGAAGCCACAGATGTATAACGTCTGCTGAATGGTCTTATGCCTACTGCCTCGGA . Bold sequences denote sense and antisense sequences . Wild type Kit expression vector was a kind gift from Ruben Kapur . G1E-ER-GATA-1 cells ( 5 x 105 ) , and primary murine fetal liver erythroid precursors ( 2 × 105 ) were spinfected ( McIver et al . , 2014 ) . 48 hr post-infection 20 × 106 G1E-ER-GATA1 cells were harvested and resuspended in 500 μl NP40 lysis buffer ( 50 mM Tris pH 8 ( Sigma ) , 150 mM NaCl ( Sigma ) , 5 mM DTT ( Sigma ) , 0 . 5% NP40 ( Sigma ) supplemented with protease inhibitors ( 0 . 2mM PMSF ( Sigma ) , 20μg/ml leupeptin ( Roche , Basel , Switzerland ) ) , 50 μg/ml RNase A ( Sigma ) and 2 μl/ml DNase I ( ThermoFisher ) . Cells were lysed on ice for 30min and insoluble cell debris were removed by centrifugation . After taking an input sample ( 25 μl ) the lysate was pre-cleared with 20 μl protein-A Agarose ( Sigma ) and 5 μl rabbit pre-immune sera for 1 hr at 4°C . After pelleting the protein-A Agarose , the supernatant was incubated with 5 μg rabbit IgG or 5 μg rabbit anti-Exosc3 ( Abcam , San Francisco CA , Ab156683 ) overnight at 4°C , before addition of a 15 μl protein-A Agarose pellet for a further 2 hr . The protein-A Agarose pellet was washed once with NP40 lysis buffer and three times with TEG buffer ( 10 mM Tris pH 7 . 6 , 50 mM NaCl , 4 mM EDTA ( Sigma ) , 5 mM DTT , 10% Glycerol ( Sigma ) ) . The protein-A Agarose was resuspended in 50 μl of 2x SDS lysis buffer ( 25 mM Tris pH 6 . 8 , 6% SDS ( Sigma ) , 4% β-mercaptoethanol ( Sigma ) , 10% glycerol , 0 . 02% bromophenol blue ( Sigma ) ) , 25 μl of 2x SDS lysis buffer was added to the input sample , and incubated at 100°C for 5 min . Proteins were resolved on an 11% SDS-PAGE gel and Exosc2 ( Abcam ab156698 ) was measured by semi-quantitative Western blotting with ECL Plus ( ThermoFisher ) . Total RNA was purified with Trizol ( ThermoFisher ) . cDNA was prepared by annealing 1 μg or 0 . 2 μg ( sorted samples ) of RNA with 250 ng of a 1:5 mixture of random hexamer and oligo ( dT ) primers ( Eurofins , Louisville , KY ) heated at 68°C for 10 min . This was followed by incubation with Murine Moloney Leukemia Virus Reverse Transcriptase ( ThermoFisher ) with 10 mM DTT , RNasin ( Promega , Madison , Wi ) , and 0 . 5 mM dNTPs at 42°C for 1 hr . The mixture was diluted to a final volume of 100 μl and heat inactivated at 95°C for 5 min . cDNA was analyzed in reactions ( 20 μl ) containing 2 μl of cDNA , appropriate primers ( Eurofins ) , and 10 μl of SYBR green master mix ( ThermoFisher ) . Product accumulation was monitored by SYBR green fluorescence . A standard curve of serial dilutions of cDNA samples was used to determine relative expression . mRNA levels were normalized to 18S rRNA . Primer sequences are found in Supplementary file 1 . For quantitation of cell surface markers , 1 × 106 cells were stained in 100 μl PBS/10% FBS ( Gemini ) with anti-mouse Ter119-APC ( RRID:AB_469474 ) ( 1:100 ) , CD71-PE ( RRID:AB_465741 ) ( 1:100 ) and Kit-PEcy7 ( RRID: AB_469644 ) ( 1:100 ) ( eBioscience , San Diego , CA ) , at 4°C for 30 min in the dark . To quantitate apoptosis after CD71/Ter119 staining , cells were washed in Annexin V Buffer ( 10 mM HEPES ( Sigma ) , 140 mM NaCl , 2 . 5 mM CaCl2 ( Sigma ) , pH 7 . 4 ) then stained with Annexin V-Pacific blue ( ThermoFisher ) ( 1:40 ) and DRAQ7 ( Abcam ) ( 1:100 ) for 20 min in the dark at room temperature . To detect intracellular phosphorylated Akt , erythroid precursor cells were expanded for 48 hr and sorted into Ter119+ and Ter119- populations using magnetic beads ( StemCell Technologies ) . Cells were serum-starved in 1% BSA ( Sigma ) in IMDM for 1 hr at 37°C before stimulation with either 10 ng/ml SCF ( Merk Millipore , Billerica , MA ) or 2 U/ml Epo ( Amgen ) for 10 min and fixed in 2% paraformaldehyde ( Alfa Aesar , Ward Hill , MA ) for 10 min at 37°C . After permeabilization overnight at -20°C in 95% methanol ( ThermoFisher ) cells were incubated for 1 hr in HBSS ( ThermoFisher ) /4% FBS at 4°C . Cells were stained with rabbit phospho-Akt or rabbit phospho-ERK ( 1:200 ) ( Cell Signaling , Danvers , MA ) for 30 min before incubation in goat anti rabbit-APC ( 1:200 ) ( Jackson ImmunoResearch , West Grove , PA ) , Kit-PEcy7 ( 1:100 ) and CD71-PE ( 1:100 ) for 30 min at room temperature . To analyze cell cycle of erythroid precursor cells , fetal liver erythroid precursors were expanded for 24 or 72 hr , and Ter119- cells were isolated . Cells were fixed/permeabilized in 70% ethanol and stained with 5 μg/ml DAPI ( Biolegend , San Diego , CA ) overnight at -20°C . Cells were washed twice in PBS before analysis . Samples were analyzed using a BD LSR II ( BD Biosciences , San Jose , CA ) or sorted into distinct populations using a BD FACSAria II . DAPI or DRAQ7 ( Abcam ) were used for apoptotic analyses , and for fixed cells , Zombie UV ( Biolegend ) staining discriminated dead cells . R1 cells were FACS-sorted 24 hr post-Exosc8-knockdown , and 5000 cells were plated in duplicate in Methocult M3434 ( StemCell Technologies ) according to the manufacturer’s instructions . CFU-E and BFU-E colonies were quantitated after culturing for 2 and 8 days , respectively , at 37°C with 5% CO2 24 hr post-Exosc8 knockdown , Ter119+ cells were depleted from the primary erythroid precursor samples . Equal numbers of cells were boiled for 10 min in SDS lysis buffer . Proteins were resolved by SDS-PAGE and incubated with rabbit anti-Kit ( Cell Signaling , D13A2 RRID:AB_1147633 ) , rabbit anti-GATA-2 or rat anti-GATA-1 ( Santa Cruz , Dallas TX , sc-265 RRID:AB_627663 ) . Primary fetal liver erythroid precursor cells ( 2 × 106 ) were crosslinked with 1% formaldehyde ( Sigma ) for 10 min . Lysates were immunoprecipitated with antibodies against phospho-Ser5 Pol II ( Covance , Princeton , NJ , H14 MMS-134R RRID:AB_10063994 ) or GATA-1 ( Grass et al . , 2006 ) using rabbit pre-immune serum or control IgM as control . For Exsoc9 ChIP , cells ( 8 × 106 ) were crosslinked , and lysates were immunoprecipitated with anti-Exosc9 antibody ( Novus Biologicals , Littleton , CO , NBP1-71702 RRID:AB_11026964 ) using control IgG as control . DNA was quantitated by real-time PCR with SYBR green fluorescence . Primers sequences used to assess protein occupancy are indicated in Supplementary file 2 . A Students T-test was used to compare experimental and control samples . When comparing multiple groups , ANOVA was conducted to identify any significant variance between samples , followed by a Tukey-Kramer test to identify statistical relationships between each pair of samples within the experiment . All analysis was conducted using JMP software ( SAS Institute Inc . Cary , NC ) . Asterisks indicate significance relative to control , *p<0 . 05 , **p<0 . 01 and ***p<0 . 001 .

Red blood cells supply an animal’s tissues with the oxygen they need to survive . These cells circulate for a certain amount of time before they die . To replenish the red blood cells that are lost , first a protein called stem cell factor ( SCF ) instructs stem cells and precursor cells to proliferate , and a second protein , known as erythropoietin , then signals to these cells to differentiate into mature red blood cells . It is important to maintain this balance between these two processes because too much proliferation can lead to cancer while too much differentiation will exhaust the supply of stem cells . Previous work has shown that a collection of proteins called the exosome complex can block steps leading towards mature red blood cells . The exosome complex controls several processes within cells by modifying or degrading a variety of messenger RNAs , the molecules that serve as intermediates between DNA and protein . However , it was not clear how the exosome complex sets up the differentiation block and whether it is somehow connected to the signaling from SCF and erythropoietin . McIver et al . set out to address this issue by isolating precursor cells with the potential to become red blood cells from mouse fetal livers and experimentally reducing the levels of the exosome complex . The experiments showed that these cells were no longer able to respond when treated with SCF in culture , whereas the control cells responded as normal . Further experiments showed that cells with less of the exosome complex also made less of a protein named Kit . Normally , SCF interacts with Kit to instruct cells to multiply . Lastly , although the experimental cells could no longer respond to these proliferation signals , they could react to erythropoietin , which promotes differentiation . Thus , normal levels of the exosome complex keep the delicate balance between proliferation and differentiation , which is crucial to the development of red blood cells . In future , it will be important to study the exosome complex in living mice and in human cells , and to see whether it also controls other signaling pathways . Furthermore , it is worth exploring whether this new knowledge can help efforts to produce red blood cells on an industrial scale , which could then be used to treat patients with conditions such as anemia .

2016

Exosome complex orchestrates developmental signaling to balance proliferation and differentiation during erythropoiesis

Alterations involving serine-threonine phosphatase PP2A subunits occur in a range of human cancers , and partial loss of PP2A function contributes to cell transformation . Displacement of regulatory B subunits by the SV40 Small T antigen ( ST ) or mutation/deletion of PP2A subunits alters the abundance and types of PP2A complexes in cells , leading to transformation . Here , we show that ST not only displaces common PP2A B subunits but also promotes A-C subunit interactions with alternative B subunits ( B’’’ , striatins ) that are components of the Striatin-interacting phosphatase and kinase ( STRIPAK ) complex . We found that STRN4 , a member of STRIPAK , is associated with ST and is required for ST-PP2A-induced cell transformation . ST recruitment of STRIPAK facilitates PP2A-mediated dephosphorylation of MAP4K4 and induces cell transformation through the activation of the Hippo pathway effector YAP1 . These observations identify an unanticipated role of MAP4K4 in transformation and show that the STRIPAK complex regulates PP2A specificity and activity . Protein phosphorylation plays a regulatory role in nearly all biological processes and dysregulation of protein phosphorylation contributes to many diseases . Both kinases and phosphatases have been implicated in the pathogenesis of specific cancers , and several small molecule kinase inhibitors are standard treatments in such cancers . In addition , several phosphatases have been identified as tumor suppressors ( Sablina and Hahn , 2007; Lawrence et al . , 2014 ) . PP2A , an abundant serine/threonine phosphatase in mammalian cells , is comprised of three subunits: A ( structural ) , B ( regulatory ) , and C ( catalytic ) . The A and C subunits form the core enzyme and interact with different B regulatory subunits to create many distinct PP2A enzymes ( Pallas et al . , 1990; Chen et al . , 2007; Cho et al . , 2007; Shi , 2009; Sents et al . , 2013 ) . Moreover , there are two A and two C isoforms , and at least four classes of B subunits B , B’ , B’’ , and B’’’ ( striatins ) , each of which exist as several different isoforms . Although the prevailing view is that the B subunits provide substrate specificity , how B subunits accomplish this regulation remains unclear ( Shi , 2009; Hertz et al . , 2016 ) . Genome characterization studies of human cancers have identified recurrent mutations and deletions involving PP2A subunits . Indeed , the PP2A Aα ( PPP2R1A ) subunit ranks among the most recurrently mutated gene across many cancer types ( Lawrence et al . , 2014 ) . Notably , mutations in Aα occur at high frequency in premalignant endometrial lesions ( Anglesio et al . , 2017 ) . PP2A is also a target of the Small T antigens ( ST ) of SV40 and other polyomaviruses including the human oncogenic Merkel cell polyomavirus ( Pallas et al . , 1990; Chen et al . , 2004; Cheng et al . , 2017 ) , and this interaction contributes to cell transformation ( Hahn et al . , 2002 ) . Structural studies have shown that ST disrupts the formation of a functional PP2A holoenzyme by displacing or hindering B subunit access to the PP2A core-enzyme ( Chen et al . , 2007; Cho et al . , 2007 ) . However , ST has a lower binding affinity in vitro for the PP2A core enzyme than B’ subunits , which suggests that ST interaction with the core enzyme may either occur prior to the B subunit binding or ST directly inhibits PP2A activity independently of subunit assembly ( Chen et al . , 2007 ) . Several investigators have used mass spectrometry to identify proteins that interact with PP2A ( Goudreault et al . , 2009; Herzog et al . , 2012 ) . These studies identified a large complex called the Striatin-interacting phosphatase and kinase ( STRIPAK ) complex ( Goudreault et al . , 2009 ) . The STRIPAK complex contains striatin family ( STRN ) proteins , several kinases , scaffolding proteins , and PP2A subunits . Indeed , striatins were initially described as non-canonical PP2A regulatory subunits ( B’’’ subunits ) ( Moreno et al . , 2000 ) . STRIPAK complexes have also been shown to associate with members of the GCKIII kinase subfamily ( MST3 , STK24 , and STK25 ) ( Kean et al . , 2011 ) . In addition , mitogen-activated protein kinase kinase kinase kinase 4 ( MAP4K4 ) , a Ste20-like kinase , although not an obligate member of the STRIPAK complex , associates with STRIPAK ( Frost et al . , 2012; Herzog et al . , 2012; Hyodo et al . , 2012 ) . We also identified members of the STRIPAK complex , including STRN3 , STRN4 , STRIP1 , and MAP4K4 in complex with SV40 ST ( Rozenblatt-Rosen et al . , 2012 ) . Although STRIPAK comprises multiple signaling enzymes , it is unclear how disruptions to the biochemical complex integrate with or disrupt phosphorylation cascades; or whether these signaling alterations synergize with ST to mediate cellular transformation . MAP4K4 is a serine/threonine kinase that was initially found to activate the c-Jun N-terminal kinase ( JNK ) signaling pathway ( Yao et al . , 1999 ) , downstream of TNF-α . MAP4K4 has also been implicated in a large number of biological processes including insulin resistance , focal adhesion disassembly , as well as cellular invasion and migration ( Collins et al . , 2006; Tang et al . , 2006; Yue et al . , 2014; Danai et al . , 2015; Vitorino et al . , 2015 ) . Recent studies have shown that MAP4K4 phosphorylates LATS1/2 , activating the Hippo tumor suppressor pathway , leading to YAP1 inactivation ( Mohseni et al . , 2014; Meng et al . , 2015; Zheng et al . , 2015 ) . Here , we investigated the role of the STRIPAK complex and MAP4K4 in human cell transformation driven by SV40 ST and found that kinase inactivation or partial suppression of MAP4K4 replace the expression of ST in the transformation of human cells . Human embryonic kidney ( HEK ) epithelial cells expressing SV40 Large T antigen ( LT ) , the telomerase catalytic subunit ( hTERT ) , and oncogenic HRAS ( referred to as HEK TER hereafter ) have served as a useful model system to identify pathways and protein complexes that can functionally substitute for SV40 ST in promoting transformation , including partial depletion of PP2A ( Chen et al . , 2004; Sablina et al . , 2010 ) . These cells , upon expression of SV40 ST or partial knockdown of PP2A Aα or Cα subunits , become tumorigenic ( Hahn et al . , 2002; Chen et al . , 2004 ) . Prior studies have shown that expression of ST , or partial inhibition of certain PP2A subunits , causes increased phosphorylation of PP2A substrates ( Sablina and Hahn , 2008; Sablina et al . , 2010 ) . To assess the serine/threonine phosphorylation events that are associated with transformation induced by ST or by partial knockdown of PP2A , we performed global Isobaric Tags for Relative and Absolute Quantitation ( iTRAQ ) phosphoproteomic profiling of HEK TER cells expressing ST ( HEK TER ST ) or in which expression of the PP2A Aα , Cα , or B56γ subunits were fully or partially suppressed using previously characterized shRNAs ( Figure 1A , Figure 1—figure supplement 1A–B ) ( Sablina et al . , 2010 ) . We also confirmed that these genetic perturbations promoted the transformation phenotype as gauged by anchorage-independent ( AI ) growth assays as previously described ( Figure 1—figure supplement 1C ) ( Sablina et al . , 2010 ) . Through mass spectrometry analysis of the phosphopeptides altered across these conditions , we identified 6025 phosphopeptides corresponding to 2428 individual proteins reproducibly detected in two replicate experiments . Processing and normalization of the raw data were performed using corresponding control experiments ( GFP control for ST , shLuciferase ( shLuc ) control for shRNAs against PP2A , see methods for details ) . We then performed comparative marker selection analysis ( Gould et al . , 2006 ) to identify candidate phosphoproteins that were most significantly correlated with the transformation phenotype ( Figure 1A ) . In consonance with previous studies ( Ratcliffe et al . , 2000; Kuo et al . , 2008 ) , we observed an increase in phosphorylation of direct or indirect targets of PP2A , including AKT1S and β-catenin ( CTNNB1 ) in cells which were transformed by either expressing ST or partial knockdown of PP2A Cα subunit in HEK TER cells ( Figure 1B ) ( Sablina et al . , 2010 ) . Conversely , we also observed decreased phosphorylation on multiple proteins in cells transformed by ST or by PP2A perturbation ( B56γ1 , Cα2 ) . Notably , the phosphorylation signature for transformation included four distinct sites on MAP4K4 ( T804 , S888 , S889 , S1272 , p<0 . 05 , Figure 1B ) . Our previous systematic analysis of SV40 ST identified MAP4K4 , in addition to PP2A and other STRIPAK components in the same complex ( STRN4 , STRN3 , CTTNBP2NL , FAM40A , MAP2K3 , STK24 , PPP2R1A ) ( Rozenblatt-Rosen et al . , 2012 ) . To confirm these interactions , we generated lentiviral C-terminal Flag-HA Tandem Affinity Purification ( CTAP ) constructs for SV40 ST as well as ST from three closely related Human Polyoma Viruses ( HPyV ) including JCPyV-CY , JCPyV-Mad1 , and BKPyV , along with GFP as a negative control . We introduced these viral proteins into HCT116 cells and performed HA-tag immunoprecipitations ( IP ) from lysates of cells expressing ST from the respective viruses . We confirmed co-complex formation between SV40 ST and MAP4K4 , as well as with STRIPAK components PP2A C , STRN3 , and STRIP1 ( Figure 1—figure supplement 1D ) . We observed that ST of JCPyV and BKPyV , the two most closely related HPyVs to SV40 , also interacted with STRN3 , STRIP1 , and PP2A C but not MAP4K4 , indicating that the interaction of SV40 ST and MAP4K4 was unique to SV40 ST . The association of ST with B’’’ subunits ( striatins ) was unexpected , because ST was previously reported to primarily bind PP2A Aα and displace most B subunits ( Pallas et al . , 1990; Chen et al . , 2007; Cho et al . , 2007; Sablina et al . , 2010 ) . These observations raised the possibility that ST modulates MAP4K4 phosphorylation via PP2A activity associated with the STRIPAK complex . To determine if MAP4K4 and other SV40 ST interacting proteins participated in cell transformation , we created and stably expressed two distinct shRNAs targeting each of several SV40 ST interacting proteins , including STRN3 , STRN4 , STRIP1 , MARCKS , MAP4K4 , and STK24 in HEK TER cells . We then assessed the ability of each of these shRNAs to promote AI growth , a readout for the transformed phenotype ( Figure 2A , Figure 2—figure supplement 1A ) . As expected , expression of SV40 ST or partial knockdown of PP2A Cα subunit in HEK TER cells induced robust AI growth ( Figure 2—figure supplement 1A ) . Among the STRIPAK components , we found that one of the two shRNAs targeting MAP4K4 ( shMAP4K4-82 ) elicited a potent transformation phenotype ( Figure 2B–C , Figure 2—figure supplement 1A ) . To ensure that the observed phenotype was specific to targeting MAP4K4 and not due to an off-target effect of this shRNA , we repeated the AI growth assay using eight different MAP4K4-targeting shRNAs including the two shRNAs used in the initial experiment ( Figure 2—figure supplement 1B ) . In addition , we found that the three shRNAs which promoted HEK TER cells to grow in an AI manner ( shRNA-82 , 92 , 93 ) only partially suppressed MAP4K4 levels ( Figure 2—figure supplement 1B ) . Specifically , we focused on shMAP4K4-82 , which promoted the most robust AI growth and knocked down MAP4K4 mRNA levels by 50% ( Figure 2—figure supplement 1B–C ) . In contrast , none of the shRNAs that induced more than 50% knockdown of MAP4K4 expression resulted in AI growth ( Figure 2—figure supplement 1B ) . This relationship between partial knockdown and cell transformation is similar to what has been reported for the knockdown of PP2A Aα and Cα subunits ( Chen et al . , 2005; Sablina et al . , 2010 ) . To further confirm these data in vivo , we performed xenograft experiments to assess tumor formation by subcutaneous injection of immunodeficient mice . Consistent with the in vitro studies , shMAP4K4-82 induced potent tumor formation when compared to the shLuc control ( Figure 2D–E ) . These observations suggest that partial knockdown , but not full depletion , of MAP4K4 , promotes both transformation and tumor formation . To understand the mechanism by which the ST/MAP4K4 axis contributes to cell transformation , we first assessed changes in interactions between MAP4K4 and its binding partners upon ST expression . Specifically , we stably expressed NTAP-MAP4K4 in HEK TER cells expressing either ST or GFP as a negative control . We used Stable Isotope Labeling with Amino Acids ( SILAC ) to encode proteins in each condition ( Figure 3A ) . We found that MAP4K4 interacted with STRIPAK components , including STRIP1 , STRN3 , STRN4 , and the PP2A Aα subunit . The interactions between MAP4K4 and the STRIPAK components were increased by 3–4-fold in cells expressing ST relative to the GFP control ( Figure 3B ) . We tested a series of ST mutants ( R21A , W147A , F148A , P132A ) that are unable to bind to PP2A Aα ( Cho et al . , 2007 ) in 293 T cells , and found that these mutant ST proteins were unable to interact with STRN3 , a core component of the STRIPAK complex ( Figure 3—figure supplement 1A ) , demonstrating that this interaction is dependent on ST binding to PP2A Aα subunit . These observations indicated that MAP4K4 interaction with STRIPAK is enhanced in cells expressing SV40 ST . To corroborate these observations , we performed IP of endogenous STRN3 , STRN4 , STRIP1 , and MAP4K4 and compared the interactions of components of the STRIPAK complex with MAP4K4 in HEK TER cells expressing either ST or GFP . Consistent with the proteomic results , we observed that the interaction of MAP4K4 with the STRIPAK complex was significantly enhanced in the presence of ST ( Figure 3C ) . We also performed these experiments in normal human fibroblasts ( IMR90 ) expressing ST or GFP ( negative control ) and confirmed the enhanced binding of MAP4K4 to STRN4 and STRIP1 ( Figure 3—figure supplement 1B ) . These observations indicate that interactions between MAP4K4 and STRIPAK components , including STRIP1 , STRN3 , and STRN4 are enhanced in the presence of SV40 ST . We next analyzed the enriched phosphopeptides from affinity-purified MAP4K4 ( Figure 3A ) to better interrogate the full phosphorylation landscape on the kinase . In two independent experiments , we quantified 17 MAP4K4 phosphorylation sites ( Figure 3D ) . The majority of these sites exhibited reduced phosphorylation in cells expressing ST . These findings further demonstrate that ST mediates dephosphorylation of several distinct MAP4K4 sites . To evaluate if MAP4K4 dephosphorylation is mediated by the STRIPAK complex , we isolated STRN4 from cells expressing ST or GFP ( Figure 3—figure supplement 1C ) and measured PP2A-specific dephosphorylation activity using synthetic phosphopeptides encompassing MAP4K4 sites S771;S775 , S876 , or S1251 . We selected these sites because they exhibited the largest change in phosphorylation upon ST expression ( Figure 3D ) . As a control , we treated parallel samples with okadaic acid ( OA ) , a potent and specific PP2A inhibitor . As expected , we observed that OA treatment eliminated phosphatase activity under all conditions ( Figure 3E ) . In contrast , co-incubation of MAP4K4 phosphopeptides with STRN4 immune complexes from ST-expressing cells led to dephosphorylation of the S771/S775 and S876 phosphopeptides by greater than twofold compared to GFP control , while we found a modest but reproducible increase of dephosphorylation of the S1251 site ( Figure 3E ) . These observations suggest that ST promotes PP2A-mediated dephosphorylation of MAP4K4 in the STRN4 complex . Since MAP4K4 phosphorylation at several sites was substantially attenuated in the presence of ST , we assessed whether this decrease in MAP4K4 phosphorylation affected MAP4K4 activity by performing an in vitro kinase assay using tandem-affinity purified MAP4K4 from cells that expressed ST or GFP control . We found that the activity of MAP4K4 was reduced in ST-expressing cells compared to cells that expressed GFP control ( Figure 4A , Figure 4—figure supplement 1A ) . To assess the relevance of MAP4K4 kinase activity to the transformation phenotype , we tested the consequences of pharmacological or genetic inhibition of MAP4K4 on AI growth . Specifically , we treated HEK TER cells with a previously described small molecule inhibitor of MAP4K4 ( compound 29 ) ( Crawford et al . , 2014 ) over a range of concentrations ( 0–2 μM ) and assessed MAP4K4 activity ( Figure 4—figure supplement 1B ) and AI cell growth ( Figure 4B ) . In consonance with what we observed with partially knocked down MAP4K4 expression , escalating doses of this MAP4K4 inhibitor led to an increase in the number of AI colonies until it reached 2 μM when MAP4K4 kinase activity was inhibited more than 90% as measured by in vitro kinase assays ( 2 μM , Figure 4B , Figure 4—figure supplement 1A ) . We found that compound 29 induced modest effects on cell proliferation over the range of tested concentrations ( Figure 4—figure supplement 1C ) . Consistent with the results from the genetic experiments ( Figure 2B–C ) , we observed that partial inhibition of MAP4K4 activity led to increased AI growth . We also tested whether inhibiting MAP4K4 by expressing a loss-of-function MAP4K4 allele promoted transformation . The kinasedead MAP4K4 K54R allele has previously been demonstrated to act as a dominant interfering mutant ( Wang et al . , 2013 ) . We created HEK TER cells stably expressing kinase-dead ( K54R ) or the wild-type ( WT ) version of MAP4K4 and confirmed the loss of kinase activity for the MAP4K4 mutant allele ( Figure 4—figure supplement 1D–E ) . When we performed AI growth assays , we observed that the introduction of MAP4K4 K54R but not WT MAP4K4 induced cell transformation ( Figure 4C ) . Together , these observations demonstrate that partial depletion or inhibition of MAP4K4 activity mimics ST in inducing transformation and that attenuation of MAP4K4 kinase activity is associated with ST-induced cell transformation . Reduction of MAP4K4 levels and activity was sufficient to drive transformation in the absence of ST; therefore , we also investigated whether members of the STRIPAK complex were required for ST-mediated oncogenic transformation ( Figure 5A ) . Specifically , we assessed the consequences of depleting components of STRIPAK in HER TER ST cell and found that knockdown of STRN4 led to a significant reduction in transformation ( Figure 5B–C ) . We tested 4 STRN4-targeting shRNAs and observed reduction in AI colonies in a manner that significantly correlated with the degree of STRN4 knockdown ( Figure 5—figure supplement 1A–B ) . To confirm that these findings were not due to an off-target effect of RNAi , we created a STRN4 allele ( STRN4-58R ) resistant to the STRN4-specific shRNA ( shSTRN4-58 ) and expressed this in HEK TER ST cells ( Figure 5—figure supplement 1C ) . We found that expression of this STRN4 allele rescued the effects of suppressing STRN4 on AI growth ( Figure 5D ) . We also deleted STRN4 using CRISPR-Cas9 gene editing and further confirmed that STRN4 expression was required for ST-induced cell transformation ( Figure 5—figure supplement 1D ) . We assessed the consequences of knocking down STRN4 in vivo and found that STRN4 knockdown significantly reduced tumor formation of HEK TER ST cells ( Figure 5E–F ) . Collectively , these observations demonstrate that STRN4 is required for ST-mediated transformation and tumor formation . To assess whether ST modulates interactions involving STRN4 , we isolated endogenous STRN4 from cells expressing either ST or a GFP control and performed a proteomic analysis of associated proteins ( Figure 6A ) . We found that STRN4 interactions with MAP4K4 were increased 1 . 6-fold , while interactions with the PP2A C subunits did not change , in cells expressing SV40 ST relative to GFP control ( Figure 6B , Figure 6—figure supplement 1A ) . Because ST promoted the interaction of MAP4K4 with STRN4 , we evaluated the role of STRN4 in organizing the STRIPAK complex . When we assessed the impact of knocking down STRN4 on the STRIPAK complex in HEK TER ST cells by Co-IP of endogenous STRIP1 , STRN4 , and MAP4K4 ( Figure 6C , Figure 6—figure supplement 1B ) with or without STRN4 suppression , we observed that interactions of MAP4K4 with other members of STRIPAK ( STRIP1 , PP2A C ) were attenuated when STRN4 was suppressed , indicating that STRN4 is required for MAP4K4 interactions with the STRIPAK complex . Prior studies have shown that Striatins act as scaffolds in the STRIPAK complex ( Chen et al . , 2014a ) . Based on these observations , we hypothesized that depletion of STRN4 in the presence of ST would lead to dissociation of MAP4K4 from the STRIPAK complex , which in turn would increase MAP4K4 activity . To test this hypothesis , we performed an in vitro kinase assay using MAP4K4 isolated from HEK TER ST cells expressing either control or STRN4-specific shRNA . We observed a modest , but statistically significant ( p<0 . 05 ) increase in MAP4K4 kinase activity when STRN4 was suppressed ( Figure 6D , Figure 6—figure supplement 1C ) . Moreover , we found that co-knockdown of MAP4K4 and STRN4 , rescued the cells from the inhibitory effect of shSTRN4 knockdown on AI growth ( Figure 6E ) . In addition , we also observed that the expression of the dominant inhibitory K54R mutant , but not wild type MAP4K4 , was able to restore the ability of these cells to form AI colonies upon STRN4 suppression ( Figure 6—figure supplement 1D ) . These observations suggest that ST inhibits MAP4K4 activity through STRN4 and the STRIPAK complex to induce transformation . To identify downstream signaling pathways affected during transformation by partial knockdown of MAP4K4 expression , we performed transcriptomic profiling of HEK TER cells expressing either shMAP4K4-82 , which induced transformation in vitro ( Figure 2B–C ) , or a control shRNA targeting luciferase . We then performed a Single-sample Gene Set Enrichment Analysis ( ssGSEA ) ( Barbie et al . , 2009 ) with the MAP4K4 knockdown gene expression signature and observed that several independent YAP1 genesets from the literature as well as two curated YAP1 and TAZ genesets from Ingenuity Pathway Analysis ( IPA ) were significantly associated with MAP4K4 knockdown ( Figure 7A ) ( Zhao et al . , 2007; Yu et al . , 2012; Hiemer et al . , 2015; Martin et al . , 2018 ) . We also observed that phosphorylation of YAP1 at S127 , a critical , negative regulatory site that blocks nuclear import of YAP1 ( Zhao et al . , 2007 ) , was decreased upon partial knockdown of MAP4K4 or expression of the MAP4K4 K54R construct in HEK TER cells ( Figure 7B ) . Consistent with prior reports on regulation of LATS1/2 by MAP4K4 ( Mohseni et al . , 2014; Meng et al . , 2015; Zheng et al . , 2015 ) , we also found that partial knockdown of MAP4K4 led to attenuation of p-LATS1 ( Figure 7C ) . In addition , we observed that mRNA and protein levels of CTGF and CYR61 , established markers of YAP1 activity , were increased upon knockdown of MAP4K4 ( Figure 7D–F ) . These observations showed that partial knockdown of MAP4K4 at levels that induce cell transformation also led to increased YAP1 activity . In contrast , we found that suppression of STRN4 in HEK TER ST cells led to an increase in pYAP1 ( Figure 7G ) , consistent with the change in MAP4K4 activity upon STRN4 knockdown ( Figure 6D ) . To evaluate the role of YAP1 in transformation induced by attenuation of MAP4K4 , we suppressed both MAP4K4 and YAP1 and tested AI colony formation ( Figure 8A ) . We found that although knockdown of MAP4K4 sufficed to promote transformation , we observed a three-fold decrease in AI colony growth when MAP4K4 was co-suppressed with YAP1 relative to an shRNA targeting luciferase as a control , indicating that transformation following partial knockdown of MAP4K4 depends on YAP1 ( Figure 8A , Figure 8—figure supplement 1A ) . To further investigate the involvement of YAP1 activity in transformation , we tested whether expression of a constitutively active YAP1 phospho-mutant allele ( 5SA ) ( Zhao et al . , 2007 ) rescued transformation when STRN4 was knocked down ( Figure 8—figure supplement 1B ) . We found that reduced levels of AI growth induced by STRN4 knockdown was rescued by expression of the wild-type or phospho-mutant YAP1 ( Figure 8B , Figure 8—figure supplement 1B ) . These observations show that the expression of YAP1 or YAP1 5SA overrides the requirement for STRN4 in transformation . To extend these observations beyond the HEK TER cells , we generated a MAP4K4 knockdown gene expression signature and assessed this signature across a large collection of cancer cell lines from the Cancer Cell Line Encyclopedia ( CCLE ) by performing ssGSEA analysis ( Figure 8C–D ) ( Barbie et al . , 2009; Barretina et al . , 2012 ) . Using the resulting Enrichment Scores ( ES ) derived from the MAP4K4 knockdown signature , we calculated information-theoretic measure , the Information Coefficient ( IC ) ( Kim et al . , 2016 ) to examine genesets that best matched the MAP4K4 knockdown signature ES across these cancer cell lines . In consonance with the findings in isogenic experiments ( Figure 7A ) , we observed that the MAP4K4 knockdown signature associated significantly with a number of YAP1 genesets derived from the literature as well as those we have generated by ectopic expression of wild-type or mutant YAP1 in immortalized human mammary epithelial cell ( YAP1 UP , YAP1 mt UP ) ( Figure 8C ) ( p-value<0 . 0001 ) ( Zhao et al . , 2008; Cordenonsi et al . , 2011; Yu et al . , 2012; Hiemer et al . , 2015; Feng et al . , 2019 ) . Furthermore , when we compared the MAP4K4 knockdown signature with gene dependency data from Project Achilles , a large-scale project involving genome-scale loss-of-function fitness screens performed in hundreds of cancer cell lines ( Aguirre et al . , 2016; Meyers et al . , 2017; Tsherniak et al . , 2017 ) , we observed significant association with dependency profiles of TAZ1 and TEAD1 ( IC = −0 . 447 , –0 . 512 , p-value=0 . 0001 , 0 . 0001 , respectively ) , which are both Hippo pathway effector molecules ( Figure 8D ) . These findings indicated that the MAP4K4 knockdown signature associated with dependencies in the Hippo/YAP1 pathway . We recently showed that systematically evaluating patterns of genetic co-dependencies across a dataset identify genes with similar function ( Pan et al . , 2018 ) . We used this same approach to examine the MAP4K4 dependency profile . The MAP4K4 dependency profile quantitatively reflected the relative effect of targeting MAP4K4 on cell proliferation/survival of 416 cell lines and is represented as ‘dependency scores’ on cell proliferation/survival during MAP4K4 inhibition ( Figure 8E ) . To assess genes that share dependency profiles with MAP4K4 , we performed an orthogonal analysis using IC-based associations to identify a group of genes whose dependency profiles were most significantly associated with MAP4K4 dependency . Consistent with the known role of MAP4K4 in regulating the Hippo pathway , we found that a number of genes whose dependency profiles were most significantly associated with those of MAP4K4 belonged to the Hippo pathway , such as LATS2 , PTPN14 , and NF2 ( Figure 8E ) ( top 25 among the 18 , 375 dependency profiles ) ( p=0 . 00001 ) . We also observed that CCM3 ( PDCD10 ) , a member of the STRIPAK complex ( Goudreault et al . , 2009 ) , was the top most significantly associated gene dependency with MAP4K4 , further supporting a link between MAP4K4 , Hippo , and the STRIPAK complex ( Figure 8E ) . These observations suggest that the gene expression associated with MAP4K4 knockdown is observed in many cancer cell lines and correlates with the Hippo signaling pathway . Several lines of evidence now implicate the disruption of specific PP2A complexes and alteration of substrate specificity by mutation , deletion , or expression of polyomavirus ST as the basis for PP2A-mediated tumor suppressor activity . These observations have led to a model in which cancer associated PP2A mutations or ST alter the composition of PP2A complexes in cells , thus altering PP2A activity toward specific substrates . However , since purified PP2A exhibits phosphatase activity towards a broad set of substrates , the mechanisms that regulate PP2A substrate specificity in cells remains incompletely understood ( Yang et al . , 1991 ) . Here , we show that STRIPAK regulates the interaction of PP2A with one substrate MAP4K4 that participates in PP2A-dependent cell transformation . These observations provide a mechanism by which phosphatase activity is regulated . Previous studies had shown that most B subunits were displaced by ST from the core enzyme and could not be detected in complex with ST ( Pallas et al . , 1990; Chen et al . , 2007; Cho et al . , 2007; Sablina et al . , 2010 ) . However , our proteomic analysis revealed that ST was bound to B’’’ subunits ( striatins ) , as well as several other STRIPAK components . Here , we evaluated whether suppressing expression of these STRIPAK components impacted ST-induced cell transformation . We found that ST expression induced increased interactions of MAP4K4 with the STRIPAK complex , which in turn reduced levels of MAP4K4 phosphorylation and activity , thus leading to increased YAP1 activity . Prior studies have connected STRIPAK with components of the Hippo pathway ( Couzens et al . , 2013 ) , and MAP4K4 has been shown to directly activate LATS1/2 kinases ( Meng et al . , 2015 ) . In addition , YAP1 has been shown to be required for SV40 ST-mediated transformation ( Nguyen et al . , 2014 ) . Recent work has shown mouse polyomavirus middle T affects YAP1 by directly binding to YAP1 and suppressing its degradation ( Hwang et al . , 2014; Rouleau et al . , 2016 ) . We propose a model for cell transformation induced by PP2A-mediated dephosphorylation of YAP1 in Figure 8—figure supplement 1C . We have shown that partial knockdown of MAP4K4 levels or inhibition of its kinase activity replaces ST in cell transformation , suggesting that MAP4K4 is a key PP2A substrate necessary for cell transformation . This observation is similar to our prior work that shows that only partial , but not complete , knockdown of PP2A Aα and Cα subunits leads to transformation ( Chen et al . , 2004; Chen et al . , 2005; Sablina et al . , 2010 ) . However , we also note that the observed effects of suppressing MAP4K4 leads to a greater increase in AI growth when compared to PP2A knockdown . We speculate that this may be due to the large repertoire of PP2A substrates that may have both pro-tumorigenic , as well as anti-tumorigenic activities . Likewise , MAP4K4 has been associated with a number of different pathways and biological processes ( e . g . , invasion , metabolism , TNF-α ) and therefore , full depletion of MAP4K4 may impact other processes that are important for transformation . Although we observed increased interactions of MAP4K4 with the STRIPAK complex in cells expressing ST , only a subset of cellular MAP4K4 interacts with STRIPAK in this context ( Figure 3C ) , further supporting the notion that MAP4K4 unbound to the STRIPAK complex may have pro-tumorigenic roles . These observations also suggest that the STRIPAK complex plays a key role in regulating PP2A activity toward specific substrates and support a model in which ST in part induces transformation by promoting interactions of the STRIPAK complex with MAP4K4 and thereby attenuating MAP4K4 kinase activity , which in turn leads to the activation of YAP1 . The mechanism by which different PP2A complexes achieve substrate specificity has long remained elusive . Recent work has shown that proteins that harbor a conserved LxxIxE motif promote interactions with B56 subunits and facilitate subsequent PP2A substrate specificity ( Hertz et al . , 2016 ) , suggesting that the substrate specificity may be achieved in part through specific interactions achieved by interactions with distinct B subunits . These findings reinforce the notion that STRIPAK serves as an organizing scaffold to bring substrates such as MAP4K4 to the PP2A complex . Indeed , recent studies have shown that MST3 , a member of the STRIPAK complex , and Ste20 kinase family member MINK1 are also substrates of the STRIPAK complex ( Gordon et al . , 2011; Hyodo et al . , 2012 ) . It will be of interest to see if these proteins also affect transformation phenotypes in other contexts . We found that the PP2A A-C complex continues to interact with non-canonical B’’’ subunits in the presence of ST . This observation confirms prior work that showed that both STRN and STRN3 binding do not overlap with canonical B subunit binding to Aα ( Moreno et al . , 2000 ) . Furthermore , ST has been shown to be unable to compete with and displace B subunits from interacting with the PP2A core enzyme ( Chen et al . , 2007 ) . Indeed , early observations involving biochemical characterization of the PP2A–ST complex showed that even in the absence of canonical B subunits , PP2A bound to ST dephosphorylated histone H1 , suggesting that ST may alter the substrate specificity of PP2A ( Kamibayashi et al . , 1994 ) . Here , we provide further evidence that ST alters substrate specificity by promoting MAP4K4 interaction with the STRIPAK complex . It is unclear if ST binding to PP2A Aα promotes active conformational changes that increase PP2A A-C subunit affinity for STRN4 , or if there is competition among the canonical and non-canonical B subunits to engage the PP2A core enzyme complex . However , it appears that ST interactions with STRIPAK are dependent on Aα , as ST mutants that failed to bind to Aα were also unable to bind to STRIPAK ( Figure 3—figure supplement 1A ) . In addition , it was recently shown that disruption of interactions between PP2A core enzyme and canonical B subunits by mutations in PP2A Aα ( P179R , R18G ) promotes PP2A interactions with members of the STRIPAK complex ( Haesen et al . , 2016 ) , reinforcing the notion that ST phenocopies the effect of cancer-associated PP2A mutations . We also observed that the total PP2A C subunit interactions with STRN4 did not change with ST expression ( Figures 3C and 6B–C , Figure 3—figure supplement 1B–C ) . This finding suggests that the increased STRIPAK activity induced by ST may not be simply due to redistribution of PP2A C subunits into the STRIPAK complex but may also elicit specific changes within the STRIPAK complex . More generally , these observations suggest that striatins act as key regulators of PP2A that impart substrate specificity . MAP4K4 is less well characterized than other members of the MAPK family but has been implicated in a number of biological processes including invasion , insulin resistance , and immunity ( Collins et al . , 2006; Tang et al . , 2006; Huang et al . , 2014; Danai et al . , 2015; Vitorino et al . , 2015 ) . Indeed , MAP4K4 has been reported to promote invasion and to act as a downstream component of TNF-α signaling ( Wright et al . , 2003; Crawford et al . , 2014; Gao et al . , 2016 ) . However , others have found evidence that MAP4K4 can also act as a candidate tumor suppressor gene ( Westbrook et al . , 2005 ) , promote apoptosis downstream of SOX2 ( Chen et al . , 2014b; Yang et al . , 2015 ) and serve as a regulator of the Hippo pathway , in part through direct phosphorylation of LATS1/2 , leading to YAP/TAZ inhibition ( Couzens et al . , 2013; Mohseni et al . , 2014; Meng et al . , 2015; Zheng et al . , 2015 ) . YAP1 is a downstream effector of the Hippo pathway and is involved in a number of important cellular processes including organ size control and cell proliferation . When the Hippo pathway is activated by upstream stimuli triggered by cell-cell contact , cell density and detachment , YAP1 is negatively regulated through a cascade of phosphorylation events causing YAP1 to reside in the cytoplasm and remain inactive . Therefore , tight regulation of the phosphorylation and dephosphorylation events that control the Hippo pathway and subsequent YAP1 activity is critical for preserving normal cellular homeostasis . YAP1 has also been shown to play prominent roles in oncogenic transformation , drug resistance and the epithelial-mesenchymal transition ( Hong et al . , 2014; Shao et al . , 2014; Wilson et al . , 2015 ) . YAP1 has also been shown to be required for KRAS and ST-mediated transformation , providing further evidence that YAP1 is critical for cancer development and maintenance ( Hong et al . , 2014; Nguyen et al . , 2014; Shao et al . , 2014 ) . Despite clear evidence for YAP1 in both cancer initiation and progression , few mutations involving YAP1 or other Hippo pathway components have been identified in cancers . Since mutations affecting PP2A subunits are commonly observed in several types of cancer , our observation that certain PP2A complexes can activate YAP1 in the context of ST-mediated transformation suggests that these cancer-associated mutations may also serve , in part , to activate YAP1 . HEK TER cells were generated from human embryonic kidney ( HEK ) cells , which were immortalized by introducing hTERT , SV40 Large-T antigen , and H-RAS G12V ( Hahn et al . , 2002 ) . These cells were cultured in MEM-alpha media ( Gibco ) supplemented with 10% FBS . 293 T cells ( ATCC ) and HCT-116 ( ATCC ) cells were cultured in Dulbecco’s modified Eagle medium ( DMEM ) ( Cellgro ) supplemented with 1% Pen-Strep ( Gibco ) , 1% Glutamax ( Gibco ) and 10% fetal bovine serum ( FBS ) ( Sigma ) . IMR90 cells ( ATCC ) were cultured in DMEM supplemented with 1% Pen-Strep , 1% Glutamax , and 1% non-essential amino acids ( Gibco ) and 15% FBS . For cell line identity confirmation , we utilize a Fluidigm genotyping to assay a set of 96 single nucleotide polymorphism ( SNP ) markers . An overlapping reference set of 42 SNPs was derived from the Affymetrix SNP6 . 0 array Birdseed genotype calls for cell lines also profiled in the Cancer Cell Line Encyclopedia ( CCLE ) project . Fingerprints ( genotypes for those same SNPs ) assayed by the Fluidigm assay for a particular cell line sample are compared to this reference set of SNPs across all CCLE lines , using the GenePattern FPmatching module at http://genepattern . broadinstitute . org/gp/ . For cells that are not part of the CCLE project , we compare all 96 SNP markers from our Fluidigm genotyping to a collection of previously performed Fluidigm assays . In addition , we determine whether pre- and post- experimental manipulation of these samples match . In some cases , we submit cell lines for authentication by short tandem repeat ( STR ) profiling at DDC Medical or ATCC . HEK TER cells expressing SV40ST or suppressed the expression of PP2A Cα , Aα or B56γ subunits were synchronized in serum-free medium for 24 hr , followed by serum stimulation ( 5 min ) and immediately harvested . Experiments were performed on two independent days as replicates . Cell pellets were solubilized by repeated pipetting using in 10 volumes of 7 . 2M guanidine HCl 0 . 1M ammonium bicarbonate . Insoluble material was pelleted for 10 min at 10 , 000 x g and the protein concentration of the supernatants quantified by bicinchoninic acid assay ( Pierce ) . Aliquots corresponding to 50 μg of each sample were transferred to new tubes and the volumes brought to 50 μl using the above solubilization buffer before further processing . Cysteine residues were reduced with 10 mM dithiothreitol ( DTT ) for 30 min at 56°C and alkylated with 22 . 5 mM iodoacetamide for 20 min at room temperature in the dark . The concentration of guanidine HCl was lowered by adding 9 volumes of 0 . 1M ammonium bicarbonate . Samples were digested overnight at 37°C using 10 μg of trypsin ( Promega ) . An additional 10 μg of trypsin was added the following morning and incubated for another 4 hr at 37°C . The resulting tryptic peptide solutions were acidified by adding trifluoroacetic acid ( TFA ) to a final concentration of 1% and desalted on a Waters C18 solid phase extraction plate ( using two consecutive passes ) . Eluted peptides were concentrated in a vacuum concentrator and reconstituted with 30 µL of 0 . 5 M triethylammonium bicarbonate . Each tube of iTRAQ reagent was reconstituted with 70 µL ethanol and added to each peptide solution . The labeling reaction was carried out for 1 hr at room temperature . Labeled peptides were combined in a tube containing 100 µL of 16 . 5 M acetic acid , concentrated by vacuum centrifugation and desalted on a Waters C18 solid phase extraction plate . Magnetic Fe-NTA agarose beads ( 300 µL of a 5% bead suspension ) were prepared as described ( Ficarro et al . , 2009 ) . The beads were added to iTRAQ labeled peptides reconstituted with 80% acetonitrile/0 . 1% TFA at a concentration of 0 . 5 µg/µL . enriched for 30 min at room temperature with end-over-end rotation . After removing the supernatant , beads were washed three times with 400 µL 80% acetonitrile/0 . 1% TFA , and once with 400 µL of 0 . 01% acetic acid . Phosphopeptides were eluted for 5 min at room temperature with 50 µL of 0 . 75M ammonium hydroxide containing 100 mM EDTA . The beads were washed once with 50 µL of water and this wash was combined with the eluate . Eluted phosphopeptides were concentrated to 10 µL by vacuum centrifugation . Ammonium formate ( pH10 ) was added to yield a final concentration of 20 mM . Enriched phosphopeptides were analyzed by multidimensional RP-SAX-RP-MS/MS ( Ficarro et al . , 2009 ) at a depth of 43 fractions on an LTQ-velos mass spectrometer . The spectrometer was operated in data dependent mode where the top 10 most abundant ions in each MS scan were subjected to alternating CAD ( electron multiplier detection , 35% normalized collision energy , q = 0 . 25 ) and HCD ( image current detection , 45% normalized collision energy ) MS/MS scans ( isolation width = 2 . 0 Da ( CAD ) and 2 . 4 Da ( HCD ) , threshold = 20 , 000 ) . Dynamic exclusion was enabled with a repeat count of 1 and exclusion duration of 30 s . ESI voltage was 2 . 2 kV . MS spectra were recalibrated using the background ion ( Si ( CH3 ) 2O ) six at m/z 445 . 12 + /- 0 . 03 and converted into a Mascot generic file format ( . mgf ) using multiplierz scripts ( Askenazi et al . , 2009; Parikh et al . , 2009 ) . CAD and HCD spectra were independently searched using both Mascot ( version 2 . 3 ) and Protein Pilot ( version 4 . 5 ) against three appended databases consisting of: ( i ) human protein sequences ( downloaded from RefSeq on 07/11/2011 ) ; ( ii ) common lab contaminants and ( iv ) a decoy database generated by reversing the sequences from these two databases . For Mascot searches , precursor tolerance was set to 1 Da and product ion tolerance to 0 . 6 Da ( CAD ) or 0 . 02 Da ( HCD ) . We used the default settings specified for CAD and HCD spectra for Protein Pilot searches ( with no precursor tolerance specified ) . Mascot search parameters included trypsin specificity , up to two missed cleavages , fixed carbamidomethylation ( C , +57 Da ) and iTRAQ8plex derivatization ( K and N-terminus ) , variable oxidation ( M , +16 Da ) and phosphorylation ( S , T , Y , +80 Da ) . Protein Pilot search parameters included trypsin specificity , fixed carbamidomethylation ( C , +57 Da ) , peptide level iTRAQ8plex labeling . mgf files corresponding to the 43 RP-SAX-RP MS/MS fractions were individually searched with Mascot and combined into one Excel file before calculating false discovery rate ( FDR ) . Peptide summaries were exported as text files from the Protein Pilot search results and imported into Excel for FDR calculation . Data files were processed to remove i ) peptide spectral matches ( PSMs ) to the reverse database; PSMs to the forward database with an FDR greater than 1 . 0% and iii ) PSMs corresponding to spectrum with no iTRAQ reporter ions . PSMs were then compared across the four search results ( Mascot CAD , Mascot HCD , Pilot CAD and Pilot HCD ) . PSMs with discordant peptide sequences were discarded . Peptide-level phosphorylation sites were selected based on a majority rule across searches for which an ID was made and used to locate protein-level phosphorylation in the SwissProt database ( downloaded 02/06/2013; note that the entry for MAP4K4 ( O95819-2 ) in this database contained a deletion at S627 ) . iTRAQ intensities were summed across all PSMs with peptide sequences overlapping the protein-level phosphorylation site . The screen was performed across two replicates after randomizing the assignment of iTRAQ channel to biological samples . Packaging and envelope plasmids were co-transfected with lentiviral or retroviral expression vectors into 293 T cells using Lipofectamine 2000 ( Life Technologies ) . Two days after transfection , 293 T cell supernatant was clarified with a 0 . 45 μm filter and supplemented with 4 μg/mL polybrene ( Santa Cruz ) before transducing recipient cells . Stable cell lines were generated after selection with 2 μg/mL puromycin ( Sigma ) , 5 μg/mL blasticidin ( Invivogen ) , 500 μg/mL G418 ( Sigma ) and 50 μg/mL hygromycin ( Santa Cruz ) as required by each vector . For MAP4K4 inhibitor experiments , dimethyl sulfoxide ( DMSO ) ( Sigma ) or inhibitor ( compound 29 ) ( Crawford et al . , 2014 ) was used at the indicated concentrations . MAP4K4 cDNA was generated by PCR-based Gateway cloning ( Invitrogen ) from HEK TER cells . NTAP-SV40 ST and NTAP-GFP have been previously described ( Rozenblatt-Rosen et al . , 2012 ) . Mutations in MAP4K4 and SV40 ST were introduced using the QuikChange XL II site-directed mutagenesis kit ( Agilent ) . Lentiviral shRNA constructs were obtained from the Genetic Perturbation Platform ( GPP ) at the Broad Institute ( Cambridge , MA ) ( http://www . broadinstitute . org/rnai/public/ ) . The following clone IDs were used for STRN3: TRCN0000365162 , TRCN0000370206 , STRIP1; TRCN0000164502 , TRCN0000162951 , MARCKS: TRCN0000197145 , TRCN0000029041 , and STK24: TRCN0000000641 , TRCN0000000644 . STRN4: TRCN0000036954 ( shSTRN4-54 ) , TRCN0000036955 ( shSTRN4-55 ) , TRCN0000036957 ( shSTRN4-57 ) , TRCN0000036958 ( shSTRN4-58 ) . Subsequent functional studies were carried out with TRCN0000036958 ( shSTRN4-58 ) and TRCN0000036955 ( shSTRN4-55 ) and as these were further confirmed to have stronger knockdown of STRN4 protein , as well as relatively less off-target effects . STRN4 open-reading frame ( ORF ) construct which is resistant to STRN4 shRNA ( TRCN0000036958 or shSTRN4-58 ) was generated by cloning of the following sequence; GCCCTTGAAGTCGAACCAATTCATGCT , which was obtained from IDT as gblocks gene fragment ( Integrated DNA Technologies ) , into STRN4 wild-type ORF in pdonr223 using Gibson assembly cloning kit ( Cat#2611 , New England Biolabs ) , followed by gateway cloning into pLX304 vectors from the GPP . For knockdown of MAP4K4 , we used clone IDs: TRCN0000220092 , TRCN0000220093 and TRCN0000195258 , TRCN0000219681 , TRCN0000219682 , TRCN0000195121 , TRCN0000199325 . For most of the study , we focused on TRCN0000219682 ( shMAP4K4-82 ) unless otherwise indicated , as described in the main text . For YAP1 shRNA , we used clone ID: TRCN0000107265 . For global phosphoproteomic experiments , we used PP2A shRNAs as previously described ( Sablina and Hahn , 2007 ) . Specifically , shRNAs targeting PP2A Cα , Aα , B56γ subunits were obtained from Genetic Perturbation Platform ( GPP ) with the clone IDs: TRCN0000002483 ( shPP2A Cα1 ) , TRCN0000002484 ( shPP2A Cα2 ) , TRCN0000002494 ( shPP2A B56γ1 ) , TRCN0000002496 ( shPP2A B56γ2 ) and TRCN0000231508 ( shPP2A Aα ) . For STRN4 CRISPR-CAS9-mediated knockout , the lentiCRISPRv2 vector was used [a gift from Feng Zhang ( Addgene plasmid # 52961 ) ( Sanjana et al . , 2014 ) . STRN4-specific sgRNA sequences were obtained from the Avana library ( Doench et al . , 2016 ) and sgRNAs were cloned according to Zhang lab protocols ( http://genome-engineering . org/gecko/wp-content/uploads/2013/12/lentiCRISPRv2-and-lentiGuide-oligo-cloning-protocol . pdf ) . Gateway-compatible cDNA entry clones were transferred from pDONR221 or pDONR223 donor vectors to the respective retro- or lentiviral Gateway destination vectors via Gateway recombinational cloning ( Life Technologies ) . The vectors MSCV-N-terminal-Flag-HA-IRES-PURO ( NTAP ) and MSCV-C-terminal-Flag-HA-IRES-PURO ( CTAP ) , as well as all HPyV cDNAs , have been previously described ( Sowa et al . , 2009; Rozenblatt-Rosen et al . , 2012; Berrios et al . , 2015 ) . Where indicated , untagged constructs were expressed in the CTAP vector with a TAA stop codon to exclude expression of the epitope tag . Wild-type or phospho-mutant YAP1 was cloned into pMSCV puro vector ( Clontech ) to generate pMSCV puro YAP1 WT or 5SA . The following plasmids were obtained from Addgene: pBabe-hygro-hTERT ( plasmid # 1773 ) ( Counter et al . , 1998 ) , pBabe-HcRed-Ras ( plasmid # 10678 ) ( Boehm et al . , 2005 ) , pBabe-neo-large T cDNA ( plasmid # 1780 ) ( Hahn et al . , 2002 ) , pWZL-Blast-ST ( plasmid # 13805 ) ( Chen et al . , 2004 ) , lentiviral packaging plasmid psPAX2 and envelope plasmid pMD2 . G ( plasmid #12260 , #12259 ) , retroviral packaging plasmid pUMVC3 ( plasmid # 8449 ) ( Stewart et al . , 2003 ) , and envelope plasmid pHCMV-AmphoEnv ( plasmid # 15799 ) ( Sena-Esteves et al . , 2004 ) . FLAG-MAP4K4 immuno-precipitates ( Adelmant et al . , 2019 ) were diluted in 100 mM Ammonium Bicarbonate containing 0 . 1% RapiGest ( final concentration ) and reduced with 10 mM DTT for 30 min at 56°C . Reduced cysteine residues were alkylated with 22 . 5 mM iodoacetamide for 20 min in the dark . Proteins were digested with 5 µg of trypsin overnight at 37°C . Tryptic peptides were purified by batch-mode reverse-phase chromatography ( POROS 50R2 , Applied Biosystems ) and subjected to immobilized metal affinity chromatography ( IMAC ) to enrich phosphopeptides as described for the global phosphoproteomics screen . Peptides from the IMAC supernatant were concentrated under vacuum and purified by batch-mode strong cation exchange chromatography ( POROS 50HS ) . Phosphopeptides were analyzed by LC-MS/MS as follow: Phosphopeptides were loaded off-line onto a precolumn ( 4 cm POROS 10R2 ) and eluted with an HPLC gradient ( NanoAcquity UPLC system , Waters; 5–40% B in 45 min; A = 0 . 2 M acetic acid in water , B = 0 . 2 M acetic acid in acetonitrile ) . Peptides were resolved on a self-packed analytical column ( 50 cm Monitor C18 , Column Engineering ) and introduced in the mass spectrometer ( QExactive HF mass spectrometer , Thermo ) equipped with a Digital PicoView electrospray source platform ( New Objective ) . The mass spectrometer was operated in data-dependent mode where the top 10 most abundant ions in each MS scan were subjected to high energy collision induced dissociation ( HCD , 27% normalized collision energy ) and subjected to MS/MS scans ( isolation width = 1 . 5 Da , intensity threshold = 25 . 000 , MS1 resolution: 120 000 ) . Dynamic exclusion was enabled with an exclusion duration of 30 s . ESI voltage was set to 3 . 8 kV . Dedicated MS/MS scans were also included to continuously monitor precursors for two phosphopeptides identified in a previous analysis . Peptides from the supernatant were separated using a 90 min HPLC gradient and analyzed using in the mass spectrometer as described above . MS spectra were recalibrated using the background ion ( Si ( CH3 ) 2O ) six at m/z 445 . 12 + /- 0 . 03 and converted into a Mascot generic file format ( . mgf ) using multiplierz scripts . Spectra were searched using Mascot ( version 2 . 6 ) against three appended databases consisting of: i ) human protein sequences ( downloaded from RefSeq on 06/26/2019 ) ; ii ) common lab contaminants and iii ) a decoy database generated by reversing the sequences from these two databases . Precursor tolerance was set to 20 ppm and product ion tolerance was set to 25 mmu . Search parameters included trypsin specificity , up to two missed cleavages , fixed carbamidomethylation ( C , +57 Da ) and variable oxidation ( M , +16 Da ) and phosphorylation ( S , T , +80 Da ) . Spectra matching to peptides from the reverse database were used to calculate a global false discovery rate and were discarded . The intensity of heavy and light SILAC features was directly retrieved from the mass spectrometry raw files using the multiplierz python environment ( Alexander et al . , 2017 ) . MAP4K4 phosphorylation sites were remapped to isoform 6 ( UniProt accession O95819-6 ) and SILAC intensities were summed for individual sites identified across overlapping peptide sequences . The SILAC intensity ratio representing the relative abundance of phosphorylation sites in cells expressing GFP or ST was normalized to correct for small difference in immunopurified MAP4K4 in the respective samples as measured in the IMAC supernatant . The relative abundance of proteins in the IMAC supernatant was calculated by summing the intensities of the heavy or light features across peptides mapping uniquely to a gene ( Askenazi et al . , 2010 ) . Two independent FLAG-MAP4K4 immunoprecipitations were performed on combined extracts of GFP and ST expressing cells metabolically encoded with heavy and light or light and heavy SILAC labels , respectively . Cell lysates were obtained using lysis buffer ( 150 mM NaCl , 50 mM Tris-HCl , 1 mM EDTA , 0 . 5% NP-40 , 10% glycerol , and protease and phosphatase inhibitor cocktail sets ( Calbiochem ) ) . Immunoprecipitations were performed with protein A/G magnetic beads ( Millipore ) mixed with immunoprecipitation antibodies . After overnight incubation at 4°C , beads were washed with high salt lysis buffer ( containing 300 mM NaCl ) , boiled in SDS sample buffer ( Boston BioProducts ) , resolved by SDS-PAGE ( Criterion TGX precast gels , Bio-Rad ) , transferred to nitrocellulose membranes ( Bio-Rad ) , blocked and incubated with the appropriate primary antibody in TBS-T overnight at 4°C . Detection of proteins was performed with horseradish-peroxidase conjugated secondary antibodies ( Rockland ) , developed using Clarity Western ECL substrate ( Bio-Rad ) , and imaged with a G:BOX Chemi detection system ( Syngene ) . HEK TER cells expressing either SV40 ST or GFP ( 30 × 15 cm diameter plates ) were harvested with lysis buffer ( 20 mM imidazole HCl , 2 mM EDTA , 2 mM EGTA , pH 7 . 0 with 10 ug/mL each of aprotinin , leupeptin , pepstatin , 1 mM benzamidine , and 1 mM PMSF ) . The clarified cell extract was incubated overnight at 4C with 20–100 ug of STRN4 antibodies ( Abcam , ab177155 ) crosslinked to 30 mg protein A agarose beads ( Thermo Scientific ) by dimethyl pimelimidate ( DMP ) . Beads were washed five times with high salt lysis buffer ( containing 300 mM NaCl ) , washed with TBS two times , and then eluted with 0 . 2 M glycine pH 3 and neutralized with 1 M Tris-HCl pH 8 . 0 . Proteins were precipitated with trichloroacetic acid ( 20% final concentration ) overnight at 4C , washed with cold acetone and processed for subsequent MudPIT analysis ( Florens and Washburn , 2006 ) . In brief , TCA-precipitated protein eluates were urea-denatured , reduced , alkylated , and digested with endoproteinase LysC followed by trypsin . The peptide mixtures were loaded onto microcapillary fused silica columns ( 100 um i . d . ) , packed with C18 reverse phase ( Aqua; Phenomenex ) , SCX ( Luna; Phenomenex ) and C18-RP , placed in-line with an Agilent 11000 quaternary pump , and analyzed by a 10-step MudPIT on linear ion traps . MS/MS datasets were searched using ProLuCID against a non-redundant human protein database ( NCBI , 2019-12-03 ) containing 44 , 080 non-redundant human proteins , 426 usual contaminants , as well as the sequences for small and large T antigens from SV40 Macaca mulatta polyomavirus 1 . To estimate false discover rates ( FDRs ) , the amino acid sequence of each non-redundant protein was randomized ( 44 , 521 shuffled proteins ) and added to the search space . Cysteine carboxylation was searched as a static modification , while methionine oxidation was searched dynamically . Peptide/spectrum matches were sorted and selected using DTASelect in combination with an in-house software , swallow , to FDRs at the peptide and protein levels of less than 1% . TAP-purified MAP4K4 eluted in standard lysis buffer with protease and phosphatase inhibitors were added to kinase assay buffer ( 25 mM Tris-HCl pH 7 . 5 , 5 mM β-glycerophosphate , 2 mM dithiothreitol , 0 . 1 mM sodium orthovanadate and 10 mM MgCl2 ) containing 20 μM ATPγS ( Abcam ) and 1 μg of myelin basic protein ( MBP ) ( Sigma ) . Where specified , ATPγS was left out of the reaction as a negative control . Kinase reactions were carried out as previously described ( Allen et al . , 2007 ) . Reactions were carried out at 30°C for 30 min . P-nitrobenzyl mesylate ( PNBM ) ( Abcam ) was then added ( 2 . 5 mM final ) and the reaction was incubated at room temperature for 2 hr , followed by addition of 6x SDS loading buffer , boiling of samples , SDS-PAGE and subsequent immunoblotting for phosphorylated MBP . Relative activity was calculated as the ratio of the band intensities ( measured with ImageJ ) between the thiophosphate ester signal ( phospho-MBP ) and HA signal ( NTAP-MAP4K4 ) . To measure PP2A phosphatase activity , we used a PP2A Immunoprecipitation Phosphatase Assay Kit ( Millipore Sigma , catalog number 17–313 ) . In brief , HEK TER GFP or ST cells were lysed in 20 mM imidazole HCl , 2 mM EDTA , 2 mM EGTA , pH 7 . 0 with 10 μg/mL each of aprotinin , leupeptin , pepstatin , 1 mM benzamidine , and 1 mM PMSF . Two milligrams of the lysates were then immunoprecipitated with 2 μg of anti-STRN4 antibody ( Abcam , ab177155 ) and 40 μl of protein-A-agarose beads at 4°C overnight . Beads were washed three times with lysis buffer followed by the Ser/Thr assay buffer . Phosphatase reactions were then performed in Ser/Thr assay buffer with a final concentration of 750 μM of MAP4K4 phosphopeptides: S771/S775 ( A-A-S-pS-L-N-L-pS-N-G-E-T-E-S-V-K ) , S876 ( L-T-A-N-E-T-Q-pS-A-S-S-T-L-Q-K ) or S1251 ( V-F-F-A-pS-V-R-S ) for 10 min at 30°C . To provide evidence that the immunoprecipitated phosphatase activity is PP2A , we treated parallel immunoprecipitates with 5 nM of okadaic acid ( Cell Signaling , #5934 ) . Dephosphorylation of the phosphopeptide was measured through malachite green phosphate detection at 650 nm . HEK TER AI growth in soft agar was performed as previously described ( Hahn et al . , 2002 ) using 6-well dishes with BactoAgar ( Gibco ) at concentrations of 0 . 3% top and 0 . 6% bottom layers . Wells were fed with top agarose once per week . After 4 to 5 weeks , cells were stained with 0 . 005% crystal violet ( Sigma ) in PBS and colonies were counted . For MAP4K4 inhibitor experiments , dimethyl sulfoxide ( DMSO ) ( Sigma ) or inhibitor ( compound 29 ) ( Crawford et al . , 2014 ) were used at the indicated concentrations in both the bottom and top soft agar layers and included in refeedings . For proliferation assays , cells were seeded in triplicate in 24-well plates ( day 0; 5 × 103 cells per well ) . Cell density was measured by crystal violet assay at intervals after plating as previously described ( Rozenblatt-Rosen et al . , 2012 ) . For in vivo xenograft experiments , 2 × 106 HEK TER ( expressing shLuc or shMAP4K4-82 ) or HEK TER ST ( expressing shLuc or shSTRN4-58 ) cells were subcutaneously injected into the top , left and right flanks of 5 female Taconic NCR-nude ( CrTac:NCr-Foxn1nu ) mice . For the shSTRN4 experiments , we re-engineered the HEK TER cells to express KRAS G12V , because other vectors containing HRAS G12V with various selection markers failed to produce sufficient levels of expression . Tumor volume was assessed via caliper measurement every week by the formula: volume = length x width2 × 0 . 5 . All procedures were performed according to protocols approved by the Institutional Animal Care and Use Committees of the Dana-Farber Cancer Institute . A total of 500 , 000 cells of either HEK TER shLuc or shMAP4K4-82 were seeded in three 15 cm dishes and allowed to grow for 48 hr . Total RNA was extracted using an RNeasy Plus Kit ( Qiagen ) . RNA sequencing libraries were prepared using a NEBNext Ultra Directional RNA Library Prep Kit for Illumina , NEB E7420 . The concentration of each cDNA library was quantified with the KAPA Illumina ABI Quantification Kit ( Kapa Biosystems ) . Libraries were pooled for sequencing using the HiSeq 2500 . Global phosphoproteomic data: Data from iTRAQ experiments were processed by first merging the two replicate datasets , which resulted in 6025 phosphopeptides corresponding to 2428 individual proteins . We then normalized the raw read counts of each sample to the corresponding control experiments ( shRNA against luciferase for shRNA experiments , and GFP for ST experiments ) followed by log2 transformation . The resulting values were further normalized by quantile normalization . We performed comparative marker selection to find phosphorylation changes which are most significantly correlated with cell transformation phenotype using signal-to-ratio statistics after 1000 permutations ( Gould et al . , 2006 ) . The transformation phenotype upon knockdown of PP2A Cα , Aα , B56γ or SV40ST expression was determined via AI growth assay described above . To facilitate direct comparison of the MAP4K4 phosphosites across different proteomic results , all MAP4K4 phosphorylation sites were mapped and compared relative to the sites in isoform 6 of the MAP4K4 protein ( O95819-6 ) Uniprot database ( https://www . uniprot . org ) . Raw mass spectrometry data files from SILAC and iTRAQ are available for free download at ftp://massive . ucsd . edu/MSV000084422/ . MudPIT mass spectrometry data files are available for download at Massive: ftp://massive . ucsd . edu/MSV000084662/ and ProteomeXchange:http://proteomecentral . proteomexchange . org/cgi/GetDataset ? ID=PXD016628 . RNAseq analysis read count was converted to Transcripts Per Million ( TPM ) using Kallisto quant functions ( https://github . com/UCSD-CCAL/ccal ) ( GRCh38 ) . Differential gene expression analysis of samples with MAP4K4 suppression vs . control was performed using mutual information . We also performed ssGSEA analysis of genesets from the literature , MsigDB ( http://software . broadinstitute . org/gsea/msigdb/index . jsp ) , as well as IPA ( https://www . qiagenbioinformatics . com/products/ingenuity-pathway-analysis ) on the samples to obtain enrichment score for each genesets ( Zhao et al . , 2007; Barbie et al . , 2009; Yu et al . , 2012; Hiemer et al . , 2015; Martin et al . , 2018 ) . Using the Information Coefficient ( IC ) ( Kim et al . , 2016 ) , we estimated the degree of association of the phenotype ( shMAP4K4 vs . shLuc ) and their significance to the genesets . ssGSEA and mutual information calculations: The FDRs were computed from empirical p-values using the standard Benjamini-Hochberg procedure . The empirical p-values were obtained from an empirical permutation test where the target profile is randomly permuted to generate a null distribution for the Information Coefficient ( IC ) values . We also generated signatures from these experiments to apply them in the CCLE RNA Seq dataset ( www . broadinstitute . org/ccle ) ( Barretina et al . , 2012 ) using ssGSEA . Using IC , we matched top gene dependencies associated with MAP4K4 knockdown signature score across CCLE using Gene dependency data from Project Achilles data portal using dataset version V3 . 12a ( www . broadinstitute . org/achilles ) ( Aguirre et al . , 2016; Meyers et al . , 2017; Tsherniak et al . , 2017 ) . Statistical analysis: All the student t-tests and p-value calculations were performed using GraphPad Prism software ( https://www . graphpad . com ) . Unless indicated , experiments were performed in triplicates and the Student's t-tests were performed between perturbation and relevant control conditions using triplicates values obtained from each experiment using parametric testing . For experiments presented in Figures 3E , 4A , 6D–E and 8B , data were first normalized to the mean of the controls and resulting mean values for each condition were plotted and error bars were calculated from standard deviation of the values .

Cells maintain a fine balance of signals that promote or counter cell growth and division . Two sets of enzymes – called kinases and phosphatases – contribute to this balance . In general , kinases “switch on” other proteins by tagging them with a phosphate molecule . This process is called phosphorylation . Phosphatases , on the other hand , dephosphorylate these proteins , switching them off . Cancer cells often have mutations that activate kinases to drive cancer growth . The same cells can have mutations that inactivate the phosphatases or reduce their abundance . The roles of phosphatases in cancer are still being studied . One major hurdle in this research is that it is not always clear how they recognize the proteins they dephosphorylate . Protein phosphatase 2A ( or PP2A for short ) is one of the phosphatases that is often mutated or deleted in human cancers . Even just reduced levels of PP2A can promote cancer . Kim , Berrios , Kim , Schade et al . used an experimental trick to decrease the phosphatase activity of PP2A in human cells growing in a dish . Biochemical analysis of these cells showed that , as expected , many proteins were now in their phosphorylated states . Unexpectedly , however , some proteins were dephosphorylated under these conditions . One of these proteins was called MAP4K4 . In the case of MAP4K4 , the dephosphorylated state contributes to the growth of the cancer cell . Kim et al . carried out further genetic and biochemical experiments to show that , in these cells , PP2A and MAP4K4 stay physically connected to one another . This connection was enabled by a group of proteins called the STRIPAK complex . The STRIPAK proteins directed the remaining PP2A towards MAP4K4 . Low levels or activity of PP2A could , therefore , promote cancer in a different way . Taken together , PP2A is not a single phosphatase that always turns proteins off , but rather is a dual switch that turns off some proteins while turning on others . Future experiments will explore to what extent these findings also apply in tumors . Information about how mutations in PP2A affect human cancers could suggest new targets for cancer drugs .

2020

STRIPAK directs PP2A activity toward MAP4K4 to promote oncogenic transformation of human cells

Tendons are extracellular matrix ( ECM ) -rich structures that mediate muscle attachments with the skeleton , but surprisingly little is known about molecular mechanisms of attachment . Individual myofibers and tenocytes in Drosophila interact through integrin ( Itg ) ligands such as Thrombospondin ( Tsp ) , while vertebrate muscles attach to complex ECM fibrils embedded with tenocytes . We show for the first time that a vertebrate thrombospondin , Tsp4b , is essential for muscle attachment and ECM assembly at myotendinous junctions ( MTJs ) . Tsp4b depletion in zebrafish causes muscle detachment upon contraction due to defects in laminin localization and reduced Itg signaling at MTJs . Mutation of its oligomerization domain renders Tsp4b unable to rescue these defects , demonstrating that pentamerization is required for ECM assembly . Furthermore , injected human TSP4 localizes to zebrafish MTJs and rescues muscle detachment and ECM assembly in Tsp4b-deficient embryos . Thus Tsp4 functions as an ECM scaffold at MTJs , with potential therapeutic uses in tendon strengthening and repair . Cellular structure and function depend on dynamic interactions with extracellular matrix ( ECM ) proteins , defects in which cause many diseases such as muscular dystrophies and osteoarthritis ( Emery , 2002; Mayer , 2003; Kanagawa and Toda , 2006; Carmignac and Durbeej , 2012; Maldonado and Nam , 2013 ) . Tendons and ligaments are especially rich in ECM proteins , predominantly laminins ( Lams ) and collagens ( Cols ) ( Hauser et al . , 1995; Kannus , 2000; Kjaer , 2004; Södersten et al . , 2007; Snow and Henry , 2009; Schweitzer et al . , 2010; Aparecida de Aro et al . , 2012; Charvet et al . , 2011 ) . These multimeric proteins assemble into extremely strong fibrillar structures capable of resisting the contractile forces of muscles and enabling movement ( Banos et al . , 2008; Thorsteinsdóttir et al . , 2011; Thorpe et al . , 2013 ) . Muscles interact with these tendon ECM proteins through integrin ( Itg ) heterodimers as well as the dystrophin-associated glycoprotein complex to form attachments at myotendinous junctions ( MTJs ) ( Kannus et al . , 1998; Kardon , 1998; Blake et al . , 2002; Bassett et al . , 2003; Henry et al . , 2005; Carmignac and Durbeej , 2012 ) . While the organization of the ECM at MTJs has been described ( Kardon , 1998; Aparecida de Aro et al . , 2012 ) , the developmental processes underlying its establishment and maintenance are poorly studied . In zebrafish embryos , early MTJs form as epithelial attachments between muscle fibers and ECM at somite boundaries ( Henry et al . , 2005; Snow and Henry , 2009 ) . Initially this ECM is rich in fibronectin ( Fn ) but accumulates other ECM proteins as it matures . Like other vertebrate tendons , these early embryonic MTJs form through transmembrane interactions between ECM proteins with Itgs and the dystroglycan-complex , thereby linking the ECM with the muscle cytoskeleton ( Henry et al . , 2001; Crawford et al . , 2003; Hall et al . , 2007; Câmara-Pereira et al . , 2009; Jacoby et al . , 2009; Goody et al . , 2010; Charvet et al . , 2011 ) Downstream components of Itg signaling such as Paxillin ( Pxn ) , Talin ( Tln ) and Integrin Linked Kinase ( ILK ) are also required for establishment and maintenance of functional MTJs in both zebrafish and mice ( Gheyara et al . , 2007; Conti et al . , 2008; Postel et al . , 2008; Câmara-Pereira et al . , 2009 ) . The arrangement of ECM proteins at MTJs also changes in architecture and composition to withstand the forces exerted by muscle contraction ( Kjaer , 2004; Câmara-Pereira et al . , 2009; Snow and Henry , 2009; Charvet et al . , 2013; Bricard et al . , 2014 ) . Lams , for example , become incorporated into collagen fibrils at MTJs , in contrast to the Lam meshwork that forms around Schwann cells and or Lam fibril networks in endothelia or lung alveolar cells ( Hamill et al . , 2009 ) . At somite boundaries in mice , Lam–Itg interactions are essential for elongation and differentiation of muscle progenitors ( Bajanca et al . , 2006 ) . Lam deposition ( as well as localization of Itgs , FAK , Paxillin ( Pxn ) , and Fn to MTJs ) also depends on Itg signaling and Rho GTPases that regulate actin cytoskeletal dynamics ( Hamill et al . , 2009 ) . Thus , bidirectional Itg signaling is required for MTJ maturation ( Parsons et al . , 2002; Crawford et al . , 2003; Snow et al . , 2008; Snow and Henry , 2009 ) . In zebrafish embryos , this includes a gradual assembly of collagen fibrils between 1–6 days post fertilization into an orthogonal arrangement at myosepta ( Bader et al . , 2009; Charvet et al . , 2011 ) . Mutations in human LAMA2 cause a congenital form of muscular dystrophy called merosin-deficient muscular dystrophy ( Tome et al . , 1994; Kanagawa and Toda , 2006; Carmignac and Durbeej , 2012 ) and COL6 mutations cause Ullrich congenital muscular dystrophy ( Bertini et al . , 2011; Bönnemann , 2011; Grässel and Bauer , 2013; Pan et al . , 2013 ) , highlighting the importance of an appropriately organized muscle ECM . However , mechanisms that control this assembly of the MTJ matrix are largely unknown . In Drosophila , the long splice form of the Itg ligand thrombospondin ( Tsp ) -TspA ( Chanana et al . , 2007; Subramanian et al . , 2007 ) , is a critical calcium-binding ECM protein secreted by tenocytes ( tendon cells ) that binds Itgs ( Position Specific ( PS ) -beta and PS-alpha2 subunits ) on myoblasts . Of the five vertebrate Tsp genes , subclass B including Tsp4 and Tsp5 have been observed in connective tissues associated with the musculoskeletal system ( Hauser et al . , 1995; Tucker et al . , 1995; Hecht et al . , 1998; Jelinsky et al . , 2010; Frolova et al . , 2014 ) and have been shown to function as homo- and hetero-pentamers through a conserved coiled-coil region ( Hauser et al . , 1995; Narouz-Ott et al . , 2000; Södersten et al . , 2006 ) . Human TSP4 levels are highly elevated in patients with Duchenne muscular dystrophy ( DMD ) , α-sarcoglycan deficiency , as well as in cardiac ECM in response to stress , and TSP5 levels increase during joint injury , osteoarthritis and cartilage degradation ( Chen et al . , 2000; Hecht et al . , 1998; Timmons et al . , 2005 ) . Purified Tsp4 and Tsp5 also interact with a multitude of ECM proteins including Lam , Col and Fn in vitro ( Narouz-Ott et al . , 2000; Chen et al . , 2007 ) . Tsp4 facilitates collagen fibril packing in mouse tendons and ECM of the heart ( Frolova et al . , 2014 ) . However , there are no known functions for Tsps during formation of muscle attachments and MTJ development in vertebrates . We have identified a novel zebrafish Tsp , Tsp4b , expressed and localized at all muscle attachment sites in the developing embryo and in adults . We show that zebrafish Tsp4b , in its pentameric form alone , interacts with ECM proteins such as Lam , activates Itg signaling within muscles , and is required for establishment and maintenance of muscle attachments . Furthermore , recombinant human TSP4 is functionally interchangeable with Tsp4b and capable of repairing and strengthening tendons when provided exogenously . Our results reveal a novel mechanism for pentameric Tsp4 in ECM protein assembly and maintenance at MTJs and a potential therapeutic approach for improving tendon strength and repair . Through an in situ expression screen for markers of muscle attachments , we identified a zebrafish tsp4 , tsp4b ( 69% similar to human TSP4; Figure 1—figure supplement 1A–C ) , which is expressed at all muscle attachment sites ( Figure 1—figure supplement 1E , F ) . In zebrafish , two genes share sequence similarity with other vertebrate Tsp4 genes—Tsp4a and Tsp4b ( previously designated as zgc: 111910 , tsp-4a and thbs4 on chromosome 21 ) . Similar to other subclass B Tsps , Tsp4b is predicted to be secreted as a pentamer as it contains a conserved CX2C motif ( CQAC—Cys-Gln-Ala-Cys ) identical to human TSP4 in its hydrophobic coiled-coil oligomerization domain ( CCD ) ( 25/36 residues are identical with human and mouse Tsp4 ) , which is required for inter-subunit disulfide linkage ( Efimov et al . , 1994 ) ( Figure 1—figure supplement 1C ) . In situ analyses revealed tsp4b mRNA throughout the differentiating myotomes of embryonic somites beginning at 16 hr post fertilization ( hpf , Figure 1—figure supplement 1D ) . Expression disappears in myoblasts as they differentiate and by 60 hpf becomes restricted to putative tendon cells near somite boundaries and along the horizontal myoseptum ( arrowheads in Figure 1—figure supplement 1E ) as well as at all muscle attachment sites of the head by 72 hpf ( arrowheads in Figure 1—figure supplement 1F ) . The expression of tsp4b resembles that of tenomodulin ( tnmd ) , a known tenocyte-specific marker ( Figure 1—figure supplement 2 ) , and their relatives are co-expressed in tendons and ligaments in humans ( Docheva et al . , 2005; Jelinsky et al . , 2010 ) . Combined fluorescent localization of tsp4b mRNA and myosin heavy chain ( MHC ) protein revealed that down regulation of tsp4b in myotome occurs abruptly as the wave of muscle differentiation passes medio-laterally through each somite ( Figure 1—figure supplement 3A–H; Devoto et al . , 1996; Henry et al . , 2005 ) . By 60 hpf , tsp4b expression was only detected in putative tenocytes along the somite boundaries , where muscle attachments have been established ( Figure 1—figure supplement 3I–P ) . A polyclonal antibody raised against the unique N-terminus of zebrafish Tsp4b revealed extracellular protein localization around the notochord and medial somite boundaries at 20 hpf ( Figure 1A–C ) where myofibers of the axial musculature first elongate and attach , and this localization progressed laterally as more lateral fibers formed functional attachments at the somite boundary . By 72 hpf , Tsp4b protein was detected at the ends of all larval axial , appendicular , pharyngeal and extraocular muscles ( Figure 1D–L ) . In the cranial region , these include muscle-cartilage attachments , inter-muscular attachments ( between segments of the sternohyoideus [SH] muscle ) and muscle-soft tissue attachments . These results suggest that during initial stages of muscle development in the trunk , myoblasts secrete their own Tsp4b to initiate attachment and MTJ assembly , and at later stages of somite muscle maturation , Tsp4b levels are maintained by secretion from tenocytes in mature MTJs . 10 . 7554/eLife . 02372 . 003Figure 1 . Zebrafish Tsp4b localizes to all muscle attachments . ( A–L ) Whole mount immunostaining of wild type embryos using anti-MHC ( A , D , G , J; green ) and anti-Tsp4b ( B , E , H , K; red ) and merged ( C , F , I , L ) . ( A–C ) 20-22 hpf and ( D–F ) 72 hpf ( lateral view ) trunk showing early Tsp4b localization around notochord and medial somite boundaries ( B and C ) and later at somite boundaries ( E and F ) . ( G–L ) Ventral ( G–I ) and lateral ( J–L ) views of 72 hpf showing Tsp4b at cranial muscle attachments . Abbreviations: AM-Adductor Mandibularis , AH-Adductor Hyoideus , AO-Adductor Operculae , DO-Dilator Operculae , HH-HyoHyal , IH-InterHyal , IMA-InterMandibularis Anterior , IMP-InterMandibularis Posterior , IO-Inferior Oblique , IR-Inferior Rectus , LAP-LevatorArcus Palatini , MR-Medial Rectus , SH-SternoHyoideus . Scale bar = 30 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 02372 . 00310 . 7554/eLife . 02372 . 004Figure 1—figure supplement 1 . Tsp4b is expressed at muscle attachments . ( A ) Phylogenetic tree showing zebrafish tsp4a and tsp4b . ( B ) Sequence and protein domains place Tsp4b in Thrombospondin subclass B . ( C ) Predicted Tsp4b functional domains: an N-terminal region ( epitope aa: 31-236 ) containing the Heparin Binding Domain ( HBD ) and Leucine Zipper Motif ( LZM ) with Coiled-Coil Domain ( CCD ) with CQAS motif ( 261 . . 264 aa ) , an integrin-binding KGD motif in the first Type 3 Calcium Binding Repeats ( TCBR ) domain ( 492-494 aa ) , Type II ( EGF-like ) Repeats ( TER ) and a Thrombospondin C-Terminal Region ( TspCTR ) . Scale bar = 10 aa . ( D–F ) Whole mount in situ for tsp4b mRNA , anterior left , showing expression at: ( D ) 18 hpf ( lateral view ) in somites , ( E ) 72 hpf ( lateral view ) at somite boundaries . ( F ) 72 hpf ( ventral head ) at cranial muscle attachments . Scale bar = 30 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 02372 . 00410 . 7554/eLife . 02372 . 005Figure 1—figure supplement 2 . tsp4b and tnmd are expressed in tenocytes at MTJs . ( A and B ) Wild type embryos at 72 hpf , labeled by fluorescent in situ hybridization and viewed laterally in whole mounts show similar patterns of expression of tsp4b ( A ) and tnmd ( B ) in groups of cells along somite boundaries ( arrowheads ) . Scale bar = 30 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 02372 . 00510 . 7554/eLife . 02372 . 006Figure 1—figure supplement 3 . Tsp4b expression is downregulated in myoblasts as they differentiate . ( A–P ) Embryos triple-labeled with anti-MHC ( A , green ) marking differentiating myoblasts , in situ hybridization for tsp4b mRNA ( B , red ) and DAPI ( C , blue ) . ( A–D ) Lateral views , anterior to the left , of embryos at 18 hpf show that differentiating myoblasts adjacent to the notochord lose tsp4b mRNA while expression is maintained in undifferentiated mesoderm cells , located laterally . At these stages no DAPI positive nuclei are located at somite boundaries where MHC-positive myofibers attach ( arrowheads ) . ( E–H ) Optical sections of the same confocal stacks in dorsal view show absence of tsp4b expression in differentiated medial myofibers . ( I–L ) Lateral views of 60 hpf embryos show tsp4b mRNA localized immediately adjacent to DAPI + nuclei ( dashed white ovals ) of putative tenocytes along somite boundaries ( arrowheads ) . ( M–P ) Optical sections in dorsal view show localized expression of tsp4b to myosepta ( arrowheads ) . Scale bars = 35 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 02372 . 006 Zebrafish embryos injected with antisense morpholino oligonucleotides ( MOs ) ( 0 . 32 ng/embryo ) targeting the translation start site of tsp4b completely lacked Tsp4b protein by 72 hpf , as determined by whole-mount immunostaining with anti-Tsp4b . tsp4b-MO injected embryos ( hereafter referred to as Tsp4b-deficient ) were slightly curved downward , but showed no defects in muscle morphology or swimming ability ( Figure 2A–C ) . However , muscle contractions induced with mild electrical stimulation caused dramatic muscle detachment in tsp4b-deficient embryos ( Figure 2D–F , J ) . While stimulation with repeated 8 millisecond pulses at 30 volts caused the occasional isolated myofiber to detach in 23% ( N = 16/68 ) of wild-type embryos , similar stimulation led to large portions of somites with detached fibers in 76% ( N = 60/79 ) of tsp4b-deficient animals ( Figure 2J ) . Furthermore , this phenotype was dependent both on the strength of stimulation as well as the dose of tsp4b-MO ( Figure 2—figure supplement 1A , B ) . This weakening of muscle attachments was specific to reduction of Tsp4b since it was partially rescued by injection of full-length tsp4b mRNA ( Figure 2K ) . Furthermore , a mosaic distribution of exogenous mRNA restored Tsp4b protein specifically at the attachment sites of rescued myofibers in a dose-dependent manner ( Figure 2G–I , Figure 2—figure supplement 2 ) . 10 . 7554/eLife . 02372 . 007Figure 2 . Tsp4b is required for muscle attachment . Whole mount immunostaining of 36 hpf Tsp4b-deficient embryos using anti-MHC ( A , D , G; green ) and anti-Tsp4b ( B , E , H; red ) and merged ( C , F , I ) . ( A–C ) Injection of 0 . 32 ng tsp4b-MO eliminates Tsp4b protein at 72 hpf but myofibers attach . ( D–F ) Electrical stimulation ( 30 V ) of these larvae causes muscle detachment . ( G–I ) Co-injection of tsp4b RNA ( 80 pg/embryo ) rescues muscle attachment and Tsp4b localization . ( J ) Histogram showing muscle detachment in 76% ( N = 79 ) of stimulated Tsp4b-deficient embryos ( Chi squared test p-value<0 . 001 ) . ( K ) Co-injection of tsp4b mRNA rescues muscle attachment in 67% ( N = 92 ) of stimulated Tsp4b-deficient embryos ( Chi squared test p-value<0 . 001 ) . ( p-value representation legend: significant *<0 . 05 , highly significant **<0 . 01 , extremely significant ***<0 . 001 , extremely significant ****<0 . 0001 ) . Scale bar = 30 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 02372 . 00710 . 7554/eLife . 02372 . 008Figure 2—figure supplement 1 . Tsp4b-deficient muscles show dose-dependent detachment upon stimulation . ( A ) Histogram quantifying the percentage of embryos with at least one detached muscle fiber in wild type ( white bars ) and tsp4b-MO injected ( gray bars ) embryos subjected to mild electrical stimulation at 48 hpf to induce muscle contractions . Tsp4b-deficient embryos were more sensitive to stimulation with 30 V than 20 V , but similar amounts of detachment were observed at higher voltages , in contrast to wild-type embryos ( N = 20 embryos respectively ) . Frequency = 4 Hz; Duration = 8 ms; Delay = 6 ms . ( B ) Higher tsp4b-MO doses cause more frequent detachment . Percentages of embryos with at least one detached myofiber after stimulation when injected with 0 . 32 ng ( 62 . 5%; N = 64 ) or 0 . 16 ng ( 28%; N = 64 ) –which reduces but does not eliminate Tsp4 . ( Chi squared test: p-Value<0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02372 . 00810 . 7554/eLife . 02372 . 009Figure 2—figure supplement 2 . Exogenous tsp4b mRNA rescues Tsp4b localization in a dose-dependent manner . Lateral views , anterior to the left , of somites in 48 hpf embryos stained with anti-MHC and anti-Tsp4b . ( A and B ) Wild types . ( C and D ) Tsp4b-deficient embryos injected with 0 . 32 ng tsp4b-MO show complete loss of Tsp4b protein . ( E and F ) Co-injection with 0 . 01 ng of full-length tsp4b mRNA restores small amounts of Tsp4b protein . ( G and H ) This is more pronounced with 0 . 02 ng of tsp4b mRNA . ( I and J ) 0 . 04 ng of tsp4b mRNA largely restores protein at somite boundaries . ( K ) Histogram shows that the dose-dependent increase in number of embryos with restored Tsp4b protein matches the concentration of tsp4b mRNA coinjected with tsp4b MO ( N = 50 embryos ) . Scale bar = 30 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 02372 . 00910 . 7554/eLife . 02372 . 010Figure 2—figure supplement 3 . Ultrastruture of MTJs in Tsp4b deficient embryos . ( A–D ) MTJ ultrastructure in wild type 72 hpf embryos before ( A ) and after stimulation ( B ) , Tsp4b-deficient embryos before ( C ) and after stimulation ( D ) . Arrowheads–electron-dense basement membrane ( BM ) at MTJs . Scale bar = 0 . 5 μm . ( E ) Histogram depicting the average spacing at MTJs , as measured in TEM images , between the BM of muscle attachments in wild type ( wt ) ( N = 6 ) , wild-type stimulated ( N = 4 ) , tsp4b-MO injected ( N = 5 ) , and tsp4b-MO injected and stimulated ( N = 4 ) . ‘N’ represents the number of somite boundaries analyzed in the embryos ( t test one-tailed , unequal variance p-value <0 . 05 ANOVA posthoc Tukey test: p-value<0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02372 . 010 In order to understand the effects of reducing Tsp4b levels on ECM organization at the MTJ , we examined MTJ ultrastructure with transmission electron microscopy ( TEM ) . While wild-type embryos at 72 hpf showed basement membranes ( BM ) flanking a tightly packed electron dense MTJ ( ∼278 nm , Figure 2—figure supplement 3A , E ) , both stimulated wild-type embryos and Tsp4b-deficient embryos prior to stimulation showed perturbation of the ECM at MTJs with intermittent separations of BM ( ∼423 nm and ∼527 nm , respectively; Figure 2—figure supplement 3B , C , E ) . In contrast , electrically stimulated Tsp4b-deficient embryos showed catastrophic defects in both ECM and BM integrity as well as a dramatic increase in MTJ separation ( ∼2668 nm , Figure 2—figure supplement 3D , E ) , while muscle fibers remained intact . These results demonstrate a requirement for Tsp4b in maintenance and strengthening of MTJs . During development of the axial musculature in zebrafish , early myofibers attach at somite boundaries by 20–24 hpf ( Devoto et al . , 1996 ) , at least a day before tendon progenitors are detectable at somite boundaries by tsp4b mRNA expression ( Figure 1—figure supplement 3 ) . To address cell autonomous roles for Tsp4b in muscle versus tendon , we used cell transplantation to create mosaic embryos in which wild type mesoderm capable of forming either muscles or tenocytes was transplanted into Tsp4b-deficient hosts . Consistent with a requirement for Tsp4b in muscle attachment , both isolated donor-derived , wild-type muscle cells ( Figure 3A–D ) , as well as putative tenocytes along somite boundaries ( Figure 3E–H ) , restored Tsp4b protein and attachment of both donor and adjacent host myofibers ( 70% , N = 73 ) , and these remained attached even when stimulated ( Figure 3I–L; 40% , N = 51 ) . This demonstrates a non cell-autonomous role of Tsp4b in maintenance of muscle attachments . 10 . 7554/eLife . 02372 . 011Figure 3 . Tsp4b has a non cell-autonomous function in muscle attachments . ( A–L ) 48 hpf ( lateral views ) of genetic mosaics generated by cell transplantation . GFP-labeled wild type muscle cells ( green ) ( A and I ) or putative tenocytes ( arrow heads ) ( green ) ( E ) were grafted into Tsp4b-deficient host embryos and stained with anti-Tsp4b ( B , F , J; red ) , and anti-MHC ( C , G , K; gray/white ) . ( I–L ) Transplants locally rescued muscle detachment after stimulation in regions where Tsp4b was restored ( white line denotes region lacking Tsp4b ) . Scale bars = 30 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 02372 . 011 Tsps function as Itg ligands . For example , interactions between vertebrate subclass A Tsps and Itgs are essential for wound healing ( Kyriakides and Bornstein , 2003; Bornstein et al . , 2004 ) . The RGD ( Arg-Gly-Asp ) motif in Drosophila Tsp is required for its interactions with muscle-specific Itg to form stable attachments ( Chanana et al . , 2007; Subramanian et al . , 2007 ) . Zebrafish Tsp4b lacks an RGD but contains a non-canonical Itg-binding ( KGD—Lys-Gly-Asp ) motif in its C-terminus ( Figure 1—figure supplement 1C; Ruoslahti , 1996; Adams , 2004 ) . We hypothesized that secreted Tsp4b in the ECM interacts with Itgs on muscle cell surfaces to promote attachment . To address this we examined effects of depleting Tsp4b on Itg activation and functional interactions with downstream Itg signaling components . Tsp4b-deficient animals showed severe reductions in key indicators of Itg signaling within myofibers . First , in 20–22 hpf embryos injected with tsp4b-MO , levels of activated focal adhesion kinase ( FAK , detected with an antibody that recognizes phosphorylation at Tyrosine 861—pFAKy861 ) were dramatically reduced ( Figure 4A–D , G ) , which was rescued by injection of tsp4b mRNA ( Figure 4E–G; Henry et al . , 2001 ) , and at later stages ( 36 hpf ) were mislocalized or lost in local regions along somite boundaries ( Figure 4H–K ) , which was also rescued by co-injection of tsp4b mRNA ( Figure 4L , M ) . In regions where pFAK was mislocalized along the somite boundaries , muscle attachments appeared disorganized . Second , simultaneous partial reduction of Tsp4b ( sub-threshold doses of tsp4b-MO ) and Itg signaling in heterozygous integrin-linked kinase ( ilk+/− ) mutants , neither of which causes any detachment on its own , led to dramatic and widespread muscle detachment upon stimulation ( Postel et al . , 2008; Figure 4N ) . Third , in Tsp4b-deficient embryos , expression and localization of a GFP-tagged Paxillin ( Pxn ) —Tg[Pxn-GFP] - was significantly reduced , indicating a severe reduction in Itg signaling within the myofibers themselves ( Figure 4—figure supplement 1A–C ) . In addition , a Tg[Itga5-RFP] transgenic line showed specific localization of RFP at MTJs , while Itga6-GFP localized more broadly along the entire length of each muscle fiber , which is reduced in Tsp4b-deficient embryos ( Figure 4—figure supplement 2A–D ) . These results are consistent with a functional role for Tsp4b as a ligand for muscle-specific Itgs at MTJs required to stabilize muscle attachments and point to Itga5 as one likely receptor . 10 . 7554/eLife . 02372 . 012Figure 4 . Tsp4b is required for muscle-specific integrin signaling at MTJ . ( A–F ) Lateral views of 20–24 hpf embryos stained with anti-phosphorylated ( Tyrosine 861 ) FAK ( pFAK; red ) and anti-MHC ( green ) . ( A and B ) pFAK localizes to the ends of early myofibers at wild-type somite boundaries . ( C and D ) Reduced pFAK levels in Tsp4b-deficient embryos . ( E and F ) pFAK levels are restored in Tsp4b-deficient embryos injected with full length tsp4b mRNA . ( G ) Fluorescence intensity measurements ( arbitrary units [A . U . ] ) for pFAK staining along somite boundaries confirm significant reductions in Tsp4b-deficient embryos , and partial rescue by co-injection of full length tsp4b mRNA ( t test: one tailed , unequal variance; p-value: wt and tsp4b-deficient <0 . 05; Tsp4b-deficient and Tsp4b-deficient + tsp4b RNA p<0 . 001 ) . ( H–M ) 36 hpf ( lateral views ) stained with anti-pFAK ( red ) , and anti-MHC ( green ) . Insets show higher magnification images of white boxed areas . ( H and I ) pFAK localizes to muscle sarcolemma . ( J and K ) In Tsp4b-deficient embryos , pFAK is reduced/discontinuous at somite boundaries . pFAK associates with ectopic muscle attachments ( arrowheads ) . ( L and M ) pFAK localization is restored in Tsp4b-deficient embryos injected with full length tsp4b mRNA . ( N ) Embryo percentages ( N = 70 embryos ) with detached muscles from an intercross between two ilk+/− heterozygotes , injected with sub-threshold amounts ( 0 . 16 ng ) of tsp4b-MO and stimulated ( 30 V ) ( Chi squared test; p-value: wt+ tsp4b-MO ( stimulated ) and ilk+tsp4b-MO ( stimulated ) p<0 . 0001 , ilk ( stimulated ) and ilk+tsp4b-MO ( stimulated ) p<0 . 0001 , ilk+ tsp4b-MO and ilk+tsp4b-MO ( stimulated ) p<0 . 0001 ) . Scale bar = 30 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 02372 . 01210 . 7554/eLife . 02372 . 013Figure 4—figure supplement 1 . Tsp4b-deficient muscles show reduced localization of Paxillin . ( A and B ) 20 hpf tg ( pxn-egfp ) embryo showing localization of the Pxn-EGFP fusion protein to the sarcolemma of myotubes at MTJs and reductions in embryos injected with tsp4b MO . ( C ) Fluorescence intensity measurements ( arbitrary units [A . U . ] ) of Pxn-EGFP at somite boundaries confirm a significant reduction in Tsp4b deficient animals . ( t test: one tailed , unequal variance; p-value<0 . 01 ) Scale bar = 30 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 02372 . 01310 . 7554/eLife . 02372 . 014Figure 4—figure supplement 2 . Itga5-RFP localizes to MTJs and Itga6-GFP localizes to muscle , and both require Tsp4b . ( A and B ) Live imaging of embryos showing localization of tg[Itga5-RFP] in wild type ( A ) and Tsp4b-knockdown ( B ) embryos at 36 hpf . ( C and D ) Embryos stained with anti-GFP show localization of Itga6-GFP in wild type ( C ) and reduced localization in Tsp4b-knockdown samples at 36 hpf ( D ) . Scale Bar = 20 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 02372 . 014 In addition to Tsp4 , the myotendinous ECM ( both early zebrafish myosepta and MTJs ) contains Lams , Dystrophin ( Dys ) , Tenascin , Fn , Cols and other Itg ligands . Lam and Dys deficiencies cause muscular dystrophies in humans and zebrafish . lam mutants show defects in myoseptum formation ( lama2 ) and muscle patterning ( lamb1 and lamc1 [Hall et al . , 2007; Peterson and Henry , 2010] ) . Many of these ECM components interact with Itgs , and the defects in Itg activation that we have observed suggest that Tsp4b may regulate assembly of some of these other ECM components ( Narouz-Ott et al . , 2000; Frolova et al . , 2012 ) . To address this hypothesis we examined Lam localization ( with a pan-Lam antibody ) in Tsp4b-deficient embryos . We found that injection of tsp4b-MO caused a reduction in localization of Lam at somite boundaries at 24 hpf ( Figure 5A–D , G , N ) , which was rescued by co-injection of tsp4b mRNA ( Figure 5E–G ) . This correlated with disorganized muscle attachments . At later stages ( 36 hpf ) , Tsp4b-deficient embryos showed discontinuities in Lam distribution at somite boundaries ( Figure 5H–K , N ) , with many fibers forming aberrant attachments at sites of localized Lam , which could be rescued by co-injection of tsp4b mRNA ( Figure 5L , M ) . lam mRNA ( lama2 , lamb1 , lamc1 , lamc2 ) levels were not significantly affected by tsp4b depletion suggesting that these effects are post-transcriptional ( Figure 5—figure supplement 1 ) and most likely due to Tsp4b-Lam interactions . In vitro binding assays using purified rat TSP4 have shown interactions with Lam , Col , and Fn among other ECM components ( Narouz-Ott et al . , 2000 ) . Thus , Tsp4b could interact with Lam in a similar heteromeric complex , and initiate an Itg signaling cascade necessary for maintenance of MTJs . 10 . 7554/eLife . 02372 . 015Figure 5 . Tsp4b is required for Laminin assembly at MTJs . ( A–F ) Lateral views of somites in 20 hpf embryos stained with anti-MHC ( green ) and anti-pan-Laminin ( Lam , red ) antibodies . Insets show higher magnification images of the areas marked by white boxes , where Lam localizes to developing myotendinous ECM ( arrowheads ) . ( A and B ) Wild-type siblings . ( C and D ) tsp4b-MO injected embryos showing local loss of Lam at 20 hpf , and ectopic muscle attachments ( arrowheads ) at remaining Lam foci . ( E and F ) Lam localization was restored in Tsp4b-deficient embryos injected with full length tsp4b mRNA . ( G ) Fluorescence intensity measurements ( arbitrary units [A . U . ] ) at somite boundaries of anti-Lam in 20–24 hpf wild type controls versus embryos injected with tsp4b-MO or co-injected with tsp4b-MO and tsp4b RNA . ( t test: one tailed , unequal variance; p-value: wt and Tsp4b-deficient <0 . 01 , Tsp4b-deficient and Tsp4b-deficient and tsp4b RNA <0 . 05 ) Scale bar = 30 microns . ( H and I ) Lateral views of somites in wild type embryos at 36 hpf stained with anti-pan-Lam ( red ) , and anti-MHC ( green ) . Insets show higher magnification images of white boxed areas . Lam localizes to myotendinous ECM ( I , arrowheads ) . ( J and K ) In Tsp4b-deficient embryos , Lam is reduced/discontinuous at somite boundaries ( K , arrowheads ) . ( L and M ) Lam localization is restored in Tsp4b-deficient embryos injected with full length tsp4b mRNA . ( N ) Embryo percentages ( 20 hpf embryos N = 30 , 72 hpf embryos N = 50 ) with reduced/mislocalized Lam at 20 and 48 hpf in wild type and Tsp4b-deficient embryos . ( Chi squared test; p-values: 20 hpf **<0 . 01 , 48 hpf ***<0 . 001 ) Scale bar = 30 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 02372 . 01510 . 7554/eLife . 02372 . 016Figure 5—figure supplement 1 . Tsp4b depletion does not alter lam transcription . Histogram showing relative fold-changes in quantitative PCR ( qRT-PCR ) assays using cDNA prepared from wild-type and Tsp4b-deficient 36hpf embryos . Tsp4b-deficient embryos show no significant difference in alpha2 ( lama2 ) , beta2 ( lamb2 ) , gamma1 ( lamc1 ) and gamma2 ( lamc2 ) , all of which are expressed in muscle . Samples are normalized against ef1alpha as a basal expression marker . rpl13a serves as an internal control . DOI: http://dx . doi . org/10 . 7554/eLife . 02372 . 016 The reduction of Itg signaling and Lam localization in Tsp4b-deficient embryos suggests that Tsp4b has both a functional role in muscle adhesion and attachment as well as a structural role in ECM assembly at the MTJ . Itg ligands commonly bind Itgs through a canonical RGD motif , which fails to bind when Aspartate ( D ) is replaced with Glutamate ( E ) ( Erb et al . , 2001; Balasubramanian and Kuppuswamy , 2003; Takahashi et al . , 2007 ) . Tsp4 contains a non-canonical KGD binding motif . To test requirements for this motif in promoting Itg activation in muscles , we designed a KGE mutant tsp4b mRNA ( Figure 6A ) . Injecting the mutant form ( KGD>KGE ) of tsp4b mRNA failed to rescue the muscle detachment phenotype in Tsp4b-deficient embryos ( Figure 6B ) , confirming a requirement for Itg binding in the function of Tsp4b . 10 . 7554/eLife . 02372 . 017Figure 6 . Integrin binding and pentamerization of Tsp4b are essential for its function . ( A ) Schematic representation of Tsp4b domains showing the location of the conserved coiled-coil region with its CQAC motif ( 70% identical across species ) . Amino acid substitutions used to study the functions of KGD and CQAC motifs are underlined . ( B ) Muscle detachment frequencies in uninjected wild-type controls , wildtypes injected with tsp4b morpholino ( MO ) , tsp4b RNA , KGE tsp4b mutant RNA , SQAS tsp4b mutant RNA , or co-injected with tsp4b MO and tsp4b RNA , KGE tsp4b mutant RNA , or SQAS tsp4b mutant RNA , after stimulation ( N = 60 embryos each ) . ( C ) Western blot performed on whole embryo protein extract using anti-GFP and anti-FLAG antibodies under non-reducing conditions . Lanes: 1–wt ( AB ) , 2–Tsp4b-deficient embryos injected with tsp4b-GFP mRNA , 3–Tsp4b-deficient embryos injected with SQAS-GFP mRNA , 4–Tsp4b-deficient embryos co-injected with tsp4b-FLAG and SQAS-GFP mRNAs . Pentameric Tsp4b-GFP ( ∼663 kDa ) and pentameric Tsp4b-FLAG ( ∼535 kDa ) bands ( black arrow head ) . Monomeric SQAS-GFP ( ∼132 kDa ) ( grey arrow head ) . The band corresponding to a 100 kDa size marker in all lanes of the blot reacted with anti-GFP is a background signal . DOI: http://dx . doi . org/10 . 7554/eLife . 02372 . 01710 . 7554/eLife . 02372 . 018Figure 6—figure supplement 1 . tsp4b-SQAS-gfp mRNA is expressed similar to wild type tsp4b-gfp mRNA . ( A , D , G ) Live embryos at tail bud ( ∼10hpf ) showing show GFP fluorescence . ( B , E , H ) embryos imaged in bright field . ( C , F , I ) Merged images . ( A–C ) Wild type embryos ( D–F ) Embryos injected with tsp4b-gfp transcript . ( G–I ) Embryos injected with SQAS-gfp mutant transcript . Scale bar = 100 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 02372 . 01810 . 7554/eLife . 02372 . 019Figure 6—figure supplement 2 . The CQAC motif is essential for Tsp4b localization and function . ( A–H ) 48 hpf embryos stained with anti-GFP ( green ) and anti-Tsp4b ( red ) show localization of all Tsp4b ( A ) vs injected Tsp4b-GFP protein ( B ) in wild type embryos . ( C and D ) Tsp4b localizes to the MTJ ( C ) but injected SQAS-GFP does not ( D ) in wild type embryos , ( E and F ) Neither Tsp4b ( E ) nor SQAS-GFP ( F ) localizes to the MTJ in Tsp4b-deficient embryos . ( G and H ) Injected Tsp4b-FLAG ( G ) localizes to the MTJ while SQAS-GFP ( H ) does not in Tsp4b-deficient embryos ( G–H ) . Scale bar = 20 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 02372 . 01910 . 7554/eLife . 02372 . 020Figure 6—figure supplement 3 . Laminin and FAK localization are dependent on pentameric Tsp4b . Fluorescence intensity measurements ( arbitrary units [A . U . ] ) at somite boundaries of anti-pFAK and anti-Lam immunohistochemical staining in wild type and Tsp4b-deficient embryos ( 48hpf ) , co-injected with tsp4b RNA , co-injected with SQAS tsp4b mutant RNA . ( t test: one tailed , unequal variance; p-values * <0 . 05 , ** <0 . 01 , ****<0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02372 . 020 Subclass B Tsps function as pentamers through the highly conserved 36 residue coiled-coil motif ( Figure 6A ) . To test requirements for oligomerization of Tsp4b in its localization we perturbed Tsp4b function by preventing it from forming the functional homo-pentameric form . To do this , we mutated the conserved CCD of Tsp4b , CQAC ( Efimov et al . , 1994; McKenzie et al . , 2006 ) to SQAS ( Ser-Gln-Ala-Ser ) to prevent formation of inter-subunit disulfide bridges that are required for pentamerization . To verify SQAS-gfp mRNA expression , we injected wild type tsp4b-gfp and SQAS-gfp into one-cell stage embryos and analyzed them for expression at gastrula stages . Both SQAS-gfp and tsp4b-gfp were expressed similarly and strongly fluorescent at 7–8 hpf , suggesting that the mutated SQAS protein is synthesized and secreted ( Figure 6—figure supplement 1 ) . However , at later stages while tsp4b-gfp localized to MTJs similar to wild-type Tsp4b , SQAS-gfp fluorescence was not localized and gradually disappeared ( Figure 6—figure supplement 2A–D ) . SQAS-gfp also failed to localize to MTJs in Tsp4b-deficient embryos , in contrast to tsp4b-FLAG mRNA ( Figure 6—figure supplement 2E–H ) . To confirm the role of CQAC motif in maintaining stable functional Tsp4b pentamers , we extracted whole embryo protein under non-reducing conditions , from 36 hpf Tsp4b-deficient embryos expressing Tsp4b-GFP , SQAS-GFP and co-expressing SQAS-GFP and Tsp4b-FLAG respectively . A western blot of the protein extract showed that Tsp4b-FLAG and Tsp-GFP predominantly existed as a pentamers , while SQAS-GFP was exclusively present as monomers , further confirming that the oligomerization function of Tsp4b mediated by CQAC motif of CCD is essential for localization of Tsp4b to the MTJ ( Figure 6C ) . To test requirements for oligomerization of Tsp4b in muscle attachment , ECM organization and Itg signaling we injected wild type embryos with the SQAS form of tsp4b mRNA . Injected larvae at 3 dpf showed significantly increased muscle detachment upon stimulation compared to controls . Furthermore , when co-injected with the tsp4b MO , the SQAS tsp4b mRNA failed to rescue muscle detachment ( Figure 6B ) , FAK activation or Lam localization at MTJs ( Figure 6—figure supplement 3 ) . These results suggest that in addition to a requirement for its own assembly and localization at MTJs , pentameric Tsp4b is a key scaffolding component that is required for MTJ ECM assembly and muscle-specific Itg signaling . Studies of bovine , equine and human TSP4 have shown that it localizes to tendons ( Hauser et al . , 1995; Chen et al . , 2000; Södersten et al . , 2006 ) . To determine if the functional roles of zebrafish Tsp4b are evolutionarily conserved , we injected recombinant human TSP4 ( 2 ng/embryo ) into the interstitial space between myofibers of 60 hpf Tsp4b-deficient embryos . Exogenous TSP4 protein localized to MTJs adjacent to the site of injection and rescued muscle detachment in Tsp4b-deficient larvae when stimulated ( 22% detached [N = 136] compared to 52% detached [N = 96] in stimulated Tsp4b-deficient embryos ) ( Figure 7A–I ) . In addition , injection of exogenous TSP4 protein reduced the occurrence of muscle detachments by ∼20% in wild-type embryos ( N = 40 embryos ) electrically stimulated with increasing voltages ( Figure 7J ) . These results suggest that TSP4 and zebrafish Tsp4b are both structurally and functionally similar , and that TSP4 also serves as a scaffold for ECM assembly at MTJs . 10 . 7554/eLife . 02372 . 021Figure 7 . Conserved functions for TSP4 in maintenance of MTJs . ( A–H ) Lateral views of trunk muscles in 72 hpf embryos stained with anti-MHC ( green ) , anti-human TSP4 ( red ) and anti-Tsp4b ( blue ) antibodies . ( A–D ) Injected recombinant human TSP4 co-localizes with zebrafish Tsp4b at muscle attachments in wild type embryos . ( E–H ) Injected TSP4 localizes to somite boundaries in Tsp4b-deficient embryos . ( I ) Injected TSP4 rescues muscle attachments in Tsp4b-deficient embryos upon stimulation ( N = 96 embryos ) ( Chi squared test , p value<0 . 001 ) . ( J ) Histogram showing percentage of embryos with detached muscles in 60 hpf wt+BSA ( white columns ) and wt+TSP4 ( shaded columns ) embryos , stimulated at 30 V , 40 V and 50 V , respectively . N = 40 embryos for each sample . ( Scale bars = 30 microns ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02372 . 021 Vertebrate tendons consist of a complex network of collagen-rich fibrils , which interact with other ECM proteins and membrane adhesion complexes on the surfaces of muscle cells to form attachments strong enough to bear contractile forces ( Ros et al . , 1995 ) . How does such a complex network of ECM proteins assemble and maintain its organization ? Muscles of the embryonic zebrafish undergo dynamic changes in ECM composition ( Henry et al . , 2001; Crawford et al . , 2003; Snow and Henry , 2009; Sen et al . , 2011 ) , in which early high levels of Fn are replaced by Lams and Cols and later by an orthogonal array of collagen fibrils ( Kannus , 2000; Câmara-Pereira et al . , 2009; Charvet et al . , 2013 ) . These changes in ECM are critical for muscle attachment and maturation of MTJs . Here we show a novel role for pentameric Tsp4 as a key scaffolding protein that orchestrates the ECM organization of MTJs necessary for muscle attachments . Because Tsps interact with other Itg ligands ( e . g . , Lam , Col , Fn ) in vitro it has been debated as to whether Tsps play instructive or merely permissive roles in ECM organization and cell-ECM interactions ( Narouz-Ott et al . , 2000; Södersten et al . , 2006; Tan and Lawler , 2009 ) . In Drosophila , by virtue of its canonical RGD motif , Tsp plays a primary role as an Itg ligand in developing tendons . In contrast , our results suggest that zebrafish Tsp4 has a dual role , both binding Itgs through its non-canonical KGD domain and organizing the tendon ECM at MTJs to maintain muscle attachments . We propose a model in which pentameric Tsp4 in vivo functions as a scaffold to assemble other ECM components , and their interactions with Itgs and Dys complexes at MTJs ( Figure 8 ) . Our mutational studies of the CQAC motif support this hypothesis and show , for the first time in vivo , that pentamerization of Tsp4 is central to its scaffolding role , which organizes collagen fibrils in tendons in response to contractile force from muscles ( Hauser et al . , 1995; Kjaer , 2004; Aparecida de Aro et al . , 2012 ) . 10 . 7554/eLife . 02372 . 022Figure 8 . Tsp4b establishes and maintains MTJ ECM organization . A model for ECM assembly at an MTJ . Pentameric assemblies of Tsp4b ( red ) associate with Lam ( green ) , Col fibrils and other ECM components . Tsp4b and Lam bind Itgs ( orange ) on the muscle cell surface , activating FAK ( green ) and recruiting Pxn ( yellow ) and Ilk ( purple ) to promote muscle specific Itg signaling and stabilize myofiber attachment . DOI: http://dx . doi . org/10 . 7554/eLife . 02372 . 022 Human TSP4 , which forms pentamers ( Lawler et al . , 1995; Frolova et al . , 2012 ) , is also capable of rescuing Tsp4b-deficient zebrafish , demonstrating that the two are functionally interchangeable . Zebrafish Tsp4b shares the conserved C-terminal domain with human TSP4 and TSP5 . The C-terminal region of mammalian Tsp5 , which also functions as a pentamer , has been suggested to interact with multiple ECM proteins in cartilage ( Hecht et al . , 1998; Chen et al . , 2007 ) and with collagen fibrils in tendons ( Chen et al . , 2007; Södersten et al . , 2007; Tan and Lawler , 2009 ) . We hypothesize that Tsp4 serves as a central organizing factor in MTJs by interacting with multiple ECM proteins , proteoglycans and receptors on muscle membranes . Other Tsps could play similar organizational roles as pentamers ( subclass B ) or trimers ( subclass A ) in contexts where they are required such as neurite outgrowth , synapse formation , wound healing and tumorigenesis ( Arber and Caroni , 1995; Kyriakides et al . , 1998 , 1999; Sid et al . , 2004; Christopherson et al . , 2005 ) . Analysis of Tsp4b expression in zebrafish reveals surprising differences in muscle-tendon cell interactions during MTJ development in different muscle groups . Axial muscles attach directly to the basement membrane at somite boundaries in the early embryo and become contractile and functional at least 12 hr before tenocytes are detected at MTJs ( Devoto et al . , 1996; Barresi et al . , 2001; Henry et al . , 2005; Snow and Henry , 2009 ) . Our results suggest that after axial muscle progenitors attach , they secrete their own Tsp4b , which they downregulate upon differentiation , while later Tsp4b is secreted and maintained by tenocytes . Tenocytes in the zebrafish trunk likely originate from the syndetome compartment in each somite , similar to their counterparts in the chick , and migrate to sites of muscle attachment ( Brent et al . , 2003; Shukunami et al . , 2006; Chen and Galloway , 2014 ) . In contrast , we observe Tsp4b expression in cranial tenocytes at MTJs but not in cranial muscles . This difference in the timing and sources of Tsp4b between cranial and axial muscles suggests that there are distinct types of interactions between myoblasts and tenocytes involved in forming different types of MTJs . Our results suggest that Tsp4b is required for both initial attachments of some early muscles , which correlates with defects in early ECM , as well as more broadly in the maintenance of all types of attachments . Tsp4−/− mice show progressive weight loss , atrophy of muscle mass and disrupted packing of collagen fibrils , but they are viable and fertile ( Frolova et al . , 2010 , 2014 ) . Similarly , Tsp4b-deficient zebrafish appear largely unaffected until their muscles are subjected to multiple contractions . Stress-induced expression appears to be a common feature of many Tsps . For example , TSP4 ( and other Tsps ) is upregulated in response to exercise , myocardial infarction and in several types of muscular dystrophy ( Chen et al . , 2000; Timmons et al . , 2005; Frolova et al . , 2012 ) . Tsp5 expression is elevated in response to biomechanical stress ( Hecht et al . , 1998 ) . Subclass A Tsps ( Tsp 1–2 ) are also elevated in response to injury or inflammation ( Bornstein et al . , 2004 ) . Elevated Tsp expression in all these scenarios could reflect an attempt by cells at ECM repair but may also be detrimental by causing tissue fibrosis . Consistent with this idea , in Tsp2−/− mutant mice , wound healing progresses rapidly without scarring , but with abnormal Col fibril structure ( Kyriakides et al . , 1998; Bornstein et al . , 2004 ) . Tsp4−/− mice show elevated levels of Col ( Col I , II , III , V ) in the heart . This elevated accumulation of Col is reversed with application of TSP4 in an in vitro cell culture system ( Frolova et al . , 2012 ) . This result is similar to our in vivo study , where injected TSP4 rescues the Tsp4b-deficient phenotype , suggesting that the scaffolding function of Tsp4 is evolutionarily conserved and functions as a modulator of ECM organization in multiple tissue types . Recent work on the functions of Tsp4 in cardiac ECM has shown interesting intracellular functions , where Tsp4 localized to the endoplasmic reticulum of cardiac fibroblasts promotes nuclear shuttling of activating transcription factor 6 alpha ( Lynch et al . , 2012 ) . While this intracellular role of Tsp4 cannot be ruled out in MTJ organization , the ability of exogenous TSP4 protein injected extracellularly to rescue a Tsp4b-deficient phenotype suggests that the primary function of Tsp4 is in ECM organization at MTJs . Muscular dystrophies are characterized by loss of muscle activity due to defects in interactions between muscles and the myomatrix ECM at MTJs . A mouse model for congenital muscular dystrophy caused by Lama2 deficiency can be treated effectively by injecting recombinant Laminin1-1-1 trimer ( Rooney et al . , 2012 ) . Hence , the strategy of treating such ‘myotendinopathies’ by injecting recombinant ECM proteins is a viable option . In addition , sports- and age-related tendon injuries are often difficult to treat and require prolonged recovery periods . Our work provides a potential new therapeutic strategy using TSP4 to promote efficient tendon and MTJ repair and correcting defects in the ECM in a whole range of myotendinopathies . Wild type embryos of the tupfel long fin ( TL ) line were used for all experiments with the exception of lost contact ( loc ) mutants , which disrupt integrin-linked kinase ( ilk ) ( Postel et al . , 2008 ) , tg ( itga5:tagrfp ) ( Lackner et al . , 2013 ) and tg ( bactin:pxn-egfp ) transgenics ( Goody et al . , 2010 ) . All embryos were raised in standard embryo medium at 28 . 5°C ( Westerfield , 2007 ) and staged as described previously ( Kimmel et al . , 1995 ) . Craniofacial muscles of zebrafish embryos were labeled as described previously ( Schilling and Kimmel , 1997 ) . Adult fish and embryos were collected and processed in accordance with the rules and protocols approved by UCI-IACUC . An antisense morpholino ( MO ) oligonucleotide—CGGCCATCCTTCAATCACAACCTTC—was designed against the translation start site of tsp4b by Gene Tools , LLC . To reduce MO-induced cell death a p53-MO ( AGAATTGATTTTGCCGACCTCCTCT ) was co-injected at 0 . 35 ng/embryo . All primers used are listed in Supplementary file 1 . The full-length tsp4b mRNA used in rescue experiments was synthesized from a full- length tsp4b cDNA clone ( Cat No . MDR1734-202795262; Fisher Scientific , Pittsburgh , PA , USA ) in pBS SK+ vector , using T3 RNA polymerase ( mMessage mMachine T3 transcription kit AM1348 , Life Technologies , Grand Island , NY , USA ) and tailed ( Poly ( A ) tailing kit , AM1350 Life Technologies , Grand Island , NY , USA ) . The pCS2-itga6-GFP clone was prepared using CloneEZ ( Cat No: L00339; GenScript USA Inc . , Piscataway , NJ , USA ) . The cloning followed a two-step protocol . In the first step , pCS2 vector was cut at ClaI site and itga6 amplicon synthesized with complementary overhangs was cloned . The following primers were used for amplifying itga6; FP: gttctttttgcaggatcccatatcgatATGGAGCTCTTTGAGAAAGCGC ( lower case—pCS2 sequence , bold—ClaI site , upper case—itga6 sequence ) RP: GAGAGGCCTTGAATTCGAatcgattgagtagctctcgttctcatcccac ( lower case—pCS2 sequence , bold—ClaI site , upper case—itga6 sequence ) . The gfp amplicon synthesized with complementary overhangs was cloned into a pCS2-itga6 clone at a BstBI site . The following primers were used for amplifying gfp from pEGFP-C1 vector; FP: CGAGAGCTACTCAATCGATTTAATGGTGAGCAAGGGCGAGG ( upper case—tsp4b sequence , bold—bridge to maintain translation frame , upper case under lined—gfp sequence ) ; RP: ccttgaattcgaatcgatggaattcCTATAGGGCTGCAGAATCTAGAGGCTCG ( lower case—pCS2 sequence , bold—EcoRI site , upper case—gfp sequence ) . tsp4b antisense morpholino ( MO ) oligonucleotide was injected at either 0 . 16 or 0 . 32 ng/embryo; the former reduced and the latter eliminated Tsp4b protein as assayed immunohistochemically . For rescue experiments zebrafish tsp4b mRNA was synthesized from a full-length cDNA clone . KGE and SQAS mutant mRNAs were synthesized from the full-length tsp4b cDNA clone using modified site-directed mutagenesis protocol ( Cat No: 200555 Quikchange II-E site-directed mutagenesis kit Agilent Technologies , Santa Clara , CA , USA ) with the following primers: KGE forward-GGGAAGGGTGAGGCATGTG; KGE reverse-CACATGCCTCACCCTTCCC; SQAS forward-CATCTCTGAGAGCCAGGCCAGCGGGCTGAG; SQAS reverse- CTCAGCCCGCTGGCCTGGCTCTCAGAGATG . KGE mutated site ( in bold underlined ) —C1482G Asp>Glu; SQAS mutated sites ( in bold underlined ) —T781A and T790A , Cys>Ser . All RNAs were injected at 100 pg/embryo . The gfp and FLAG tagged constructs were constructed using Gibson Cloning on pCS2 vector backbones ( Gibson et al . , 2009 ) . The following primers were used to amplify the pCS2 vector backbone , tsp4b mRNA , gfp and pCS2-3XFLAG vector backbone . pCS2 vector for tsp4b-gfp fusion: FP—GAGCCTCTAGATTCTGCAGCCCTATAGgaattcaaggcctctcgagcctctag ( upper case under lined is gfp sequence , lower case is pCS2 vector sequence ) ; RP—GAGGAGATGCATTGTGCCGGCCATgaatcgatgggatcctgcaaaaagaacaag ( upper case is tsp4b sequence , lower case is pCS2 vector sequence ) . tsp4b for gfp fusion: FP—gttctttttgcaggatcccatcgattcATGGCCGGCACAATGCATCTCC ( upper case is tsp4b sequence , lower case is pCS2 vector sequence ) ; RP—GCTCCTCGCCCTTGCTCACCATCAAGGGGTCCATGCCATGTTGTGTACTG ( upper case under lined is gfp sequence , upper case is tsp4b sequence ) . The same set of primers was used to amplify the SQAS mutant form of cDNA from the pCS2 clone . gfp for tsp4b fusion: FP—GTACACAACATGGCATGGACCCCTTGATGGTGAGCAAGGGCGAGGAGC ( upper case is tsp4b sequence , upper case underlined is gfp sequence ) ; RP—ctagaggctcgagaggccttgaattcCTATAGGGCTGCAGAATCTAGAGGCTC ( upper case under lined is gfp sequence , lower case is pCS2 vector sequence ) . tsp4b for FLAG fusion: FP—caagctacttgttctttttgcaggatcATGGCCGGCACAATGCATCTCCTC ( upper case is tsp4b sequence , lower case is pCS2-3XFLAG vector sequence ) ; RP—cgtcatggtctttgtagtccatgtcCAAGGGGTCCATGCCATGTTG ( upper case is tsp4b sequence , lower case is pCS2-3XFLAG vector sequence ) . pCS2-3XFLAG for tsp4b fusion: FP—CAACATGGCATGGACCCCTTGgacatggactacaaagaccatgacg ( upper case is tsp4b sequence , lower case is pCS2-3XFLAG vector sequence ) ; RP—GAGGAGATGCATTGTGCCGGCCATgatcctgcaaaaagaacaagtagcttg ( upper case is tsp4b sequence , lower case is pCS2-3XFLAG vector sequence ) . The synthesized mRNA was purified and concentrated using RNA clean and concentrator-5 kit ( Cat No: R1015; Zymo Research , Irvine , CA , USA ) . Recombinant human TSP4 ( TSP4 ) protein solution was injected at 1 mg/ml , using a glass microelectrode , into the interstitial space between myofibers of anesthetized 48 hpf embryos . Embryos were allowed to recover for 12 hr in normal embryo medium before fixation for further analysis . To assess muscle attachment strength we designed an electrical stimulation protocol to induce muscle contraction . Teflon-insulated platinum microelectrodes ( D10PW; Plastics One Inc . , USA ) , with the insulation stripped from the ends , were connected to a Grass SD-5 stimulator ( Grass Instruments , Warwick , RI , USA ) with which we could vary pulse strength ( volts ) , duration , frequency ( λ ) , and delay . Electrodes were positioned at the anterior and posterior ends of individual wild type or Tsp4b-deficient embryos and stimulated at varying strengths—0 V , 20 V , 30 V , 40 V , 50 V—at duration = 8 ms , λ = 4 pulses/s , and delay = 6 ms . Embryos were anaesthetized with Tricaine ( ethyl 3-aminobenzoate methanesulfonate; Cat No . A5040 , Sigma-Aldrich , Milwaukee , WI , USA ) and placed on a silicone plate with embryo medium and stimulated for 2 . 5 min before transfer into fixative . 30 V was found to be most effective in causing muscle detachment but allowing embryos to remain otherwise healthy with this protocol ( 10% of wild type embryos show any detached muscles , while 60% of Tsp4b-deficient embryos show widespread detachment in somites ) . To determine the dose–response , tsp4b-MO embryos injected with different amounts were subjected to muscle stimulation–28% ( N = 7/25 ) injected with 0 . 16 ng/embryo of tsp4b-MO showed detachment , 63% ( N = 40/64 ) injected with 0 . 32 ng/embryo showed detachment . Zebrafish tsp4b ( NM_173226 ) expression was visualized using an antisense RNA probe synthesized against a 626 bp ( 10–636 bp ) region of tsp4b cDNA using T7 RNA polymerase ( Cat No: 10881767001 , Roche Diagnostic Corporation Indianapolis , IN , USA ) . Zebrafish tnmd ( NM_001114413 ) expression was visualized using an antisense RNA probe synthesized to recognize a 602 bp ( 109–710 bp ) region of tnmd cDNA using T7 RNA polymerase . Whole mount in situ hybridization was performed according to standard protocol and developed using NBT/BCIP ( NBT-11383213001 , BCIP- 11383221001 , Roche Diagnostics Corporation , Indianapolis , IN , USA ) or Fast Red ( Cat No: 11496549001 , Roche Diagnostics Corporation , Indianapolis , IN , USA ) ( Thisse and Thisse , 2008 ) . A zebrafish-specific Tsp4b antibody was generated against 618 bp ( 91–708 bp ) of the unique N-terminal region . This was cloned into pGEX-4T-2 expression vector , expressed as a GST tagged peptide , purified as per standard protocol and this fusion protein ( 206 aa ) was used to raise antibodies in rabbits at Thermo Fischer/Open Biosystems , Rockford , IL , USA ( Ring et al . , 2002 ) . Wild-type embryos stained with pre-immune serum from host showed absence of specific signal . Antibody specificity was verified by staining wild-type and Tsp4b deficient embryos with anti-Tsp4b . All embryos used for immunofluorescence experiments were fixed in 95% methanol and 5% glacial acetic acid for 4–6 hr at −20°C . They were rehydrated in Phosphate Buffered Saline ( PBS ) with 2% dimethyl sulfoxide ( DMSO ) , 0 . 5% Triton ( PBDT ) and permeabilized with cold acetone for 10–15 min at −20°C . Following permeabilization , a standard antibody staining protocol with PBDT was used . Primary antibodies and concentrations used: rabbit anti-Tsp4b ( 1:500 ) , rabbit monoclonal anti-human TSP4 ( 1:300; Abcam , Cambridge , MA , USA ) , mouse anti-myosin heavy chain ( MHC ) ( A1025; Developmental Studies Hybridoma Bank , Iowa City , IA , USA - 1:250 ) , rabbit anti-pan laminin ( Peterson and Henry , 2010 ) ( RB-082-A0; 1:250; Thermo Scientific Inc . , Waltham , MA , USA ) , rabbit anti-FAK [pY861 ( Henry et al . , 2001 ) [44-626G] , 1:250; Life Technologies , Grand Island , NY , USA] , and chicken anti-GFP ( ab13970; 1:1000; Abcam , Cambridge , MA , USA ) . DiAmino PhenylIndole ( DAPI ) ( D1306; 1:1000; Life Technologies , Grand Island , NY , USA ) was used to mark cell nuclei . Preabsorbed secondary antibodies were all obtained from Jackson Immunoresearch , West Grove , PA , USA and used for indirect immunofluorescence at 1:1000 , including: Alexa Fluor 488 conjugated donkey anti-mouse IgG ( 715-546-150 ) , DyLight 549 conjugated donkey anti-rabbit IgG ( 711-506-152 ) , Alexa Fluor 488 conjugated donkey anti-chicken IgY ( 703-486-155 ) , Alexa Fluor 488 conjugated donkey anti-mouse IgG ( 715-546-150 ) . After staining , embryos were mounted in 1% low melt agarose in PBS and imaged . The quantity of embryos chosen was dependent on the variance observed in the experiment . Experiments that involved stimulation of muscles after various treatments were performed on a larger data set to account for variability in the quantity and quality of treatments . Embryos that exhibited gross morphological defects associated with RNA toxicity or morpholino toxicity or other idiopathic growth defects were excluded from the analysis . In order to avoid background effects of adult parents , embryos from different tanks with adults of different age groups were used to collect the embryos for the analysis . Tsp4b forms pentamers through inter-chain interactions at the CCD and formation of inter-chain disulfide bonds with cysteine residues of the CQAC motif . In order to visualize the pentameric form on the blot it was necessary to perform the protein extraction and western blot protocol in a non-reducing condition ( Narouz-Ott et al . , 2000 ) . Whole protein extract was prepared from 36 hpf embryos , de-yolked in Ringer's solution by rapid flushing using a pipette tip . The de-yolked embryos were pelleted at 1500 RPM to remove the yolk granules in the supernatant . After three washes in Ringer's solution , a modified radio immunoprecipitation assay ( RIPA ) buffer was added ( 2 μl per embryo ) . RIPA composition: 0 . 05 M Tris pH 7 . 4 , 0 . 15 M NaCl , 1% Triton X100 , 1 . 25 mM CaCl2 , 0 . 9 mM MgCl2 , 20 μM MnCl2 , 0 . 2 mM ZnCl2 and protease inhibitors ( PMSF and Leupeptin ) . The tissue was homogenized using a pestle and DNase added to the homogenate and incubated on ice for 30 min . A 3–10% gradient gel was cast using a gradient former . A non-reducing protein loading buffer was added to the protein extract and 20 μl of this protein extract was loaded on to the gel . A prestained PageRuler protein ladder ( Cat No . 26616; Thermo Scientific Inc . , Waltham , MA , USA ) was used as a molecular weight reference . Western analysis was performed on the blot using mouse monoclonal anti-FLAG antibody ( Clone M2; Cat No . F1804; Sigma-Aldrich , Milwaukee , WI , USA ) and mouse monoclonal anti-GFP antibody ( Clone JL-8; Cat No . 632381; Clontech ) at 1:1000 dilution and secondary antibody anti-mouse-HRP ( 715-035-150; Jackson Immunoresearch , West Grove , PA , USA ) at 1:10 , 000 dilution . Signal development was performed using a chemiluminescent assay with Luminol ( Cat No . 123072 , Sigma-Aldrich , Milwaukee , WI , USA ) and Coumaric acid ( Cat No . C9008 , Sigma-Aldrich , Milwaukee , WI , USA ) . Whole embryo RNA was extracted from wild-type and Tsp4b-deficient embryos were collected at 60 hpf according to standard protocols using Trizol ( 15596-018; Life Technologies , Grand Island , NY , USA ) . The experiment was designed in accordance with guidelines to conduct and report quantitative PCR ( Bustin et al . , 2009 ) . The RNA quality was assessed on an Agilent bioanalyzer 2100 and a RIN value >9 . 0 was obtained for the samples . RNA concentration was normalized between samples and used as a template for cDNA synthesis . cDNA was synthesized with oligo dT primers using the standard protocol of ProtoScript M-MuLV First Strand cDNA Synthesis Kit ( #E6300S; New England BioLabs Inc . , Ipswich , MA , USA ) . The synthesized cDNA was diluted to 1:20 and used as a template for qRT-PCR using the protocol for LightCycler 480 SYBR Green I Master kit ( 04707516001; Roche ) . The reaction was run on LightCycler 480 II Real time-PCR Instrument ( Roche Diagnostics Corporation , Indianapolis , IN , USA ) and analyzed using LightCycler 480 Software release 1 . 5 . 0 SP3 . The histogram was plotted on Microsoft Office–Excel . Lyophilized recombinant human TSP4 without the signal peptide ( Ala22-Asn961 , accession # P35443 ) tagged with a C-terminal His tag was derived from Chinese Hamster Ovary ( CHO ) cell line ( 2390-TH-050 Lot #MLO0612091; RD systems Inc . , Minneapolis , MN , USA ) It was reconstituted in a tris buffer ( Narouz-Ott et al . , 2000 ) ( 15 mM Tris pH 7 . 4 , 75 mM NaCl , 1 mM ZnCl2 , 1 mM CaCl2 ) at a concentration of 1 mg/ml as per the manufacturer's instructions . Control experiments were performed by injecting BSA solution in the same buffer at a 1 mg/ml concentration . Wild-type and Tsp4b-deficient 36 hpf embryos were anesthetized in tricaine and embedded in 2% low melting agarose ( prepared using embryo medium ) . Using a microelectrode , 2 ng of human TSP4 protein is injected into the extracellular space between muscle fibers . Embryos were removed from the agarose and transferred to embryo medium and allowed to recover for 12 hr in 28°C . Approximately half of wild-type and Tsp4b-deficient embryos injected with the human TSP4 protein were subjected to mild electrical stimulation and fixed for staining . Unstained 72 hpf wild-type and Tsp4b-deficient embryos ( both before and after stimulation ) were fixed in a 4% PFA ( Cat # 15710; Electron Microscopy Sciences [EMS] , Hatfield , PA , USA ) and 2% Glutaraldehyde ( Cat # 16020; EMS , Hatfield , PA , USA ) cocktail for 6 hr at 4°C . Embryos were washed in PBS+0 . 1% Triton and embedded in 5% low melt agarose ( Cat #20-104; Apex Chemicals ) . Vibratome sections ( Leica VT1000S ) were cut in cold PBS . Lateral sections ( 75 µm ) were cut at 0 . 75 mm/s and a vibration frequency of 70 Hz . The sections were washed in 0 . 15 M Sodium Cacodylate ( Cat # 12300; EMS , Hatfield , PA , USA ) , stained sequentially with 1% Osmium tetroxide ( Cat No . RT 19100 , EMS , Hatfield , PA , USA ) and 1% Uranyl acetate ( Cat No . RT 22400-1 , EMS , Hatfield , PA , USA ) , and were embedded in Durcupan for ultrathin sections . The Imaging was performed on a JEOL 1200EX TEM ( NCMIR , UCSD ) . Whole mount in situ stained embryos were photographed using a Zeiss Axioplan-2 microscope using Volocity image acquisition software ( Improvision , Perkin Elmer Inc . ) . Embryos processed for fluorescent immunohistochemistry were imaged using either: ( 1 ) an Olympus Fluoview FV100 confocal system with an IX81 inverted microscope using Plan-Apo 20X/0 . 75 NA and Plan-Apo 40X/1 . 3 NA ( oil ) objectives , respectively , or ( 2 ) a Zeiss LSM780 confocal system with an Observer Z1 inverted microscope and a Plan-Apo 20X/0 . 8 NA objective . Confocal stacks were analyzed using ImageJ software ( Hartig , 2013 ) . Multiple sequence alignment of protein sequences was performed using ClustalW2 ( http://www . ebi . ac . uk/Tools/msa/clustalw2/ ) . Phylogenetic tree of Tsp genes was constructed using MegAlign on DNASTAR Lasergene suite . Quantification of data from muscle stimulation studies was performed with Microsoft Excel ( Office 2007 ) . Statistical significance was calculated using a Chi-Squared test . Standard error calculated from the data set and plotted on the histogram .

Tendons , the tough connective tissues that link muscles to bones , are essential for lifting , running and other movements in animals . A matrix of proteins , called the extracellular matrix , connects the cells in a tendon , giving it the strength it needs to prevent muscles from detaching from bones during strenuous activities . To achieve this strength , extracellular matrix proteins bind to one another and to receptors on the muscle cell surface that are linked to its internal scaffolding , thereby organizing other proteins into a structure called a myotendinous junction . However , despite the essential roles of tendons , scientists do not fully understand how this organization occurs , or how it can go awry . Subramanian and Schilling screened zebrafish for genes that are essential for proper muscle attachment , and zeroed in on a gene encoding a protein called Thrombospondin-4b ( Tsp4b ) . A similar protein helps to connect muscle and tendon cells in fruit flies . Without Tsp4b , zebrafish are able to form connections between muscles and tendons , but the muscles detach easily during movement . This weakened connection is caused by disorganization of the proteins in the extracellular matrix , which results in reduced signaling from the muscle cell receptors . When a human form of this protein was injected into zebrafish embryos lacking Tsp4b , it settled into the junctions between muscle and tendon cells . The human protein repaired the detached muscles and restored the proper organization of the matrix . This improved the strength of the muscle-tendon attachment in the treated fish embryos , suggesting that similar injections could also help to strengthen and repair muscles and tendons in people .

2014

Thrombospondin-4 controls matrix assembly during development and repair of myotendinous junctions

CpG islands ( CGIs ) are associated with most mammalian gene promoters . A subset of CGIs act as polycomb response elements ( PREs ) and are recognized by the polycomb silencing systems to regulate expression of genes involved in early development . How CGIs function mechanistically as nucleation sites for polycomb repressive complexes remains unknown . Here we discover that KDM2B ( FBXL10 ) specifically recognizes non-methylated DNA in CGIs and recruits the polycomb repressive complex 1 ( PRC1 ) . This contributes to histone H2A lysine 119 ubiquitylation ( H2AK119ub1 ) and gene repression . Unexpectedly , we also find that CGIs are occupied by low levels of PRC1 throughout the genome , suggesting that the KDM2B-PRC1 complex may sample CGI-associated genes for susceptibility to polycomb-mediated silencing . These observations demonstrate an unexpected and direct link between recognition of CGIs by KDM2B and targeting of the polycomb repressive system . This provides the basis for a new model describing the functionality of CGIs as mammalian PREs . The capacity to segregate and functionalize regulatory elements within large and complex vertebrate genomes relies on the activity of chromatin modifying and epigenetic systems ( Kouzarides , 2007 ) . Nevertheless , it remains very poorly understood how the enzymes responsible for initiating these functional chromatin states are targeted to and recognize defined regions of the genome . Understanding the molecular mechanisms that underpin these processes is becoming increasingly important given the mounting genetic evidence implicating misregulation of chromatin modifying activities in human disease and cancer ( Greer and Shi , 2012; Butler et al . , 2012 ) . In mammals , the majority of CpG dinucleotides are methylated on the 5 position of cytosine and this epigenetic modification can be stably transmitted across cell divisions ( Klose and Bird , 2006 ) . DNA methylation is thought to be a highly stable modification that helps maintain intergenic and heterochromatic regions in a transcriptionally inert chromatin state ( Bird , 2002; Klose and Bird , 2006 ) . A major exception to this pervasive methylation is short contiguous regions of the genome , called CpG islands ( CGIs ) , which lack DNA methylation and are associated with roughly two-thirds of human gene promoters ( Illingworth and Bird , 2009 ) . It is thought that CGIs function to buffer gene regulatory elements from the repressive effects of surrounding DNA methylation and therefore play a central role in the regulatory capacity of the genome ( Blackledge and Klose , 2011 ) . Although for the past two decades CGIs were considered to play a largely passive role in promoter and regulatory element function , we and others recently discovered that the non-methylated CpG dinucleotides found within CGIs act as a binding site for proteins that contain the ZF-CxxC DNA binding domain ( Blackledge et al . , 2010; Thomson et al . , 2010 ) . ZF-CxxC domain-containing proteins are found in chromatin modifying complexes and therefore play an unexpected and proactive role in specifying unique chromatin modification architecture at CGI associated gene promoters . For example , the ZF-CxxC domain containing protein CFP1 is a core component of the SET1 histone H3 lysine 4 ( H3K4 ) methyltransferase complex ( Lee and Skalnik , 2005 ) and targets this transcriptionally permissive modification to CGIs genome-wide . Similarly , the KDM2A histone H3K36 demethylase enzyme contains a ZF-CxxC domain that targets the enzyme directly to CGIs where it removes histone H3 lysine 36 ( H3K36 ) dimethylation ( Blackledge et al . , 2010 ) . In mammals H3K36me1/me2 is found broadly across the genome ( Peters et al . , 2003; Robin et al . , 2007; Blackledge et al . , 2010 ) and may be inhibitory to transcriptional initiation ( Strahl et al . , 2002; Carrozza et al . , 2005 ) . Specific targeting of KDM2A to CGIs appears to remove H3K36me2 at CGIs as part of a mechanism to mark these regions with transcriptionally permissive chromatin ( Blackledge et al . , 2010 ) . Importantly , the targeting of these modifications to CGI associated gene promoters generally occurs independently of the transcriptional state of the associated gene ( Blackledge et al . , 2010 ) . This suggests that ZF-CxxC mediated chromatin modifications function upstream of transcription to create a permissive chromatin environment at gene promoters and demarcate these regions from surrounding non-regulatory regions of the genome ( Blackledge and Klose , 2011; Deaton and Bird , 2011 ) . Unlike the majority of CGIs which are marked with a permissive chromatin architecture controlled by the ZF-CxxC dependent chromatin modification system , a subset of CGIs can exist in an alternative more repressed chromatin state . This is dictated by the action of the two central polycomb repressive complexes ( PRCs ) in mammals , called PRC1 and PRC2 ( Mikkelsen et al . , 2007; Simon and Kingston , 2009 ) . PRC1 complexes have E3 ubiquitin ligase activity that mono-ubiquitylates H2A on position 119 ( H2AK119ub1 ) ( Wang et al . , 2004b; de Napoles et al . , 2004 ) and PRC2 complexes have histone H3 lysine 27 ( H3K27 ) methyltransferase activity ( Cao et al . , 2002; Czermin et al . , 2002; Kuzmichev et al . , 2002; Müller et al . , 2002 ) . In contrast to the permissive and accessible chromatin state found at most CGIs , polycomb mediated chromatin modifications are thought to generate a more compact and inhibitory chromatin environment , limiting transcription of associated genes ( Shao et al . , 1999; Lavigne et al . , 2004; Francis et al . , 2004; Eskeland et al . , 2010 ) . In Drosophila , which lack CGIs , polycomb repressive complexes are targeted to regions of the genome called polycomb response elements ( PREs ) through the action of transcription factors ( reviewed in ( Schuettengruber et al . , 2007; Ringrose and Paro , 2007; Simon and Kingston , 2009 ) ) . Interestingly , in mammals this situation is very different in that PREs predominantly correspond to CGIs ( Ku et al . , 2008; Woo et al . , 2010 ) and there is mounting evidence that chromatin features specific to CGIs may play an important role in guiding PRC complexes to these particular regions of the genome ( Mendenhall et al . , 2010; Kanhere et al . , 2010; Lynch et al . , 2012 ) . At the mechanistic level , how CGIs can function as mammalian PREs remains unknown , constituting a major gap in our understanding of how CGIs are functionally linked to the polycomb repressive systems in mammals . In addressing this question , here we demonstrate that KDM2B , a paralogue of KDM2A , plays an important role in targeting the polycomb repressive complex 1 ( PRC1 ) to CGIs and regulating gene expression in mouse embryonic stem cells ( ESCs ) . This provides the first evidence linking direct recognition of CGIs with recruitment of a polycomb repressive complex in mammals . We recently demonstrated that the KDM2A protein binds to CpG islands genome-wide ( Blackledge et al . , 2010 ) . However , its paralogue , KDM2B , was reported to be concentrated in the nucleolus . Therefore , to examine the properties of endogenous KDM2B and understand how it functions in comparison to KDM2A , an antibody against KDM2B was raised . In contrast to previous reports , we observe that KDM2B is distributed throughout the nucleus and excluded from nucleoli ( Figure 1—figure supplement 1 ) . Based on these observations , and considering that KDM2B also encodes a ZF-CxxC domain , we set out to examine whether KDM2B had a similar role to KDM2A in binding non-methylated DNA and CGIs ( Blackledge et al . , 2010 ) . In agreement with previous work , recombinant KDM2B was found to bind non-methylated DNA in vitro ( Koyama-Nasu et al . , 2007; Yamagishi et al . , 2008; Blackledge et al . , 2012 ) in a manner that is comparable to that of KDM2A ( Figure 1A , B ) . This suggests that KDM2B , like KDM2A , may recognize non-methylated DNA in vivo and bind CGI elements . To examine this possibility a chromatin immunoprecipitation sequencing ( ChIP-seq ) analysis was carried out to determine the binding profiles of KDM2A and KDM2B genome-wide in mouse ESCs . A visual inspection of these profiles suggests that both KDM2A and KDM2B localize to promoter associated CGIs and are excluded from non-CGI associated promoters ( Figure 1C ) . When KDM2A and KDM2B ChIP-seq signal was plotted over all transcription start sites ( TSSs ) that had been segregated based on their classification as CGI or non-CGI associated there was a very specific enrichment of both KDM2A and KDM2B at the CGI associated class of transcription start site ( Figure 1D ) . Furthermore , when KDM2A and KDM2B signal was plotted over all CGIs , including those away from gene promoters , there was a very specific spatial localization of both proteins at CGIs with signal intensity peaking at the centre of the island ( Figure 1E ) . Together these observations demonstrate in vivo that KDM2A and B are both bound to CGIs genome-wide . 10 . 7554/eLife . 00205 . 003Figure 1 . KDM2B binds to non-methylated CpG DNA and localises to CGIs genome-wide . ( A ) Electrophoretic mobility shift assay ( EMSA ) demonstrating that the recombinant KDM2B ZF-CxxC domain binds to a DNA probe containing non-methylated CpGs ( left panel ) in a concentration-dependent manner . Binding is abrogated by CpG methylation ( right panel ) . ( B ) An analogous EMSA to part ( A ) using the known CGI binding factor KDM2A for comparison . ( C ) Input , KDM2A , and KDM2B ChIP-seq profiles over a region of the genome containing CGI and non-CGI associated genes . Bio-CAP-seq profiles are shown to indicate the location of non-methylated DNA ( Blackledge et al . , 2012 ) . Above the sequencing traces individual genes are shown with the arrow indicating the transcription start site and vertical black lines corresponding to exons . The location of CGIs are indicated by green bars . Both KDM2A and KDM2B associate specifically with CGIs containing non-methylated DNA . ( D ) KDM2A and KDM2B ChIP-seq signal segregates specifically with CGI associated gene promoters ( left panel ) and is excluded from gene promoters not associated with CGIs ( right panel ) . ( E ) KDM2A and KDM2B ChIP-seq signal is centred over CGIs , in agreement with their capacity to recognize non-methylated DNA at these sites . DOI: http://dx . doi . org/10 . 7554/eLife . 00205 . 00310 . 7554/eLife . 00205 . 004Figure 1—figure supplement 1 . KDM2B is found throughout the nucleus and not concentrated in the nucleolus . ( A ) Endogenous KDM2B immunofluorescence in five different mammalian cell lines . KDM2B signal is in red ( left ) and DAPI signal in blue ( right ) . In all cell types KDM2B is a broadly distributed throughout the nucleus and not concentrated in the nucleolus as previously reported ( Frescas et al . , 2007 ) . ( B ) Immunofluorescence in HeLa cells for Flag-tagged KDM2B and endogenous fibrillarin , a nucleolus marker . KDM2B is largely excluded from nucleoli . ( C ) The KDM2A and KDM2B antibodies specifically recognize their respective antigens . Empty expression vector , Flag-KDM2A , and Flag-KDM2B were transiently transfected into 293T cells . Whole cell extracts were made and probed with anti-KDM2A ( left panel ) , anti-KDM2B ( centre panel ) , or anti-Flag ( right panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00205 . 004 KDM2A and KDM2B profiles at CGI associated genes appeared largely similar ( Figure 1 ) . However when the levels of KDM2A and B were directly compared at individual CGIs genome-wide , a subset of CGIs was found that were preferentially enriched for KDM2B and depleted of KDM2A ( Figure 2A ) . To begin examining whether there is a specific feature shared amongst KDM2B enriched CGIs , the genes associated with these CGIs were extracted and subjected to gene ontology ( GO ) enrichment analysis ( Figure 2B ) . Interestingly , these genes were significantly associated with GO terms related to embryo development , morphogenesis , and cellular differentiation . In early development and embryonic stem cells , these classes of genes are often targeted by the polycomb repressive complexes and maintained in a poised but silent state until their expression is induced during lineage commitment ( Lee et al . , 2006; Boyer et al . , 2006; Bracken et al . , 2006; Mikkelsen et al . , 2007 ) . The enrichment of KDM2B at these specific classes of genes suggests that KDM2B may play a unique role in their regulation perhaps related to polycomb mediated repression . 10 . 7554/eLife . 00205 . 005Figure 2 . KDM2B is enriched at polycomb repressed CGIs . ( A ) An MA-like plot depicting the relative enrichment of KDM2A and KDM2B at all gene associated CGIs in mouse ESC cells . The log2 mean read intensity is displayed on x-axis and the log2 relative enrichment of KDM2B compared to KDM2A is displayed on the y-axis . The subset of CGIs highly enriched for KDM2B is coloured blue . ( B ) A histogram displaying fold enrichment values for GO term analysis of the genes which are over twofold enriched for KDM2B ( blue data points , part ( A ) ) at a FDR <5% . ( C ) A heat map illustrating ChIP-seq signal at all CGI transcription start sites ( TSSs ) in mouse ESC cells indicating depletion of KDM2A and enrichment of KDM2B at polycomb target CGIs marked by RING1B and EZH2 . The scatter plot ( far right in red ) illustrates the log ratio of KDM2B to KDM2A enrichment at the same intervals depicted in the heat map as a scatter plot . ( D ) ChIP-seq profiles for KDM2A and KDM2B at the Adss non-polycomb target CGI gene indicating similar binding of both KDM2A and B ( upper panels ) . ( E ) ChIP-seq for KDM2A , KDM2B , RING1B ( PRC1 ) , and EZH2 ( PRC2 ) at the Lhx4 gene ( lower panels ) . KDM2B is specifically enriched and KDM2A depleted at this polycomb repressed CGI . In all cases Bio-CAP-seq indicates the location of underlying non-methylated DNA and clearly depicts the spatial relationship between KDM2B , polycomb group proteins , and non-methylated DNA . DOI: http://dx . doi . org/10 . 7554/eLife . 00205 . 00510 . 7554/eLife . 00205 . 006Figure 2—figure supplement 1 . KDM2B is enriched at polycomb associated CGIs . ( A ) – ( C ) ChIP-seq profiles for KDM2A and KDM2B over a series of non-polycomb associated CGIs . Similar enrichment of KDM2A and B is observed at these sites . ( D ) – ( F ) ChIP-seq profiles for KDM2A , KDM2B , RING1B , and EZH2 over a series of polycomb associated CGIs . KDM2B is enriched and KDM2A depleted at these sites . DOI: http://dx . doi . org/10 . 7554/eLife . 00205 . 006 To examine this possibility , the binding profiles of KDM2A , KDM2B , and components of the polycomb repressive complexes were examined in more detail . In mammals , there are two general polycomb repressive complexes ( PRC ) called PRC1 and PRC2 ( Shao et al . , 1999; Cao et al . , 2002; Czermin et al . , 2002; Kuzmichev et al . , 2002; Müller et al . , 2002; Levine et al . , 2002 ) . Previously , ChIP-seq based analysis has been used extensively to examine the location of PRC1 ( via RING1B ) ( Tavares et al . , 2012 ) and PRC2 ( via EZH2 ) genome-wide ( Peng et al . , 2009 ) . Using this information , KDM2A and KDM2B ChIP-seq was compared to the polycomb repressive complexes in mouse ESCs by examining the signal intensity around all CGI associated gene promoters ( Figure 2C ) . From the heat maps it was immediately apparent that KDM2A and KDM2B bind in a seemingly equal manner to most CGI associated genes , but a subset of genes are specifically enriched for KDM2B and depleted of KDM2A . In keeping with the gene ontology analysis , the KDM2B enriched target genes segregate almost exclusively with genes enriched for polycomb repressive complex members RING1B and EZH2 . Perhaps more surprisingly , the profile of the PRC1/2 components over CGI-associated target genes did not map precisely to the gene TSS but instead tracked with KDM2B binding and the underlying non-methylated DNA signal at these same regions . This striking spatial relationship is apparent at individual genes and also more generally when PRC1/2 components are examined at all polycomb bound genes ( Figure 2C–E and Figure 2—figure supplement 1 ) . Together these observations suggest that KDM2B , unlike KDM2A and CFP1 ( Thomson et al . , 2010 ) , is enriched at polycomb associated CGIs . Furthermore , the clear spatial relationship between polycomb repressive proteins , KDM2B , and non-methylated DNA indicates there may be a mechanistic relationship between recognition of non-methylated DNA at CGIs and polycomb repressive complex nucleation . Based on the clear enrichment of KDM2B at polycomb associated CGIs in ESCs in vivo , we sought to understand if KDM2B is part of a protein complex in ESCs that might contribute to this localization . To achieve this we isolated stable ESC lines expressing epitope-tagged KDM2B and carried out affinity purification from nuclear extract followed by mass spectrometry to identify associated proteins ( Figure 3A–C ) . To ensure that any identified interactions were not mediated through DNA or non-specific interactions with the affinity matrix , we also carried out parallel purifications in which the extract had been pre-treated with nuclease to remove any DNA contamination and from a cell line containing only the empty expression vector . In the purifications from extracts containing epitope-tagged KDM2B a series of proteins in addition to KDM2B were identified ( Figure 3B , C ) . These included RING1B , YAF2 , RYBP , PCGF1 , BCOR , and BCORL1 . Importantly RING1B is the E3 ubiquitin ligase that functions as the catalytic core of the PRC1 complex and is responsible for mono-ubiquitylation of H2A at position 119 ( H2AK119ub1 ) . To verify that endogenous KDM2B interacts with RING1B , KDM2B was immunoprecipitated from ESC nuclear extract and the presence of RING1B in the immunoprecipitates verified by western blotting ( Figure 3D ) . Therefore , KDM2B exists as part of a variant PRC1 complex in mouse embryonic stem cells . This observation is in agreement with recent work in cancer cell lines that identified KDM2B as part of a similar variant PRC1 complex ( Gearhart et al . , 2006; Sánchez et al . , 2007; Gao et al . , 2012 ) . 10 . 7554/eLife . 00205 . 007Figure 3 . KDM2B forms a variant PRC1 complex in mouse ESCs containing RING1B and PCGF1 . ( A ) To purify KDM2B and associated proteins , a mouse ESC cell line stably expressing Flag-2XStrepII-tagged KDM2B was generated . Nuclear extract was isolated from this cell line , KDM2B affinity purified , and the purified proteins subject to mass spectrometry . ( B ) Purified KDM2B fractions were resolved by gradient SDS-PAGE and visualized by SyproRuby staining . The purifications were performed in the absence and presence of benzonase to exclude DNA-mediated interactions and a cell line containing only the empty vector was used to control for non-specific binding to the affinity matrix . The elutions were probed by western blot for KDM2B as indicated . ( C ) Elutions from the KDM2B affinity purification were directly analysed by tryptic digestion followed by peptide identification by LC–MS/MS . The Mascot scores and peptide coverage are shown for the respective affinity purifications . KDM2B in ESCs associates with a variant PRC1 complex containing RING1B , BCOR/BCORL1 , PCGF1 , RYBP , YAF2 and SKP1 . ( D ) Western blot analysis of endogenous KDM2B immunoprecipitation from ESC nuclear extract , verifying that KDM2B interacts with RING1B , but not the PRC2 component EZH2 . Flag immunoprecipitation was performed as negative control . ( E ) Reciprocal affinity purifications and subsequent LC-MS/MS for RING1B , YAF2 , RYBP , and PCGF1 confirm the interaction between KDM2B and these PRC1 components . This analysis further indicates that PCGF1 is unique to the KDM2B-PRC1 complex . Protein identification scores and sequence coverage ( Cov [%] ) are indicated . ( F ) A schematic representation of the variant KDM2B PRC1 complex ( left panel ) in comparison to canonical PRC1 complexes ( right panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00205 . 00710 . 7554/eLife . 00205 . 008Figure 3—figure supplement 1 . KDM2B forms a variant PRC1 complex in mouse ESCs containing RING1B and PCGF1 . ( A ) – ( D ) Stable cell lines expressing Flag-2XStrepII tagged RING1B ( A ) , YAF2 ( B ) , RYBP ( C ) , and PCGF1 ( D ) were generated and each protein affinity purified from mouse ESC nuclear extract . Purified fractions were resolved by gradient SDS-PAGE and visualized by SyproRuby staining . In each case the input , flowthrough , and elution is indicated above . The elutions were probed by western blot against the tagged protein as indicated below each SyproRuby stained gel . DOI: http://dx . doi . org/10 . 7554/eLife . 00205 . 008 To understand in more detail the composition and specificity of the KDM2B/PRC1 complex in ESCs , we generated individual stable cell lines expressing epitope tagged RING1B , YAF2 , RYBP , and PCGF1 ( Figure 3E and Figure 3—figure supplement 1 ) . Nuclear extracts were prepared from each cell line , epitope-tagged factors were affinity purified , and the associated proteins identified by mass spectrometry . Importantly KDM2B was identified in all purifications validating the association between KDM2B and these PRC1 components . Recently it has become apparent that different PRC1 complexes form in a manner that is dependent on a central dimerization partner that interacts with the RING1A/B catalytic core ( Gao et al . , 2012 ) . These dimerization partners , called polycomb group ring fingers ( PCGFs ) , act as bridging molecules that convey unique subunit composition to RING1B containing complexes and are required for H2AK119ub1 catalysis ( Cao et al . , 2005; Li et al . , 2006; Ben-Saadon et al . , 2006; Buchwald et al . , 2006 ) ( Figure 3F ) . There are six characterized RING1B dimerization partners PCGF1 ( NSPC1 ) , PCGF2 ( MEL18 ) , PCGF3 ( DONG1 ) , PCGF4 ( BMI1 ) , PCGF5 ( RNF159 ) , and PCGF6 ( MBLR ) . In purifying RING1B complexes we identified five of the six known PCGF subunits and most of the major auxiliary factors known to interact with these complexes ( Figure 3E ) . This supports previous observations that RING1B is the central enzymatic component of most PRC1 complexes in ESCs ( Illingworth et al . , 2012 ) . Although YAF2 and RYBP were both identified in our KDM2B purification , when these proteins were purified independently it became apparent that they are also found in multiple other PCGF containing PRC1 complexes ( Figure 3E ) . This indicates that both YAF2 and RYBP can form part of the KDM2B-PRC1 variant complex , but also function as part of other PRC1 complexes ( Figure 3E ) . YAF2 and RYBP are thought to interact directly with RING1B ( Wang et al . , 2010 ) suggesting that they are likely to be recruited to the KDM2B-PRC1 complex via RING1B . In the KDM2B purification the only PCGF identified was PCGF1 , suggesting that this factor is unique to the KDM2B-PRC1 variant complex . When the PCGF1 complex was purified ( Figure 3E ) its composition was almost identical to the KDM2B complex , with the exception that it also included RING1A , an alternative catalytic core of PRC1 that is highly sequence-similar to RING1B but is expressed at greatly reduced levels in ESCs . Therefore , based on our detailed reciprocal complex purifications , KDM2B forms a PRC1 variant complex that includes BCOR , BCORL1 , PCGF1 , RING1 , and YAF2 or RYBP ( Figure 3F ) . Interestingly , in agreement with PRC1 purifications from cancer cell lines ( Gao et al . , 2012 ) , the PCGF1-containing complex associates with RYBP or YAF2 but fails to integrate chromobox domain-containing ( CBX ) proteins that recognize H3K27me3 ( Cao et al . , 2002; Wang et al . , 2004a ) . CBX proteins and RYBP/YAF2 are thought to interact with RING1B in a mutually exclusive manner ( Wang et al . , 2010 ) . This suggests that PCGF1 may play an important role in specifying the capacity of RYBP and YAF2 to associate with RING1B in the KDM2B variant PRC1 complex . Importantly , the inclusion of KDM2B in a PRC1 complex in ESCs suggest that the relationship between KDM2B and polycomb repressed CGIs may be mediated through its direct interaction with PRC1 components . Both KDM2A ( Figure 2C ) and CFP1 ( Thomson et al . , 2010 ) are inhibited from binding to polycomb repressed CGIs in vivo despite the presence of non-methylated CpG at these sites . This may be related to the compact nature of polycomb-repressed regions and the distinct requirement for the ZF-CxxC domain to bind accessible linker DNA ( Zhou et al . , 2012 ) . However , KDM2B differs from KDM2A and CFP1 in that it is enriched at polycomb repressed CGIs ( Figure 2 ) . Based on our biochemical purifications that indicate KDM2B in ESCs is part of a variant PRC1 complex ( Figure 3 ) , it remained possible that KDM2B was enriched at polycomb repressed CGIs via a ZF-CxxC domain independent mechanism relying on PRC1 . To test this hypothesis we obtained mouse ESCs which lack RING1A and have a floxed allele of RING1B that can be conditionally removed by addition of the drug tamoxifen ( Figure 4A ) ( Endoh et al . , 2008 ) . Removal of the catalytic core of PRC1 is known to destabilize the remainder of the core PRC1 complex ( Leeb and Wutz , 2007; Endoh et al . , 2008 ) . Therefore , we first isolated chromatin from wild type and RING1A/B deleted ESC cells and carried out chromatin immunoprecipitation for RING1B and KDM2B . As would be expected , there were high levels of RING1B at polycomb repressed CGI containing genes and the RING1B signal was lost following tamoxifen treatment ( Figure 4B ) . We then analysed the levels of KDM2B at a series of polycomb repressed and non-polycomb repressed CGI associated genes by ChIP and observed only small or no changes in KDM2B binding in the absence of RING1A/B ( Figure 4C ) . This suggests that an intact PRC1 complex is not required to target KDM2B to polycomb repressed sites in vivo . 10 . 7554/eLife . 00205 . 009Figure 4 . KDM2B ZF-CxxC DNA binding domain is required for CGI binding . ( A ) Schematic representation showing removal of RING1B in ESC Ring1a−/− Ring1bfl/fl cells by tamoxifen treatment . ( B ) ChIP analysis indicating that RING1B is enriched at polycomb target CGIs in untreated cells ( red bars ) . After 48 hr of tamoxifen treatment RING1B binding is lost at polycomb targets ( blue bars ) . Error bars represent SEM of three biological replicates . ( C ) ChIP analysis demonstrating that removal of RING1B does not lead to loss of KDM2B binding at regular or polycomb associated target CGIs ( compare red and green bars ) . This demonstrates that RING1B is not required to recruit KDM2B to polycomb repressed sites . Error bars represent SEM of three biological replicates . ( D ) ChIP analysis in ESC cells stably expressing tagged wild-type ( WT ) KDM2B or a mutant KDM2B that disrupts its DNA-binding capacity ( K643A ) . The ZF-CxxC domain of KDM2B is required for KDM2B binding to CGIs regardless of their polycomb status . Error bars represent SEM of two biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 00205 . 009 Given that KDM2B does not appear to rely on an intact PRC1 complex to localize to polycomb CGIs , we hypothesized its recognition of polycomb targets must be achieved at least in part through the function of its ZF-CxxC non-methylated DNA binding domain . To test this hypothesis we generated stable ESC lines expressing either epitope tagged wild type KDM2B or a version of KDM2B containing a point mutation in the ZF-CxxC domain that ablates binding to CpG dinucleotides . We then isolated chromatin from these cell lines and analysed binding of exogenous KDM2B to CGI targets ( Figure 4D ) . Importantly , WT KDM2B localized specifically to CGIs but a ZF-CxxC mutant failed to bind any class of CGI . This demonstrates that although KDM2B is part of a PRC1 complex its localisation to regular and polycomb repressed CGIs requires the recognition of non-methylated DNA via the ZF-CxxC domain . The capacity of KDM2B to target CGIs in the absence of PRC1 components suggests that KDM2B could function as targeting module for PRC1 and may contribute to polycomb mediated repression . Based on the clear enrichment of KDM2B at polycomb repressed genes and the inclusion of KDM2B in a PRC1 complex we examined whether depletion of KDM2B would affect the expression of polycomb repressed gene targets in mouse ESCs . To achieve this we stably knocked down KDM2B in ESCs using a lentiviral mediated shRNA approach . Knockdown was approximately by 80% at the RNA level ( Figure 5A ) and there was a clear reduction of KDM2B at the protein level ( Figure 5B ) . Importantly , depletion of KMD2B did not result in destabilization of either the PRC1 or PRC2 complexes allowing us to specifically examine defects resulting from reduced KDM2B ( Figure 5B ) . Microarray based gene expression profiling was then carried out to compare gene expression in the scrambled and KDM2B shRNA lines ( Figure 5C ) . Gene expression profiling revealed a substantial number of expression changes with genes being both down-regulated and up-regulated . In total 653 genes changed expression by more than 1 . 5-fold: 329 had reduced gene expression and 324 had increased gene expression . Based on the association of KDM2B with polycomb repressed genes we then focussed in on the set of 324 genes that are up-regulated in the KDM2B depletion line . Interestingly 24% of these genes are characterized by RING1B occupancy . This is significantly greater than would be expected by chance ( 1 . 9-fold enrichment , hypergeometric p=1 . 3 × 10−8 ) and is similar to the percentage of up-regulated genes in the RING1B null ESC cells that are PRC1 targets ( Leeb et al . , 2010 ) . Furthermore , when all gene expression changes are depicted in a volcano plot there is a clear propensity for RING1B occupied genes to have increased gene expression when KDM2B is depleted ( Figure 5C ) . In order to validate the observed up-regulation of polycomb targets in the KDM2B depletion cell lines we analysed a panel of 12 genes determined to be up-regulated based on the microarray using quantitative RT-PCR ( Figure 5D ) . In this analysis we observed corresponding increases in transcription in the KDM2B depletion line at these polycomb targets . Together these findings indicate that as with the removal of RING1B , depletion of KDM2B leads to up regulation of a subset of polycomb repressed target genes in mouse embryonic stem cells . 10 . 7554/eLife . 00205 . 010Figure 5 . KDM2B depletion in ESCs results in the up-regulation of a subset of polycomb repressed target genes . ( A ) RT-PCR showing reduction of KDM2B at the mRNA level in the knockdown ( KD ) but not control cell line . The levels of the closely related KDM2A mRNA were not affected . Error bars represent SEM of three biological replicates . ( B ) Western blot analysis showing that KDM2B knockdown depletes KDM2B but not the closely related KDM2A protein . Importantly depletion of KDM2B does not destabilize other polycomb group proteins including RING1B , EZH2 or EED . Lamin A/C indicates equal loading . ( C ) A volcano plot illustrating the gene expression changes in the KDM2B depletion line compared to the control cell line . The x-axis corresponds to fold change and the y-axis to the significance level . RING1B associated genes are coloured red and show preferential enrichment in the genes up-regulated upon KDM2B knockdown . ( D ) A panel of polycomb target genes identified in microarray analysis that were up-regulated upon depletion of KDM2B were validated by RT-PCR analysis . In most cases the level of up-regulation was more pronounced when analysed by RT-PCR . Error bars represent SEM of four biological replicates . Values are normalized to Gapdh and expression in the control line set to 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00205 . 010 The up-regulation of a subset of polycomb repressed genes following KDM2B depletion and its physical association with a PRC1 complex suggests that KDM2B may play a role in polycomb function at CGIs , perhaps acting as a targeting factor for the PRC1 complex via its capacity to recognize CGI DNA . To examine this possibility we carried out ChIP-seq analysis for RING1B and KDM2B in the control and KDM2B knockdown cell lines . As expected KDM2B occupancy was greatly reduced in the knockdown cell line , in fitting with the global reduction of KDM2B at the RNA and protein level ( Figure 6A ) . Next we examined RING1B occupancy in the control cell line . We confirmed high levels of RING1B at previously identified RING1B sites ( Figure 6B e . g . Atf3 gene ) . Interestingly however , based on our highly optimized RING1B ChIP protocol we also observed a large number of lower magnitude binding events ( ‘novel’ sites ) that did not correspond to sites previously characterized as being occupied by RING1B ( ‘established’ sites ) ( Figure 6B ) . To verify that these are bona fide RING1B binding events we carried out conventional ChIP followed by quantitative PCR on a series of these novel peaks in cells where we can conditionally delete RING1B and observed a clear loss of RING1B signal following deletion ( Figure 6—figure supplement 1 ) . This indicates these lower magnitude peaks correspond to novel RING1B binding events . Interestingly , 68% of the 15 , 740 newly identified RING1B binding sites overlapped with CGIs , and 77% of these were coupled to transcription start sites . Importantly , a minority of these novel RING1B occupied CGI sites showed detectable EZH2 ( 23% ) or H3K27me3 ( 20% ) , indicating that these sites are largely devoid of appreciable PRC2 . Including the already characterized RING1B associated CGIs , this means that clear RING1B peaks are detectable at nearly 70% of the 22 , 849 mouse CGIs genome-wide ( Illingworth et al . , 2010 ) , and 88% of RING1B occupied CGIs also are bound by detectable levels of KDM2B . 10 . 7554/eLife . 00205 . 011Figure 6 . Depletion of KDM2B causes a reduction in RING1B occupancy . ( A ) KDM2B ChIP-seq signal was plotted at CGIs in the control ( solid blue line ) and knockdown cell line ( dotted blue line ) . Sequencing signal in the input samples over the same regions are indicated as solid or dotted black lines . KDM2B ChIP-seq signal is specifically lost over CGIs genome-wide in the KDM2B knockdown cell line . ( B ) A snapshot showing RING1B ChIP-seq signal in the control cell line illustrating a previously identified high magnitude RING1B binding sites ( i . e . Atf3 ) and novel low magnitude RING1B binding sites . Input sequencing traces over the same region are shown in black . Bio-CAP-seq signal indicates regions containing non-methylated DNA . Above the sequencing traces gene promoters are show by black arrows and exons by vertical black lines . CGIs are shown as green bars with previously identified RING1B peaks and novel RING1B peaks indicated with purple boxes . ( C ) RING1B ChIP-seq signal from the control ( solid purple line ) and KDM2B knockdown cell line ( dotted purple line ) were plotted over previously identified CGI associated RING1B binding sites . Input sequencing signal was plotted over same regions . In the KDM2B knockdown line there is a specific reduction of RING1B binding over the CGI . ( D ) The same ChIP-seq signal as in part ( C ) was plotted over novel low magnitude binding sites identified in part ( B ) . In the KDM2B knockdown cell line there is an even more severe loss of RING1B binding at these novel RING1B occupied sites . ( E ) , ( F ) Examples of ChIP-seq profiles at individual high ( E ) and low ( F ) magnitude RING1B binding sites showing clear reduction of RING1B following KDM2B knockdown . DOI: http://dx . doi . org/10 . 7554/eLife . 00205 . 01110 . 7554/eLife . 00205 . 012Figure 6—figure supplement 1 . Conditional removal of RING1B validates the identity of novel peaks identified in the RING1B ChIP-seq . Eight genes corresponding to novel RING1B peaks identified in ChIP-seq were examined by ChIP quantitative-PCR ( bar graphs ) in the RING1B conditional ESC cells before ( blue bars ) and after ( red bars ) deletion of RING1B by addition of tamoxifen . In all cases deletion of RING1B leads to a loss of RING1B ChIP signal by quantitative-PCR , indicating that these are bona fide RING1B occupied sites . RING1B ChIP-seq traces are shown in each case for comparison . Error bars indicate the SEM for three biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 00205 . 01210 . 7554/eLife . 00205 . 013Figure 6—figure supplement 2 . KDM2B depletion results in a loss of RING1B at polycomb associated CGIs . ( A ) ChIP q-PCR analysis demonstrating that KDM2B depletion results in a clear reduction of KDM2B occupancy at polycomb and non-polycomb CGIs . Error bars represent SEM of three biological replicates ( B ) ChIP analysis demonstrating that KDM2B depletion does not result in major effects on KDM2A occupancy . Error bars represent SEM of three biological replicates . ( C ) ChIP analysis demonstrating that KDM2B depletion causes a reduction in RING1B occupancy at polycomb associated CGIs . Error bars represent SEM of four biological replicates . In all cases ChIP material was analysed by qPCR using primers specific for ( i ) non-CGI promoters , ( ii ) gene bodies , ( iii ) non-PcG target CGIs , ( iv ) PcG target CGIs , and ( v ) PcG target CGIs of genes upregulated in KDM2B knockdown cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00205 . 01310 . 7554/eLife . 00205 . 014Figure 6—figure supplement 3 . Genes associated with novel RING1B CGIs are on average expressed at higher levels than genes associated with established RING1B CGIs . Average microarray signal intensity for genes characterized as associated with established RING1B CGIs ( top panel ) , novel RING1B CGIs ( middle panel ) , and non-RING1B associated CGIs ( bottom panel ) . On average , novel RING1B associated CGI associated genes are expressed at similar levels to non-RING1B associated CGI genes ( median signal of 10 . 47 vs 10 . 24 ) . In contrast , established RING1B associated CGIs are on average expressed at lower levels ( median signal of 8 . 47 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00205 . 014 To examine whether KDM2B is required for RING1B association with CGIs , RING1B ChIP-seq signal from the control and KDM2B KD cell lines was initially plotted over previously identified RING1B CGI associated target sites ( Figure 6C ) . This revealed a specific reduction in RING1B signal at these sites following KDM2B knockdown in fitting with a role for KDM2B in RING1B targeting . This reduction in RING1B binding was evident when the ChIP-seq signal was visualized at individual target genes ( Figure 6E ) and similar changes in RING1B occupancy were observed by conventional ChIP q-PCR analysis at selected RING1B bound CGIs ( Figure 6—figure supplement 2 ) . Importantly , at sites where RING1B occupancy was affected , loss of binding correlated with the underlying non-methylated DNA signal and KDM2B occupancy ( Figure 6E ) . These observations indicate that depletion of KDM2B causes a loss of normal RING1B targeting at polycomb occupied CGI regions of the genome , in accordance with the observed reactivation of a subset of polycomb repressed genes in the KDM2B knockdown line . Based on the observation that RING1B is more widely associated with CGI elements than previously appreciated ( Figure 6B ) , we next examined whether KDM2B contributes to RING1B binding at these novel sites by plotting RING1B signal in the control and KDM2B knockdown cell lines over these regions ( Figure 6D , F ) . Interestingly , depletion of KDM2B resulted in an even more dramatic effect on RING1B occupancy at these novel low magnitude sites , suggesting that these RING1B binding events are more dependent on KDM2B . Importantly , the observed reduction in RING1B occupancy at high and low magnitude sites was not due to destabilization of polycomb repressive systems , as the protein levels of both PRC1 and PRC2 components were similar in the wild type and KD cell lines ( Figure 5B ) . Together these observations indicate that KDM2B plays a widespread role in targeting RING1B in vivo both at previously identified established polycomb-repressed CGI-associated genes and a novel class of lower magnitude RING1B occupied CGI sites . A major function of the PRC1 complexes is to mono-ubiquitylate histone H2A on position 119 ( H2AK119ub1 ) and this activity is essential for PRC1 mediated gene repression ( Endoh et al . , 2012 ) . A monoclonal antibody for H2AK119ub1 ( Vassilev et al . , 1995 ) has been widely used to evaluate H2AK119ub1 in vivo . Unfortunately , this antibody performs poorly under some experimental conditions , notably chromatin immunoprecipitation . To overcome this limitation we generated a novel antibody against a branched peptide encompassing the isopeptide junction between H2AK119 and the Ub chain . Following immunization , a double affinity purification strategy yielded a highly specific H2AK119ub1 antibody that recognizes native H2AK119ub and functions in ChIP ( Figure 7—figure supplements 1–3 ) . Using this novel H2AK119ub1 specific antibody we set out to examine whether depletion of KDM2B affected PRC1 catalysed chromatin modification in vivo . To achieve this we first compared the global level of H2AK119ub1 in the WT and KD cell lines ( Figure 7A , B ) . Interestingly , depletion of KDM2B results in an approximately 40% global reduction of this modification , indicating that KDM2B contributes significantly to H2AK119ub1 at the genome scale , most likely due to its widespread role in guiding RING1B occupancy . This was not the result of more general changes in chromatin modification as other histone marks including H3K4me3 and H3K27me3 were largely unaffected ( Figure 7C ) . 10 . 7554/eLife . 00205 . 015Figure 7 . Depletion of KDM2B causes a reduction in H2AK119ub1 . ( A ) KDM2B knockdown results in a global loss of H2AK119ub1 , as demonstrated by western blot for H2AK119ub1 in KDM2B knockdown and control cells . Western blot for total H2A is shown as a loading control . ( B ) Quantification of H2AK119ub1 levels by fluorescence based quantitative western blotting . The levels of H2AK119ub1 are approximately 40% lower in KDM2B knockdown compared to control cells . The western blot signal for H2AK119ub1 was quantified relative to H2A and the error bars represent the SEM of six biological replicates . ( C ) KDM2B knockdown does not cause global changes in levels of H3K27me3 , H3K4me3 or H3K36me2 , as demonstrated by western blot in the KDM2B knockdown and control cells . Western for total Histone H3 is shown as a loading control . ( D ) KDM2B knockdown cells show locus specific depletion of H2AK119ub1 at genes up-regulated in the KDM2B knockdown cell line . Comparatively there are only small changes in H3K27me3 and H3K4me3 at these same sites . ChIP material was analysed by qPCR using primers specific for ( i ) non-CGI promoters , ( ii ) gene bodies , ( iii ) Non-PcG target CGIs , ( iv ) PcG target CGIs , and ( v ) PcG target CGIs of genes up-regulated in KDM2B knockdown cells . Error bars represent the SEM of four biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 00205 . 01510 . 7554/eLife . 00205 . 016Figure 7—figure supplement 1 . Purification of an H2AK119ub1 antibody . A rabbit was immunized with a synthetic branched peptide containing H2AK119ub1 . Serum from the immunized animal was tested against histone extracts from cells that contain a conditional allele of Ring1b that can be deleted by the addition of the drug tamoxifen . Drug treatment removes the RING1B H2AK119 E3 ligase and eliminates H2AK119ub1 . In the crude serum from the immunized animal there is a clear H2AK119ub1 signal in the RING1B containing cell line that is lost upon drug treatment ( panel i ) , however there was some cross-reactivity with unmodified H2A . To remove the H2A cross-reactivity the serum was first depleted of H2A reactivity ( panel ii ) and the flowthrough ( panel iii ) was then affinity purified on a column containing the H2AK119ub1 antigen ( panel iv ) yielding a highly specific antibody against H2AK119ub1 when compared to the commercially available H2AK119ub1 monoclonal antibody ( panel v ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00205 . 01610 . 7554/eLife . 00205 . 017Figure 7—figure supplement 2 . The purified H2AK119ub1 antibody recognizes a Xist inactivated chromosome . The H2AK119ub1 antibody recognizes native H2AK119ub1 in cells as indicated by specific immunofluorescence staining of a Xist inactivated chromosome in mouse embryonic stem cells . H3K27me3 specific antibodies are used to identify the Xist inactivated chromosome and the merged image shows a clear overlap between H3K27me3 and H2AK119ub1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00205 . 01710 . 7554/eLife . 00205 . 018Figure 7—figure supplement 3 . The purified H2AK119ub1 antibody works in chromatin immunoprecipitation . The H2AK119ub1 antibody works specifically in chromatin immunoprecipitation as indicated by a loss of H2AK119ub1 ChIP signal following removal of RING1B in the conditional mouse ESCs . DOI: http://dx . doi . org/10 . 7554/eLife . 00205 . 018 To understand whether these global reductions in H2AK119ub1 were also evident at polycomb repressed target genes , we analysed H2AK119ub1 at a panel of CGI associated genes . Interestingly reductions in H2AK119ub1 appeared to be restricted to the polycomb target CGIs that were up-regulated in the absence of KMD2B ( Figure 7D ) . This also correlated with a more subtle reduction in H3K27me3 but largely unchanged levels of H3K4me3 ( Figure 7D ) . This suggests that although KDM2B plays a role in the nucleation of RING1B at most polycomb associated CGIs , the level of RING1B reduction at individual genes likely dictates the effects on H2AK119ub1 and gene expression . This is in accordance with the recent observation that H2AK119ub1 is important for gene repression mediated by PRC1 ( Endoh et al . , 2012 ) . Interestingly , the level of H2AK119ub1 reduction at individual loci did not appear as dramatic as the global reductions observed by western blotting . One possibility for this difference may be that pervasive KDM2B dependent targeting of PRC1 to CGIs leads to low level deposition of H2AK119ub1 away from established polycomb sites . One can envisage how loss of KDM2B would more dramatically affect this additional pool of H2AK119ub1 resulting in significant global reduction in its abundance . Together these observations suggest that some polycomb repressed sites rely more specifically on KDM2B mediated RING1B occupancy for normal H2AK119ub1 and polycomb mediated gene repression ( Endoh et al . , 2012 ) . Here we demonstrate that KDM2B recognizes CGI elements via its ZF-CxxC domain and in contrast to its paralogue , KDM2A , is preferentially enriched at polycomb-repressed CGIs in mouse ESCs ( Figure 2 ) . By virtue of its physical association with a variant PRC1 complex ( Figure 3 ) and through the functionality of its ZF-CxxC domain ( Figure 4 ) , KDM2B is required for normal targeting of the catalytic component of the PRC1 complex , RING1B , to CGI elements ( Figure 6 ) . Depletion of KDM2B results in a reduction of RING1B at polycomb repressed CGIs and leads to diminished global H2AK119ub1 ( Figure 7 ) . This causes reactivation of a subset of polycomb repressed target genes ( Figure 5 ) . Surprisingly , we also uncover a more widespread association of RING1B with CGIs that is also dependent on KDM2B ( Figure 6 ) . Together these observations provide an unexpected and direct link between CGIs , recognition of non-methylated DNA by KDM2B , and binding of the PRC1 complex to target sites in vivo . The mechanisms underpinning how polycomb repressive complexes are recruited to target sites in mammals have remained largely elusive , with the exception that CGIs appear to function as the mammalian equivalent of the Drosophila polycomb response element ( Simon and Kingston , 2009 ) . In most species PRC1 is thought to be mechanistically recruited in a hierarchical manner through chromo-domain mediated recognition of H3K27 methylation at pre-existing PRC2 modified sites ( Cao et al . , 2002; Min et al . , 2003; Wang et al . , 2004b ) . This largely places PRC1 as a subservient reader of silencing events initiated by PRC2 . However , the singularity of the hierarchal recruitment mechanism has recently been challenged by a series of elegant experiments in mammals demonstrating that although genetic perturbation of the PRC2 complexes reduces PRC1 localization to polycomb repressed CGIs ( Tavares et al . , 2012 ) , H3K27 independent PRC1 targeting mechanisms are sufficient to maintain H2AK119ub1 and repression at many PRC1 bound targets ( Schoeftner et al . , 2006; Pasini et al . , 2007; Leeb et al . , 2010; Tavares et al . , 2012 ) . The specific molecular mechanisms underpinning H3K27me3 independent PRC1 targeting nevertheless remain unclear . For example , a PRC1 variant complex containing RYBP and lacking chromobox domain-containing proteins appears to be involved in this process but it is unclear how RYBP would achieve specificity for polycomb targets as it has no inherent chromatin or DNA binding specificity ( Arrigoni et al . , 2006; Neira et al . , 2009; Tavares et al . , 2012; Gao et al . , 2012; Hisada et al . , 2012 ) . Furthermore RYBP depletion appears to destabilize the RING1B protein ( Tavares et al . , 2012 ) , reducing the effective pool of the enzyme and making it difficult to determine the specific contribution of RYBP to targeting . Based on our observations that KDM2B is important for RING1B binding to CGI elements , the KDM2B-PRC1 complex provides an alternative mechanism for PRC1 recognition of polycomb repressed CGIs in mammals . KDM2B forms a variant PRC1 complex lacking CBX proteins but interestingly contains RYBP or the related YAF2 protein . In contrast to depletion of RYBP , reduction of KDM2B did not destabilize RING1B , allowing us to demonstrate that KDM2B contributes to the localization of RING1B to target sites in vivo . This targeting event appears to play an important role in the repression of a subset of polycomb target genes through H2AK119 ubiquitylation . Together these observations provide the first mechanistic evidence for the existence of a PRC1 targeting pathway that does not rely on recognition of H3K27me3 but instead interrogates the underlying non-methylated DNA sequence present at CGIs . Based on this work we propose an alternative to the hierarchical model of PRC1 targeting whereby CGIs are recognized by a variant PRC1 complex containing KDM2B through direct binding to non-methylated DNA . Other direct recognition mechanisms based on interactions between site specific DNA binding transcription factors and PRC1 have recently been identified further supporting the concept that DNA information at polycomb repressed sites is important for PRC1 targeting in vivo ( Ren and Kerppola , 2011; Dietrich et al . , 2012; Yu et al . , 2012 ) . Together , the hierarchical recruitment system and alternative direct targeting mechanisms likely function in a complementary and in some cases redundant fashion at many polycomb targets in vivo . This is supported by the observation that KDM2B depletion in embryonic stem cells results in up-regulation of only a subset of PRC1 target genes . Therefore , one can envision that polycomb repression , which must be maintained over cell divisions and developmental time scales , may rely on the combined activity of these separate pathways as an important stability component for maintenance of the silenced state . KDM2B is enriched at polycomb repressed sites in vivo , but genome-wide analysis revealed it is also more broadly associated with CGIs . Based on this observation , an interesting question remains regarding how KDM2B specifically contributes to PRC1 recognition of polycomb targets , while also binding to CGIs throughout the genome . Our surprising observation that RING1B is detectable at nearly 70% of all CGIs is supportive of a model whereby KDM2B may act as a sampling module allowing PRC1 to be recruited at very low levels to most or all CGIs by virtue of its capacity to recognize CGIs . This would then imply that productive PRC1 enrichment and polycomb mediated silencing is in turn only achieved at a subset of these sampled targets . Fittingly , novel RING1B associated CGIs do not differ in expression from the average gene ( Figure 6—figure supplement 3 ) , whereas established polycomb CGIs show reduced expression . This is largely in agreement with recent mechanistic work in mammalian systems studying the capacity of CGIs to attract polycomb mediated repression , where it has been suggested that the default state for CGIs is indeed to acquire polycomb mediated silencing but local chromatin features and other activating marks appear to function to counteract this process ( Mendenhall et al . , 2010; Lynch et al . , 2012 ) . Conceptually this parallels the long held view from genetic and molecular studies in Drosophila that indicate a continual antagonism exists between the function of positively acting trithorax chromatin modifying systems and repressive polycomb systems at PREs ( Poux et al . , 2002; Klymenko and Müller , 2004 ) . In Drosophila , ultimately this balance is resolved by the local activity of transcriptional regulators . In analogy to the situation at Drosophila PREs , the KDM2B-PRC1 complex may therefore function to continually recruit PRC1 to CGIs at low levels but ultimately only establish the fully PRC1/PRC2 occupied and polycomb repressed state if the appropriate local chromatin and transcriptional environment is met . It is tempting to speculate that PRC1 , and perhaps even PRC2 via its poorly characterized DNA binding components ( Cao et al . , 2002; Kim et al . , 2009; Li et al . , 2010 ) , may be continually interfacing CGIs in a dynamic manner in order to constantly sample their susceptibility to polycomb repression . Once the conditions are met for polycomb mediated silencing , one can envision a simple feed forward loop may exist where the combined activities of PRC1 and PRC2 stabilize each other through canonical PRC1 complexes that contain CBX proteins recognising H3K27me3 placed by PRC2 ( Cao et al . , 2002; Wang et al . , 2004b ) and PRC2 activity favouring the increasingly compacted chromatin state associated with PRC1 occupancy ( Francis et al . , 2004; Eskeland et al . , 2010; Yuan et al . , 2012 ) . This could account for the remarkable spatial correlation between mammalian polycomb repression domains and CGIs at established polycomb repressed sites ( Ku et al . , 2008 ) . This speculative model is appealing as it would in part explain the co-occupancy of PRC1 and PRC2 at most established polycomb sites in vivo and could reconcile why only certain CGIs are susceptible to acquiring full polycomb complex nucleation and silencing in vivo . In order to fully realize the complexities and interplay between polycomb silencing activities , more detailed molecular and biochemical understanding of this system is clearly required . For example , our study has focussed on ESCs where polycomb factors are highly abundant , so it will be imperative to understand if similar KDM2B dependent PRC1 targeting mechanisms function in more committed cell types . Also , it is known that H3K27me3 and CBX dependent PRC1 targeting activities play an important role in RING1B occupancy . Therefore , KDM2B dependent targeting of the PCGF1-PRC1 complex on its own is insufficient to account for the functionality of PRC1 in vivo . This suggests that a complex set of PRC1 targeting mechanisms , perhaps utilizing specific PRC1 complexes , and based on both direct and indirect recognition of CGI associated sites is required for normal PRC1 function . In agreement with this idea , knockdown of KDM2B leads to reactivation of only a subset of polycomb repressed genes . Although this may be due to incomplete removal of KDM2B by RNAi , it may also be related to compensatory function of other PRC1 complexes . Nevertheless , our demonstration that KDM2B links PRC1 to recognition of CGIs provides an important new piece of the puzzle and demonstrates for the first time a direct mechanism for polycomb recognition of CGIs elements in vivo . Importantly , it also provides a novel conceptual framework on which to consider initiation of polycomb mediated silencing . The full length human KDM2B , mouse RING1B , mouse YAF2 ( IMAGE clone 3488691 ) , human PCGF1 ( IMAGE clone 3621400 ) , mouse RYBP ( IMAGE clone 8861345 ) were PCR-amplified and inserted via ligation independent cloning ( LIC ) into a pCAG-IRES-puro eukaryotic expression vector that has been modified to express a N-terminal Flag and 2XStrep2 tag and contain LIC cloning sites . All PCR generated constructs were verified by sequencing . Mutation of the KDM2B ZF-CxxC DNA binding domain ( K643A ) was introduced into the wild type KDM2B construct by Quikchange mutagenesis XL kit ( Stratagene , Santa Clara , CA ) . A polyclonal antibody against KDM2B was generated by immunizing a rabbit ( PTU/BS Scottish National Blood Transfusion Service ) with a Hisx6-tag fusion protein encoding amino acids 755–917 of human KDM2B protein . KDM2B antigen was coupled to AffiGel10 resin ( BioRad , Hemel Hempstead , UK ) and the antibody was affinity purified . The RING1B mouse monoclonal antibody has been described previously ( Atsuta et al . , 2001 ) . To obtain a highly pure RING1B antibody preparation , hybridoma cells were cultured and the antibody was purified from tissue culture supernatant by protein A agarose ( RepliGen , Waltham , MA ) based affinity chromatography . A rabbit polyclonal antibody against Flag-2XStrepII sequence was generated by immunization of a rabbit with a synthetically synthesized peptide conjugated to mariculture KLH carrier protein ( ThermoScientific , Waltham , MA ) . Flag-2XStrepII peptide was covalently immobilized on SulfoLink resin ( ThermoScientific ) and antibody was affinity purified prior to use . Polyclonal antibody against histone H3 and histone H3 lysine 4 trimethylation ( H3K4me3 ) were generated using the synthetic peptides CIQLARRIRGERA , and ART ( K ) QTARKSTGGC ( where brackets indicate position of trimethyl mark ) , respectively . For both antibodies , peptides were conjugated to mariculture KLH carrier protein ( ThermoScientific ) prior to rabbit immunization , and after obtaining serum from immunized animals the antibodies were affinity purified using the respective peptide antigen coupled to SulfoLink resin ( ThermoScientific ) . A polyclonal antibody against histone H3 lysine 36 dimethylation ( H3K36me2 ) has been described previously ( Blackledge et al . , 2010 ) , and commercially available antibodies were used to detect Histone H3K27 trimethylation ( H3K27me3 ) ( Diagenode , Liege , Belgium; 069-050 ) and total histone H2A ( Millipore , Billerica , MA; 07-146 ) . A branched peptide was synthesised ( by GL Biochem , Shanghai , China ) with the sequence H2N-CVLLPKK ( H2N-LRGG ) TESHHK-NH2 , where the H2N-LRGG branch is coupled to the Lys119 residue through a Lys–Gly isopeptide bond . This branched peptide was conjugated to maleimide-activated KLH ( Pierce Protein Biology Products , Rockford , IL ) and a rabbit was immunised with the peptide-conjugate KLH . Following immunisation H2AK119ub1 specific antibodies were purified in a two-step process . First , a peptide of the sequence H2N-CVLLPKKTESHHK-NH2 , corresponding to amino acids 114–125 of histone H2A , was immobilised on SulfoLink Resin ( Pierce Protein Biology Products ) . Serum was purified over this resin to remove antibodies cross-reactive with unmodified H2A . Secondly , the flow-through from this resin was then applied to a second Sulfolink Resin on which was immobilised the original branched peptide ( H2N-CVLLPKK [H2N-LRGG]TESHHK-NH2 ) used for immunisation . Murine V6 . 5 ESC cells ( C57BL/6 ( F ) × 129/sv ( M ) ) were cultured on Mitomycin C inactivated mouse embryonic fibroblasts ( MEFs ) in DMEM ( Gibco , Carlsbad , CA ) supplemented with 15% FBS , 10 ng/mL leukemia-inhibiting factor ( LIF ) , penicillin/streptomycin , beta-mercaptoethanol , l-glutamine and non-essential amino-acids . Prior to use in ChIP experiments , V6 . 5 cells were cultured for two passages under feeder-free conditions . Feeder-independent E14 and EFC-1 mouse embryonic stem cells were grown on 0 . 1% gelatine . Ring1A−/−; RING1Bfl/fl; Rosa26::CreERT2 ESCs were allowed to settle overnight without feeders and cultured in the absence or presence of 4-hydroxy tamoxifen ( 800 μM ) for 48 hr to deplete RING1B . For generation of stable cell lines , 3 μg expression constructs were transfected into E14 or EFC-1 cells using Lipofectamine 2000 ( Invitrogen , Carlsbad , CA ) , and stable integrants were selected using 1 . 5 μg/mL puromycin . Staining was performed in HeLa , 293T , IMR90 and U2OS cells , which were cultured in DMEM ( Lonza , Basel , Switzerland ) supplemented with 10% FCS and penicillin/streptomycin . Transient transfection in HeLa cells was achieved using Fugene HD reagent ( Roche , Basel , Switzerland ) and 2 μg of Flag-tagged KDM2B expression plasmid . Briefly , cells seeded on cover slips were fixed with 4% paraformaldehyde for 20 min , permeabilized with 0 . 5% Triton X-100 for 10 min and blocked using 3% BSA for 30 min . Fixation was done approximately 24 hr after transfections . Cells were incubated with KDM2B rabbit polyclonal primary antibody for 2 hr or the Flag mouse monoclonal M2 antibody ( Sigma , St Louis , MO ) in case of exogenous expression , followed by incubation in secondary antibody conjugated with Rhodamine or FITC ( Jackson ImmunoResearch Laboratories , West Grove , PA ) for 1 hr . All intermediary washing steps and dilutions were performed in 1× PBS . After 4 , 6-diamidino-2-phenylindole dihydrochloride ( DAPI ) nuclear staining , the cover slips were mounted on glass slides in fluorescent mounting medium ( DAKO , Glostrup , Denmark ) and allowed to dry overnight . Images were aquired using a Zeiss AxioSkop fluorescent microscope . Recombinant ZF-CxxC constructs encompassing human KDM2A ( encoding amino acids 558–704 ) and KDM2B ( encoding amino acids 600–750 ) were engineered to include an N-terminal 6-his tag followed by a tobacco etch virus ( TEV ) protease cleavage site as previously described ( Blackledge et al . , 2010; Blackledge et al . , 2012 ) . The ZF-CxxC protein was isolated by Ni-NTA-mediated purification as described before and TEV-cleavage as described previously ( Blackledge et al . , 2012; Klose and Bird , 2004 ) . EMSA probes were generated and labelled and EMSA was performed as previously described with samples analysed on a 1 . 3% agarose gel ( Blackledge et al . , 2010 ) . Scrambled or KDM2B shRNAs were cloned into pLKO . 1puro ( Addgene , Cambridge , MA ) and sequence verified . For producing recombinant lentiviral particles , the shRNA constructs were co-transfected with psPAX2 packaging plasmid and pMD2 . G envelope helper plasmid into 293T cells using FuGene ( Roche , Basel , Switzerland ) . The 21mer shRNA sequences used were: ( Kdm2b ) 5′-CGCTGTGGAAATATCTGTCAT-3′ , ( control ) 5′-CCTAAGGTTAAGTCGCCCTCG-3′ . Lentiviral infection of E14 cells was performed overnight in the presence of 4 μg/mL polybrene . The following day , cells were diluted into fresh growth media and allowed to settle onto gelatine-coated dishes . Puromycin selection ( 1 . 5 μg/mL ) was started 48 hr following transduction and stable lines were isolated and expanded . Cells were harvested by scraping , washed with 1× PBS and equilibrated in 10 pellet volumes of buffer A ( 10 mM Hepes pH 7 . 9 , 1 . 5 mM MgCl2 , 10 mM KCl , 0 . 5 mM DTT , 0 . 5 mM PMSF , and complete protease inhibitors [Roche , Basel , Switzerland] ) . The cells were incubated on ice for 10 min and recovered by centrifugation at 1500×g for 5 min . The pellet was resuspended in 3 volumes of buffer A supplemented with 0 . 1% NP-40 and incubated on ice for 10 min with gentle agitation . Nuclei were recovered by centrifugation at 1500× rpm for 5 min . Recovered nuclei were then resuspended in 1× pellet volume of buffer B ( 5 mM Hepes [pH 7 . 9] , 26% glycerol , 1 . 5 mM MgCl2 , 0 . 2 mM EDTA , and complete protease inhibitors [Roche] , 0 . 5 mM DTT ) supplemented with 400 mM NaCl . The nuclei were first resuspended in a buffer with 250 mM salt . The volume of the buffer and the cells was taken into account , and the salt concentration was increased to 400 mM by gently adding concentrated 5 M NaCl dropwise while mixing . The extraction was allowed to proceed for 1 hr on ice with occasional agitation , and then the nuclei were pelleted at 13 , 000 rpm for 20 min at 4°C . The supernatant was taken as the nuclear extract . For purification of KDM2B-Flag/StrepII , the salt concentration of the nuclear extract was first reduced to 150 mM NaCl by dilution into nuclear extraction buffer B without salt . Between 10–15 mg of nuclear extract were used for each large-scale affinity purification . Avidin ( IBA , Goettingen , Germany; 20 μg/1 mg extract ) was added to remove biotinylated molecules which would bind unspecifically to the matrix . If benzonase treatment was performed , 75 U/mL nuclear extract benzonase nuclease ( Novagen , Darmstadt , Germany ) was added at this point . Extracts were pre-cleared for 30 min at 4°C , followed by a max speed centrifugation ( 13000× rpm ) for 5 min . Cleared extract was then added to 50 μL packed StrepTactin superflow high-capacity resin ( IBA ) and allowed to rotate gently for 3 hr at 4°C . Eight wash steps were performed ( 20 mM Tris pH 8 . 0 , 500 mM NaCl , 0 . 2% NP-40 , 1 mM DTT ( fresh ) and 5% glycerol ) in 1 . 5 mL low bind tubes ( Eppendorf , Hamburg , Germany ) . Bound material was eluted in buffer containing 20 mM Tris pH 8 . 0 , 150 mM NaCl , 0 . 2% NP-40 , 1 mM DTT ( fresh ) , 5% glycerol and 10 mM D-desthiobiotin ( IBA ) . The same experimental setup was used to identify associated polypeptides for RING1B , YAF2 , RYBP and PCGF1 . The immunoprecipitated solution was desalted using chloroform-methanol extraction , followed by overnight in-solution tryptic digestion and subsequent peptide purification using Waters C18 Sep-Pak cartridges . The analysis of immunoprecipitated digested material was performed by LC–MS/MS using an Orbitrap Velos ( Thermo mass spectrometer ) coupled to a nano-UPLC system ( NanoAcquity , Waters , Milford , MA ) using a reversed phase 75 μm × 250 mm , 1 . 7μm particle size column , as described ( Fischer et al . , 2012 ) . MS/MS spectra were searched against the UniProt SwissProt Mus database ( v2011 . 11 . 18 , 16 , 460 sequences ) in Mascot v2 . 3 . 01 , allowing one missed cleavage and 20 ppm/0 . 5 Da mass deviations in MS/MSMS , respectively . Carbamidomethylation of cysteine was a fixed modification . Oxidation of methionine , and deamidation of asparagine and glutamine were used as variable modifications . Protein assignment was based on at least two peptides identified . For each immunoprecipitation , 0 . 5 mg nuclear extract was used in buffer BC150 ( 150 mM KCl , 10% glycerol , 50 mM HEPES [pH 7 . 9] , 0 . 5 mM EDTA , 0 . 5 mM DTT [added fresh] ) . Approximately 3 μg of antibody of interest was added and the reaction allowed to incubate overnight at 4°C . The next morning , 20 μL of packed protein A beads ( RepliGen , Waltham , MA ) were added , and the samples rotated for 1 hr at 4°C . Following washing with BC300 ( 300 mM KCl , 10% glycerol , 50 mM HEPES [pH 7 . 9] , 0 . 5 mM EDTA , 0 . 5 mM DTT [added fresh] ) , the beads were resuspended in a small volume of SDS-PAGE loading dye and boiled at 95°C for 5 min . After spinning at 1000×g for 5 min , the supernatant was used for western blotting . Total RNA was extracted using the Qiagen RNeasy Mini kit . Approximately 10 μg nucleic acid were treated with Turbo DNase ( Ambion , Carlsbad , CA ) at 37°C for 30 min , according to the manufacturer's instructions . Genomic DNA-free RNA samples were further purified using the RNeasy kit RNA cleanup protocol . Samples were run on an 1% agarose gel to check quality of RNA preparation and integrity of 18S and 28S rRNA bands . For subsequent RT-PCR analysis , cDNA was synthesized with the ImProm-II Reverse Transcription System ( Promega , Madison , WI ) . Quantitative real-time PCR was performed in duplicate by using Quantace SYBR Green master mix , using Gapdh as housekeeping gene . For microarray studies , RNA integrity was assessed on a BioAnalyzer; all samples had a RNA Integrity Number ( RIN ) ≥9 . 5 ( Agilent Laboratories , Santa Clara , CA ) . Sense ssDNA was generated from 300 ng starting RNA with the Ambion WT Expression Kit according to the manufacturer's instructions . Sense ssDNA was fragmented and labeled using the GeneChip WT Terminal Labeling and Controls Kit . The fragmented peak size was measured on the BioAnalyzer and was in the expected 40–70 nt range . The labeled ssDNA from four biological replicates was hybridized to Affymetrix Mouse Gene 1 . 0 ST Array ( Affymetrix , Santa Clara , CA ) . Chips were processed on an Affymetrix GeneChip Fluidics Station 450 and Scanner 3000 . Cel files were generated using Command Console ( Affymetrix ) . Affymetrix microarray probe intensities were normalised using RMA normalisation via the Bioconductor/R package oligo ( version 1 . 18 . 1 ) ( Carvalho and Irizarry , 2010 ) and differences tested using the Bioconductor/R package limma ( version 3 . 10 . 3 ) ( Smyth , 2004 ) . In order for expression to have been considered changed , a majority of the probe sets targeting the gene needed a Benjamini and Hochberg corrected FDR of less than 0 . 05 and the most significant probe required a fold change of at least 1 . 5-fold between KDM2B knockdown and scrambled control samples . Enrichments in RING1B marked genes were tested by calculating the proportion of genes significantly changed compared to the fraction of genes on the array marked . Differences were tested using the hypergeometric test . Odds ratio for the RING1B marked genes to increase in expression vs decrease was tested using Fisher's exact test . Chromatin immunoprecipitation was performed as previously described ( Schmidt et al . , 2009 ) , with minor modifications . For KDM2A , KDM2B and RING1B ChIP , cells were fixed for 1 hr in 2 mM EGS , followed by 15 min in 1% formaldehyde , while for histone modification ChIP cells were fixed for 10 min in 1% formaldehyde alone . In both cases , formaldehyde was quenched by the addition of glycine to a final concentration of 125 μM . Sonication was performed using a BioRuptor sonicator ( Diagenode , Liege , Belgium ) to produce fragments of approximately 0 . 5–1 kb . Immunoprecipitation was performed overnight at 4°C with approximately 3 µg of antibody and chromatin corresponding to 5 × 106 cells . Antibody bound proteins were isolated on protein A agarose beads ( RepliGen , Waltham , CA ) or protein A magnetic Dynabeads ( Invitrogen , Carlsbad , CA ) , washed extensively , eluted , and cross-links reversed according to the Upstate protocol . Samples were then sequentially treated with RNase and proteinase K before being purified on a PureLink PCR micro column ( Invitrogen ) . Real-time qPCR was performed using Sybr Green ( Quantace , London , UK ) on a Rotor-Gene 6000 ( Corbett , Hilden , Germany ) or sequencing libraries were generated as described previously ( Blackledge et al . , 2010 ) and sequenced on the Illumina HiSeq2000 platform with 51 bp reads . Published datasets for CpG island ( CGI ) intervals were obtained from ( Illingworth et al . , 2010 ) . Published RING1B ChIP-seq and input reads from mouse ESCs were obtained from GSM585229 ( Tavares et al . , 2012 ) and intervals were called as described below . EZH2 ChIP-seq from mouse ESCs was obtained from GSM480161 ( Peng et al . , 2009 ) . TSS annotation and GO slim categories were obtained from Ensembl 66 . All sequence was mapped using Bowtie ( Langmead et al . , 2009 ) version 0 . 12 . 7 against the mouse genome ( mm9 ) . Up to two mismatches were allowed and only uniquely mapping reads were kept . Mapped reads were then de-duplicated using Picard to remove potential PCR-duplicates . Where appropriate , sets of mapped reads were normalized by random sampling of the larger set to the size of the smaller . Peak intervals were generated by MACS 1 . 4 . 2 ( Zhang et al . , 2008 ) using normalized chip and input samples . A p-value threshold of 1 × 10−5 and a fold enrichment over input threshold of 6 was used to filter the resulting peak intervals . Intervals closer than 200 bp were merged . Profile plots were generated using the custom script bam2geneprofile by counting reads mapping over each base of a pre-defined set of intervals and averaging the depth . In the case of Figure 6C , interval lengths were normalised . In all other cases a fixed window size about a point was used . Where two samples are shown on the same plot , the numbers of mapping reads has been normalised . Results are normalised over the number of intervals in the input set . To calculate enrichment of KDM2B over KDM2A at CGIs overlapping TSSs , the coverage of each factor which overlapped a 1 kb window around any TSS was calculated using the coverageBed tool from BedTools ( Quinlan and Hall , 2010 ) on normalised read sets . For each CGI the ratio of KDM2B to KDM2A reads was calculated . The genes associated with CGIs that showed a greater than twofold enrichment of KDM2B over KDM2A were used to perform GO analysis . Enrichment of GO terms from the Biological Function category of the GO slim hierarchy was tested using the custom script GO . py . The significance of each category was tested using a hypergeometric test and FDRs calculated using the method of Benjamini and Hochberg ( Benjamini and Hochberg , 1995 ) . Terms with an FDR <0 . 05 and fold enrichment greater than 3 . 5 are shown . ChIP-seq heatmaps were generated using the custom script bam2peakshape . py . Briefly , each CGI window was divided into 25 bp bins and the number of reads mapping to each bin counted . Reads were then normalised to the total number of reads mapping across all intervals . Intervals were then ranked by the number of reads in the RING1B set that map to that interval . Source code for custom scripts described is available in the mercurial repository at www . cgat . org/hg/cgat . The change-set used was 114addb46882 . Sequencing and microarray data can be accessed via the geo accession GSE41267 .

Gene expression in eukaryotic cells can be controlled in a number of different ways , including various epigenetic mechanisms that do not involve making changes to DNA sequences that define the genes themselves . A well-known epigenetic mechanism for silencing genes in vertebrates is DNA methylation—the addition of a methyl group ( CH3 ) to cytosine , which is one of the four bases found in the DNA . Methylation is thought to silence genes by preventing transcription factors from binding to the DNA , and also by recruiting proteins that inhibit the transcription of DNA . DNA methylation occurs naturally throughout the genome , mostly at positions where cytosine is bonded to guanine to form a CpG dinucleotide . While the cytosine bases in most CpG dinucleotides are methylated , there are short stretches of DNA known as CpG islands that contain a high proportion of unmethylated CpG dinucleotides . These islands contain a large number of cytosine and guanine bases , and they are often found at or near transcription start sites . The lack of methylation at CpG islands has long been assumed to have a passive role in gene expression , leaving the DNA easily accessible and available for transcription factors to bind and initiate transcription . However , recent work suggests that CpG islands may have a more active role . In particular , it has been shown that specific proteins bind to CpG islands to create chromatin environments that are more favourable for the initiation of gene expression . Moreover , a subset of CpG islands can also bind polycomb-group proteins , including the polycomb repressive complex 1 ( PRC1 ) that silence gene expression . These complexes have an important role in the regulation of genes during early development in animals , but the mechanism by which PRC1 recognizes CpG islands in mammals has remained enigmatic . Farcas et al . now reveal that a protein , KDM2B ( FBXL10 ) , can recognize CpG islands and recruit PRC1 to them . To achieve this , KDM2B encodes a DNA binding domain that specifically recognizes non-methylated CpG dinucleotides . By interacting biochemically with a variant PRC1 complex , KDM2B then nucleates PRC1 at CpG islands , and PRC1 activity silences certain polycomb target genes in embryonic stem cells . Surprisingly , Farcas et al . also find low but appreciable levels of PRC1 at most CpG islands genome-wide , in addition to the high levels of PRC1 at selected islands: this suggests that KDM2B may sample the whole genome to find CpG islands where PRC1 can establish silencing . An improved understanding of the polycomb repressive system , and the role of CpG islands within it , could lead to new insights into the role of epigenetic mechanisms in mammalian development .

2012

KDM2B links the Polycomb Repressive Complex 1 (PRC1) to recognition of CpG islands

Neural computations underlying cognitive functions require calibration of the strength of excitatory and inhibitory synaptic connections and are associated with modulation of gamma frequency oscillations in network activity . However , principles relating gamma oscillations , synaptic strength and circuit computations are unclear . We address this in attractor network models that account for grid firing and theta-nested gamma oscillations in the medial entorhinal cortex . We show that moderate intrinsic noise massively increases the range of synaptic strengths supporting gamma oscillations and grid computation . With moderate noise , variation in excitatory or inhibitory synaptic strength tunes the amplitude and frequency of gamma activity without disrupting grid firing . This beneficial role for noise results from disruption of epileptic-like network states . Thus , moderate noise promotes independent control of multiplexed firing rate- and gamma-based computational mechanisms . Our results have implications for tuning of normal circuit function and for disorders associated with changes in gamma oscillations and synaptic strength . Cognitive processes are mediated by computations in neural circuits and are often associated with gamma frequency oscillations in circuit activity . Gamma activity and cognitive performance often co-vary within tasks and between individuals , while cognitive deficits in psychiatric disorders such as autism and schizophrenia are linked to altered gamma frequency network dynamics ( Uhlhaas and Singer , 2012; Spellman and Gordon , 2014 ) . Such disorders are also linked to changes in the efficacy of excitatory glutamatergic and inhibitory GABAergic synapses ( Rubenstein and Merzenich , 2003; Lewis et al . , 2012 ) . A critical and unresolved issue is the mechanistic relationship between gamma oscillations , the strength of excitation and inhibition , and circuit computations . On the one hand , neural codes based on firing rates may be sufficient for circuit computations ( Shadlen and Newsome , 1994; Histed and Maunsell , 2014 ) . In this scenario gamma oscillations might index circuit activation , but would not be required for computation . Evidence that rate coded computations and gamma oscillations arise from shared circuit mechanisms could be interpreted to support this view ( Lundqvist et al . , 2010; Pastoll et al . , 2013 ) , which predicts that when synaptic properties of a circuit are altered then gamma activity and the output of the rate-coded computation will co-vary . Alternatively , gamma oscillations , while sharing cellular substrates with rate-coded computations , may nevertheless support independent or multiplexed computational modes . For example , according to the communication through coherence hypothesis , tuning of gamma frequency activity may facilitate selective interactions between distant brain regions ( Fries , 2009 ) . In this scenario independent control of rate coded computation and gamma activity would be beneficial , for example by allowing tuning of coherence without disrupting multiplexed rate-coded computations . However , it is unclear how this could be achieved in circuits where gamma and rate-coded computations share common synaptic mechanisms , as this would require variation in synaptic properties to differentially affect gamma activity and the rate coded computation . We address these issues using a model that accounts , through a common synaptic mechanism , for gamma oscillations and spatial computation by neurons in layer 2 of the medial entorhinal cortex ( MEC ) ( Pastoll et al . , 2013 ) . The rate-coded firing of grid cells in the MEC is a well-studied feature of neural circuits for spatial cognition ( Moser and Moser , 2013 ) . During exploration of an environment individual grid cells are active at multiple locations that together follow a hexagonal grid-like organization . At the same time MEC circuits generate periods of activity in the high gamma frequency range ( 60–120 Hz ) nested within a slower theta ( 8–12 Hz ) frequency network oscillation ( Chrobak and Buzsáki , 1998 ) . Analysis of spatial correlations in grid firing , of manipulations to grid circuits , and recording of grid cell membrane potential in behaving animals , collectively point towards continuous two-dimensional network attractor states as explanations for grid firing ( Bonnevie et al . , 2013; Domnisoru et al . , 2013; Schmidt-Hieber and Häusser , 2013; Yoon et al . , 2013 ) . In layer II of the MEC , which has the highest known density of grid cells ( Sargolini et al . , 2006 ) , stellate cells that project to the dentate gyrus of the hippocampus are the major population of excitatory neurons ( Gatome et al . , 2010 ) . These excitatory ( E ) neurons do not appear to influence one another directly but instead interact via intermediate inhibitory ( I ) neurons ( Dhillon and Jones , 2000; Couey et al . , 2013; Pastoll et al . , 2013 ) . Models that explicitly incorporate this recurrent E-I-E connectivity can account for grid firing through velocity-dependent update of network attractor states ( Pastoll et al . , 2013 ) . When these models are implemented with excitable spiking neurons they also account for theta-nested gamma frequency network oscillations ( Pastoll et al . , 2013 ) . The influence in these , or other classes of attractor network models , of the strength of E to I or I to E connections on gamma oscillations and grid firing , or other attractor computations , has not been systematically investigated . We find that while gamma oscillations and grid firing are both sensitive to the strength of excitatory and inhibitory connections , their relationship differs . Although their underlying synaptic substrates are identical , gamma activity nevertheless provides little information about grid firing or the presence of underlying network attractor states . Thus , gamma activity is not a good predictor of rate-coded computation . Unexpectedly , we find the range of E- and I- synaptic strengths that support gamma and grid firing is massively increased by moderate intrinsic noise through a mechanism involving suppression of seizure-like events . In the presence of moderate noise differences in synaptic strength can tune the amplitude and frequency of gamma across a wide range with little effect on grid firing . We obtain similar results in implementations of E-I models in which connectivity is probabilistic and in models extended to include additional I to I and E to E connections . Our results suggest constraints for extrapolation of differences in gamma activity to mechanisms for cognition , identify noise as a critical factor for successful circuit computation , and suggest that tuning of excitatory or inhibitory synaptic strength could be used to control gamma-dependent processes multiplexed within circuits carrying out rate coded computations . What happens to grid firing patterns when the strengths of excitatory and/or inhibitory synaptic connections in the model are modified ? To address this we first evaluated grid firing while simulating exploration within a circular environment with a network from which noise sources were absent ( Figure 2A ) . When we reduce the strength of connections from I cells by threefold and increase the strength of connections from E cells by threefold we find that grid firing is abolished ( Figure 2Ab vs Figure 2Aa ) . Exploring the parameter space of gE and gI more systematically reveals a relatively restricted region that supports grid firing ( Figure 2D and Supplementary file 1A–D ) . Rather than the required gI and gE being proportional to one another , this region is shifted towards low values of gI and high gE . Thus , the ability of recurrently connected networks to generate grid fields requires specific tuning of synaptic connection strengths . 10 . 7554/eLife . 06444 . 005Figure 2 . Noise increases the range of synaptic strengths that support grid firing . ( A–C ) Example spatial firing fields ( left ) and spatial autocorrelation plots ( right ) of E and I cells for networks without noise ( A; σ = 0 pA ) , with noise level set to σ = 150 pA ( B ) , and noise level set to σ = 300 pA ( C ) and with the strengths of recurrent synaptic connections indicated by arrows in ( D–F ) . Maximal firing rate is indicated to the top right of each spatial firing plot . The range of spatial autocorrelations is normalized between 0 and 1 . ( D–F ) Gridness score as a function of gE and gI for networks with each noise level . Each item in the color plot is an average gridness score of four simulation runs . Arrows indicate the positions of grid field and autocorrelation examples from simulations illustrated in ( A–C ) . Simulations that did not finish in a specified time interval ( 5 hr ) are indicated by white color . ( G ) Difference between gridness scores of networks with σ = 150 pA and networks with σ = 0 pA plotted as a function of gE and gI . ( H ) Gridness score plotted as a function of the standard deviation of intrinsic noise . Each noise level comprises simulations from a neighborhood of gE and gI surrounding a center point in the parameter space ( center included ) indicated by arrows in ( D–F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 00510 . 7554/eLife . 06444 . 006Figure 2—figure supplement 1 . Sensitivity of grid firing to changes in feedback inhibition , excitation and noise levels in networks with connection probability between pairs of neurons drawn according to the synaptic profile functions in Figure 1B . ( A–C ) Example spatial firing fields ( left ) and spatial autocorrelation plots ( right ) of E and I cells for networks without noise ( A; σ = 0 pA ) , with noise set to σ = 150 pA ( B ) , and noise set to σ = 300 pA ( C ) and with the strengths of recurrent synaptic connections indicated by arrows in ( D–F ) . Maximal firing rate is indicated in the top right of each spatial firing plot . The range of spatial autocorrelations is normalized between 0 and 1 . ( D–F ) Gridness score as a function of gE and gI for networks with each noise level . Each item in the color plot is an average gridness score of two simulation runs . Arrows indicate the positions of grid field and autocorrelation examples from simulations illustrated in ( A–C ) . Simulations that did not finish in a specified time interval ( 5 hr ) are indicated by white color . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 00610 . 7554/eLife . 06444 . 007Figure 2—figure supplement 2 . Spatial information and sparsity of firing fields of E and I cells . ( A ) Spatial information of E ( top ) and I ( bottom ) cells as a function of gE and gI in networks from Figure 2 . ( B ) Same as ( A ) , but the color plots show spatial sparsity of E and I cells . Black lines indicate the region from Figure 2D–F where the gridness score = 0 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 00710 . 7554/eLife . 06444 . 008Figure 2—figure supplement 3 . Gridness scores of I cells . Colour plots show gridness score as a function of gE and gI for networks without noise ( A ) , with noise standard deviation σ = 150 pA ( B ) , and σ = 300 pA ( C ) . Data are from simulations of networks with feedback inhibition only ( E-I networks; Figure 2 ) . Black lines indicate the region from Figure 2D–F where the gridness score of E cells = 0 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 00810 . 7554/eLife . 06444 . 009Figure 2—figure supplement 4 . Spatial firing fields in networks with uncorrelated spatial input applied to each I cell . ( A ) Examples of firing fields of E and I cells . Gridness score and maximal firing rate of the firing field is indicated in the top left and right parts of each firing field , respectively . ( B ) Distributions of spatial sparsity ( left ) , spatial information ( centre ) and gridness score ( right ) of 100 randomly selected cells from each population of neurons . Each simulation run was repeated 10 times with different random seeds . Network parameters were gE = 3 nS and gI = 1 nS . Each I cell received connections from three randomly selected neurons with a place like spatial firing field . Properties of place cells: rmax = 100 Hz , σfield = 80 cm ( cf . Appendix 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 009 Because neural activity in the brain is noisy ( Shadlen and Newsome , 1994; Faisal et al . , 2008 ) , we wanted to know if the ability of the circuit to compute location is affected by noise intrinsic to each neuron ( Figure 1C ) . Given that continuous attractor networks are often highly sensitive to noise ( Zhang , 1996; Eliasmith , 2005 ) , we expected that intrinsic noise would reduce the parameter space in which computation is successful . In contrast , when we added noise with standard deviation of 150 pA to the intrinsic dynamics of each neuron , we found that both configurations from Figure 2Aa , b now supported grid firing patterns ( Figure 2Ba , b ) . When we considered the full space of E and I synaptic strengths in the presence of this moderate noise we now found a much larger region that supports grid firing ( Figure 2E and Supplementary file 1E–H ) . This region has a crescent-like shape , with arms of relatively high gI and low gE , and low gI and high gE . Thus , while tuning of gI and gE continues to be required for grid firing , moderate noise massively increases the range of gE and gI over which grid fields are generated . When intrinsic noise was increased further , to 300 pA , the parameter space that supports grid firing was reduced in line with our initial expectations ( Figure 2Ca , b , F and Supplementary file 1I–L ) . To systematically explore the range of gE and gI over which the network is most sensitive to the beneficial effects of noise we subtracted grid scores for simulations with 150 pA noise from scores with deterministic simulations ( Figure 2G ) . This revealed that the unexpected beneficial effect of noise was primarily in the region of the parameter space where recurrent inhibition was strong . In this region , increasing noise above a threshold led to high grid scores , while further increases in noise progressively impaired grid firing ( Figure 2H ) . In probabilistically connected networks , the range of gE and gI supporting grid firing was reduced , but the shape of the parameter space and dependence on noise was similar to the standard networks ( Figure 2—figure supplement 1 ) , indicating that the dependence of grid firing on gE and gI , and the effects of noise , are independent of the detailed implementation of the E-I attractor networks . How closely does the firing of I cells in the simulated networks correspond to inhibitory activity in behaving animals , and to what extent is the pattern of I cell firing affected by gE , gI and noise ? While there is little data on the spatial firing of interneurons in the MEC , recent evidence indicates that the majority of parvalbumin positive interneurons have firing fields with significant spatial stability , but low spatial sparsity and grid scores compared to excitatory grid cells ( Buetfering et al . , 2014 ) . A possible interpretation of these data is that parvalbumin positive cells are unlikely to fulfill the roles of I cells predicted in E-I models . However , in networks that we evaluate here in which E cells have grid firing fields in the presence of moderate noise , I cell firing fields also have a much lower spatial information content and spatial sparsity than the corresponding E cell firing fields ( E cells: spatial sparsity 0 . 788 ± 0 . 061 , spatial information: 1 . 749 ± 0 . 32 bits/spike; I cells: spatial sparsity 0 . 239 ± 0 . 018 , spatial information 0 . 243 ± 0 . 024 bits/spike; p < 10−16 for comparisons of both spatial sparsity and information; paired t-test; data range is indicated as mean ± standard deviation ) ( Figure 2A–C and Figure 2—figure supplement 2 ) . Spatial autocorrelograms of simulated I cell firing fields also do not contain the six hexagonally organized peaks that are characteristic of grid fields ( Figure 2A–C ) . Nevertheless , I cell spatial autocorrelograms produce positive grid scores ( 0 . 39 ± 0 . 16; Figure 2—figure supplement 3 ) , although these are reduced compared to scores for the E cells in the same networks ( E cells: 0 . 796 ± 0 . 157; p < 10−16; paired t-test; mean ± SD ) and in many networks are below the threshold considered previously to qualify as grid like ( cf . Figure 4B of Buetfering et al . , 2014 ) . When we evaluated the dependence of I cell spatial firing on gE , gI and noise , it appeared to be similar to that of E cells ( Figure 2—figure supplement 3 ) . To assess whether grid scores of I cells can be reduced further in E-I networks while maintaining grid firing by E cells , we investigated networks in which uncorrelated spatial input is applied to each I cell ( Figure 2—figure supplement 4 ) . In these simulations E cells had grid scores of 0 . 57 ± 0 . 25 , spatial sparsity of 0 . 78 ± 0 . 03 and spatial information of 1 . 69 ± 0 . 18 bits/spike , whereas I cells had grid scores of 0 . 16 ± 0 . 2 ( p < 10−16 , paired t-test ) , spatial sparsity of 0 . 21 ± 0 . 01 ( p < 10−16 , paired t-test ) and spatial information of 0 . 2 ± 0 . 01 bits/spike ( p < 10−16 , paired t-test; range of all data sets is mean ± SD ) . Thus , spatial firing of I cells has a similar dependence on noise , gE and gI to grid cells , conventional indices of spatial firing are nevertheless much lower for I cells in E-I networks compared to E cells , and grid firing by E cells in E-I networks is relatively robust to disruption of the rotational symmetry of I cell firing fields . Together these simulations demonstrate that attractor circuit computations that generate grid firing fields require specific tuning of gE and gI . In the absence of noise grid firing is supported in relatively restricted regions of parameter space . Optimal levels of noise , which produce single cell membrane potential fluctuations of a similar amplitude to experimental observations ( Domnisoru et al . , 2013; Pastoll et al . , 2013; Schmidt-Hieber and Häusser , 2013 ) , promote grid firing by reducing the sensitivity of grid computations to the strength of recurrent synaptic connections , particularly when inhibition is relatively strong and excitation is weak . Is the sensitivity of gamma frequency oscillations to synaptic strength and to noise similar to that of grid firing ? To evaluate gamma activity we recorded synaptic currents from single E and I cells across multiple theta cycles ( Figure 3A–C ) . For the network configurations illustrated in Figure 2Aa , b and in which intrinsic noise is absent , we observed synaptic events entrained to theta cycles ( Figure 3Aa , b ) . However , the timing and amplitude of synaptic events typically differed between theta cycles and no consistent gamma rhythm was apparent . In contrast , in the presence of noise with standard deviation 150 pA we observed nested gamma frequency synaptic activity with timing that was consistent between theta cycles ( Figure 3Ba ) . In this condition the frequency of the gamma oscillations was reduced and their amplitude increased by raising gI and lowering gE ( Figure 3Bb ) . With a further increase in noise to 300 pA , gamma activity remained entrained to theta cycles , but became less ordered ( Figure 3Ca , b ) . 10 . 7554/eLife . 06444 . 012Figure 3 . Differential sensitivity of gamma oscillations and grid fields to changes in the strength of E and I synapses . ( A–C ) Examples of inhibitory ( red ) and excitatory ( blue ) synaptic currents recorded respectively from excitatory and inhibitory neurons from simulations highlighted by arrows in panels ( D–F ) . ( D–F ) Top: Correlation value at the first local maximum of the autocorrelation of inhibitory synaptic currents ( I → E cells , 25 randomly selected E cells ) , plotted as a function of gE and gI , for networks without noise ( D ) , with noise level set to σ = 150 pA ( E ) , and noise level set to σ = 300 pA ( F ) . Each point is an average over five simulation trials . In these simulations velocity and place cell inputs were disabled . The duration of simulations was 10 s . Bottom: Frequency corresponding to the peaks of the autocorrelation functions for simulations in the top panels . Black lines in ( E ) indicate the region from Figure 2E where the gridness score = 0 . 5 . ( G ) Scatter plots show gridness score as a function of gamma oscillation strength ( top ) and frequency ( bottom ) for simulations with noise absent ( green ) , with an intermediate level of noise ( red ) and highest simulated noise level ( blue ) . Each dot represents data from a single network configuration . ( H ) Top: Gamma oscillation strength plotted as a function of standard deviation of the noise current . Grey color indicates simulations with gE = 3 nS , gI = 1 nS ( A ) . Red color indicates simulations with gE = 1 nS , gI = 3 nS ( B ) . Bottom: Frequency corresponding to the detected autocorrelation peak . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 01210 . 7554/eLife . 06444 . 013Figure 3—figure supplement 1 . Sensitivity of gamma oscillations to changes in the strength of E and I synapses in networks with connection probability between pairs of neurons drawn according to the synaptic profile functions in Figure 1B . ( A–C ) Examples of inhibitory ( red ) and excitatory ( blue ) synaptic currents recorded respectively from excitatory and inhibitory neurons from simulations highlighted by arrows in panels ( D–F ) . ( D–F ) Top: Correlation value at the first local maximum of an autocorrelation of inhibitory synaptic currents ( I → E cells , 25 randomly selected E cells ) , plotted as a function of gE and gI , for networks without noise ( D ) , with noise set to σ = 150 pA ( E ) , and noise set to σ = 300 pA ( F ) . Each point is an average over five simulation trials . In these simulations velocity and place cell inputs were disabled . The duration of simulations was 10 s . Bottom: Frequency corresponding to the peaks of the autocorrelation functions for simulations in the top panels . Black lines in ( E ) indicate the region from Figure 2—figure supplement 1 where the gridness score = 0 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 01310 . 7554/eLife . 06444 . 014Figure 3—figure supplement 2 . Scatter plots of gridness score as a function of the amplitude of gamma oscillations . ( A–C ) The plots show relationships between grid field computations ( gridness score ) and the power of nested gamma oscillations for deterministic networks ( A ) , networks with moderate noise ( B ) and networks with the highest simulated noise level ( C ) . Noise level is indicated by σ . The strength of the oscillation was obtained by computing autocorrelation functions of inhibitory currents impinging onto 25 randomly selected E cells in the network and detecting their first local maxima . The correlation value at the first local maximum is plotted on the abscissa . Color coding determines the values of gE and gI , as shown in the 2D colorbar . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 01410 . 7554/eLife . 06444 . 015Figure 3—figure supplement 3 . Scatter plots of gridness score as a function of the detected oscillation frequency . ( A–C ) The plots show relationships between grid field computations ( gridness score ) and the frequency of gamma oscillations for deterministic networks ( A ) , networks with moderate noise ( B ) and networks with the highest simulated noise level ( C ) . Noise level is indicated by σ . The frequency of the oscillation was obtained by computing autocorrelation functions of inhibitory currents impinging onto 25 randomly selected E cells in the network and detecting their first local maxima . The time lag at the first local maximum yielded the frequency of the oscillation , which is plotted on the abscissa . Color coding determines the values of gE and gI , as shown in the 2D colorbar . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 01510 . 7554/eLife . 06444 . 016Figure 3—figure supplement 4 . Amplitude and frequency of gamma oscillations in the gE and gI parameter regions where grid fields are robust . Amplitude ( top ) and frequency ( bottom ) of detected gamma oscillations for simulations in which gridness score is greater than 0 . 5 , in deterministic networks ( A ) , networks with an intermediate level of noise ( B ) and in networks with the highest simulated level of noise ( C ) . The data in this figure are from simulations in Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 016 To explore gamma activity across a wider range of gI and gE we automated quantification of the strength and frequency of oscillatory input to E cells ( see ‘Materials and methods’ ) . In the absence of noise gamma frequency activity only occurred for a narrow range of gI and gE ( Figure 3D ) . Strikingly , following addition of moderate noise the region of parameter space that supports gamma activity was massively expanded ( Figure 3E ) . Within this space , the amplitude of gamma increased with increasing inhibition , whereas the frequency was reduced . As noise is increased further the amplitude and frequency of gamma oscillations are reduced ( Figure 3F ) . We found a similar dependence of gamma oscillations on noise , gE and gI in networks with probabilistic connectivity ( Figure 3—figure supplement 1 ) . Thus , intrinsic noise modifies the amplitude and frequency of nested gamma oscillations . To determine whether there is a systematic relationship between values of gE and gI that generate gamma and grid firing we compared the gridness score and gamma scores for each circuit configuration ( Figure 3G , Figure 3—figure supplements 2 , 3 ) . We found this relationship to be complex and highly sensitive to noise . However , we did not find any evidence for strong linear relationships between gamma amplitude or gamma frequency and grid score ( R2 < 0 . 12 for all comparisons ) , while gamma amplitude and frequency provided only modest amounts of information about grid scores ( 0 . 27 < MIC < 0 . 33 and 0 . 27 < MIC < 0 . 37 respectively ) . The relationship between noise intensity and gamma differed from that for grid computations . Whereas , grids emerged above a sharp noise threshold ( Figure 2H ) , for the same regions in parameter space the frequency and amplitude of gamma oscillations varied smoothly as a function of noise ( Figure 3H ) . Thus , neither the frequency nor the power of gamma appears to be a good predictor of grid firing . When we considered only regions of parameter space that generate robust grid fields ( grid score >0 . 5 ) , we found circuits generating almost the complete observed range of gamma amplitudes ( 0 . 02 < autocorrelation peak < 0 . 59 ) and frequencies ( 31 Hz < frequency < 102 Hz ) ( Figure 3—figure supplement 4 ) . For example , considering the crescent shaped region of E-I space that supports grid firing in the presence of intermediate noise ( the region within the isocline in Figure 3E ) , when gI is high and gE low then the amplitude of gamma is relatively low and the frequency high . Moving towards the region where gI is high and gE is low , the amplitude of gamma is increased and the frequency is reduced . Thus , variation of synaptic strength across this region of E-I space can be used to tune the properties of gamma activity while maintaining the ability of the network to generate grid fields . Together these data indicate that an optimal level of noise promotes emergence of gamma oscillations , while the properties of oscillations may depend on the relative strength of synaptic connections . The relationship between gamma and synaptic strength differs to that for grid computations . Strikingly , while gamma activity provides relatively little information about grid firing , differential sensitivity of gamma and grid firing to gE and gI provides a mechanism for circuits to tune gamma frequency activity while maintaining the ability to compute rate coded grid firing fields . Given the emergence of a large parameter space that supports grid firing following introduction of moderate noise , we were interested to understand how noise influences the dynamics of the E-I circuits . One possibility is that in networks that fail to generate grid firing fields network attractor states form , but their activity bumps are unable to track movement . In this scenario disrupted grid firing would reflect incorrect control of network activity by velocity signals . Alternatively , deficits in grid firing may reflect failure of network attractor states to emerge . To distinguish these possibilities we investigated formation of activity bumps in network space over the first 10 s following initialization of each network ( Figure 4 ) . 10 . 7554/eLife . 06444 . 010Figure 4 . Noise promotes formation of continuous attractors . ( A ) Examples of snapshots of network activity of E cells from simulations in which velocity and place cell inputs are inactivated . Each row shows a simulation trial with a value of gE and gI highlighted by an arrow in panel ( B ) . The corresponding probability of bump formation ( P ( bumps ) ) and the maximal firing rate is indicated to the left and right , respectively . ( B ) Color plots show probability of bump formation ( P ( bumps ) ) , for the simulated range of gE and gI and the three simulated noise levels . Each color point is an average of five 10 s simulation runs . Arrows show positions in the parameter space of examples in ( A ) . Black lines indicate the regions where the gridness score = 0 . 5 ( cf . Figure 2D–F ) . ( C ) Relationship between gridness score computed from the grid field simulation runs ( Figure 2D–F ) and the probability of bump formation ( B ) . ( D ) Relationship between gamma oscillation strength ( Figure 3D–F ) and the probability of bump formation ( B ) . Each color in ( C and D ) represents one noise level and each dot in the scatter plots corresponds to simulations of a single pair of values of gE and gI . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 01010 . 7554/eLife . 06444 . 011Figure 4—figure supplement 1 . Sensitivity of bump attractor spontaneous drift to variations in gE and gI and noise levels . ( A ) Schematic of the bump attractor drift estimation procedure . The first 500 ms of a simulation trial are used to initialize the bump attractor . Onset of theta modulated input current was at 500 ms . The estimated centers of bump attractors measured by the least squares fit of symmetric Gaussians were at 1 s ( initial position ) and 9 s ( final position ) . The drift was then estimated as the distance on twisted torus between the initial and final position . Simulation time was 10 s . ( B ) Color plots show bump attractor drifts averaged over five simulation trials , for the simulated ranges of excitatory and inhibitory synaptic strengths and levels of noise . Networks without noise can form stable bump attractors in a subset of their parameter region . Networks with noise suffer from attractor drift in majority of the parameter region . Black lines in ( B ) indicate the region from Figure 2D–F where gridness score = 0 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 011 Our analysis suggests that the deficit in grid firing in deterministic compared to noisy networks reflects a failure of attractor states to emerge . For deterministic simulation of the points in parameter space considered in Figure 2Aa , which are able to generate grid patterns , we found that a single stable bump of activity emerged over the first 2 . 5 s of simulated time ( Figure 4Aa ) . In contrast , for deterministic simulation of the point considered in 2Ab , which in deterministic simulations did not generate grid patterns , a single stable bump fails to emerge ( Figure 4Ab ) . Quantification across the wider space of gE and gI values ( see ‘Materials and methods’ ) indicated that when gI is low there is a high probability of bump formation as well as grid firing , whereas when gI is high the probability of both is reduced ( Figure 4B ) . In contrast to the deterministic condition , for circuits with intrinsically noisy neurons activity bumps emerged in the first 1 . 25 s following initialization of the network ( Figure 4Ac–e ) and the area of parameter space that supported bump formation was much larger than that supporting grid firing ( Figure 4B ) . Plotting gridness scores as a function of bump probability indicated that bump formation was necessary , although not sufficient for grid formation ( Figure 4C ) , while plotting the first autocorrelation peak as a function of bump probability supported our conclusion that grid computation and gamma activity are not closely related ( Figure 4D ) . Together , these data indicate that noise promotes formation of attractor bumps in network activity and in deterministic simulations the failure of the circuit to generate attractor states largely accounts for disrupted grid firing . In noisy networks the presence of low grid scores for networks with high bump scores ( Figure 4C ) is explained by sensitivity of these network configurations to noise-induced drift . This is illustrated by the region of parameter space from Figure 2Ab , where gI is relatively high and gE relatively low , and which in deterministic simulations fails to generate bumps or grids . With moderate noise , this point generates bumps that show little drift ( Figure 4Ac ) , whereas as noise is increased further the bump begins to drift ( Figure 4Ae ) . In contrast , at the point illustrated in Figure 2Aa , which forms grids and bumps in the presence or absence of noise , activity bumps are relatively stable in each condition ( Figure 4Aa , d ) , although drift increases with greater noise ( Figure 4—figure supplement 1 ) . Thus , intrinsic noise has two opposing effects on bump formation . For much of the parameter space we consider moderate noise promotes emergence of bumps and grids , while across all of parameter space noise reduces bump stability leading to deterioration of grids . To investigate how addition of noise promotes emergence of network attractor states we investigated the dynamics of neurons in the simulated circuits . We focus initially on the point in parameter space identified in Figure 2Ab , where grids are found in the presence of moderate noise , and bumps are found when noise is moderate or high . When we examined times of action potentials generated by all neurons in this circuit , we find that in the absence of noise the network generates hyper-synchronous seizure-like states at the start of each theta cycle ( Figure 5A and Figure 5—figure supplement 1A ) . The number of E cells active on each theta cycle differs , but their activity is typically restricted to the rising phase of theta , and there is no consistent structure in the pattern of activated neurons . The number of simultaneously active I cells is also greatest at the start of each theta cycle . The I-cells continue to fire over the theta cycle , but their synchronization declines . When moderate noise is added to the circuit only a subset of E-cells are active on each theta cycle , forming an activity bump ( Figure 5B and Figure 5—figure supplement 1B ) . The I-cells are active at gamma frequency and the formation of an activity bump in the E-cell population is reflected by an inverted bump in the I-cell population activity ( Figure 5B ) . With increased noise there is a similar overall pattern of activity , but spike timing becomes more variable , causing the bumps to drift and reducing the degree of synchronization at gamma frequencies ( Figure 5C and Figure 5—figure supplement 1C ) . 10 . 7554/eLife . 06444 . 017Figure 5 . Noise opposes generation of seizure-like states . ( A–C ) Raster plots show activity of all neurons in the excitatory ( red ) and inhibitory ( blue ) populations for the duration of two theta cycles ( top ) , along with the average population firing rates for both populations ( center and bottom; calculated with a sliding rectangular window with 2 ms duration and 0 . 5 ms time step ) , for networks where noise is absent ( A; σ = 0 ) , with noise set to σ = 150 pA ( B ) , and with noise set to σ = 300 pA ( C ) . Simulations were performed in the absence of animal movement and place cell input; gE = 1 nS and gI = 3 nS . ( D ) Maximal average population firing rate of E cells estimated from the whole simulation run ( 10 s; 500 ms at the beginning of the simulation excluded ) for each simulated level of noise . Each point is an average of maxima from five simulation runs . ( E ) Probability of the maximal population-average firing rate during each theta cycle exceeding 300 Hz , that is , at least 60% of E cells firing synchronously within a time period of 2 ms in the parameter space of gE and gI when σ = 0 pA . Black lines indicate regions where gridness score equals 0 . 5 . ( F ) Scatter plots show the relationship between gridness score and the maximal firing rate during the simulation ( left ) and the probability of the maximal population-average firing rate during each theta cycle exceeding 300 Hz ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 01710 . 7554/eLife . 06444 . 018Figure 5—figure supplement 1 . Examples of activity in the network . ( A–C ) Top: Mean maximal firing rate per theta cycle ( average over five trials ) , outlining the average activity during theta cycles , in the parameter space of gE and gI . Center and bottom: Raster plots ( center ) and population-average firing rates ( bottom ) of all cells in selected locations of the E-I parameter space during 16 consecutive θ cycles . Action potentials and firing rates of E and I cells are colored red and blue , respectively . An arrow highlights the position in the parameter space . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 018 To determine whether these changes in network dynamics are seen across wider regions of parameter space we first quantified the presence of seizure like events from the maximum population firing rate in any 2 ms window over 10 s of simulation time ( E-ratemax ) . Strikingly , we found that in the absence of noise epochs with highly synchronized activity were found for almost all combinations of gE and gI , whereas these seizure-like events were absent in simulations where noise was present ( Figure 5D ) . Interestingly , while grids emerge in deterministic networks in regions of E-I space where E-ratemax is relatively low , there is a substantial region of parameter space in which E-ratemax is >400 Hz , but grids are nevertheless formed . It is possible that seizure-like states may be rare in this region of parameter space and so do not interfere sufficiently with attractor dynamics to prevent grid firing . To test this we calculated for each combination of gE and gI the proportion of theta cycles having events with population-average rate >300 Hz ( PE-rate > 300 ) . For values of gE and gI where grid fields are present PE-rate > 300 was relatively low , indicating that seizure-like events are indeed rare ( Figure 5E ) . Consistent with this , when we plotted grid score as a function of PE-rate > 300 , we found that PE-rate > 300 was relatively informative about the gridness score in networks without noise ( MIC = 0 . 624 ) and a low value of PE-rate > 300 was necessary for grid firing ( Figure 5F ) . In contrast , E-ratemax was less informative of grid firing ( 0 . 392 ≤ MIC ≤0 . 532 ) and a wide range of values were consistent with grid firing ( Figure 5F ) . Thus , while grid firing is compatible with occasional seizure-like events , when seizure-like events occur on the majority of theta cycles then grid firing is prevented . Because seizure-like events tend to initiate early on the depolarizing phase of each theta cycle , we asked if synchronization by theta frequency drive plays a role in their initiation . When theta frequency input was replaced with a constant input with the same mean amplitude , the power of gamma oscillations was still dependent on the levels of noise and changes in gE and gI ( Figure 6—figure supplement 1 ) . However , in contrast to simulations with theta frequency input ( Figure 5D , E ) , noise-free networks without theta exhibited hyper-synchronous firing only when gE was <0 . 5 nS ( Figure 6A ) and generated grid firing fields almost in the complete range of gE and gI ( Figure 6D , G ) . Addition of noise in the absence of theta had mostly detrimental effects on grid firing ( Figure 6E , F , H , I and Figure 6—figure supplement 2 ) . Interestingly , with intermediate levels of noise , the subregion with high gridness scores ( >0 . 5 ) retained its crescent-like shape ( Figure 6E , H ) , but was smaller when compared to the networks with theta frequency inputs ( size of regions with and without theta: 488/961 vs 438/961 ) , while the range of gamma frequencies present was much lower than in networks containing theta drive . Together , these data indicate that moderate noise prevents emergence of seizure like states by disrupting synchronization of the attractor network by the shared theta frequency drive . In networks with moderate noise theta drive promotes grid firing and enables a wide range of gamma frequencies to be generated without disrupting grid firing . 10 . 7554/eLife . 06444 . 019Figure 6 . Seizure-like states and grid firing fields in networks without theta frequency inputs . ( A–C ) Maximal average population firing rate of E cells estimated from the whole simulation run ( 10 s; 500 ms at the beginning of the simulation excluded ) for each simulated level of noise indicated by σ , in networks with theta frequency inputs replaced with a constant input with the same mean amplitude . Each point is an average of maxima from five simulation trials . Black lines indicate the regions from ( G–H ) where gridness score = 0 . 5 . ( D–F ) Example spatial firing fields ( left ) and autocorrelation plots ( right ) for the specific values of gE and gI indicated by arrows in ( G–I ) , corresponding to the three simulated noise levels . Maximal firing rate is indicated at the top right of each spatial firing plot . The range of spatial autocorrelations is normalized between 0 and 1 . ( G–I ) Gridness score as a function of gE and gI , for each simulated level of noise . Each item in the color plot is an average gridness score of three simulation runs of 600 s duration . Arrows indicate the positions of grid field and autocorrelation examples from simulations illustrated in ( D–F ) . Simulations that did not finish in a specified time interval ( 5 hr ) are indicated by white color . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 01910 . 7554/eLife . 06444 . 020Figure 6—figure supplement 1 . Effect of replacing theta frequency inputs by a constant input with an equal mean amplitude . ( A–C ) Amplitude ( top ) and frequency ( bottom ) of detected gamma oscillations ( ‘Materials and methods’ ) in deterministic networks ( A ) , networks with an intermediate level of noise ( B ) and in networks with the highest simulated level of noise ( C ) . Each point is an average of five simulation runs . Data are from the same simulation set . White color indicates simulation runs in which no autocorrelation peaks were detected ( cf . ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 02010 . 7554/eLife . 06444 . 021Figure 6—figure supplement 2 . Effect of noise on gridness scores in networks without theta frequency inputs . The plot shows a difference between gridness scores of networks with σ = 150 pA and networks with σ = 0 pA plotted as a function of gE and gI when theta inputs were replaced with a constant input with an equal mean amplitude . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 02110 . 7554/eLife . 06444 . 022Figure 6—figure supplement 3 . Firing rates of E cells . ( A ) Average firing rate of all E cells during simulations of animal movement as a function of gE and gI . Black lines outline the region from Figure 2D–F where gridness score = 0 . 5 . ( B ) Relationship between gridness score and firing frequency of E cells . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 02210 . 7554/eLife . 06444 . 023Figure 6—figure supplement 4 . Calibration of the gain of the velocity inputs . ( A–C ) Bump attractor speed as a function of the strength of the velocity current for the three simulated levels of noise . 10 simulation runs were performed for each level of noise ( blue markers ) . In each run the speed of the bump was calculated in response to the injected velocity input and the data were used to fit a linear relationship using an estimation procedure outlined in Appendix 1 ( black line ) . ( D–F ) Slope of the estimated velocity gain of the attractor networks as a function of gE and gI for all simulated levels of noise . ( G–I ) Same as in ( D–F ) but the plots show error of fit for the estimated linear relationships . Arrows show locations of the data plotted in ( A–C ) . Black lines in ( D–I ) indicate the region from Figure 2D–F where gridness score = 0 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 02310 . 7554/eLife . 06444 . 024Figure 6—figure supplement 5 . Effectivity of the place cell resetting mechanism as a function of gE and gI and noise levels . ( A ) Illustration of the procedure to estimate the difference between the bump position induced by place cells and actual estimated position of the bump state , by using a sliding window with 250 ms duration and 125 ms time step . The resulting distance from the reset position , in one simulation run , was then an average over all sliding windows . ( B ) Color plots show the effectivity of place cell mechanism for an average of five simulation runs with 10 s duration . Place cells are most effective in networks with an intermediate amount of noise . Black lines in ( B ) indicate the region from Figure 2D–F where gridness score = 0 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 024 Our analysis points towards suppression of seizure-like events as the mechanism by which moderate noise promotes grid firing , while interactions between noise and theta appear important for the capacity to multiplex grid firing with a wide range of gamma frequencies . However , we wanted to know if other factors might contribute to these beneficial roles of noise . Grid fields may also fail to form if overall activity levels are too low , in which case neurons with grid fields instead encode head direction ( Bonnevie et al . , 2013 ) . This observation is unlikely to explain our results as the mean firing rate of E cells in networks that generated grid firing fields ( grid score >0 . 5 , networks with gE or gI set to 0 excluded ) was in fact lower than the firing rate of networks without grid fields ( 1 . 2; 1 . 0; 1 . 0 Hz grid fields vs 3 . 0; 2 . 7; 1 . 2 Hz no grid fields , in networks with σ = 0; 150; 300 pA respectively ) . There was also no systematic relationship between grid score and firing frequency ( Figure 6—figure supplement 3 ) . We also wanted to know if other properties of grid fields vary as a function of gE and gI . Parameters used to calibrate velocity integration by the grid network varied very little with changes in gE and gI ( Figure 6—figure supplement 4 ) , whereas drift increased with gI ( Figure 4—figure supplement 1 ) and place cell input was most effective in opposing attractor drift in noisy networks with high gridness scores ( Figure 6—figure supplement 5 ) . These data are consistent with suppression of seizure like events as the mechanism by which noise promotes grid firing , while interactions between noise and theta frequency inputs profoundly influence the dynamics of attractor networks that generate grid fields . Our analysis so far focuses on E-I attractor networks as simple models of grid firing that are compatible with the finding that synaptic interactions between stellate cells in layer 2 of the MEC are mediated via inhibitory interneurons ( Dhillon and Jones , 2000; Couey et al . , 2013; Pastoll et al . , 2013 ) . However , there is evidence that interneurons active during theta-nested gamma activity make connections to one another as well as to stellate cells ( Pastoll et al . , 2013 ) . To establish whether this recurrent inhibition substantially modifies our conclusions from simpler E-I networks , we extended the E-I model to also include synapses between interneurons ( see ‘Materials and methods’ ) . In the resulting E-I-I networks , in the absence of noise , grid firing emerges across a much larger region of parameter space compared to E-I networks ( Figure 7A , Figure 7—figure supplements 1–4 ) . However , as in E-I networks occasional seizure like activity was present across a wide range of gE and gI ( Figure 7—figure supplement 5 ) , and gamma frequency activity was largely absent ( Figure 7D , G ) . Following addition of noise with standard deviation of 150 pA to E-I-I networks , grid firing was maintained , seizure like activity was abolished , and gamma like activity emerged ( Figure 7B , E , H and Figure 7—figure supplement 5 ) . Increasing the noise amplitude to 300 pA reduced grid firing and interfered with the emergence of gamma oscillations ( Figure 7C , F , I and Figure 7—figure supplements 1–5 ) . Importantly , just as in E-I networks , the presence of moderate noise in E-I-I networks enables tuning of gamma activity by varying gE and gI while maintaining the ability of the networks to generate grid firing fields . Gamma activity had a higher frequency in E-I-I compared to E-I networks , with a greater proportion of the parameter space supporting gamma frequencies >80 Hz . This higher frequency gamma is similar to fast gamma observed experimentally in the MEC ( cf . Chrobak and Buzsáki , 1998; Colgin et al . , 2009; Pastoll et al . , 2013 ) . Thus , by including additional features of local circuits in layer 2 of the MEC , E-I-I models may more closely recapitulate experimental observations . Nevertheless , E-I-I networks maintain the ability , in the presence of moderate noise , for variation in gE and gI to tune gamma oscillations without interfering with grid firing . 10 . 7554/eLife . 06444 . 025Figure 7 . Gridness scores and gamma activity in networks with recurrent inhibition . ( A–C ) Gridness score as a function of gE and gI for networks without noise ( A; σ = 0 pA ) , with noise level set to σ = 150 pA ( B ) , and noise level set to σ = 300 pA ( C ) . Simulations that did not finish in a specified time interval ( 5 hr ) are indicated by white color . ( D–F ) Examples of inhibitory ( red ) and excitatory ( blue ) synaptic currents recorded respectively from excitatory and inhibitory neurons from simulations highlighted by arrows in panels ( G–I ) . ( G–I ) Top: Correlation value at the first local maximum of an autocorrelation of inhibitory synaptic currents ( I → E cells , 25 randomly selected E cells ) , plotted as a function of gE and gI , for networks without noise ( G ) , with noise level set to σ = 150 pA ( H ) , and noise level set to σ = 300 pA ( I ) . Each point is an average over five simulation trials . In these simulations velocity and place cell inputs were disabled . The duration of simulations was 10 s . Bottom: Frequency corresponding to the peaks of the autocorrelation functions for simulations in the top panels . Black lines in ( H ) indicate the regions from ( B ) where gridness score = 0 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 02510 . 7554/eLife . 06444 . 026Figure 7—figure supplement 1 . Spatial firing fields in networks that contain recurrent I → I synapses . ( A–C ) Example spatial firing fields ( left ) and spatial autocorrelation plots ( right ) for networks with gE = 3 nS and gI = 1 nS ( A ) and networks with gE = 1 nS and gI = 3 nS ( B ) , corresponding to the three simulated noise levels indicated by σ . Maximal firing rate is indicated to the top right of each spatial firing plot . Range of spatial autocorrelations is normalized between 0 and 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 02610 . 7554/eLife . 06444 . 027Figure 7—figure supplement 2 . Continuous attractors in networks that contain direct I → I synapses . ( A ) Examples of E cell population firing rate snapshots from simulations in which velocity and place cell inputs are inactivated . Each row shows a simulation trial with a value of gE and gI highlighted by an arrow in panel ( B ) . The corresponding probability of bump formation ( P ( bumps ) ) is indicated to the left . Maximal firing rate for each set of snapshots is indicated to the right . ( B ) Color plots show probability of bump formation ( P ( bumps ) ) , for the simulated range of gE and gI and the three simulated noise levels . Each color point is an average of five 10 s simulation runs . Black lines in ( B ) indicate the region from Figure 7A–C where gridness score = 0 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 02710 . 7554/eLife . 06444 . 028Figure 7—figure supplement 3 . Sensitivity of bump attractor spontaneous drift to variations in gE , gI and noise levels in networks that contain direct I → I synapses . ( A ) Schematic of the bump attractor drift estimation procedure . The first 500 ms of a simulation trial are used to initialize the bump attractor . Onset of theta modulated input current was at 500 ms . The estimated centers of bump attractors measured by the least squares fit of symmetric Gaussians were at 1 s ( initial position ) and 9 s ( final position ) . The drift was then estimated as the distance on twisted torus between the initial and final position . Simulation time was 10 s . ( B ) Color plots show bump attractor drifts averaged over five simulation trials , for the simulated ranges of excitatory and inhibitory synaptic strengths and levels of noise . Networks without noise can form stable bump attractors in a subset of their parameter region . Networks with noise suffer from attractor drift in the majority of the parameter region . Black lines in ( B ) indicate the region from Figure 7A–C where gridness score = 0 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 02810 . 7554/eLife . 06444 . 029Figure 7—figure supplement 4 . Calibration of the gain of the velocity inputs in networks that contain direct I → I synapses . ( A–C ) Bump attractor speed as a function of the strength of the velocity current for the three simulated levels of noise indicated by σ . Values of gE and gI are indicated by arrows in ( D–I ) . 10 simulation runs were performed for each level of noise ( blue markers ) . In each run the speed of the bump was calculated in response to the injected velocity input and the data were used to fit a linear relationship using an estimation procedure outlined in Appendix 1 ( black line ) . ( D–F ) Slope of the estimated velocity gain of the attractor networks as a function of gE and gI for all simulated levels of noise . ( G–I ) Same as in ( D–F ) but the plots show error of fit for the estimated linear relationships . Arrows in ( D–I ) show locations of the data plotted in ( A–C ) . Black lines in ( D–I ) indicate the region from Figure 7A–C where gridness score = 0 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 02910 . 7554/eLife . 06444 . 030Figure 7—figure supplement 5 . Seizure-like states in networks that contain direct I → I synapses . ( A–C ) Raster plots show activity of all neurons in the excitatory ( red ) and inhibitory ( blue ) populations for the duration of two theta cycles ( top ) , along with the average population firing rates for both populations ( center and bottom; calculated with a sliding rectangular window with 2 ms duration and 0 . 5 ms time step ) , for networks where noise is absent ( A; σ = 0 ) , with noise set to σ = 150 pA ( B ) , and with noise set to σ = 300 pA ( C ) . Simulations were performed in the absence of animal movement and place cell input; gE = 1 nS and gI = 3 nS . ( D ) Maximal average population firing rate of E cells estimated from the whole simulation run ( 10 s; 500 ms at the beginning of the simulation excluded ) for each simulated level of noise . Each point is an average of maxima from five simulation runs . ( E ) Probability of the maximal population-average firing rate during each theta cycle exceeding 300 Hz , that is , at least 60% of E cells firing synchronously within a time period of 2 ms , in the parameter space of gE and gI when σ = 0 pA . Black lines indicate the regions from Figure 7A–C where gridness score equals 0 . 5 . ( F ) Scatter plots show the relationship between gridness score and the maximal firing rate during the simulation ( left ) and the probability of the maximal population-average firing rate during each theta cycle exceeding 300 Hz ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 03010 . 7554/eLife . 06444 . 031Figure 7—figure supplement 6 . Sensitivity of grid firing to changes in inhibition and excitation in networks that contain direct E → E synapses . ( A–C ) Example firing fields ( left ) and spatial autocorrelation plots ( right ) for the strengths of recurrent synaptic connections indicated by arrows in ( D–F ) for networks without noise ( A; σ = 0 pA ) , with noise set to σ = 150 pA ( B ) , and noise set to σ = 300 pA ( C ) . ( D–F ) Gridness score as a function of gE and gI for networks with each noise level . Each item in the color plot is an average gridness score of two simulation runs . Arrows indicate the positions of grid field and autocorrelation examples from simulations illustrated in ( A–C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 03110 . 7554/eLife . 06444 . 032Figure 7—figure supplement 7 . Sensitivity of gamma oscillations to changes in inhibition and excitation in networks that contain direct E → E synapses . ( A–C ) Examples of inhibitory ( red ) and excitatory ( blue ) synaptic currents recorded respectively from excitatory and inhibitory neurons from simulations highlighted by arrows in panels ( D–F ) . ( D–F ) Top: Correlation value at the first local maximum of an autocorrelation of inhibitory synaptic currents ( I → E cells , 25 randomly selected E cells ) , plotted as a function of gE and gI , for networks without noise ( D ) , with noise set to σ = 150 pA ( E ) , and noise set to σ = 300 pA ( F ) . Each point is an average over five simulation trials . In these simulations velocity and place cell inputs were disabled . The duration of simulations was 10 s . Bottom: Frequency corresponding to the peaks of the autocorrelation functions for simulations in the top panels . Black lines in ( E ) indicate the region from Figure 7—figure supplement 6 where the gridness score = 0 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 03210 . 7554/eLife . 06444 . 033Figure 7—figure supplement 8 . Continuous attractors in networks that contain direct E → E synapses . ( A ) Examples of E cell population firing rate snapshots from simulations in which velocity and place cell inputs are inactivated . Each row shows a simulation trial with a value of gE and gI highlighted by an arrow in panel ( B ) . The corresponding probability of bump formation ( P ( bumps ) ) is indicated to the left . Maximal firing rate for each row is indicated to the right . ( B ) Color plots show probability of bump formation ( P ( bumps ) ) , for the simulated range of gE and gI and the three simulated noise levels indicated by σ . Each color point is an average of five 10 s simulation runs . Arrows show positions in the parameter space of examples in ( A ) . Black lines indicate the region from Figure 7—figure supplement 6 where the gridness score = 0 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 03310 . 7554/eLife . 06444 . 034Figure 7—figure supplement 9 . Seizure-like states in networks that contain direct E → E synapses . ( A–C ) Raster plots show activity of all neurons in the excitatory ( red ) and inhibitory ( blue ) populations for the duration of two theta cycles ( top ) , along with the average population firing rates for both populations ( center and bottom; calculated with a sliding rectangular window with 2 ms duration and 0 . 5 ms time step ) , for networks where noise is absent ( A; σ = 0 ) , with noise set to σ = 150 pA ( B ) , and with noise set to σ = 300 pA ( C ) . Simulations were performed in the absence of animal movement and place cell input; gE = 1 nS and gI = 3 nS . ( D ) Maximal average population firing rate of E cells estimated from the whole simulation run ( 10 s; 500 ms at the beginning of the simulation excluded ) for each simulated level of noise . Each point is an average of maxima from five simulation runs . ( E ) Probability of the maximal population-average firing rate during each theta cycle exceeding 300 Hz , that is , at least 60% of E cells firing synchronously within a time period of 2 ms in the parameter space of gE and gI when σ = 0 pA . ( F ) Scatter plots show the relationship between gridness score and the maximal firing rate during the simulation ( left ) and the probability of the maximal population-average firing rate during each theta cycle exceeding 300 Hz ( right ) . Black lines in ( D and E ) indicate the region from Figure 7—figure supplement 6 where the gridness score = 0 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 03410 . 7554/eLife . 06444 . 035Figure 7—figure supplement 10 . Probability of bump formation and network activity plots in networks with structured E → E and unstructured E → I and I → E connections . Since the presence of bump attractors is necessary for grid computation , we tested whether networks with only structured E-E connections can generate activity bumps . We used the Gaussian fitting procedure ( cf . ‘Materials and methods’ ) to estimate the presence of bump attractors in these networks . ( A ) Probability of bump formation as a function of the E-E synaptic scaling factor ( gE → E ) and the width of the synaptic profile ( σE → E ) . Arrow highlights the position in the parameter space corresponding to the raster plots ( center ) and network activity snapshots ( bottom ) for E and I cells . Firing rate in the network activity color plots are in the range of 0 ( dark blue ) to the maximum firing rate indicated to the right of the plot ( dark red ) . In these networks gE = 1 nS and gI = 0 . 1 nS . ( B ) Same as ( A ) but gE = 3 nS and gI = 1 nS . ( C ) Same as ( A ) and ( B ) but in these simulations the synaptic scaling factor of E-E connections and the width of the synaptic profile were fixed ( gE → E = 3 nS and σE → E = 0 . 0833 ) and gE and gI varied in the range of 0–6 nS . Simulations that produced excessive spiking activity and did not finish in a specified time limit ( 3 hr ) are indicated by white color . Many networks suffer from runaway excitation and inhibition ( A ) or generate only background synaptic activity characterized by low firing rates of E and I cells ( B and C ) . The Gaussian fitting procedure used to estimate the probability of bump formation can nevertheless yield a high bump score due to the fact that this procedure can also give a high score to intermittent pockets of activity ( A ) or pockets of background activity of E cells ( B and C ) . This activity , however , is not stable enough to generate grid firing fields . DOI: http://dx . doi . org/10 . 7554/eLife . 06444 . 035 Finally , we asked if addition of synaptic connections between excitatory cells modifies the relationship between gamma , noise , gE and gI . While the E-I model is consistent with the connectivity between stellate cells in layer 2 of the MEC , adjacent pyramidal cells may also have grid firing properties . Unlike stellate cells , pyramidal cells interact with one another directly via excitatory connections and indirectly via inhibitory interneurons ( Couey et al . , 2013 ) . To assess the impact of E-E connections , we first extended the E-I model to allow each E cell to excite other E cells that are nearby in neuron space . The dependence of grid firing , gamma oscillations , and bump formation on noise , gE and gI was similar to E-I networks ( Figure 7—figure supplements 6–9 ) . We also attempted to evaluate networks in which E-E connections were structured , but E-I and I-E connections were uniformly distributed . However , in these networks we were unable to identify parameters that support formation of stable activity bumps ( Figure 7—figure supplement 10 ) . This is consistent with instability of simpler network attractors based on E-E connections ( Seung et al . , 2000 ) . We investigated the relationship between rate coded spatial computations and nested gamma oscillations in attractor network models of grid firing . While in the models we consider rate coding and gamma oscillations share the same neural substrate , that is projections from a population of E cells to an I cell population , which in turn projects back to the E cell population , we find that their sensitivity to variations in excitatory and inhibitory synaptic strengths nevertheless differs . A moderate level of noise promotes generation of both grid fields and nested gamma oscillations , primarily by the disruption of epileptic-like firing of E and I cells in the network . When the strength of E or I connections is varied in the presence of moderate noise a wide range of gamma frequency and power can be obtained without affected grid firing . Thus , noise can be beneficial for computations performed by the nervous system , while the frequency and power of multiplexed gamma oscillations can be tuned independently of rate-coded grid computations , suggesting a mechanism for differential control of multiplexed neural codes . Our results suggest a novel beneficial role for noise . In general noise in the nervous system is believed to distort the fidelity of transmitted signals ( Faisal et al . , 2008 ) . Exceptions are stochastic resonance phenomena in which noise promotes detection of small amplitude signals by individual neurons ( Longtin et al . , 1991; Benzi et al . , 1999; Shu et al . , 2003 ) , improvements in signal coding through desynchronization of neuronal populations ( Hunsberger et al . , 2014 ) and emergence of stochastic weak synchronization in interneuron networks ( Tiesinga and Jose , 2000 ) . The beneficial role for noise that we identify here differs from these phenomena in that it emerges through interactions between populations of neurons and because the grid cell attractor network performs a computation—generation of a spatial code from velocity inputs—rather than propagating input signals . We find that by opposing emergence of hyper-sychronous seizure-like states noise allows the network to generate stable bump attractor states . Noise prevents the seizure-like states by desynchronizing neuronal responses to common theta input . We were able to identify this role for noise because spiking and synaptic dynamics are explicitly represented in the simulated network . These dynamics are absent from other attractor network models of grid firing ( Fuhs and Touretzky , 2006; Guanella et al . , 2007; Burak and Fiete , 2009 ) . They are also absent from other models of theta-nested gamma oscillations that simulate two-dimensional dynamical systems of E and I populations with theta modulated inputs to the network ( Onslow et al . , 2014 ) . Thus , intrinsic cellular and synaptic dynamics in conjunction with noise sources may be important in accounting for computations and oscillatory activity in neural networks . The distinct control of rate coded grid computations and gamma oscillations by noise , gE and gI was independent of the detailed implementation of the E-I models we considered and was maintained in more complex models incorporating I-I and E-E coupling . Current available experimental data appears to be insufficient to distinguish between these different models . For example , our analysis of interneuron firing indicates that while E-I models predict that interneurons will have spatial firing fields , they have lower spatial information content , spatial sparsity and grid scores than E cells and therefore may be difficult to detect in existing experimental datasets and with current analysis tools . Thus , evidence previously interpreted to argue against E-I based mechanisms for grid firing may in fact not distinguish these from other possible mechanisms . Indeed , we found that grid firing by E cells can be maintained during spatial input that distorts the spatial firing pattern of I cells ( Figure 2—figure supplement 4 ) . While these simulations establish in principle that E-I based attractor networks can generate grid outputs even when spatial firing of many E and I cells in the network is not clearly grid-like , the extent to which these networks can account for additional details of experimental observations , for example weak periodic patterns in the spatial autocorrelation of the firing fields of some PV interneurons ( cf . Buetfering et al . , 2014 , Figure 4a ) , is not yet clear . Our results are consistent with local synaptic connections , in addition to those between E cells and I cells , having important functional roles . For example addition of synapses between interneurons to E-I networks causes an overall increase in the frequency of gamma activity and in the stability of grid firing . Nevertheless , we find that in these modified networks moderate noise still enables variation in gE and gI to tune gamma oscillations independently from grid firing . An intriguing aspect of our results is that they suggest novel approaches to suppressing seizures and to promoting normal cognitive function . Seizures have previously been suggested to result from deficits in inhibition or from alterations in intrinsic excitability of neurons ( Lerche et al . , 2001; Treiman , 2001 ) . We show that seizures can be induced when these properties are held constant simply by reducing levels of noise within a circuit . A future experimental challenge for dissecting the contribution of intrinsic noise to seizures will be to target biological noise sources . In the brain noise arises from ion channel gating and from background synaptic activity . It is therefore difficult to manipulate noise sources without also affecting intrinsic excitability or excitation-inhibition balance . However , it may be feasible to add noise to circuits through transcranial magnetic stimulation ( Ruzzoli et al . , 2010 ) . In this case our simulations predict that addition of noise may restore epileptic circuits to normal activity . This mechanism may explain why focal electrical stimulation of the entorhinal cortex in patients with seizures leads to an enhancement of memory performance ( Suthana et al . , 2012 ) . While correlations between gamma oscillations and various cognitive and pathological brain states are well established , the proposed computational roles of gamma oscillations have been difficult to reconcile with rate-coded representations with which they co-exist . We were able to address this issue directly by analyzing a circuit in which gamma oscillations and rate-coded computations arise from a shared mechanism . Rather than gamma serving as an index of rate-coded computation , we find instead that there is a substantial parameter space across which rate-coded computation is stable , while the amplitude and frequency of theta-nested gamma oscillations varies . Our analysis leads to several new and testable predictions . First , tuning of recurrent synaptic connections could be used to modify gamma oscillations without affecting rate-coded computation . If multiple networks of the kind we simulate here correspond to grid modules providing input to downstream neurons in the hippocampus ( Stensola et al . , 2012 ) , then adjusting gE or gI would alter gamma frequency with minimal effect on the grid firing pattern of each module . If the downstream neurons integrate input at the gamma time scale , then this should lead to re-mapping of their place representation in the absence of any change in either the strength of their synaptic inputs or the information they receive from upstream grid cells . Adjustment of gE and gI could be achieved dynamically through actions of neuromodulators ( Marder , 2012 ) , or on slower developmental time scales ( Widloski and Fiete , 2014 ) . Second , subtle differences in gamma could be a sensitive index of network pathology at stages before deficits in rate coded computation are apparent . If cognitive deficits in psychiatric disorders reflect a failure of rate coded computation , then our analysis predicts that a change in noise within a circuit , in addition to synaptic modification , may be necessary for deficits to emerge . From this perspective it is intriguing that seizure phenotypes are often associated with disorders such as autism ( Deykin and MacMahon , 1979 ) . Alternatively , cognitive deficits may result from a failure to coordinate gamma frequency synchronization of circuits that converge on downstream targets . In this case we expect cognitive deficits to be phenocopied by manipulations that affect gamma frequency or power without influencing rate-coded computations ( Sigurdsson et al . , 2010; Spellman and Gordon , 2014 ) . In conclusion , our systematic exploration of three dimensions of parameter space ( gE , gI and intrinsic noise ) illustrates the complexity of relationships between rate-coded computation , gamma frequency oscillations and underlying cellular and molecular mechanisms . Our results highlight the challenges in straightforward interpretation of experiments in which these parameters are correlated to one another , ( cf . Wang and Krystal , 2014 ) . While there are parallels to investigations of pace-making activity in invertebrate circuits ( Marder and Taylor , 2011 ) , which demonstrate that many parameter combinations can account for higher order behavior , there are also critical differences in that the models we describe account for multiplexing of rate-coded computation and oscillatory activity , while the number of neurons and connections in the simulated circuit is much larger . Future experimentation will be required to test our model predictions for unexpected beneficial roles of noise and for control of gamma oscillations independently from grid firing by modulating the strength of excitatory and inhibitory synaptic connections . The model comprised a network of exponential integrate and fire neurons ( Fourcaud-Trocmé et al . , 2003 ) implemented as a custom-made module of the NEST simulator ( Gewaltig and Diesmann , 2007 ) . The network investigated in the majority of simulations ( Figures 1–6 ) is modified from that in Pastoll et al . ( 2013 ) and consists of excitatory ( E ) and inhibitory ( I ) populations of neurons that were arranged on a twisted torus with dimensions of 34 × 30 neurons . In networks where connection strengths were generated probabilistically instead of in an all-to-all way , the synaptic weights from E to I cells and vice versa were constant , while the probability of connection between the pre- and post-synaptic neuron was drawn according to Figure 1B . In addition , some networks also included direct uniform recurrent inhibition between I cells ( Figure 7; referred to as E-I-I networks ) or direct structured recurrent excitation between E cells ( Figure 7—figure supplements 6–10 ) . When recurrent excitation was present , synaptic weights between E cells followed the connectivity profile in which the strongest connection was between cells that were close to each other in network space ( Figure 1B ) and the weights between E and I cells were generated either according to synaptic profiles from Figure 1B ( Figure 7—figure supplements 6–9 ) or the E-I connectivity was uniform with a probability of connection of 0 . 1 ( Figure 7—figure supplement 10 ) . E and I cells also received the theta current drive which was the sum of a constant amplitude positive current and a current with sinusoidal waveform ( 8 Hz ) . The constant component of the drive was required to activate the circuit , while the theta drive frequency was chosen to reflect the frequency of theta oscillations in behaving animals . The amplitude ( cf . Appendix 1 ) was chosen to produce theta modulation of I cell firing similar to that observed in behaving animals ( cf . Chrobak and Buzsáki , 1998 ) and ex-vivo models of theta-nested gamma activity ( cf . Pastoll et al . , 2013 ) . In order to oppose drift of the activity bump in networks that simulated exploration of the arena E cells received input from cells with place-like firing fields simulated as Poisson spiking generators with their instantaneous firing rate modeled as a Gaussian function of the animal position . Full details of the connectivity and network parameters are in Appendix 1 . In all simulations the networks were parameterized by the standard deviation of noise ( σ ) injected independently into each E and I cell and by synaptic scaling parameters ( gE and gI ) . Noise was sampled from a Gaussian distribution with standard deviation either set to σ = 0 , 150 or 300 pA , or alternatively in the range of 0–300 pA in steps of 10 pA ( Figures 2H , 3H ) . The peaks of the synaptic profile functions ( Figure 1B ) were determined by the gE and gI parameters that appropriately scaled the maximal conductance values of the excitatory and inhibitory connections respectively . Gridness scores were estimated by simulating exploration in a circular arena with a diameter of 180 cm . For each value of gE and gI the simulations consisted of two phases . In the first phase , animal movement with constant speed and direction ( vertically from bottom to top ) was simulated in order to calibrate the gain of the velocity input to achieve 60 cm spacing between grid fields in the network . In the second phase , the calibrated velocity input gains were used during a simulation of realistic animal movements with duration of 600 s ( Hafting et al . , 2005 ) . Each simulation was repeated 1–4 times . For each trial , gridness score was then estimated from an E or I cell located at position ( 0 , 0 ) on the twisted torus . In simulations where interneurons received uncorrelated spatial inputs ( Figure 2—figure supplement 4 ) , gridness scores were estimated from 100 randomly selected E and I cells on the twisted torus . For the analysis of bump attractor properties and gamma oscillations a separate set of simulations were run . For each value of gE , gI and noise level , there were five trials of 10 s duration during which the velocity and place cell inputs were deactivated . For each trial spiking activity of all cells was recorded . In addition , inhibitory synaptic currents of 25 randomly selected E cells were saved and used for further analysis . The strength and frequency of gamma oscillations were estimated from the inhibitory synaptic currents recorded from E cells . The currents were first band-pass filtered between 20 and 200 Hz . For each trace , its autocorrelation function was computed and the first local maximum was detected using a peak detection algorithm which was based on calculating the points in the autocorrelation function where the first difference of the signal changed sign from positive to negative and thus approximated the points where the first derivative was zero and the second derivative was negative . The strength and frequency of gamma oscillations was estimated from the correlation value and lag at the position of the first local maximum respectively . Properties of bump attractors were estimated by fitting symmetric Gaussian functions onto successive snapshots of firing rates of each cell in the E population . For each snapshot this procedure yielded the position of the bump center and its width . The probability of bump formation was then estimated as a proportion of population-activity snapshots that were classified as bump attractors , that is , those fitted Gaussian functions whose width did not exceed the shorter side of the twisted torus . Other properties of bump attractors were estimated by analyzing successive positions of the bump attractor centers . Action potential raster plots of E and I populations ( Figure 5A–C , Figure 5—figure supplement 1 and Figure 7—figure supplement 10 ) show neuron indices that are flattened in a row-wise manner with respect to the two-dimensional twisted torus . Data points with white color in Figure 5D , E and Figure 5—figure supplement 1A have been excluded from analysis since the maximal firing rate of E cells exceeded 500 Hz/2 ms window . The calculation of the maximal information coefficient ( MIC ) for the relationship between gridness score , gamma and bump scores was estimated by applying the MIC measure using the minepy package ( Albanese et al . , 2013 ) . Calculations of spatial information were carried out according to ( Skaggs et al . , 1996 ) . Spatial sparsity was calculated by following the procedure outlined in ( Buetfering et al . , 2014 ) . All other data analysis and simulations were performed in Python .

When electrodes are placed on the scalp , or lowered into the brain itself , rhythmic waves of electrical activity are seen that reflect the coordinated firing of large numbers of neurons . The pattern of the waves varies between different brain regions , and according to what the animal or person is doing . During sleep and quiet wakefulness , slower brain waves predominate , whereas faster waves called gamma oscillations emerge during cognition—the act of processing knowledge . Gamma waves can be readily detected in a region of the brain called the medial entorhinal cortex ( MEC ) . This brain region is also known for its role in forming the spatial memories that allow an individual to remember how to navigate around an area they have previously visited . Individual MEC cells increase their firing rates whenever an individual is at specific locations . When these locations are plotted in two dimensions , they form a hexagonal grid: this ‘grid cell map’ enables the animal to keep track of its position as it navigates through its environment . To determine how MEC neurons can simultaneously encode spatial locations and generate the gamma waves implicated in cognition , Solanka et al . have used supercomputing to simulate the activity of more than 1 . 5 million connections between MEC cells . Changing the strength of these connections had different effects on the ability of the MEC to produce gamma waves or spatial maps . However , adjusting the model to include random fluctuations in neuronal firing , or ‘noise’ , was beneficial for both types of output . This is partly because noise prevented neuronal firing from becoming excessively synchronized , which would otherwise have caused seizures . Although noise is generally regarded as disruptive , the results of Solanka et al . suggest that it helps the MEC to perform its two distinct roles . Specifically , the presence of noise enables relatively small changes in the strength of the connections between neurons to alter gamma waves—and thus affect cognition—without disrupting the neurons' ability to encode spatial locations . Given that noise reduces the likelihood of seizures , the results also raise the possibility that introducing noise into the brain in a controlled way could have therapeutic benefits for individuals with epilepsy .

2015

Noise promotes independent control of gamma oscillations and grid firing within recurrent attractor networks

Humans and other species have perceptual mechanisms dedicated to estimating approximate quantity: a sense of number . Here we show a clear interaction between self-produced actions and the perceived numerosity of subsequent visual stimuli . A short period of rapid finger-tapping ( without sensory feedback ) caused subjects to underestimate the number of visual stimuli presented near the tapping region; and a period of slow tapping caused overestimation . The distortions occurred both for stimuli presented sequentially ( series of flashes ) and simultaneously ( clouds of dots ) ; both for magnitude estimation and forced-choice comparison . The adaptation was spatially selective , primarily in external , real-world coordinates . Our results sit well with studies reporting links between perception and action , showing that vision and action share mechanisms that encode numbers: a generalized number sense , which estimates the number of self-generated as well as external events . Animals , including humans , estimate spontaneously and reasonably accurately the approximate quantity of arrays of objects , without recourse to other forms of representation , such as density ( Cicchini et al . , 2016 ) . Even newborn infants of less than 3 days show selective habituation to number ( Izard et al . , 2009 ) . There is now very good evidence in both human and non-human primates that number is encoded by intraparietal and prefrontal cortex ( Castelli et al . , 2006; Dehaene et al . , 2003; Harvey et al . , 2013; Nieder , 2005 , 2012 , 2016; Nieder et al . , 2006; Nieder and Miller , 2004; Piazza and Eger , 2016; Piazza et al . , 2004 , 2007 ) , even in numerically naive monkeys ( Viswanathan and Nieder , 2013 ) . All these studies point to the existence of a visual sense of number within a parietal–frontal network ( Dehaene , 2011 ) . A truly abstract sense of number should be capable of encoding the numerosity of any set of discrete elements , displayed simultaneously or sequentially , in whatever sensory modality . Some evidence exists for such a generalized number sense . Neurons in the lateral prefrontal cortex ( lPFC ) of behaving monkeys encode numerosity for both auditory and visual sensory modalities , suggesting supra-modal numerosity processing ( Nieder , 2012 ) . Another study reported separate populations of neurons in the intraparietal sulcus ( IPS ) responding selectively to sequential or simultaneous numerical displays , while a third set of neurons showed numerosity selectivity for both simultaneous and sequential presentations , suggesting that the information about spatial and temporal numerosity converges to a more abstract representation ( Nieder et al . , 2006 ) . There is also evidence from functional imaging in humans for a right lateralized fronto-parietal circuit activated by both auditory and visual number sequences , and that right IPS is involved in processing both sequential and simultaneous numerosity formats ( Castelli et al . , 2006; Piazza et al . , 2006 ) . Psychophysical evidence showing little cost in cross-modal or cross-format matching also points to a common number sense spanning sensory modalities and formats . For example , human adults are very efficient in making cross-modal and cross-format judgments , with very little cost in either accuracy or reaction times when comparing auditory with visual temporal sequences or dot arrays ( Barth et al . , 2003; Brannon , 2003 ) . Developmental work also show similar accuracy in pre-schoolers for comparing spatial array of dots either with other spatial arrays , or with sequences of sounds ( Barth et al . , 2005 ) . Preferential-looking studies show infants prefer to look at screens displaying adults faces numerically matched with the soundtrack of adult voices ( Jordan and Brannon , 2006 ) , and abstract visual ensembles ( shapes ) numerically matched with on-going sequence of sounds ( Izard et al . , 2009 ) . However , not all agree: Tokita & Ishiguchi ( Tokita and Ishiguchi , 2012 ) reported significantly lower precision for cross-format number comparisons in adults , than for within format comparisons . That there is little or no cost in these matches is certainly indicative of efficient transfer of information between senses , but says little about the mechanisms involved . The match is made at the decision level , so the interaction could be at any stage up to and including decision mechanisms . One of the more powerful psychophysical techniques to probe mechanisms is adaptation ( Mollon , 1974; Thompson and Burr , 2009 ) . Number , like most other primary visual attributes , is also highly susceptible to adaptation ( Schwiedrzik et al . , 2016; Burr and Ross , 2008 ) : visually inspecting for a few seconds a large number of items results in the perceived numerosity of a subsequent ensemble to be strongly underestimated , and vice-versa after adaptation to low numbers ( Burr and Ross , 2008 ) . More recently we have shown that adaptation to numerosity also occurs with sequentially presented stimuli , and that the adaptation effects are both cross-modal and cross-format ( Arrighi et al . , 2014 ) : adapting to sequences of tones affects the perceived numerosity of a subsequently presented series of flashes ( and vice versa ) , and adapting to sequences of flashes affects the perceived numerosity of spatial arrays of items . Importantly , the adaptation was spatially selective , in external rather than eye-centered coordinates , suggesting it has a perceptual rather than cognitive basis . fMRI studies have demonstrated BOLD activity selective for adaptation to numerosity in the human parietal sulcus ( Piazza et al . , 2004 , 2007; Castaldi et al . , 2014; He et al . , 2015 ) . All these studies strongly suggest that the number sense is a high-generalized system , capable of combining numerical information from different senses , and across different formats . Numerical information is also relevant for the production of specific action sequences , from dance routines to more simple repetitive behavioral tasks . A few studies point to an interconnection between numerosity and motor control . For example , neurons in area 5 of the superior parietal lobule of monkey show a clear selectivity for the number of self-produced actions , and inactivation of the area impedes number-based tasks ( Sawamura et al . , 2002 , 2010 ) . Other work has shown that the left ventral premotor cortex is activated by counting successive sensory stimuli ( Kansaku et al . , 2007 ) , and that the human cerebellum shows strong activation for simple numerical calculations ( Arsalidou and Taylor , 2011 ) . The existence of anatomical and functional connections between number and action-generation systems raise the possibility that number-for-action could be encoded within a truly abstract numerosity mechanism . To test this idea , we measured cross-adaptation between motor repetitions and perception of numerosity . The results show that adapting to self-generated action does affect the representations of numerosity of external events , both sequential ( series of flashes ) and simultaneous ( dots ensembles ) , and that the adaptation is spatially selective in external , not hand-centered coordinates . Each trial began with a motor adaptation phase in which participants performed tapping movements for six seconds , under two different conditions ( tested on separate sessions ) : 'high adaptation' , where participants were asked to tap as quickly as possible ( average 5–6 taps/s ) ; and 'low adaptation' where they tapped more slowly ( average 1 . 12 taps/s: see Materials and methods and Figure 5 ) . After the adaptation phase , the test stimulus – either a sequence of flashes or a cloud of dots ( tested on separate sessions ) – was randomly displayed either to the same side of the screen where the hand had been tapping , or to the symmetrically opposite side . Participants estimated the numerosity of the test stimulus , which varied randomly from trial to trial within the range 6–14 . To minimize sensory feedback , participants were placed in a dark room and wore soundproof headphones , and tapped in mid-air behind the computer screen without touching any surface . The results are shown in Figure 1 . Panels A & B show numerosity estimates averaged over all subjects as a function of the physical numerosities displayed . When the test stimulus was displayed on the right side of the screen ( where the adaptation had occurred ) , rapid tapping caused a consistent underestimation of the numerosity of the test , while slow tapping caused an overestimation . The adaptation effects were similarly strong for when the test was a sequence of flashes ( Figure 1A ) as when it was an array of dots presented simultaneously ( Figure 1B ) . Interestingly , the effect occurred only when the stimuli were presented on the same side as the tapping hand ( the right side ) : when presented on the other ( left ) side , adaptation produced no consistent effect ( Figure 1A and B open symbols ) . 10 . 7554/eLife . 16161 . 003Figure 1 . Effects of motor adaptation on perceived numerosity . ( A and B ) Average perceived numerosity as a function of physical numerosity for slow tapping ( downward triangles ) and fast tapping ( upward triangles ) , for sequential ( left ) and simultaneous ( right ) formats . Filled symbols indicate the conditions in which stimuli were spatially congruent with the tapping region , small open symbols to estimates obtained for the unadapted location ( left-hand side ) . ( C and D ) . Adaptation magnitudes for individual subjects when test and tapping were spatially congruent , plotted against the spatially incongruent condition . Stars reports averages , squares single subject data . Error bars refer to ± 1 SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 16161 . 003 We defined adaptation magnitude as the percentage difference in perceived numerosity after adaptation to fast or slow tapping , averaged over all numerosities . For sequential and simultaneous presentations , the adaptation magnitude averaged across subjects ( filled symbols in Figure 1C and D ) was around 20 and 25% respectively for stimuli presented to the adapted location , a very strong effect . For stimuli presented to the unadapted location , the average effect was only 4 & 2% . We also calculated adaptation magnitude for individual subjects . Figure 1C and D plot adaptation magnitudes for the congruent condition ( where the visual stimuli were presented to the right side ) , against the incongruent condition ( stimuli to the left side ) . All subjects showed a significant effect in the congruent condition ( error bars 1 sem ) , but very little effect in the incongruent condition . ANOVA showed that the congruent conditions were highly significant ( F ( 1 , 32 ) = 70 . 219 , p = 0 . 001 , η2 = 0 . 29 , Cohen’s d = 1 . 278 and F ( 1 , 48 ) = 47 . 176 , p = 0 . 0004 , η2 = 0 . 217 , Cohen’s d = 1 . 062 for sequential and simultaneous condition respectively ) , while the non-congruent conditions were weak and insignificant ( ≈ 4% effect , F ( 1 , 32 ) = 1 . 403 , p = 0 . 302 , η2 = 0 . 007 , Cohen’s d = 0 . 167 and ≈ 2% effect , F ( 1 , 48 ) = 0 . 919 , p = 0 . 375 , η2 = 0 . 008 , Cohen’s d = 0 . 179 ) . That the adaptation is spatially specific suggests it is of a perceptual rather than cognitive nature , and unlikely to result from a response bias or any other generalized artifact . This first experiment revealed two clear results: that motor adaptation affects visual estimates of numerosity , for both sequential and simultaneous displays; and that the adaptation is spatially specific . The spatial specificity suggests that the effect is not a high-level , cognitive phenomenon ( such as 'internal counting' ) , but perceptual in nature , mediated by neural mechanisms with circumscribed receptive fields . To verify the robustness of the spatial selectivity , and to understand it better , we repeated the experiment with a new subject pool , changing the tapping hand and location . In this experiment we tested only the simultaneous presentation , as this is the most revealing ( and surprising ) result . The violet symbols of Figure 2A replicate the results of the previous experiment , tapping with the right ( dominant ) hand and testing on both right and left sides ( randomly interleaved ) : the adaptation effect was again strong for stimuli presented on the same side ( filled symbols ) , and non-existent for stimuli on the other side ( open symbols ) ( F ( 1 , 40 ) = 70 . 207 , p = 0 . 000397 , η2 = 0 . 116 , Cohen’s d = 0 . 724; F ( 1 , 40 ) = 2 . 036 , p=0 . 213 , η2 = 0 . 0019 , Cohen’s d = 0 . 0873; for adapted and unadapted location respectively ) . The red symbols of Figure 2B show the results for tapping on the left with the left ( non-dominant ) hand: again the effects occurred only for visual stimuli presented on the congruent side ( left ) , although they were somewhat weaker ( F ( 1 , 40 ) = 9 . 305 , p = 0 . 028 , η2 = 0 . 05 , Cohen’s d = 0 . 4588; F ( 1 , 40 ) = 0 . 265 , p = 0 . 629 , η2 = 0 . 001 , Cohen’s d = 0 . 0633; for adapted and unadapted location respectively ) . Figure 2C shows results for tapping with the dominant ( right ) hand on the left side of the screen . Here , adaptation was found only for stimuli presented to the left side of the screen , suggesting that it is spatially selective in external rather than hand-centered coordinates ( F ( 1 , 40 ) = 36 . 840 , p = 0 . 002 , η2 = 0 . 104 , Cohen’s d = 0 . 6814; F ( 1 , 40 ) = 1 . 380 , p = 0 . 293 , η2 = 0 . 0023 , Cohen’s d = 0 . 096; for adapted and unadapted location respectively ) . Figure 2D shows the results for all six subjects . There is some variability between subjects , particular in the crossed condition , where one subject showed adaptation to stimuli on the right after tapping on the left with the right hand , but by and large the individual data reinforce the group data . 10 . 7554/eLife . 16161 . 004Figure 2 . Reference frame of motor adaptation . ( A ) Average perceived numerosity as a function of physical numerosity for the slow-and fast-tapping conditions ( downward and upward triangles respectively ) , for right-hand tapping . Filled symbols refer to trials when the stimuli were presented in the spatial region where the subjects had tapped ( right side ) small open symbols to trials when the stimuli were presented on the other side . This data replicates Figure 1B with a fresh subject pool . ( B ) Same as A , except subjects tapped with their left hands . Filled symbols refer to testing in the same spatial region where the subjects had tapped ( left side ) , small open symbols to the right side . Other conventions like A . ( C ) Same as A , except the right hand tapped on the left side of the screen . Filled symbols refer to testing on the same spatial region where the subjects had tapped ( left side ) , small open symbols to the right side . ( D ) Adaptation magnitudes for individual subjects when test and tapping were spatially congruent , plotted against the spatially incongruent condition . Color-coding as for A , B and C ( purple: right hand , right side; red: left hand , left side; orange: right hand , left side ) . Stars reports averages , squares single subject data . Error bars refer to ± 1 SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 16161 . 004 In the previous experiment , subjects tapped in mid air to minimize sensory feedback . In the next series of experiments we manipulated the amount of sensory feedback in the adaptation phase to examine interactions between sensory and motor signals . In the first condition ( tactile only ) , subjects tapped a mouse behind the monitor , allowing for tactile feedback ( Figure 3A ) . The adaptation effect in this condition was strong , around 20% ( F ( 1 , 40 ) = 743 . 738 , p = 0 . 0001 , η2 = 0 . 203 , Cohen’s d = 1 . 009 ) . In the next condition ( visual and tactile ) , the monitor accompanied each mouse-tap with a flash , to give visual as well as tactile feedback . Despite the extra feedback , adaptation remained around 20% , ( F ( 1 , 40 ) = 36 . 746 , p = 0 . 002 , η2 = 0 . 184 , Cohen’s d = 0 . 949 ) as shown in panel B of Figure 3 . The last adaptation condition ( visual only ) comprised a sequence of visual flashes whose rates were determined by the adapting motor routine of the previous conditions ( visual and tactile ) . Again , the adaptation effect was found to be strong ( F ( 1 , 40 ) = 61 . 740 , p = 0 . 001 , η2 = 0 . 230 , Cohen’s d = 1 . 093 ) , and similar to the other conditions , around 20% ( Figure 3C ) , making these three adaptation conditions equally effective as tapping in mid-air ( F ( 4 , 29 ) = 0 . 475 , p = 0 . 754 , η2 = 0 . 07 , Cohen’s d = 0 . 548: see Figure 3D ) . 10 . 7554/eLife . 16161 . 005Figure 3 . Role of sensory feedback of motor adaptation on perceived numerosity . ( A ) , ( B ) , ( C ) Average responses as a function of physical numerosity for slow adaptation ( downward triangles ) , fast adaptation ( upward triangles ) and no adaptation ( diamonds ) , for the three different conditions . ( D ) Bar graphs report the average adaptation effect for all adapting conditions ( tactile only - red; visual and tactile - blue , visual only – black and the 2 conditions of Exp 1: sequential-green and simultaneous-violet ) . Open symbols show single subject data . Error bars report ± 1 SEM . All the conditions provided significant effects ( all p-values < 0 . 05 ) . The magnitude of the effect does not differ between conditions ( p > 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16161 . 005 We also verified the results with a two-alternative forced-choice technique . Subjects adapted to high or low tapping rates , as in the first experiment ( no tactile or visual feedback ) , then two clouds of dots were simultaneously presented to the right ( adapted ) and left ( unadapted ) positions . The numerosity of each stimulus varied from trial to trial over the range 5–20 , and subjects indicated which stimulus appeared more numerous . Figure 4A plots average responses as a function of the difference between the right and the left stimulus ( normalized to the average of the two numerosities ) , to yield psychometric functions . The effect of adaptation was again clear: adapting to low tapping rates shifts the curve to the left ( compared with baseline ) , consistent with an overestimation of the perceived numerosity ( t ( 5 ) = 3 . 285 , p = 0 . 021 , Cohen’s d = 1 . 101 ) and high tapping rates caused the opposite effect , even if weaker ( t ( 5 ) = 1 . 237 , p = 0 . 27 , Cohen’s d = 0 . 558 ) . The differences in the points of subjective equality ( PSEs , given by the 50% point of the curves ) of the two adapting conditions again gives an index of magnitude of adaptation , around 15% . Figure 4B shows the PSEs for adaptation to the two conditions . Despite some variability amongst subjects the effects are quite robust and statistical significant as shown by a two-tailed paired t-test: t ( 5 ) = 3 . 56 , p = 0 . 029 , Cohen’s d = 1 . 612 . This experiment confirms the main results with a different technique , and also confirms the spatial selectivity of the adaptation: if adaptation was not spatially selective , it would work equally on the presentations to the left and right sides , annulling the effect . 10 . 7554/eLife . 16161 . 006Figure 4 . Forced-choice measurement of motor adaptation . ( A ) Psychophysical functions for pooled data ( 6 subjects ) after adaptation to fast ( light violet circles ) , slow ( dark violet triangles ) or no ( black squares ) tapping . The curves indicate the proportion of trials when the test ( presented on the right , the same side of tapping ) was seen as more numerous than the unadapted stimulus ( presented on the left ) , as a function of the numerosity difference ( normalized by the averaged of the two stimuli ) . Adaptation to slow tapping shifted the curve leftwards , showing that subjects were biased to perceive the stimulus as more numerous that it was; and adaptation to fast tapping shifted it rightwards . The point where the best-fitting curves pass 50% is considered the point of subjective equality ( PSE , indicated by the coloured arrows ) . ( B ) PSEs for individual subjects after adaptation to fast tapping ( ordinate ) against those after adaptation to low motor repetitions ( abscissa ) . The filled star shows results for data averaged across subjects . Error bars report ± 1 SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 16161 . 006 This study shows that estimates of numerosity , both sequential and simultaneous , are strongly biased after adapting to repetitive finger tapping: rapid tapping decreases apparent numerosity , slow tapping increases it . The effect is spatially selective , primarily in external rather than hand-centered coordinates . There has been a long-standing debate as to whether adaptation effects operate on numerosity per se , or via texture-density mechanisms ( Burr and Ross , 2008; Anobile et al . , 2014 , 2015 , 2016; Bell et al . , 2015; Dakin et al . , 2011; Durgin , 1995 , 2008; Morgan et al . , 2014; Ross and Burr , 2012; Ross , 2010; Tibber et al . , 2012 , 2013 ) . A similar argument could be made here: that the adaptation was to temporal frequency , rather than to numerosity . As with spatial adaptation , there are many reasons to suggest that this is unlikely . However , the cross-format adaptation ( adapt to tapping sequence and test on dot array ) clearly rules out this possibility: the spatial arrays are not temporally modulated . It is numerosity that is being adapted , not temporal frequency . The current results reinforce the many previous studies ( Izard et al . , 2009; Nieder , 2012; Nieder et al . , 2006; Barth et al . , 2003; Brannon , 2003; Barth et al . , 2005; Jordan and Brannon , 2006; Arrighi et al . , 2014; Jordan et al . , 2005 ) discussed in the introduction that point to the existence of a generalized sense of number . Most of these studies relied principally on cross-modal comparisons of number , which could occur at any processing stage , up to and including decision mechanisms . The spatial selectivity shown in our study suggests that the interaction is perceptual rather than cognitive: adapting on the left side did not affect stimuli on the right , and vice versa . Importantly , the specificity was in external coordinates , as adapting the left field with the right hand caused adaptation for visual stimuli presented to the left , not the right visual field . This complements nicely the result of our previous study ( Arrighi et al . , 2014 ) , where we showed that adaptation to visual sequences affects number perception of both sequential and simultaneous presentations , in a spatially selective manner . Interspersing an eye-movement between adaptation and test showed that the adaptation was spatially specific in external rather than eye-centered coordinates: as the current study shows the selectivity is external , not hand-centered . It would be interesting to look at the spatial tuning of the adaptation on a finer grain , to define the size of the adaptation field . The present study shows that the adaptation is at least broadly tuned , confined to a particular hemifield . It would be very informative to determine whether there was also selectivity within each hemifield , and on how fine a grain . Some may find the spatial selectivity of the adaptation difficult to reconcile with the concept of a generalized , abstract sense of number . However , cross-modal effects can also show spatial selectivity . For example , cross-modal integration of visual and auditory ( or tactile ) information occurs only if the stimuli are spatially coincident ( within certain bounds ) ( Slutsky and Recanzone , 2001 ) . Similarly event time , which certainly transcends modalities , and also seems to be coded in parietal cortex ( Leon and Shadlen , 2003 ) , is affected by motion adaptation , in a spatially selective manner ( Burr et al . , 2007; Fornaciai et al . , 2016 ) . Interestingly , the spatial selectivity of the adaptation is in external – not eye-based – coordinates as we observed for number , here and in the previous study ( Arrighi et al . , 2014 ) . We tested adaptation to action under various feedback conditions: visual and tactile , only visual , only tactile , and minimal feedback . All conditions produced similar amounts of adaptation . In the 'minimal feedback' conditions , where subjects tapped in mid-air , there was no tactile feedback from hitting a surface . We could not , however , remove all forms of kinaesthetic feedback , and therefore cannot be certain whether the adaptation signal was the intension to move , or the sensory proprioceptive feedback from the finger . But both are signals about action , whether they are 'inflow' or 'outflow' . It is interesting that this condition with reduced perceptual feedback produced the same amount of adaptation , as did the conditions with visual and/or tactile feedback . It is also interesting that the vision-only condition produced similar adaptation . Many studies have suggested that vision and action are linked ( Arrighi et al . , 2011; Goodale and Milner , 1992 ) . This study is a further clear example of their interconnection , in the encoding the numerosity of internally generated actions and externally generated events . A total of 15 adults ( 13 naïve to the purpose of the study , 2 author; mean age 27 , all right-handed with normal or corrected-to-normal vision ) participated in the numerosity estimation experiments . Six of them were tested in the sequential condition ( test stimuli: sequences of flashes ) and 7 of them in the 'simultaneous condition' in which test stimuli consisted of array of dots simultaneously presented . Three of these ( 2 author and 1 naïve subject ) also participated , together with 3 additional naïve subjects ( mean age of group: 28 ) , in the second experiment investigating the reference frame of the motor adaptation after-effect . Eventually , six subjects ( mean age of group: 28 ) were tested in the experiment concerning forced-choice discrimination of numerosity . All participants gave written informed consent . Experimental procedures were approved by the local ethics committee ( Comitato Etico Pediatrico Regionale—Azienda Ospedaliero-Universitaria Meyer—Firenze FI ) and are in line with the declaration of Helsinki . Stimuli were created and presented with Psychophysics toolbox for Matlab and displayed on a 60 Hz - 17” , touch screen monitor ( LG-FLATRON L1732P ) placed at a subjects view distance of 57 cm . To eliminate auditory feedback , participants wore soundproof headphones . In some conditions , hand movements were monitored by an infrared motion sensor device ( Leap motion controller - https://www . leapmotion . com/ ) running at 60 Hz . During the design of the experiments we computed an appropriate sample size to confidently report an effect of motion adaptation on perceived numerosity . Sample size was measured by means of a one-sample t-test assuming a value of 0 ( no effect ) as a Null Mean and retrieving a value for alternative mean and standard deviation from a previous study of our group ( see Figures 4 and 5 in Arrighi , et al . [Arrighi et al . , 2014] ) . The analysis revealed that with a sample size of 4 , a power of 0 . 95 was achieved with an alpha level of 0 . 01 . For this reason in all or experiments , we always tested a number of participants greater than 4 ( see below for details ) . We did not set any inclusion criteria for subject selection or their data: all data , for all experimental conditions , were analyzed and reported . In all conditions where subjects estimated numerosity we tested statistical significance with a 2 × 9 repeated measures ANOVA with test numerosity ( 9 levels for numerosity , range 6–14 ) and adaptation type ( low and high ) as main factors . Difference in the adaptation effects between the several adaptation conditions ( visual , tactile , visual-tactile , and the two conditions with minimal feedback ) were measured by a one-way ANOVA . In the numerosity discrimination task , difference in the adaptation effects for high and low adaptation were tested for statistical significance by mean of two-tailed paired t-test . For t-test analyses we measured Cohen's d . For repeated measures ANOVA and regression analyses , we reported both Cohen's d and η2 . Here Cohen's d was measured transforming η2 into Cohen's d ( Cohen , 1988 ) . All data are publicly available at Figshare ( Anobile et al . , 2016 ) . Figure 5 plots the tapping rate for the fast against the slow adaptation conditions , expressed as actions per second ( Hz ) . Different colors and symbols refer to different experimental conditions ( see caption ) . On average ( across trials and conditions ) , when asked to tap quickly , participants tapped at a frequency of 5–6 Hz ( for a total number of 30–36 tapping repetitions ) with almost no difference between the adapting conditions: mean 5 . 33 ± 0 . 9; 5 . 48 ± 0 . 5; 5 . 2 ± 0 . 7; 5 . 54 ± 0 . 8; 6 . 19 ± 0 . 37; 5 . 69 ± 0 . 38; 5 . 67 ± 0 . 31 for the ‘sequential’ , ‘simultaneous’ , ‘visual and tactile’ , ‘tactile only’ , ‘adapt with the right hand in the right space’ , ‘adapt with the left hand in the left space’ and ‘adapt with the right hand in the left space’ respectively . Also tapping frequencies for the condition in which subjects tapped slowly were similar across adapting conditions with all values ranging between 0 . 7 and 1 . 3 Hz ( mean 1 . 31 ± 0 . 4; 1 . 29 ± 0 . 3; 1 . 12 ± 0 . 4; 0 . 7 ± 0 . 3; 1 . 18 ± 0 . 12; 1 . 07 ± 0 . 13; 1 . 18 ± 0 . 17 for the ‘sequential’ , ‘simultaneous’ , ‘visual and tactile’ , ‘tactile only’ , ‘adapt with the right hand in the right space’ , ‘adapt with the left hand in the left space’ and ‘adapt with the right hand in the left space’ respectively ) . These data clearly indicate that regardless the tapping routine to be performed on a rigid surface or in mid-air , the tapping temporal dynamics were always very similar . We also tested whether there was a correlation between faster tapping rate and adaptation effects . There was a slight , but non-significant tendency for faster tapping rates to be associated with lower adaptation . But as the correlation was not significant , we assume that variable tapping rates was not a cause for concern for the results of these experiments .

Humans and many other animals have the ability to make spontaneous and rapid estimates of the approximate number of items that they can see . This sense of number , or “numbersense” , is particularly important in humans , as evidence suggests that it lays the groundwork for acquiring mathematical skills . Researchers have many questions about numbersense . Is it a kind of perception ? Or does it require more active thought , like counting ? Do people have the same sense of number when they view , hear or touch items that depict the same number ? Having a sense of number is essential for carrying out certain actions , like the following the steps in a dance , but the connection between action and numbersense is not entirely clear . A process called adaptation means that viewing specific stimuli for a period of time can affect what people think they see subsequently . For example , viewing large numbers of dots makes subsequent smaller groups of dots seem like they contain fewer dots than they actually do . Anobile , Arrighi et al . have now investigated the link between action and numbersense by asking volunteers to tap one hand either rapidly or slowly in one spot for a short time . The volunteers were then shown a series flashes or a cloud of dots in the region where they had been tapping and asked to estimate the number of flashes or dots . After fast tapping , the volunteers greatly underestimated the numbers of flashes or dots that they saw; after slow tapping , they overestimated the numbers . However , if the images were shown far away from where the volunteers had been tapping , their estimates were more accurate . Overall , the results suggest that adaptation is controlled by space-specific sensory mechanisms rather than some kind of active counting . Furthermore , numbersense appears to have a generalized form that is shared by the brain regions responsible for perception and action . Because numbersense and mathematical ability are linked , this strong connection between action and number perception may have important implications for understanding and treating math-related learning disabilities . Anobile , Arrighi et al . next plan to study how movement-driven adaptation affects numbersense in children and adults with these conditions .

2016

A shared numerical representation for action and perception

Malaria parasites use the RhopH complex for erythrocyte invasion and channel-mediated nutrient uptake . As the member proteins are unique to Plasmodium spp . , how they interact and traffic through subcellular sites to serve these essential functions is unknown . We show that RhopH is synthesized as a soluble complex of CLAG3 , RhopH2 , and RhopH3 with 1:1:1 stoichiometry . After transfer to a new host cell , the complex crosses a vacuolar membrane surrounding the intracellular parasite and becomes integral to the erythrocyte membrane through a PTEX translocon-dependent process . We present a 2 . 9 Å single-particle cryo-electron microscopy structure of the trafficking complex , revealing that CLAG3 interacts with the other subunits over large surface areas . This soluble complex is tightly assembled with extensive disulfide bonding and predicted transmembrane helices shielded . We propose a large protein complex stabilized for trafficking but poised for host membrane insertion through large-scale rearrangements , paralleling smaller two-state pore-forming proteins in other organisms . Malaria parasites evade host immunity by replicating within vertebrate erythrocytes . In humans , the virulent Plasmodium falciparum pathogen uses multiple ligands for erythrocyte invasion ( Cowman et al . , 2012 ) and then remodels its host cell to achieve tissue adherence and nutrient acquisition ( Goldberg and Cowman , 2010; Wahlgren et al . , 2017; Desai , 2014 ) . Remarkably , a single protein complex , termed RhopH , contributes to each of these activities despite their separate timings and cellular locations ( Gupta et al . , 2015; Goel et al . , 2010 ) . The three subunits of the RhopH complex , known as CLAG , RhopH2 , and RhopH3 , are conserved and restricted to Plasmodium spp . ; none have significant homology to proteins in other genera ( Kaneko , 2007 ) , suggesting that these proteins and the complex they form evolved to meet the specific demands of bloodstream parasite survival . While RhopH2 and RhopH3 are single-copy genes in all Plasmodium spp . , CLAG proteins are encoded by a multigene family with variable expansion in malaria parasite species infecting humans and other vertebrates including birds , rodents , and primates ( Kaneko et al . , 2001; Cortés et al . , 2007; Rovira-Graells et al . , 2015 ) . Each of these subunits is transcribed in mature schizont-infected erythrocytes ( Figure 1A; Ling et al . , 2004 ) ; during translation , these proteins assemble with unknown stoichiometries into a complex that is packaged into rhoptry organelles ( Ito et al . , 2017 ) . Upon host cell rupture , RhopH3 , but not CLAG or RhopH2 subunits , facilitates invasion of the next erythrocyte . Some 18 hr later , CLAG3 , a paralog encoded by the parasite chromosome 3 , inserts in the host erythrocyte membrane to form the plasmodial surface anion channel ( PSAC ) for nutrient uptake ( Desai et al . , 2000; Nguitragool et al . , 2011; Pillai et al . , 2012 ) ; other paralogs may also contribute to PSAC ( Gupta et al . , 2020 ) or , in the case of CLAG9 , to cytoadherence ( Trenholme et al . , 2000; Goel et al . , 2010; Nacer et al . , 2011 ) . RhopH2 and RhopH3 also traffic to the host membrane and are required for PSAC activity ( Ito et al . , 2017; Counihan et al . , 2017 ) . Because these proteins have no homologs in other genera , how they traffic within infected cells and serve these multiple roles is unknown . Our data reveal essential features of the RhopH complex . We combine mass spectrometry , single-particle cryo-electron microscopy ( cryo-EM ) and biochemical studies using conditional knockdown of protein export to determine that the RhopH is initially produced as a soluble complex that functions in erythrocyte invasion . The complex remains soluble in extracellular merozoites and , upon completed invasion , is deposited into the parasitophorous vacuole surrounding the intracellular parasite . A protein translocon on the parasitophorous vacuolar membrane , PTEX ( de Koning-Ward et al . , 2009 ) , contributes to RhopH export via an unknown mechanism ( Ito et al . , 2017 ) . Our high-resolution de novo RhopH complex structure and biochemical studies suggest large-scale conformational changes for eventual conversion to an integral form at the host erythrocyte membrane . This conversion is PTEX dependent and enables channel-mediated uptake of host plasma nutrients . To address these questions , we sought to recover well-behaved RhopH complexes . Alkaline Na2CO3 extraction but not hypotonic treatment partially released CLAG3 from infected cell membranes ( Figure 1B , top row ) , implicating both integral and peripheral membrane pools . We found that simple freeze-thaw also releases some CLAG3 from the peripheral pool ( bottom row ) ; although Na2CO3 extraction releases a larger amount , freeze–thaw is gentler and does not denature many proteins . Neither of these treatments is expected to release integral membrane proteins . Using multiple C-terminal tags engineered into the single clag3h gene of the KC5 line ( Gupta et al . , 2018; CLAG3-tv2; Figure 1C , bottom ) , we effectively harvested this minor fraction from human blood cultures . This CLAG3 remained associated with RhopH2 and RhopH3 ( Figure 1C ) and yielded monodisperse protein complexes in negative stain imaging ( Figure 1D ) . Native mass spectrometry ( MS ) yielded a molecular weight of 433 , 790 ± 10 Da ( Figure 1E , Figure 1—figure supplement 1 ) , matching the expected mass for a heterotrimeric complex with a 1:1:1 stoichiometry; a 0 . 6% mass error may reflect post-translational modification and/or proteolytic processing , as reported for RhopH3 ( Ito et al . , 2017 ) . A smaller 333 , 232 ± 3 Da fraction corresponded to a minor CLAG3–RhopH2 heterodimer . Thus , freeze–thaw permits gentle , detergent-free harvest of this essential complex . RhopH complexes segregated into 2D classes with two primary views ( two-lobe and side views , Figure 1F , top row ) . We next used a green fluorescent protein ( GFP ) -derivative-tagging approach ( Ciferri et al . , 2012 ) , confirmed integrity of each variant ( Figure 1—figure supplement 2A ) , and detected densities reflecting addition of this globular epitope tag . This independently confirmed single copies of each subunit and established an orthogonal arrangement for CLAG3 and RhopH2 ( arrows , Figure 1F ) . Three-dimensional reconstruction provided a low-resolution image of the entire complex and established a two-lobed structure ( Figure 1G ) . A similar two-lobed structure was obtained for RhopH complexes recovered from spent media without protease inhibitors , detergents , or freeze–thaw , implicating a highly stable complex . Finally , the purified RhopH complex resisted aggregation and unfolding at temperatures above those seen in malaria fevers ( Figure 1—figure supplement 2B , C ) . We submit that a thermostable RhopH complex is well-equipped for transit through diverse subcellular environments . We next determined the complex’s de novo structure using cryo-EM and concentrated protein from sequential coimmunoprecipitation ( 0 . 8–2 mg/mL , FLAG and His10 tags ) . Initial analyses with 2D and 3D classifications yielded a two-lobed structure with a 3 . 3 Å resolution ( Figure 2 , Figure 2—figure supplement 1; Table 1 ) ; per-particle contrast transfer function ( CTF ) estimation and motion correction improved overall resolution to 2 . 9 Å . Soluble RhopH is a heterotrimeric complex consisting of single CLAG3 , RhopH2 , and RhopH3 subunits ( Figure 2A ) , as predicted above . CLAG3 mediates subunit associations through independent contacts with RhopH2 and RhopH3 , which do not directly interact with one another . The visualized complex assumes a ‘shallow bowl with a short base’ appearance due to an out-of-plane orientation of RhopH2 relative to CLAG3 ( Figure 2B ) . On the opposite face , a CLAG3 mid-section protrudes to create a short ‘base’ that includes a critical amphipathic α-helix proposed to line the PSAC pore at the host membrane , as described below . The bowl’s opposite rim is formed by globular α-helices from CLAG3 and RhopH3 . From other angles , an asymmetric two-lobed architecture is apparent , with a well-resolved large lobe that enabled confident de novo model building for CLAG3 and RhopH3 ( Figure 2C , Figure 2—figure supplement 1C ) . In contrast , the small lobe was initially not well-resolved . We hypothesized that both lobes have defined structures that undergo relative movement and therefore used multi-body refinement ( Nakane et al . , 2018 ) to identify rigid but mobile substructures . Assuming two bodies joined by a CLAG3 stem , we refined each lobe separately and improved the small lobe’s resolution ( Figure 2—figure supplement 1C ) . The small lobe’s hammer-shaped ends were now clearly visualized , improving model building from 225 to 513 residues for RhopH2 . Excluding their flexible N- and C-terminal tails , ≥90% of CLAG3 and RhopH3 residues were also confidently localized . Multibody refinement also defined the directions and extent of motion between the two lobes ( Figure 2—figure supplement 2; Videos 1 and 2 ) . Interestingly , consideration of protein energy landscapes using normal mode analysis ( Suhre and Sanejouand , 2004 ) predicted remarkably similar motions ( Videos 3–7 ) . Although the biological significance of this mobility is uncertain , conservation of the stem sequence and length in P . falciparum CLAG paralogs and among other Plasmodium spp . supports an important role ( Figure 2—figure supplements 3 and 4; 48% bridge region identity between divergent human P . falciparum and P . vivax CLAGs ) . CLAG3 contains three visually distinct domains ( Figure 2C ) : an N-terminal sphere , an elongated central bridge for binding RhopH2 , and a C-terminal bundle encasing an amphipathic helix that later integrates in the host erythrocyte membrane ( Sharma et al . , 2015 ) . The N- and C-terminal domains hold RhopH3 tightly through bidentate interactions via a ‘1300 loop’ and a ‘300 region’ that form orthogonal pincer-grasp interactions . We illustrate these high-confidence interactions between CLAG3 and the other subunits in Figure 2D . The 1300 loop packs against RhopH3 with discrete foci of hydrophobic and salt-bridge interactions ( formed by CLAG3 residues Y1331 , F1338 , F1347 , L1348 and D1334 , D1339 , D1340 , E1345 , respectively , Figure 2E , left panels ) . These CLAG3 residues and the cognate RhopH3 residues are highly conserved ( Figure 2—figure supplement 4B and D ) , implicating essential roles in stabilizing the complex . The less strictly conserved 300 region consists of three α-helices , with two helices ( 10 and 11 ) interacting with RhopH3 residues 397–412 and 575–588 to create a hydrophobic core with a convergence of aromatic side chains ( core formed by F314 , F322 , V330 , M333 , M394 , Y399 , Figure 2E , bottom right ) . The third CLAG3 helix ( helix 14 ) and an upstream loop are closely apposed to RhopH3 through complex interactions . Together , the 300 region and 1300 loop produce an extensive 3700 Å2 CLAG3 interface with RhopH3 . The 2005 Å2 CLAG3–RhopH2 interface is much more fragmented ( residues 706–715 , 787–805 , and 920–939 of CLAG3 and 414–435 , 580–594 , 682–708 , and 760 from RhopH2 ) . Interestingly , the CLAG3 backbone threads back and forth through the bridge domain ( Figure 2F , bottom left ) to form a wall-like interface; both surfaces are enriched with hydrophobic , conserved residues that form stable interactions ( Figure 2F , bottom right; Figure 2—figure supplements 3 and 4A , C ) . Each subunit has numerous conserved cysteines that contribute to tight assembly of this large complex through the formation of observed and possible disulfide bonds ( Figure 2G , H ) . Although we did not detect bonding between subunits , several cysteines were not visualized and could form such interactions . Conserved cysteines are a common feature of rhoptry proteins ( Kaneko , 2007 ) ; they presumably contribute stability during egress and erythrocyte invasion and may also be critical for RhopH enzymatic activity at its final erythrocyte membrane destination ( Carter , 1973 ) . CLAG3’s central position in the structure , together with its surface exposure on erythrocytes and immune selection ( Iriko et al . , 2008 ) , likely accounts for expansion of the clag gene family in all Plasmodium spp . We examined CLAG phylogeny and found that P . falciparum paralogs cluster into well-supported groups containing species infecting other mammals ( Figure 2—figure supplement 5A ) . CLAG9 clustered independently and represented an older lineage . Sequences from Plasmodium spp . -infecting birds formed a separate group ( labeled ‘Clade F’ in Figure 2—figure supplement 5A , based on taxonomy proposed by Galen et al . , 2018 ) . These sauropsid CLAG sequences are split into two well-supported orthologous groups , one that is basal to the CLAG2/CLAG3/CLAG8 orthologs and one that is basal to the CLAG9 orthologs . This pattern suggests an ancient split into two paralogs in the common ancestor of sauropsid and mammalian Plasmodium spp . , with subsequent diversification of mammalian paralogs . This diversification and ongoing gene family expansion ( Otto et al . , 2018 ) may yield distinct RhopH complexes capable of divergent functions including erythrocyte invasion , cytoadherence , and nutrient uptake . Expansion may also permit fine-tuning of PSAC permeabilities to allow nutrient uptake in both malnourished and well-fed hosts ( Mira-Martínez et al . , 2017 ) . Structural similarity searches of the Protein Data Bank ( PDB ) revealed weakly significant hits for each subunit that may guide structure–function studies of this Plasmodium-restricted complex ( Figure 2—figure supplement 5B and C ) . RhopH3 exhibited the greatest structural similarity with alignment to domains from SepL , a regulator of type III translocon-based secretion in bacteria ( Burkinshaw et al . , 2015 ) . RhopH2 partially aligned with Bcl-xL , an anti-apoptotic protein that also regulates membrane permeabilization ( Finucane et al . , 1999 ) . Both hits from our structural similarity searches raise the tantalizing possibility that RhopH2 and RhopH3 function to regulate PSAC . Such regulation could produce the unprecedented selectivity of this channel , which imports diverse nutrients including purines , amino acids , sugars , and some vitamins while maintaining very low Na+ permeability to prevent host cell osmotic lysis ( Cohn et al . , 2003 ) . Biochemical studies point to a direct contribution of the RhopH complex in PSAC-mediated nutrient uptake ( Gupta et al . , 2018; Gupta et al . , 2020 ) , with a single confidently predicted CLAG3 transmembrane domain distal to a 10–30 residue hypervariable region ( HVR , Figure 3A ) . Site-directed mutagenesis of a conserved A1215 residue in this transmembrane domain ( α-helix 44 in our structure ) alters channel gating , selectivity , and conductance , supporting a pore-lining helix ( Sharma et al . , 2015 ) . Notably , a PDB structure search identified this and several neighboring helices with a significant alignment to APH-1 , an integral membrane component of human γ-secretase ( Figure 3—figure supplement 1A ) . The corresponding APH-1 α-helix makes stable interactions with phospholipid in that structure ( Bai et al . , 2015 ) , further supporting membrane insertion of CLAG3 α-helix 44 . This important helix is buried within a CLAG3 C-terminal bundle ( Figure 3B–D ) , paralleling buried hydrophobic helices in some much smaller pore-forming proteins ( Dal Peraro and van der Goot , 2016; Figure 3—figure supplement 1B ) . Transverse and longitudinal views establish that multiple Phe side chains segregate to one surface of helix 44 and that polar side chains line up at the opposite face ( Figure 3E ) , as expected for a helix that lines an aqueous pore ( Sharma et al . , 2015 ) . Although its physicochemical properties are conserved in CLAG orthologs , helix 44 exhibits little primary sequence conservation ( Figure 3F ) . In contrast to this helix , the nearby HVR was poorly ordered , consistent with an unstructured extracellular loop that functions as an immune decoy ( Figure 3D ) . The single predicted transmembrane domains on RhopH2 and RhopH3 are also buried in the soluble structure ( helices defined by V740-D757 and G595-Y622 of these subunits , respectively; Figure 3—figure supplement 1C–F ) . Thus , known and predicted transmembrane domains are shielded in the trafficking RhopH complex ( Figure 3G ) , implicating large-scale protein rearrangements for their membrane insertion . The peripheral and integral membrane pools ( Ito et al . , 2017 ) of the RhopH complex may both be formed during protein synthesis . Alternatively , the complex may be produced exclusively as a soluble form for trafficking and membrane insertion at a later point in the cell cycle . To distinguish between these models , we performed fractionation studies with synchronous cultures at defined developmental stages . During stage-specific synthesis in schizont-infected cells ( Ling et al . , 2004 ) , both peripheral and integral membrane pools were reproducibly detected ( Figure 4A , top row ) . This finding’s interpretation is complicated by preexisting CLAG3 derived from the preceding cycle and trafficked to the infected cell surface ( Figure 1A ) . To address this uncertainty , we treated early schizont-stage cultures with protease to identify prior-cycle CLAG3 inserted at the erythrocyte membrane . As the integral pool was quantitatively proteolyzed ( Figure 4A , B ) , we conclude that the integral pool in these cells reflects protein made in the previous cycle; the larger carbonate-extractable pool represents newly synthesized protein . Fractionation studies using purified merozoites revealed carbonate-extractable CLAG3 and undetectable levels of integral protein ( Figure 4C ) , consistent with packaging of newly synthesized RhopH complex into rhoptries and jettisoning of the prior-cycle integral host membrane pool upon schizont rupture; the host membrane marker , Band3 , is also discarded at egress . Thus , CLAG3 is synthesized as a soluble protein that associates with other RhopH subunits to interact peripherally with membranes in rhoptries; whorls seen in rhoptries may provide a membranous surface for transfer of these proteins to the next erythrocyte ( Bannister et al . , 1986 ) . We then tracked this newly synthesized pool through the parasite bloodstream cycle and found that merozoites transfer their peripheral CLAG3 pool to immature ring-stage parasites , which also carry negligible amounts of the integral form ( Figure 4D , E , rings ) . With parasite maturation , CLAG3 transitions from a primarily extractable form upon synthesis in schizonts into a growing integral pool after transfer into new erythrocytes ( Figure 4D , E , trophozoites ) . During this conversion , CLAG3 remains associated with other RhopH subunits and eventually localizes to the infected host cell membrane ( Vincensini et al . , 2008; Nguitragool et al . , 2011; Ahmad et al . , 2020 ) . How does this 440 kDa soluble RhopH complex convert into an integral form ? Upon erythrocyte invasion , these and other rhoptry proteins are deposited into the parasitophorous vacuole . The PTEX protein translocon exports proteins secreted by the intracellular parasite into host cytosol ( de Koning-Ward et al . , 2009; Beck et al . , 2014; Ho et al . , 2018 ) . It may therefore also export RhopH proteins into host cytosol; such transfer would be novel as it has not been established for other merozoite proteins deposited in the vacuole . While two studies have obtained conflicting results about whether RhopH proteins are exported via this translocon , both reported that PTEX knockdown abolishes activation of PSAC-mediated nutrient uptake at the host membrane ( Beck et al . , 2014; Ito et al . , 2017 ) . To examine membrane insertion , we performed CLAG3 fractionation using 13F10 , a conditional PTEX knockdown parasite ( Beck et al . , 2014 ) whose protein export requires trimethoprim ( TMP , Figure 4F ) . We found that CLAG3 transitions to an integral form in this parasite normally in the presence of TMP , but that PTEX knockdown produces a loss of integral CLAG3 ( -TMP , p = 0 . 01 , n = 3 ) . CLAG3 that failed to insert into the membrane was more readily solubilized ( -TMP , soluble lane ) , possibly due to protein crowding as a result of blocked export from the parasitophorous vacuole . Thus , RhopH membrane insertion is dependent on PTEX activity . Stage-dependent membrane insertion was further evaluated in CLAG3-tv2 parasites with protease susceptibility studies . Both the freeze–thaw released and carbonate-extractable pools of CLAG3 were unaffected by extracellular protease , but the integral pool at the host membrane yielded a C-terminal cleavage product that remained membrane embedded ( Figure 4G ) . α-Helix 44 is within this cleavage fragment and likely provides the responsible transmembrane anchor . Collectively , these findings indicate that CLAG3 is synthesized and trafficked in a soluble RhopH complex that undergoes marked rearrangements during its export to enable insertion at the host membrane . We propose that RhopH evolved as a modular three-protein complex suited for essential and divergent functions at separate points in the bloodstream parasite cycle ( Figure 5 ) . A soluble form , packaged into rhoptry secretory organelles , facilitates RhopH3 contribution to erythrocyte invasion through still unknown mechanisms that presumably involve surface interactions . A large exposed surface area of ~32 , 000 Å2 and globular architecture of RhopH3 provide candidates for inquiry . Our structure similarity searches found that RhopH3 residues 434–665 align with domains 2 and 3 of SepL; because domain three mediates interaction with the Tir receptor ( Burkinshaw et al . , 2015 ) , one possibility is that RhopH3 interacts with an unidentified host cell receptor at this site . The RhopH3 C-terminus provides another surface for the presumed interactions , as suggested by site-directed mutagenesis of serine 804 and by studies with a monoclonal antibody against a 134 aa recombinant fragment ( Doury et al . , 1994; Ekka et al . , 2020 ) . This entire region ( residues 716–897 ) is not resolved in our structure and appears to be flexible . Invasion-inhibiting antibodies that bind here may directly or indirectly prevent essential interactions with a cognate receptor . These findings and recent structural studies of the Rh5-CyRA-Ripr ( Wright et al . , 2014; Wong et al . , 2019 ) should enable structure-guided therapies targeting erythrocyte invasion , an Achilles heel in the parasite’s bloodstream cycle . A soluble RhopH complex may also facilitate transfer to new erythrocytes for a second role in PSAC-mediated nutrient uptake ( Nguitragool et al . , 2011 ) . We determined that the complex is transferred to the new host cell and deposited in the parasitophorous vacuole in a soluble form . The member subunits may then be exported into host cell cytosol via PTEX , as suggested by confocal immunofluorescence assays showing blocked export of each RhopH subunit in PTEX knockdown parasites ( Ito et al . , 2017 ) . Forward and reverse coimmunoprecipitation experiments also suggest that the RhopH complex directly interacts with PTEX to enter host cell cytosol ( de Koning-Ward et al . , 2009; Counihan et al . , 2017 ) . We next show that CLAG3 membrane insertion occurs via a PTEX-dependent mechanism ( Figure 4F ) . Insertion may occur either concurrently with or after export . Because exported chaperones are thought to facilitate refolding of exported proteins and subsequent transit to specific host cell sites , failed CLAG3 membrane insertion may result from blocked export of multiple effector proteins . Our structure reveals several intriguing and unique problems faced by the RhopH complex during its export and host membrane insertion . How this large complex crosses the parasitophorous vacuolar membrane remains unclear . If it transits directly through PTEX , this tightly assembled ternary complex with numerous disulfide bonds would require carefully coordinated unfolding and disassembly by HSP101 and possibly other vacuolar activities before translocation ( Beck et al . , 2014; Ho et al . , 2018; Matthews et al . , 2019 ) . Subsequent reassembly in host cytosol may be even more complicated , with largely uncharacterized machinery needed to reform a stable complex without denaturation . Another dilemma exposed by these studies is the precise mechanism by which one or more RhopH subunits become integral to the host erythrocyte membrane while remaining strictly associated with each other ( Ito et al . , 2017; Ahmad et al . , 2020 ) . Although membrane insertion during transit through PTEX would follow the precedent of Sec translocon-mediated membrane insertion in bacteria and other eukaryotes ( Denks et al . , 2014 ) , PTEX appears to lack a lateral gate , as used by other translocons to transfer cargo proteins into the adjacent lipid bilayer ( Egea and Stroud , 2010; Corey et al . , 2019; Ho et al . , 2018 ) . We tend to favor membrane insertion after transfer into host cytosol . In this scenario , the energetically demanding process of conformational rearrangement to expose and insert specific α-helical domains into the host membrane may be facilitated by interactions with parasite-derived chaperones and Maurer’s cleft organelles ( Proellocks et al . , 2016 ) . Although various studies support a role of the RhopH complex in PSAC formation and nutrient uptake ( Nguitragool et al . , 2011; Mira-Martínez et al . , 2019; Sharma et al . , 2013; Ito et al . , 2017; Counihan et al . , 2017 ) , whether this ternary complex directly forms the aqueous pore in the host erythrocyte membrane remains debated . In vitro selection previously implicated a short , but critical amphipathic CLAG3 motif in solute selectivity and PSAC single-channel gating ( Lisk et al . , 2008; Sharma et al . , 2015 ) . Our de novo structure establishes that this motif indeed forms an α-helix with hydrophobic and polar side chains segregated to opposite faces ( helix 44 , Figure 3 ) , supporting a pore-lining helix in the host membrane . While studies suggest that CLAG3 oligomerizes at the host membrane and has an surface-exposed variant region ( Figure 5; Gupta et al . , 2018; Nguitragool et al . , 2014 ) , RhopH2 and RhopH3 are not exposed based on protease susceptibility studies ( Ito et al . , 2017 ) . The CLAG3 helix 44 and the individual predicted transmembrane domains on RhopH2 and RhopH3 are separated from one another by 46–101 Å in the soluble structure ( Figure 3G ) . If all three helices come together to form the eventual nutrient pore , a remarkable rearrangement of the complex will be required during its conversion from a soluble to a membrane-inserted form . While our findings suggest interactions with PTEX or exported chaperone proteins , these rearrangements may also be facilitated by post-translational modifications such as site-specific phosphorylation and lysine acetylation ( Cobbold et al . , 2016; Pease et al . , 2013 ) . Our findings provide a framework for understanding two unique and essential functions in bloodstream malaria parasites . Structure-guided development of therapies can now be pursued against a strictly conserved target exposed to plasma at two key points in the parasite cycle . P . falciparum laboratory strains were grown in O+ human erythrocytes ( Interstate Blood Bank ) using standard methods and maintained at 5% hematocrit under 5% O2 , 5% CO2 , 90% N2 at 37°C . CRISPR-Cas9 gene editing was used to produce engineered P . falciparum lines using the KC5 laboratory clone carrying a single clag3h gene to avoid epigenetic switching ( Gupta et al . , 2018 ) . Transfections were performed by electroporation of pUF1-Cas9 and modified pL6 plasmids for homologous replacement of the genomic site as described ( Ghorbal et al . , 2014 ) ; 1 . 5 µM DSM1 and 2 nM WR99210 were used to select for integrants , which were detected by PCR . All experiments were performed with limiting dilution clones that were confirmed with DNA sequencing . Primary protein purifications used the edited CLAG3-tv2 clone , in which a C-terminal multiple affinity tag consisting of His10-FLAG-thrombin-TEV-HA-twinstrept-BC2 nanobody binding site was appended to an otherwise unmodified CLAG3h . The CLAG3-GFP incorporates a C-terminal His8-monomeric GFP-FLAG-Twin strept tag on CLAG3h . The CLAG3-tv1+RhopH2-mV strain contains a C-terminal 3xFLAG-3xHA-His8-Strept II tag on CLAG3 and a monomeric Venus tag at the RhopH2 C-terminus; this parasite was produced by sequential CRISPR-Cas9 editing of the two genomic loci and used for negative stain imaging of mVenus-tagged RhopH2 . Up to 1 mL of enriched schizont-stage parasites were harvested by the percoll–sorbitol method and frozen in liquid nitrogen at 20% v/v in 200 mM NaCl , 10 mM Tris , pH 7 . 5 with 1 mM phenylmethylsulfonyl fluoride ( PMSF ) . Frozen parasites were thawed at room temperature , and insoluble debris was pelleted at 20 , 000 × g for 10 min at 4°C . NaCl was added to 500 mM before overnight incubation of the clarified lysate with anti-FLAG M2 affinity agarose resin ( Sigma–Aldrich ) at 4°C with gentle agitation . The resin was subsequently washed with 1–5 mL of 10 mM Tris , pH 7 . 5 and 500 mM NaCl before elution in 10 mM Tris , pH 7 . 5 , 200 mM NaCl and 0 . 15 mg/mL 3xFLAG peptide . The eluate was concentrated for native mass spectrometry and cryo-EM studies via a second affinity purification on Ni-NTA agarose resin ( Qiagen ) and small volume elution in 200 mM NaCl , 300 mM imidazole , 10 mM Tris , pH 7 . 5 . After overnight dialysis to remove imidazole , purified RhopH complex was further concentrated by ultracentrifugation at 150 , 000 × g for 1 hr , yielding 0 . 8–2 mg/mL protein in 30 µL . Purified RhopH complex was buffer-exchanged into native mass spectrometry ( MS ) solution ( 200 mM ammonium acetate , pH 7 . 5 , 0 . 01% Tween-20 ) using Zeba microspin desalting columns with a 40 kDa cut-off ( ThermoScientific; Olinares et al . , 2016; Olinares and Chait , 2020 ) . Buffer-exchanged sample ( 3 µL ) was loaded into a locally prepared gold-coated quartz emitter and electrosprayed into an Exactive Plus EMR instrument ( ThermoFisher Scientific ) with a modified static nanospray source ( Olinares and Chait , 2020 ) . The MS parameters used include spray voltage , 1 . 2–1 . 3 kV; capillary temperature , 150–250°C; in-source dissociation , 10 V; S-lens RF level , 200; resolving power , 17 , 500 at m/z of 200; AGC target , 1 × 106; maximum injection time , 200 ms; number of microscans , 5; injection flatapole , 8 V; interflatapole , 4 V; bent flatapole , 4 V; high-energy collision dissociation , 200 V; ultrahigh vacuum pressure , 7–8 × 10−10 mbar; total number of scans , ≥100 . Mass calibration in positive extended mass range ( EMR ) mode was performed using cesium iodide . The acquired MS spectra were visualized using Thermo Xcalibur Qual Browser ( versions 3 . 0 . 63 and 4 . 2 . 47 ) . Spectra deconvolution was performed either manually or using the software UniDec versions 3 . 2 and 4 . 1 ( Marty et al . , 2015; Reid et al . , 2019 ) . The resulting deconvolved spectrum from UniDec was plotted using the m/z software ( Proteometrics LLC ) . Experimental masses were reported as the mean ± SD across all calculated mass values within the observed charge state series . Mass accuracies were calculated as the % difference between the measured and expected masses relative to the expected mass . Thermal denaturation of the RhopH complex was evaluated with two methods . ThermoFluor assays were performed with 20 µL of 0 . 4 mg/mL freeze–thaw extracted RhopH complex and a 1× dilution of SYPRO Orange . Fluorescence intensity was continuously monitored during a thermal ramp from 25°C to 95°C in 0 . 5°C/10 s increments . Raw fluorescence and first-derivative plots were used to assess unfolding . RhopH complex aggregation was also evaluated using sizing with thermal ramp application on Uncle ( Unchained Labs ) and duplicate samples of 8 . 9 µL of 0 . 1 mg/mL RhopH complex . Aggregation was measured by monitoring static light scattering at 266 and 473 nm with a ramp from 20°C to 80°C at a constant rate of 1 . 0°C/min for 1 hr with measurements at 0 . 5°C increments . Purified RhopH protein ( 4 . 8 μL of a 0 . 05 mg/mL solution ) was applied to carbon film grids ( CF200-Cu , Electron Microscopy ) and stained with 4 . 8 μL of 0 . 75% uranyl formate for 30 s . After drying , grids were loaded onto a ThermoFischer Tecnai 12 electron microscope with a Gatan Ultra Scan camera operating at 120 kV . Images were collected using EPU software ( ThermoFischer ) at 67 , 000× magnification for a pixel size of 1 . 77 Å . The datasets consisted of between 69 and 142 micrographs ( culture-media RhopH , 69 micrographs; complexes containing RhopH2-mV , 109; CLAG3-tv1 , 124; CLAG3-GFP , 142 ) . All negative stain image processing was performed using RELION 2 . 0 ( Scheres , 2012 ) . Micrographs were processed without CTF correction . Initial auto-picking was performed using a Gaussian blob . Well-behaved classes from 2D classification of Gaussian blob-picked particles were used for template-based auto-picking . Further 2D classification was performed to clean the particle set . For datasets with GFP derivative tagging , additional density for the bulky epitope was visible is several 2D classes . For freeze–thawed solubilized and spend-media RhopH , an initial model was generated and used for 3D auto-refinement in RELION . Three-dimensional models represent views in Chimera ( Pettersen et al . , 2004 ) . 2 . 5 μL of 0 . 8 mg/mL RhopH was applied to glow-discharged Quantifoil Cu 300 mesh grids ( 1 . 2/1 . 3 ) , blotted for 3 s , and plunge frozen in liquid ethane cooled by liquid nitrogen using a Vitrobot plunge freezing instrument ( FEI/ThermoFisher ) . The blotting chamber was maintained at 20°C and 100% humidity . One thousand three hundred and ten micrographs were collected on a Titan Krios ( ThermoFisher ) transmission electron microscope operated at 300 kV . Images were recorded on a K2 Summit camera ( Gatan Inc ) operated in super-resolution counting mode and a physical pixel size of 0 . 84 Å . The detector was placed at the end of a GIF Quantum energy filter ( Gatan Inc ) , operated in zero-energy-loss mode with a slit width of 20 eV . Each image was fractionated into 58 frames with a frame exposure of 0 . 4 s and a dose rate of 3 e–/Å2/s , giving a total accumulated dose of 70 e–/Å2 over the 23 . 2 s exposure . All data was collected using the Latitude S software ( Gatan Inc ) . All cryo-EM image processing was performed in RELION 3 . 0 . Movies were motion corrected and dose-weighted using MotionCor2 ( Zheng et al . , 2017 ) . Contrast transfer function ( CTF ) parameters were determined using the Gctf ( Zhang , 2016 ) wrapper in RELION . Initial particle picking was performed with the Laplacian-of-Gaussian ( LoG ) picker in RELION . Subsequent 2D classes from the LoG-picked particles were used for template-based auto-picking performed in RELION resulting in 311 , 390 particles . After two rounds of 2D classification , the initial collection was cleaned to 214 , 233 particles and used to generate an initial 3D model . Three-dimensional classification using five classes with regularization parameter T = 4 resulted in one well-resolved class of 68 , 216 particles . Three-dimensional auto-refinement of these particles resulted in a 3 . 26 Å map . Two rounds of particle polishing and one round of CTF refinement further improved the resolution to 2 . 92 Å . Although the large lobe was well-resolved and permitted de novo model building , the small lobe and C-terminal bundle of CLAG3 were resolved to lower resolution inhibiting interpretation . Further 3D classification did not improve small subunit interpretability . To better resolve RhopH2 and CLAG3 C-terminal domain , multibody refinement was performed ( Nakane et al . , 2018 ) . Multibody refinement using masked region 1 of the large subunit and masked region two as the small subunit and the bridge between the large and small subunit resulted in better EM density for mobile elements of the small subunit although a lower overall resolution for the second masked region . Multibody analysis also yielded the top components of motion . Model building was performed in Coot ( Emsley et al . , 2010 ) . EM density maps were generated in RELION by post-processing with a constant B factor or locally sharpened regions of the maps in Local Resolution . Initially , a poly-alanine model was built for well-ordered regions of the RhopH complex in Coot . The sequence registry was determined by a combination of manual examination of side-chain density and the PHENIX assign_sequence program ( Liebschner et al . , 2019 ) , which predicts sequence registry based on side-chain density . Regions of the map with low resolution were built through a combination EM density interpretation and secondary structure prediction performed in JPred ( Drozdetskiy et al . , 2015 ) . Real space refinement with secondary structure restraints was performed in PHENIX real space refine ( Afonine et al . , 2018 ) . Structural figures were generated in PyMOL 2 . 1 . 0 ( Schrödinger ) or Chimera . Prediction of motion in the final model was performed using the elNémo server ( Suhre and Sanejouand , 2004 ) . CLAG DNA sequences were downloaded from PlasmoDB ( http://PlasmoDB . org ) and aligned using the MAFFT server ( Katoh et al . , 2019 ) with default parameters . Sequences shorter than 2000 nucleotides in length were removed to maximize sequence overlap . The multiple-sequence alignment was corrected manually to preserve the reading frame . Phylogenetic analysis of the remaining 147 sequences was performed using the MEGA X software ( Kumar et al . , 2018; Stecher et al . , 2020 ) . A phylogenetic tree was inferred using the neighbor-joining method ( Saitou and Nei , 1987 ) based on pairwise distances computed using the maximum composite likelihood method ( Tamura et al . , 2004 ) , with the rate variation among sites modeled with a gamma distribution ( shape parameter = 1 ) . To assess how well the data supported the groups in the tree , 250 bootstrap replicates were performed ( Felsenstein , 1985 ) . The ConSurf server ( https://consurf . tau . ac . il/; Ashkenazy et al . , 2016 ) was used to generate per-residue conservation scores and map conservation values on the 3D RhopH complex structure . Non-redundant sequences of RhopH subunits from Plasmodium spp . were identified through the use of PlasmoDB and NCBI Protein BLAST . The sequences were aligned using Clustal Omega . ConSurf was then used to evaluate evolutionary conservation of amino acid residues; the resulting conservation scores were used for color-coding residues in PyMOL . The Dali server ( http://ekhidna2 . biocenter . helsinki . fi/dali/; Holm , 2019 ) was used to search for proteins with 3D structures like that of the RhopH complex . Exhaustive PDB database searches revealed significant matches to specific domains from individual RhopH subunits , as defined by Dali Z-scores ≥ 3 . 0 . PyMOL alignments of RhopH domains and PDB structures of corresponding hits were used to evaluate biological significance . Synchronization for stage-dependent membrane fractionation assays utilized two 5% sorbitol treatments ~6 hr apart . Ring-stage infected cells were harvested immediately without enrichment . Trophozoite- and schizont-stage-infected cells were then harvested 18 hr and 40 hr after sorbitol treatment , respectively , and enriched through the percoll–sorbitol method . Cells infected with 13F10 growth with or without TMP were harvested without enrichment as these cells lack PSAC activity ( Beck et al . , 2014; Ito et al . , 2017 ) . Freed merozoite studies were performed with 3D7 parasites using synchronous schizonts enriched using the percoll–sorbitol method . Purified schizonts were cultured with 25 µM E64D at 7 . 5 × 107 cells/mL and closely monitored for 4–5 hr for the development of segmenters containing fully formed merozoites . Cells were then washed , adjusted to 2 . 5 × 107 cells/mL in complete media , and allowed to recover at 37°C for 15 min . Freed merozoites ( 2 . 5 × 108 cells/mL ) were obtained by sequential passage through two 1 . 2 µm syringe filters to rupture the mature segmenters . A hemocytometer was used to confirm that merozoites were free of contaminating intact erythrocytes before pelleting ( 4500 × g , 5 min ) and freezing along with matched intact schizonts . Fractionation studies were performed using matched cell pellets resuspended in lysis buffer ( 7 . 5 mM Na2HPO4 , 1 mM EDTA , pH 7 . 5 ) at 3 . 5% hematocrit; this cell lysate corresponded to the ‘total’ input . Cellular debris and membranes were pelleted by ultracentrifugation at 100 , 000 × g for 1 hr at 4°C . The supernatant was kept as the ‘soluble’ fraction . Membranes were resuspended and incubated in 200 µL of 100 mM Na2CO3 , pH 11 at 4°C for 30 min before ultracentrifugation ( 100 , 000 × g , 1 hr , 4°C ) to separate peripheral from integral membrane proteins . Samples were neutralized with 1 M HCl and solubilized in a modified Laemmli buffer with a final 6% sodium dodecyl sulfate ( SDS ) concentration . Protease susceptibility experiments used percoll–sorbitol-enriched cells . Infected cells were treated with Pronase E in phosphate-buffered saline ( PBS ) supplemented with 0 . 6 mM CaCl2 and 1 mM MgCl2 for up to 1 hr at 37°C . They were then extensively washed in PBS with 1 mM PMSF prior to membrane fractionation . Samples were prepared in a modified Laemmli buffer with a final 6% SDS concentration . Proteins were separated on a 4–15% Mini-PROTEAN TGX gel ( Bio-RAD ) and transferred to nitrocellulose . After blocking , antibodies against CLAG3 ( Nguitragool et al . , 2011 ) , Band3 ( Santa Cruz ) , HA epitope tag ( Sigma–Aldrich ) , EXP2 ( European Malaria Reagent Repository ) , or aldolase ( Abcam ) were applied and visualized as described ( Ito et al . , 2017 ) . Band intensities were quantified using ImageJ and analyzed in Prism ( GraphPad ) . Statistical significance for numerical data was calculated by unpaired Student’s t-test or one-way ANOVA . Significance was accepted at p < 0 . 05 or indicated values .

Malaria is an infectious disease caused by the family of Plasmodium parasites , which pass between mosquitoes and animals to complete their life cycle . With one bite , mosquitoes can deposit up to one hundred malaria parasites into the human skin , from where they enter the bloodstream . After increasing their numbers in liver cells , the parasites hijack , invade and remodel red blood cells to create a safe space to grow and mature . This includes inserting holes in the membrane of red blood cells to take up nutrients from the bloodstream . A complex of three tightly bound RhopH proteins plays an important role in these processes . These proteins are unique to malaria parasites , and so far , it has been unclear how they collaborate to perform these specialist roles . Here , Schureck et al . have purified the RhopH complex from Plasmodium-infected human blood to determine its structure and reveal how it moves within an infected red blood cell . Using cryo-electron microscopy to visualise the assembly in fine detail , Schureck et al . showed that the three proteins bind tightly to each other over large areas using multiple anchor points . As the three proteins are produced , they assemble into a complex that remains dissolved and free of parasite membranes until the proteins have been delivered to their target red blood cells . Some hours after delivery , specific sections of the RhopH complex are inserted into the red blood cell membrane to produce pores that allow them to take up nutrients and to grow . The study of Schureck et al . provides important new insights into how the RhopH complex serves multiple roles during Plasmodium infection of human red blood cells . The findings provide a framework for the development of effective antimalarial treatments that target RhopH proteins to block red blood cell invasion and nutrient uptake .

2021

Malaria parasites use a soluble RhopH complex for erythrocyte invasion and an integral form for nutrient uptake

Complex interspecies relationships are widespread among metazoans , but the evolutionary history of these lifestyles is poorly understood . We describe a fossil beetle in 99-million-year-old Burmese amber that we infer to have been a social impostor of the earliest-known ant colonies . Promyrmister kistneri gen . et sp . nov . belongs to the haeteriine clown beetles ( Coleoptera: Histeridae ) , a major clade of ‘myrmecophiles’—specialized nest intruders with dramatic anatomical , chemical and behavioral adaptations for colony infiltration . Promyrmister reveals that myrmecophiles evolved close to the emergence of ant eusociality , in colonies of stem-group ants that predominate Burmese amber , or with cryptic crown-group ants that remain largely unknown at this time . The clown beetle-ant relationship has been maintained ever since by the beetles host-switching to numerous modern ant genera , ultimately diversifying into one of the largest radiations of symbiotic animals . We infer that obligate behavioral symbioses can evolve relatively rapidly , and be sustained over deep time . A pervasive feature of colony-forming insect societies is the profusion of intruder arthropods that have evolved to exploit their rich resources ( Kistner , 1979; Kistner , 1982; Hölldobler and Wilson , 1990; Parker , 2016 ) . The diversity of such organisms is impressive , with ~10 , 000 species hypothesized to target or profit from ant nests alone ( Elmes , 1996 ) . Hostility of ant workers to virtually all non-nestmate organisms has selected for defensive or host-deceptive adaptations in myrmecophiles which are often phenotypically remarkable , involving changes in anatomy , chemical ecology and behavior ( Kistner , 1979; Kistner , 1982; Hölldobler and Wilson , 1990; Parker , 2016 ) . In a number of cases , traits have arisen that enable the myrmecophile to manipulate worker behavior , circumventing aggression and enabling social interactions to evolve that assimilate the symbiont into colony life . Such relationships rank among the most behaviorally intimate interactions known between animal species ( Kistner , 1979; Hölldobler and Wilson , 1990; Parker , 2016 ) , and are typically achieved by the myrmecophile’s capacity to mimic the chemical and/or tactile cues involved in nestmate recognition ( Kistner , 1979; Hölldobler and Wilson , 1990; Parker , 2016 ) . The clown beetle family Histeridae includes multiple lineages that have independently evolved myrmecophily ( Parker , 2016; Kovarik and Caterino , 2005 ) , including Haeteriinae , a subfamily of ~335 described species comprising possibly the single largest radiation of myrmecophiles known within the Coleoptera ( Parker , 2016; Kovarik and Caterino , 2005; Helava et al . , 1985 ) . We report the discovery of a crown-group haeteriine in Upper Cretaceous Burmese amber , revealing that the clown beetle-ant interaction has an exceptionally deep evolutionary history . To our knowledge , the relationship constitutes the most ancient behavioral symbiosis known in the Metazoa . Symbiotic relationships in which different animal species interact socially with each other have arisen sporadically across the metazoan tree of life . Such relationships encompass a spectrum of dependency , from transient , facultative associations seen in mixed-species groups of insectivorous birds ( Sridhar et al . , 2009 ) , cetaceans and ungulates ( Stensland et al . , 2003; Goodale et al . , 2017 ) , to obligate symbiotic lifestyles typified by brood parasitic cuckoos , cowbirds ( Johnsgard , 1997 ) , mutualistic cleaner fish ( Grutter , 1999 ) and oxpeckers ( Nunn et al . , 2011 ) . Of all animal groups , however , the complex societies of ants play host to the greatest diversity of behavioral symbionts . Several major radiations of myrmecophiles are known , each containing hundreds of symbiotic species , including the lycaenid butterflies ( Pierce et al . , 2002 ) , eucharitid wasps ( Murray et al . , 2013 ) , paussine ground beetles ( Moore and Robertson , 2014 ) and multiple lineages of rove beetles ( Kistner , 1979; Kistner , 1982; Parker , 2016; Parker and Grimaldi , 2014; Seevers , 1965; Maruyama and Parker , 2017 ) . The diversity and often-broad geographic ranges of these clades imply that their relationships with ants are evolutionarily ancient ( Parker and Grimaldi , 2014; Yamamoto et al . , 2016 ) . Although fossil myrmecophiles are known from as far back as the Eocene ( Parker and Grimaldi , 2014; Wasmann , 1929 ) , ant eusociality is known to be at least twice as old , with the earliest definitively social ants occurring in Upper Cretaceous Burmese amber ( Barden and Grimaldi , 2016 ) . Whether their colonies were targeted by myrmecophiles has , however , been unclear: ants are comparatively scarce in Cretaceous ambers ( Grimaldi and Agosti , 2000; LaPolla et al . , 2013; Barden , 2016; Barden , 2018 ) , and myrmecophilous invertebrates typically live at densities orders of magnitude lower than their hosts ( Kistner , 1979 ) . The unlikely discovery of a myrmecophile clown beetle in Burmese amber reveals that a major radiation of ant symbionts has its origins in Mesozoic ant societies . Analysis of Promyrmister’s morphology and phylogenetic position indicates the new genus represents an extinct lineage within the crown-group of Haeteriinae , a clade of obligate myrmecophiles ( Figure 2A , B; Figure 2—figure supplements 2 and 3; see Materials and methods ) . In haeteriine taxa for which detailed behavioral observations exist , the beetles have been shown to engage in intimate behaviors with ants , involving stomodeal trophallaxis ( mouth-to-mouth feeding ) ( Wheeler , 1908; Henderson and Jeanne , 1990; Akre , 1968 ) , grooming workers with their appendages ( and being groomed or licked by hosts in return ) ( Akre , 1968 ) , physically grasping onto ants ( phoresis ) ( Akre , 1968; von Beeren and Tishechkin , 2017 ) , or being carried around nests by workers ( Kistner , 1982 ) . Mimicry of colony cuticular hydrocarbons occurs ( Lenoir et al . , 2012 ) , as well as chemical manipulation of host ants via ‘appeasement’ substances exuded from gland openings on the margins of the prothorax ( Kistner , 1982; Seyfried , 1928 ) or in the postcoxal regions of the beetle’s underside ( Figure 2—figure supplement 1C ) . Promyrmister appears to be closely allied to the extant genus Haeterius ( Figure 2A–C; Figure 2—figure supplement 1 ) . This genus and a handful of closely related taxa including Eretmotus , Sternocoelis and Satrapes comprise the only group of Haeteriinae known to occur in the Palaearctic , consistent with the Eurasian palaeolocality of Promyrmister in Burmese amber . Like all of these genera , Promyrmister exhibits classical haeteriine attributes that are thought to be true adaptations for myrmecophily , including broad expansions of the tibiae with spines on the outer margin ( Figure 2C; Figure 2—figure supplement 1E ) , short tarsi received on the outer face of each tibia ( Figure 2—figure supplement 1E ) , a triangular antennal scape ( Figure 2—figure supplement 1D ) , pronounced antennal cavities on the prothoracic hypomeron ( Figure 2—figure supplement 1D ) and a broad proventral lobe to fully embrace the retracted head ( Figure 2—figure supplement 1B–D ) . Those features are thought to be protective modifications that enable myrmecophile beetles to withstand handling by ant mandibles ( Parker , 2016 ) . We and others have previously described rove beetles ( Staphylinidae ) in the Burmese palaeofauna that were putative symbionts of termite colonies ( Yamamoto et al . , 2016; Cai et al . , 2017 ) . These specimens exhibit a defensive ecomorphology and are thought to have been persecuted intruders that were not behaviorally integrated into their host’s societies ( Yamamoto et al . , 2016; Cai et al . , 2017 ) . In contrast , Haeteriinae embody a form of true behavioral symbiosis , where the relationship with host ants can involve social interactions ( Figure 2D , E; Figure 4 ) . The prothoracic glandular openings of Haeteriinae that secrete putative appeasement compounds are challenging to demonstrate even in extant taxa , but in the Promyrmister holotype , a large globule of possible exudate originates from the left margin of the prothorax , consistent with the position of such glands ( Figure 1A , G ) . Additionally , Promyrmister possesses clear postcoxal secretory glands ( Figure 1H; Figure 1—figure supplement 2A , B , D ) , with a globule of possible exudate emanating from the postcoxal gland opening on the right side of the body ( Figure 1H; Figure 1—figure supplement 2B , D ) . Beyond Promyrmister’s phylogenetic position within the Haeteriinae clade , the fossil’s anatomy implies a chemical strategy to become accepted or at least tolerated inside colonies ( hypothetical reconstruction in Figure 2F ) , akin to modern haeteriine species that have so far been examined ( Akre , 1968; Lenoir et al . , 2012; Seyfried , 1928 ) . What were the Cretaceous host ants of Promyrmister ? All ants thus far described from Burmese amber belong to stem-group Formicidae , including members of the extinct subfamily Sphecomyrminae and three other genera , Gerontoformica , Myanmyrma and Camelomecia that similarly lack crown-group features but are placed incertae sedis within Formicidae ( Barden , 2016; Barden , 2018 ) . In contrast , fossils of definitive crown-group ant subfamilies are absent , or vanishingly rare , among the thousands of ant inclusions now recovered from this amber deposit ( Barden , 2016; McKellar et al . , 2013 ) ( P . Barden , personal communication ) . Crown-group ants are also unknown from contemporaneous Charentese amber ( Barden , 2016 ) . We posit that the overwhelming prevalence of stem-group ants in Burmese amber implies that they were potential hosts of Promyrmister ( Figure 2F ) . Such a scenario entails that haeteriines may not have originated with the modern ant groups that host them today; instead myrmecophily evolved first in stem-group ant colonies , with the beetles later switching to crown-group ants . We cannot , however , rule out an alternative scenario , that an as-yet undiscovered diversity of crown-group ants were , in fact , present in the Burmese palaeofauna , and it was these that selected for the early evolution of myrmecophily . Molecular dating indicates that crown-group ants had originated by this time ( Brady et al . , 2006; Moreau and Bell , 2013; Borowiec et al . , 2017a ) ( see dotted lines in Figure 3 ) . If present in this ancient ecosystem , perhaps their cryptic biologies limited their entrapment in amber . Whether haeteriines evolved in stem- or crown-group ant colonies , their original hosts are presumably long-extinct . The present-day host associations of haeteriines imply that these myrmecophiles have host-switched between many modern ant lineages ( Figure 3 ) . The beetles have been recorded in colonies of ant species scattered across the subfamilies Dolichoderinae , Dorylinae , Formicinae , Myrmicinae and Ponerinae ( Helava et al . , 1985; Tishechkin , 2007 ) ( Figure 3 ) . We suggest that it is this capacity for host switching that may explain the great longevity of the clown beetle-ant symbiosis . Through host switching , the clade as a whole has circumvented potential coextinction with host ant lineages that disappeared from the Cretaceous to the present ( Barden and Grimaldi , 2016; Barden , 2016 ) . Moreover , in some cases , the beetles have radiated dramatically with certain ant groups: the vast majority of the contemporary species richness of Haeteriinae is found in taxa that have adapted to colonies of Neotropical army ants ( Ecitonini ) , including at least 30 genera associated with Eciton army ants alone ( Parker , 2016; Helava et al . , 1985; von Beeren and Tishechkin , 2017; Tishechkin , 2007 ) . Some of these haeteriines have remarkable adaptations for life in colonies of those nomadic ants ( Figure 4 ) . Neotropical army ants are thought to have begun diversifying approximately in the Oligocene ( Brady et al . , 2014; Borowiec et al . , 2017b ) , implying that the bulk of haeteriine cladogenesis occurred within this window too , long after the beetles originated in the Cretaceous . An ancient association between histerids and ants is further suggested by inquiline-like morphology in two other Cretaceous clown beetle fossils ( Caterino et al . , 2015; Caterino and Maddison , 2018 ) , although unlike Promyrmister , the taxonomic affinities of these specimens are ambiguous and they are not definitive members of wholly symbiotic lineages . We infer that Haeteriinae was a relatively diverse clade by at least the beginning of the Upper Cretaceous , and likely originated and began undergoing basal cladogenesis very soon after the inferred Early Cretaceous emergence of ant eusociality ( Barden and Grimaldi , 2016; Grimaldi and Agosti , 2000; Barden , 2016; Brady et al . , 2006; Moreau and Bell , 2013; Borowiec et al . , 2017a ) . A time-calibrated molecular phylogeny of Haeteriinae may provide a more precise estimate of this temporal window . However , based on such analyses for other myrmecophile taxa , rapid evolution of specialized symbiotic phenotypes appears to be a common feature to clades of social insect symbionts ( Moore and Robertson , 2014; Parker and Grimaldi , 2014 ) , and presumably results from intense selection pressures inside colonies ( Kistner , 1979; Parker , 2016 ) . Promyrmister adds further support to the view that the earliest-known ants were socially complex ( Barden and Grimaldi , 2016 ) . Evidently , their colonies were also resource rich enough for exploitation by impostor myrmecophiles , which we conclude have been an unremitting part of ant biology . Despite their phenotypic intricacy and obligate dependency on other species , complex behavioral relationships between animals can be extraordinarily ancient , and persist over deep evolutionary time . This study is based on a single specimen of Burmese amber ( CNU-008021 ) collected from Noije Bum , Tanaing , Kachin , Myanmar . The specimen is housed at Key Laboratory of Insect Evolution and Environmental Changes , Capital Normal University , Beijing . The holotype of the new genus and species is embedded in a cuboid amber piece . The holotype was examined under a Leica M205C dissecting microscope and photographed using a Visionary Digital BK Lab Plus system ( Austin , Texas ) . The source images were aligned and stacked in Helicon Focus ( Ukraine ) . Fluorescence images of the fossil were made on a Zeiss LSM 880 ( with Airyscan ) confocal microscope ( Germany ) with a 488 nm laser . Scanning electron microscopic images of Haeterius were obtained using a Tabletop Hitachi Microscope TM3030Plus ( Japan ) . Morphological terminology follows Ślipiński and Mazur ( 1999 ) , Zhou et al . ( 2018 ) , and Caterino and Tishechkin ( 2015 ) . We scored Promyrmister for 259 external morphological characters used by Caterino and Tishechkin ( 2015 ) in a study investigating relationships among the tribe Exosternini , which is closely related to Haeteriinae . From the original matrix , we selected all taxa from the nearest sister clades of Haeteriinae , including 35 taxa belonging to Exosternini , including all species of Yarmister ( apparently the closest genus to Haeteriinae; Caterino and Tishechkin , 2015 ) . We also included representatives of four other tribes: Omalodini , Histerini , Hololeptini and Platysomatini , and assigned Hister unicolor as the primary outgroup , following Caterino and Tishechkin ( 2015 ) . The final taxon list is presented in Supplementary file 2A . We also enlarged our data matrix by adding one more taxon ( Haeterius ferrugineus ) two more characters ( 260 and 261 ) , and one more state for Character 14: 260: Epistoma: ( 1 ) without depressions receiving scapes in repose , occasionally with small depressions but without sharp arched-inwards epistomal striae; ( 2 ) with large depressions receiving scapes in repose , often defined by sharp arched-inwards epistomal striae . 261: Arched-inwards epistomal striae: ( 1 ) convergent , but separated from each other in the middle; ( 2 ) convergent , and meeting each other in the middle; ( 3 ) inapplicable . 14: Epistoma , surface: 6 ) deeply depressed , with lateral ridges ( =raised epistomal striae ) aligned with frontal stria . The complete matrix of 46 taxa , 261 characters was constructed in Mesquite v . 3 . 20 ( Maddison and Maddison , 2016 ) ; the matrix is provided in the nexus file ( Supplementary file 1 ) . Bayesian analysis was carried out using MrBayes 3 . 2 . 6 ( Ronquist et al . , 2012 ) accessed via the CIPRES Science Gateway Version 3 ( Miller et al . , 2010 ) ( phylo . org ) . The Mkv model of character evolution was used with a gamma distribution , and two MCMC were executed with four chains for 100 million generations . Convergence was judged to have occurred when the standard deviation of split frequencies dropped below 0 . 005 , and by ESS values higher than 200 in Tracer v1 . 7 . 0 ( Rambaut et al . , 2018 ) , indicating adequate estimation of the posterior . The first 25% of trees were discarded as burn-in . We used Treeannotator ( Bouckaert et al . , 2014 ) to obtain the maximum clade credibility tree from post burn-in trees ( ESS > 200 ) ( Figure 2A ) , and added the estimated nodal Bayesian posterior probability ( BPP ) in FigTree v1 . 4 . 3 ( https://github . com/rambaut/figtree/ ) . Parsimony analysis was conducted in TNT Version 1 . 5 ( Goloboff and Catalano , 2016 ) using Traditional Search without , and with implied weighting setting ( function K = 13 in Figure 2—figure supplements 2 and 3 ) . A consensus tree ( Figure 2B; L = 1604 , CI = 25 , RI = 42 ) was obtained from four shortest-length trees ( L = 1483 , CI = 28 , RI = 48 ) and the branch support was also calculated using 10 , 000 bootstrap replicates . Mapping character state changes onto the tree was performed in WinClada ( Nixon , 2002 ) . A list of haeteriine host ant genera was obtained from the literature ( Helava et al . , 1985; Yélamos , 1997; Tishechkin , 2007; Lapeva-Gjonova , 2013; Mazur , 1981 ) . To estimate ages of stem-group and Recent host ant taxa in Figure 3a , data for earliest-known fossils were obtained from Barden ( 2016 ) ( Barden , 2016; Barden , 2018 ) , and molecular age estimates of crown-groups were taken from recent taxon-specific phylogenetic studies ( Borowiec et al . , 2017a; Borowiec et al . , 2017b; Blaimer et al . , 2015; Ward et al . , 2015; Ferguson-Gow et al . , 2014; Ward et al . , 2010; Schmidt , 2013 ) . Data are presented in Supplementary file 2B . This published work and the nomenclatural acts it contains have been registered in ZooBank , the online registration system for the International Code of Zoological Nomenclature . The ZooBank LSIDs ( Life Science Identifiers ) can be resolved and the associated information viewed through any standard web browser by appending the LSID to the prefix ‘http://zoobank . org/’ . The LSIDs for this publication are to be found at: The specific LSIDs for new nomenclatural acts: All data generated or analyzed during this study are included in this published article ( and its Supplementary Information files ) . The holotype specimen of Promyrmister kistneri is housed at Key Laboratory of Insect Evolution and Environmental Changes , Capital Normal University , Beijing ( accession number CNU-008021 ) .

Many animals live lives that are closely intertwined with those of other species . While a clown fish sheltering within the tentacles of a sea anemone may be a textbook example , ‘symbiotic’ interactions that occur inside ant nests are among some of the most dramatic . Known as myrmecophiles – after the Greek for ‘ant lovers’ , many insects , spiders and mites have evolved to live alongside ants in one way or another . Some of these animals display elaborate behaviors – like mouth-to-mouth feeding or grooming of worker ants – which assimilates them into the nest society; some even release chemicals that mimic the ants’ own scents to avoid being detected as an intruder . The earliest examples of ancestral ants are found encapsulated in 99-million-year-old amber from a mine in northern Myanmar ( Burma ) . Zhou et al . have now discovered an ancient beetle , perfectly preserved in the same amber deposits , that may have also lived within the colonies of those earliest-known ants . Based on its appearance , the beetle – named Promyrmister kistneri – belongs within a subfamily of clown beetles ( called the Haeteriinae ) that are all specialized nest intruders with dramatic behavioral and chemical adaptations that help them to infiltrate ant colonies . The ancient clown beetle shares several of features with its modern relatives – including thick , spiked legs and well-protected head and antennae – which are believed to help the beetles withstand handling by the ants’ jaws . The specimen also has glands near the base of its legs , implying that it also released chemical signals that may have helped it to deceive or pacify the ancient ants . The fact that this extinct clown beetle is as old as the earliest-known ants implies that the close relationship between these insects has been sustained for an exceptionally long time . It is potentially the oldest known example of a symbiotic interaction in the animal kingdom that depends on social interactions between the two organisms . However , the host ants of Promyrmister are believed to be long-extinct , suggesting that symbiotic clown beetles had to switch to living inside colonies of modern ants to circumvent their own extinction . This flexibility to adapt to new partner species may be a critical feature that allows some symbiotic organisms to persist throughout evolution .

2019

A Mesozoic clown beetle myrmecophile (Coleoptera: Histeridae)

The joint activity of neural populations is high dimensional and complex . One strategy for reaching a tractable understanding of circuit function is to seek the simplest dynamical system that can account for the population activity . By imaging Aplysia’s pedal ganglion during fictive locomotion , here we show that its population-wide activity arises from a low-dimensional spiral attractor . Evoking locomotion moved the population into a low-dimensional , periodic , decaying orbit - a spiral - in which it behaved as a true attractor , converging to the same orbit when evoked , and returning to that orbit after transient perturbation . We found the same attractor in every preparation , and could predict motor output directly from its orbit , yet individual neurons’ participation changed across consecutive locomotion bouts . From these results , we propose that only the low-dimensional dynamics for movement control , and not the high-dimensional population activity , are consistent within and between nervous systems . The increasing availability of large scale recordings of brain networks at single neuron resolution provides an unprecedented opportunity to discover underlying principles of motor control . However , such long-sought data sets are revealing a new challenge - the joint activity of large neural populations is both complex and high dimensional ( Ahrens et al . , 2012; Cunningham and Yu , 2014; Yuste , 2015 ) . Population recordings have as many dimensions as neurons , and each neuron’s activity can have a complex form . What strategies can we use to expose the hoped-for simplifying principles operating beneath the turbulent surface of real-world brain activity ? One route is dimension reduction ( Briggman et al . , 2006; Cunningham and Yu , 2014; Kobak et al . , 2016 ) , which focuses on identifying the components of activity that co-vary across the members of a neural population , shifting the focus from the high dimensional recorded data to a low-dimensional representation of the population . Such low-dimensional signals within joint population activity have been described in neural circuits for sensory encoding ( Mazor and Laurent , 2005; Bartho et al . , 2009 ) , decision-making ( Briggman et al . , 2005; Harvey et al . , 2012; Mante et al . , 2013 ) , navigation ( Seelig and Jayaraman , 2015; Peyrache et al . , 2015 ) , and movement ( Levi et al . , 2005; Ahrens et al . , 2012; Kato et al . , 2015 ) . Implicit in such dimension reduction approaches is the hypothesis that the high-dimensional population activity being recorded , while highly heterogenous , is derived from a simpler , consistent low-dimensional system ( Brody et al . , 2003; Churchland et al . , 2010; Kato et al . , 2015; Miller , 2016 ) . We sought to directly test this hypothesis by identifying the simplest dynamical system that can account for high dimensional population activity . A useful model to address these questions is the neural control of movement . Movement arises from the mass action of neuron populations ( Georgopoulos et al . , 1986; Getting , 1989; Ahrens et al . , 2012; Portugues et al . , 2014; Yuste , 2015; Petersen and Berg , 2016 ) . While individual neuron activity can correlate with specific aspects of movement ( Chestek et al . , 2007; Hatsopoulos et al . , 2007; Churchland et al . , 2010 , 2012 ) , the embedded low dimensional signals in population recordings ( Briggman et al . , 2005; Levi et al . , 2005; Kato et al . , 2015 ) and the intermittent participation of individual neurons across repeated movements in both vertebrates ( Carmena et al . , 2005; Huber et al . , 2012 ) and invertebrates ( Hill et al . , 2010 , 2015 ) together suggest that only the collective population activity , and not specifics of single neuron firing , are key to movement control . If so , then finding the underlying dynamical system will be necessary for a parsimonious theory of the neural control of movement ( Briggman and Kristan , 2008 ) . In order to identify the simplest dynamical system underlying population activity in movement control , we imaged large populations at single-neuron , single-spike resolution in the pedal ganglion of Aplysia during fictive locomotion ( Figure 1A ) . The pedal ganglion presents an ideal target for testing hypotheses of movement control as it contains the pattern generator ( Jahan-Parwar and Fredman , 1979 , 1980 ) , motorneurons ( Hening et al . , 1979; Fredman and Jahan-Parwar , 1980 ) and modulatory neurons ( Hall and Lloyd , 1990; McPherson and Blankenship , 1992 ) underlying locomotion . Moreover , its fictive locomotion is sustained for minutes , ideal for robustly characterising population dynamics . Using this model system , here we find its low-dimensional , underlying dynamical system , test if the low-dimensional signal encodes movement variables , and determine the contribution of single neurons to the low-dimensional dynamics . 10 . 7554/eLife . 27342 . 003Figure 1 . Population dynamics during fictive locomotion . ( A ) Voltage-sensitive dye recording of the pedal ganglion ( Pd ) network in an isolated central nervous system preparation ( top ) using a photodiode array ( blue hexagon ) . The array covered the dorsal surface of the ganglion ( bottom ) . Ce: cerebral ganglion; Pl: pleural ganglion; Pd9/10: pedal nerve 9/10 . ( B ) Stimulus protocol . Three escape locomotion bouts were evoked in each preparation by stimulation of tail nerve Pd9 . Parameters are given for the stimulus pulse train . ( C ) Example population recording . Raster plot of 160 neurons before and after Pd9 nerve stimulation . Neurons are grouped into ensembles of similarly-patterned firing , and ordered by ensemble type ( colors ) - see Materials and methods . ( D ) Power spectra of each population’s spike-trains , post-stimulation ( grey: mean spectrum of each bout; black: mean over all bouts ) . ( E ) Network firing rate over time ( grey: every bout; black: mean; red bar: stimulation duration . Bins: 1 s ) . ( F ) Terminology and schematic illustration of the necessary conditions for identifying a periodic attractor ( or ‘cyclical’ attractor ) . Left: to characterise the dynamics of a N-dimensional system , we use the joint activity of its N units at each time-point t – illustrated here for N=2 units . The set of joint activity points in time order defines the system’s trajectory ( black line ) . Right: the three conditions for identifying a periodic attractor . In each panel , the line indicates the trajectory of the joint activity of all units in the dynamical system , starting from the solid dot . The manifold of a dynamical system is the space containing all possible trajectories of the unperturbed system – for periodic systems , we consider the manifold to contain all periodic parts of the trajectories ( grey shading ) . In ( condition 3 ) , the dashed line indicates where the normal trajectory of the system would have been if not for the perturbation ( red line ) . See Figure 1—figure supplement 1 for a dynamical model illustrating these conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 27342 . 00310 . 7554/eLife . 27342 . 004Figure 1—figure supplement 1 . Necessary and sufficient conditions of a periodic attractor . Here we demonstrate the essential properties of a periodic attractor network . ( A ) Top: a simple neuronal implementation of a periodic attractor network using three mutually inhibitory neurons . Bottom: model neurons are not endogenous oscillators - if uncoupled , they do not oscillate in response to constant input . The oscillatory activity we show here is an emergent property of the model network . ( B ) The underlying periodic attractor is revealed by the recurrence of the neurons’ joint activity . Top: given sustained input , the three neurons rapidly settle into an alternating oscillatory rhythm . Bottom: projection of the three neurons’ firing rates onto the first two principal components ( PCs ) . Starting from time zero ( black dot ) , the joint activity of the network rapidly converges to a recurrent trajectory , where each corner of the triangle - extending into the short diagonal line - corresponds to the maximum rate of one neuron . The space containing the repeated trajectory is the attractor’s manifold . ( C ) Convergence to the manifold from different initial conditions . From three different initial conditions for the neurons ( coloured dots ) , each corresponding joint activity converges to the attractor manifold . ( D ) Transient perturbations cause a divergence and return to the attractor’s manifold . Top: A brief increase in input to one neuron ( black bar ) perturbs the ongoing oscillation , which is rapidly recovered . Bottom: in the PC projection , the initial recurrent trajectory ( in black ) is moved away from the manifold by the perturbation ( red ) ; removing the perturbing input rapidly moves the joint activity back to the manifold ( blue ) . ( E ) Sustained perturbations move the network to a new attractor . Top: A sustained change in input ( black bar ) causes one neuron ( purple ) to become silent and the others to change their oscillation pattern . Bottom: in the PC projection , the perturbation causes the joint activity of the network to move from the initial stable manifold ( black ) to a new stable manifold ( blue: note this is a repeating line ) . ( F ) Decaying input causes a spiral . Top: linearly decaying input to all neurons changes the amplitude , but not the frequency of the oscillatory activity . Bottom: in the PC projection , this decay is shown by the spiral of the joint activity’s trajectory to the stable point at ( 0 , 0 ) – which corresponds to no activity . DOI: http://dx . doi . org/10 . 7554/eLife . 27342 . 004 We show that evoking fictive locomotion caused heterogenous population spiking activity , but under which always lay a low-dimensional , slowly decaying periodic orbit . This periodic trajectory met the convergence and perturbation criteria for an attractor . Crucially , we identify the attractor as a stable , decaying spiral in every preparation . We decoded motorneuron activity directly from the low-dimensional orbit , showing that it directly encodes the relevant variables for movement . Yet we found that individual neurons varied their participation in the attractor between bouts of locomotion . Consequently , only the low-dimensional signal and not the high-dimensional population activity was consistent within and between nervous systems . These findings strongly constrain the possible implementations of the pattern generator for crawling in Aplysia; and by quantifying the attractor they make possible future testing of how short- and long-term learning change properties of that attractor . Collectively , these results provide experimental support for the long-standing idea that neural population activity is a high-dimensional emergent property of a simpler , low-dimensional dynamical system . We first established that under the heterogenous population activity evoked by the tail-nerve stimulation there was a low dimensional periodic trajectory , consistent with there being a periodic attractor in the pedal ganglion . Projections of a population’s joint activity into three dimensions typically showed that stimulation caused a strong deviation from the spontaneous state , which then settled into repeated loops ( Figure 2A ) . Capturing a significant proportion ( 80% ) of the population variance generally required 4–8 embedding dimensions ( Figure 2B ) , representing a dimension reduction by more than a factor of 10 compared to the number of neurons . Thus , throughout our analysis , we projected each evoked program into the number of embedding dimensions needed to capture at least 80% of the variance in population activity ( 4–8 dimensions: inset of Figure 2B ) . However , we cannot directly visualise this space; therefore we could not tell by visual inspection if the low-dimensional trajectory repeatedly returned to the same position , and so was truly periodic . 10 . 7554/eLife . 27342 . 005Figure 2 . Population dynamics form a low-dimensional periodic orbit . ( A ) Projection of one evoked population response into three embedding dimensions , given by its first three principal components ( PCs ) . Dots: start of recording ( black ) and stimulation ( pink ) ; spontaneous activity is shown in grey . Smoothed with 2 s boxcar window . ( B ) Proportion of population variance explained by each additional embedding dimension , for every evoked population response ( n=30; light-to-dark grey scale indicates stimulations 1 to 3 of a preparation ) . We chose a threshold of 80% variance ( red line ) to approximately capture the main dimensions: beyond this , small gains in explained variance required exponentially-increasing numbers of dimensions . Inset: Histogram of the number of PCs needed to explain 80% variance in every recorded population response . ( C ) Quantifying population dynamics using recurrence . Population activity at some time t is a point in N-dimensional space ( black circle ) , following some trajectory ( line and open circles ) ; that point recurs if activity at a later time t+n passes within some small threshold distance θ . The time n is the recurrence time of point t . ( D ) Recurrence plot of the population response in panel A . White squares are recurrence times , where the low-dimensional dynamics at two different times passed within distance θ . We defined θ as a percentile of all distances between points; here we use 10% . Stimulation caused the population’s activity to recur with a regular period . Projection used 4 PCs . ( E ) Histograms of all recurrence times in each population response ( threshold: 10% ) , ordered top-to-bottom by height of normalised peak value . Vertical line indicates the minimum time we used for defining the largest peak as the dominant period for that population response . Right: density of time-points that were recurrent , and density of recurrence points with times in the dominant period . ( F ) Periodic orbit of each evoked population response , estimated as the mean recurrence time from the dominant period . DOI: http://dx . doi . org/10 . 7554/eLife . 27342 . 00510 . 7554/eLife . 27342 . 006Figure 2—figure supplement 1 . Range of dynamics in the evoked locomotion programs . Here we quantify the complexity of the dynamics in each evoked program using the structure of the recurrence plots ( five example plots are shown ) . Each white dot is a pair of time-points at which the low-dimensional trajectory recurred . A diagonal line indicates a contiguous period of recurrence of the population dynamics . We thus quantified the complexity of each program by the duration of the longest diagonal line ( measuring the predictability of the system ) and by the proportion of recurrent points falling in long ( >5 s ) diagonal lines ( measuring the cleanness of the system’s dynamics ) . We plot the joint distribution of these two quantities here , one point per program; colours indicate programs from the same preparation . These show how the range of dynamics captured by our dataset stretches from quasi-noise ( bottom left ) , though stuttering/pausing oscillations ( top left ) , to continuous oscillation ( right ) . Despite this heterogeneity , almost all evoked programs revealed the same underlying dynamical system ( Figures 2–4 , main text ) . DOI: http://dx . doi . org/10 . 7554/eLife . 27342 . 00610 . 7554/eLife . 27342 . 007Figure 2—figure supplement 2 . Robustness of the periodic orbits . Here we show that the periodic orbits , defined by recurrence , are robust to the choice of threshold for defining ‘close’ . We defined the threshold as a percentile of all distances between all points in the post-stimulation trajectory . We show here the effects of choosing a threshold of 10% ( A ) , 5% ( B ) , and 2% ( C ) . Left column: histograms of all recurrence times in each program , ordered top-to-bottom by height of normalised peak value . Vertical line indicates the minimum time we used for defining the largest peak as the dominant period for that program . Middle column: density of time-points that were recurrent , and density of recurrence points with times in the dominant period . Right column: Periodic orbit of each program , estimated as the mean recurrence time from the dominant period . We use a threshold of 10% in the main text as it allows a denser sampling of potential trajectory points that are noisy realisations of the underlying attractor . DOI: http://dx . doi . org/10 . 7554/eLife . 27342 . 007 To determine whether population activity in higher dimensions reached a stable periodic orbit , we made use of the idea of recurrence ( Lathrop and Kostelich , 1989; Marwan et al . , 2007 ) . For each time-point in the low-dimensional trajectory of the population’s activity , we check if the trajectory passes close to the same point in the future ( Figure 2C ) . If so , then the current time-point recurs , indicating that the joint activity of the population revisits the same state at least once . The time between the current time-point and when it recurs gives us the period of recurrence . A strongly periodic system would thus be revealed by its population’s trajectory having many recurrent points with similar recurrence periods; random or chaotic dynamics , by contrast , would not show a single clustered recurrence period . Plotting recurrent time-points showed that the evoked low-dimensional population activity typically recurred with a regular period ( example in Figure 2D ) . We found strongly periodic recurrence on the scale of 10–15 s in many but not all of the 30 evoked population responses ( Figure 2E , F ) . This reflected the range of stimulation responses from strongly periodic activity across the population to noisy , stuttering , irregular activity ( Figure 2—figure supplement 1 ) . Nonetheless , despite this heterogeneity across stimulus responses , the activity of almost all populations was dominated by a single periodic orbit ( Figure 2E ) , robust to the choice of threshold for defining recurrence ( Figure 2—figure supplement 2 ) . The trajectory of a periodic dynamical system remains within a circumscribed region of space – the manifold – that is defined by all the possible states of that system . ( We schematically illustrate a manifold by the grey shading in Figure 1F ( condition 2 ) , and demonstrate the manifold of our model periodic attractor network in panel C of Figure 1—figure supplement 1 ) . If the population responses of the pedal ganglion are from an underlying periodic attractor , then the population’s joint activity should rapidly reach and stay on its manifold when evoked; reach the same manifold every time it is evoked; and return to the manifold when perturbed ( these three conditions are schematically illustrated in Figure 1F; see Figure 1—figure supplement 1 for the corresponding examples from the dynamical model ) . We found that almost all evoked population responses quickly reached a state of high recurrence , within one oscillation period ( Figure 3A ) , and were thereafter dominated by recurrence , indicating they quickly reached and stayed on the manifold . 10 . 7554/eLife . 27342 . 008Figure 3 . Low dimensional population dynamics meet the conditions for a periodic attractor . ( A ) Distribution of the time the population dynamics took to coalesce onto the attractor from the stimulation onset , and the subsequent stability of the attractor ( measured by the proportion of recurrent points ) . Colours indicate evoked responses from the same preparation . The coalescence time is the mid-point of the first 5 s sliding window in which at least 90% of the points on the population trajectory recurred in the future . ( B ) Projection of three sequential population responses from one preparation onto the same embedding dimensions . Dots are time of stimulus offset . ( C ) Sequential population responses fall onto the same manifold . Dots indicate distances between pairs of population responses in the same preparation; color indicates preparation . Control distances are from random projections of each population response onto the same embedding dimensions - using the same time-series , but shuffling the assignment of time series to neurons . This shows how much of the manifold agreement is due to the choice of embedding dimensions alone . The two pairs below the diagonal are for response pairs ( 1 , 2 ) and ( 1 , 3 ) in preparation 4; this correctly identified the unique presence of apparent chaos in response 1 ( see Figure 3—figure supplement 1 ) . ( D ) Distances between pairs of population responses from the same preparation in three states: the end of spontaneous activity ( at stimulus onset ) ; between stimulation onset and coalescence ( the maximum distance between the pair ) ; and after both had coalesced ( both reaching the putative attractor manifold; data from panel C ) . ( E ) Example neuron activity similarity matrices for consecutively evoked population responses . Neurons are ordered according to their total similarity in stimulation 2 . ( F ) Correlation between pairs of neuron similarity matrices ( Data ) compared to the expected correlation between pairs of matrices with the same total similarity per neuron ( Control ) . Values below the diagonal indicate conserved pairwise correlations between pairs of population responses within the same preparation . The two pairs on the diagonal are response pairs ( 1 , 3 ) and ( 2 , 3 ) in preparation 7; this correctly identified the unique presence of a random walk in response 3 ( see Figure 3—figure supplement 1 ) . ( G ) Spontaneous divergence from the trajectory . For one population response , here we plot the density of recurrence points ( top ) and the mean recurrence delay in 5 s sliding windows . Coalescence time: grey line . The sustained ‘divergent’ period of low recurrence ( grey shading ) shows the population spontaneously diverged from its ongoing trajectory , before returning . Black dot: pre-divergence window ( panel I ) . ( H ) Breakdown of spontaneous perturbations across all population responses . Returned: population activity became stably recurrent after the perturbation . ( I ) Returning to the same manifold . For each of the 17 ‘Returned’ perturbations in panel H , the proportion of the recurrent points in the pre-divergence window that recurred after the divergent period , indicating a return to the same manifold or to a different manifold . DOI: http://dx . doi . org/10 . 7554/eLife . 27342 . 00810 . 7554/eLife . 27342 . 009Figure 3—figure supplement 1 . Convergence and non-convergence to the same manifold . ( A ) Robustness of convergence to the same manifold . Here we repeat the analysis of Figure 3A , but using different low-dimensional projections . In the main text Figures 2–4 , we analyse low-dimensional projections of each individual program; the comparison of distances between sequentially evoked programs was then done by projecting all three programs onto the axes of the first program . Here we derive a common set of axes , by first applying PCA to the concatenated time-series of the three programs . We then project each program onto those common axes , before computing distances between data and control ( random projection ) pairs of programs ( dots ) . We obtain the same results: sequentially-evoked programs converge to the same manifold . ( B ) Apparent chaotic motion in the non-convergent program . Here we show the recurrence plot for program 1 of preparation 4 , the only program that was further from the other programs in the same preparation than random projections . We see that this program seemed to reach an apparently chaotic attractor about 45 s after stimulation: quasi-periodic activity , with no regular structure . ( C ) Apparent random-walk in the non-correlated program . Here we show the recurrence plot for program 3 of preparation 7 , the only program that did not recapitulate the pair-wise correlations of neural activity of the other programs in the same preparation . We see that this program had no clear periodic activity , but seemingly random periods of locally-similar activity , consistent with a dynamical system undergoing a random walk . DOI: http://dx . doi . org/10 . 7554/eLife . 27342 . 00910 . 7554/eLife . 27342 . 010Figure 3—figure supplement 2 . Spontaneous perturbations of ongoing programs . We plot here examples of each type of spontaneous perturbation we observed , visualised using the recurrence plot ( top ) , and the raster plot of population activity ( bottom ) sorted into detected ensembles . ( A ) Spontaneous perturbation away from the periodic attractor . Blue line indicates the final 5 s window with more than 50% recurrence , long before the end of the recording . ( B ) Spontaneous perturbation returning to the same manifold of the periodic attractor . The orange lines indicates the point of maximum divergence from the manifold . The recurrence plot shows that same trajectory was recovered thereafter ( white patches in upper right corner ) , visible as the return of the phasically firing ensembles ( raster plot ) . ( C ) Spontaneous perturbation moving to a new periodic attractor . The recurrent trajectory up to 30 s post-stimulus did not recur thereafter ( black region in upper right of the recurrence plot ) , but was replaced with a new recurrent trajectory . The raster plot shows this was an apparent transition in the oscillation period , and/or firing pattern of the ensembles . DOI: http://dx . doi . org/10 . 7554/eLife . 27342 . 010 But does each population response from the same preparation reach the same manifold ? The key problem in analysing any putative attractor from experimental data is identifying when the experimentally-measured dynamics are or are not on the attractor’s manifold , whether due to perturbations of the system or noise in the measurements . Moreover , we cannot directly compare time-series between evoked responses because , as just demonstrated , each response may reach the manifold at different times ( see also panel C in Figure 1—figure supplement 1 ) . Thus the set of recurrent time-points allowed us to identify when the joint population activity was most likely on the attractor’s manifold , and then to make comparisons between population responses . To determine if sequentially-evoked responses from the same preparation reached the same manifold , we projected all 3 population responses into the same set of embedding dimensions , using only the recurrent points ( Figure 3B; Figure 3—figure supplement 1 shows these results are robust to other projections ) . Falling on the same manifold would mean that every recurrent point in one population response’s trajectory would also appear in both the others’ trajectories , if noiseless . Consequently , the maximum distance between any randomly chosen recurrent point in population response A and the closest recurrent point in population response B should be small . We defined small here as being shorter than the expected distance between a recurrent point in A and the closest point on a random projection of the activity in the same embedding dimensions . Despite the inherent noise and limited duration of the recordings , this is exactly what we found: pairs of evoked population responses from the same preparation fell close to each other throughout ( Figure 3C ) , well in excess of the expected agreement between random projections of the data onto the same embedding dimensions . We also checked that this convergence to the same manifold came from different initial conditions . The initiating stimulation is a rough kick to the system – indeed a fictive locomotion bout can be initiated with a variety of stimulation parameters ( Bruno et al . , 2015 ) – applied to ongoing spontaneous activity . Together , the stimulation and the state of spontaneous activity when it is applied should give different initial conditions from which the attractor manifold is reached . We found that the stimulation caused population responses within the same preparation to diverge far more than in either the spontaneous activity or after coalescing to the manifold ( Figure 3D ) . Thus , a wide range of initial driven dynamics in the pedal ganglion population converged onto the same manifold . Previous studies have used the consistency of pairwise correlations between neurons across conditions as indirect evidence for the convergence of population activity to an underlying attractor ( Yoon et al . , 2013; Peyrache et al . , 2015 ) . The intuition here is that neurons whose activity contributes to the same portion of the manifold will have simultaneous spiking , and so their activity will correlate across repeated visits of the population’s activity to the same part of the manifold . To check this , we computed the pairwise similarity between all neurons within an evoked population response ( Figure 3E ) , then correlated these similarity matrices between responses from the same preparation . We found that pair-wise similarity is indeed well-preserved across population responses in the same preparation ( Figure 3F ) . This also shows that the apparent convergence to the same manifold is not an artefact of our choice of low-dimensional projection . In many population responses , we noticed spontaneous perturbations of the low-dimensional dynamics away from the trajectory ( examples in Figure 3—figure supplement 2 ) , indicated by sudden falls in the density of recurrent points ( Figure 3G ) . That is , perturbations could be detected by runs of contiguous points on the population trajectory that were not recurrent . As each spontaneous perturbation was a cessation of recurrence in a trajectory accounting for 80% of the co-variation between neurons , each was a population-wide alteration of neuron activity ( see example rasters in Figure 3—figure supplement 2 ) . In most cases ( 90% ) , the population dynamics returned to a recurrent state after the spontaneous perturbation ( Figure 3H; Figure 3—figure supplement 2 , panel B ) , consistent with the pertubation being caused by a transient effect on the population The two perturbations that did not return to a recurrent state were consistent with the end of the evoked fictive locomotion and a return to spontaneous activity ( Figure 3—figure supplement 2 , panel A ) . Of those that returned , all but three clearly returned to the same manifold ( Figure 3I ) ; for those three , the spontaneous perturbation appeared sufficient to move the population dynamics into a different periodic attractor ( Figure 3—figure supplement 2 , panel C ) . Potentially , these are the known transitions from the escape gallop to normal crawling ( Flinn et al . , 2001 ) . The low dimensional dynamics of the pedal ganglion thus meet the stability , manifold convergence , and perturbation criteria of a periodic attractor network . While these results show the existence of a periodic orbit on an attractor in the evoked population responses , they cannot address whether these arise from the same putative attractor within and , crucially , between animals . To determine if there is a common underlying attractor despite the heterogeneity in spiking patterns across the population responses ( Figure 2—figure supplement 1 ) , we introduced a statistical approach to quantifying the low-dimensional trajectory . We first fitted a linear model of the local dynamics around each time point in the low-dimensional projection ( see Materials and methods ) . For each N-dimensional point P⁢ ( t ) in this projection , we fitted the N-dimensional model P*˙=𝐀⁢P* to the trajectory forwards and backwards in time from point P⁢ ( t ) . In this model , the change in the trajectory over time P*˙ in the neighbourhood of point P⁢ ( t ) is determined by the values of the N×N matrix 𝐀 . The maximum eigenvalue of A thus tells us whether the trajectory around point P⁢ ( t ) is predominantly expanding or contracting in the N-dimensional projection , and whether or not it is rotating ( Strogatz , 1994 ) . By fitting the linear model to each point on the trajectory we obtained time-series of the maximum eigenvalues , describing the local dynamics at each point along the trajectory . The time-series of eigenvalues typically showed long periods of similar magnitude eigenvalues , corresponding to the recurrent points ( Figure 4A ) . Consequently , by then averaging over the eigenvalues obtained only for recurrent points , we could potentially capture the dynamics of the underlying attractor . Doing so , we found that the evoked population responses had highly clustered maximum eigenvalues ( Figure 4B , C ) , and thus highly similar underlying dynamics despite the apparent heterogeneity of spike-train patterns between them . The dominance of negative complex eigenvalues implies the pedal ganglion network implements a contracting periodic orbit - it is a stable spiral attractor ( Figure 4D ) . 10 . 7554/eLife . 27342 . 011Figure 4 . The pedal ganglion contains a spiral attractor . ( A ) Example time-series from one population response of the real ( top ) and imaginary ( bottom ) component of the maximum eigenvalue for the local linear model . Points are averages over a 5 s sliding window . Red bar indicates stimulus duration . ( B ) Dominant dynamics for each evoked population response . Dots and lines give means ±2 s . e . m . of the real and imaginary components of the maximum eigenvalues for the local linear model . Colours indicate responses from the same preparation . Black dot gives the mean over all population responses . Grey shaded regions approximately divide the plane of eigenvalue components into regions of qualitatively different dynamics: fixed point attractor; stable spiral ( bottom-right schematic ) ; unstable spiral ( top-right schematic ) . ( C ) As panel B , converted to estimates of orbital period and rate of contraction . ( Note that higher imaginary eigenvalues equate to faster orbital periods , so the ordering of population responses is flipped on the x-axis compared to panel B ) . ( D ) A preparation with a visible spiral attractor in a three-dimensional projection . Each line is one of the three evoked population responses , colour-coded by time-elapsed since stimulation ( grey circle ) . The periodicity of the evoked response is the number of loops in the elapsed time; loop magnitude corresponds to the magnitude of population activity . The approximate dominant axis of the spiral’s contraction is indicated . Bottom: corresponding raster plot of one evoked response . Neurons are clustered into ensembles , and colour-coded by the change in ensemble firing rate to show the dominance of decreasing rates corresponding to the contracting loop in the projection . ( E ) As panel D , but for a preparation with simultaneously visible dominant spiral and minor expansion of the low-dimensional trajectory . The expansion corresponds to the small population of neurons with increasing rates . DOI: http://dx . doi . org/10 . 7554/eLife . 27342 . 01110 . 7554/eLife . 27342 . 012Figure 4—figure supplement 1 . Further properties of the spiral attractor . ( A ) Other eigenvalues . We plot here the mean real eigenvalue for all unique dimensions . ( We plot each dimension that has a unique real eigenvalue , indicating a possible change in whether the magnitude of the trajectory is contracting or expanding: complex eigenvalues come in conjugate pairs of dimensions , so each pair only defines one unique dimension ) . Dimensions are arranged in decreasing order of explained variance . Almost all population responses contained solely contracting trajectories in all dimensions , corresponding to a dominant decrease in neuron firing rates . A few population responses contained low-variance dimensions in which the trajectory was expanding , corresponding to a fraction of neurons with increasing firing rates ( as illustrated in Figure 4E ) . Lines are colour-coded by preparation; one line per population response . ( B ) Proportion of local trajectories that were dominated by rotation in each preparation . This is measured as the proportion of either the top two eigenvalues of the local linear models that were complex: by looking at the top two dimensions , we accounted for some evoked responses in which the dominance of contraction-only ( real eigenvalue ) and contraction-and-rotation ( complex eigenvalue ) dynamics alternated betwen the top two dimensions over time . Population responses were dominated by rotation throughout . DOI: http://dx . doi . org/10 . 7554/eLife . 27342 . 012 In most population responses , the low-dimensional trajectory had negative , complex eigenvalues in all embedding dimensions , meaning that the spiral attractor completely characterised the population dynamics ( Figure 4—figure supplement 1 ) . Intriguingly , a few population responses had a positive real eigenvalue in one low-variance dimension ( Figure 4—figure supplement 1 ) , implying a simultaneous minor expansion of the population trajectory . This corresponded to the appearance of a small sub-set of neurons with increasing firing rates ( Figure 4E ) . The identification of a stable spiral makes a clear prediction for what should and should not change over time in the dynamics of the population . The negative complex eigenvalues mean that the magnitude of the orbit decays over time , corresponding to the decreasing population spike rate in most evoked responses ( Figure 1E ) . However , a stable spiral indicates only a decrease in magnitude; it does not mean the orbital period is also slowing . Consequently , the presence of a stable spiral attractor predicts that the magnitude and period of the orbit are dissociable properties in the pedal ganglion network . We checked this prediction using the linear model . The linear model estimated a mean orbital period of around 10 s ( Figure 4C ) , consistent with the directly-derived estimate from the recurrent points ( Figure 2F ) . This indicated the linear model was correctly capturing the local dynamics of each program . But our linear model also gave us a time-series of estimates of the local orbital period ( Figure 5A ) , which we could use to check whether the orbital period was changing during each evoked response . We found that the population responses included all possible changes in periodic orbit: slowing , speeding up , and not changing ( Figure 5B ) . As predicted there was no relationship between the contraction of the periodic orbit and its change in period ( Figure 5C ) . 10 . 7554/eLife . 27342 . 013Figure 5 . The spiral attractor dissociates changes in oscillation period and firing rate . ( A ) Example of a change in the local estimate of the periodic orbit over a population response; here , slowing over time ( n=57 points are each averages over a 5 s sliding window; ρ is weighted Spearman’s rank correlation - see Materials and methods; P from a permutation test ) . Changes in the periodic orbit were assessed only after coalescence to the manifold ( grey line ) . ( B ) Histogram of correlations between time elapsed and local estimate of the periodic orbit for each population response ( positive: slowing; negative: speeding up ) . Red bars correspond to population responses with P<0 . 01 ( permutation test ) . Number of local estimates ranged between 31 and 72 per population response . ( C ) Relationship between the change in periodic orbit over time and the rate of contraction for each population response ( Pearson’s R; n=30 responses ) . DOI: http://dx . doi . org/10 . 7554/eLife . 27342 . 013 Collectively , these periodic , decaying dynamics are ethologically consistent with locomotion that comprises a repeated sequence of movements that decays in intensity over time ( Jahan-Parwar and Fredman , 1979; Flinn et al . , 2001; Marinesco et al . , 2004 ) . If this putative low-dimensional periodic attractor is the ‘motor program’ for locomotion , then we should be able to decode the locomotion muscle commands from its trajectory . In 3 of the 10 preparations we were able to simultaneously record activity from the P10 nerve that projects to the neck muscles ( Xin et al . , 1996 ) for all three evoked population responses . The spiking of axons in this nerve should correspond to the specific neck contraction portion of the cyclical escape locomotion . We thus sought to decode the spiking of P10 directly from the low-dimensional population trajectory ( Figure 6A ) . 10 . 7554/eLife . 27342 . 014Figure 6 . Motor output can be decoded directly from the low-dimensional trajectory of population activity . ( A ) An example two-dimensional projection of one population’s response trajectory , color-coded by simultaneous P10 firing rate . In this example pair of dimensions , we can see nerve P10 firing is phase-aligned to the periodic trajectory of population activity . ( B ) Example fit and forecast by the statistical decoding model for P10 firing rate . Grey bar indicates stimulation time . ( C ) For the same example P10 data , the quality of the forecast in the 10 s after each fitted 40 s sliding window . Match between the model forecast and P10 data was quantified by the fits to both the change ( R: correlation coefficient ) and the scale ( MAE: median absolute error ) of activity over the forecast window . ( D ) Summary of model forecasts for all nine population responses with P10 activity ( main panel ) . Dots and lines show means ±2 s . e . m . over all forecast windows ( N=173 ) . Three examples from the extremes of the forecast quality are shown , each using the fitted model to the first 40 s window to forecast the entire remaining P10 time-series . The bottom right example is from a recording in which the population response apparently shutdown half-way through . Inset , lower left: summary of model fits in the training windows; conventions as per main panel . DOI: http://dx . doi . org/10 . 7554/eLife . 27342 . 01410 . 7554/eLife . 27342 . 015Figure 6—figure supplement 1 . Ruling out P10 motorneurons in the recorded population . We aimed to decode the spiking activity of the motorneuron axons in the P10 nerve directly from the low-dimensional trajectory of population activity in the pedal ganglion . If any P10 motorneurons had been captured in our population recording , these could have made a disproportionate contribution to the decoding . We thus first sought to detect them , and remove them if necessary . We reasoned that any putative P10 motorneuron should have close to a 1:1 locking between its spikes in the population activity recordings and corresponding spikes in the P10 nerve recording ( panel A shows two example P10 unit recordings ) ; in other words P ( spike10|spike ) ≈1 . Moreover , we expect a putative P10 motorneuron to show ∼1:1 locking consistently across all three programs from the same preparation . For each neuron in each recorded program with simultaneous P10 activity , we computed P ( spike10|spike ) separately for a range of delays ( 0 to 25 ms ) between neuron and P10 spikes , allowing for ±2ms jitter at each delay . We plot the maximum probability for each neuron in programs 1 and 2 ( B ) , and programs 1 and 3 ( C ) ; symbol size is proportional to the mean number of spikes for that neuron in the two programs ( as few spikes would indicate an unlikely motorneuron ) . Colours indicate neurons from the three different preparations . We see that no neuron in our dataset comes close to P ( spike10|spike ) ≈1 at any delay , and there are no neurons with consistently high locking between programs . We concluded that there are no P10 motorneurons in our population recordings . DOI: http://dx . doi . org/10 . 7554/eLife . 27342 . 01510 . 7554/eLife . 27342 . 016Figure 6—figure supplement 2 . Increasing the dimensionality of state-space did not improve the P10 decoding model . Here we show the effects of increasing the embedding dimensions for the population activity before using the decoding model for P10 activity . We used enough dimensions to capture at least 90% of the variance; we used 80% in the main text . ( A ) Comparison of the number of embedding dimensions needed to explain 80% and 90% variance for the nine recordings with simultaneous P10 activity . A 10% increase in variance needed at least double the number of dimensions . ( B ) Example fit and forecast by the statistical decoding model for P10 firing rate , when using enough embedding dimensions to capture 90% variance . Grey bar indicates stimulation time . ( C ) Summary of model forecasts for all nine population responses with P10 activity , when using enough embedding dimensions to capture 90% variance . Dots and lines show means ±2 s . e . m . over all forecast windows ( N=173 ) . MAE: median absolute error ( error in scale ) of forecast; R2: variance explained of forecast . ( D ) Comparison of the mean MAE in forecasting P10 activity when using embedding dimensions explaining 80% and 90% variance . Increasing the number of dimensions did not systematically lower the error in forecasting the scale of the P10 activity . ( E ) Comparison of the mean R2 in forecasting P10 activity when using embedding dimensions explaining 80% and 90% variance . Increasing the number of dimensions did not systematically improve the forecasting of the variance of the P10 activity . DOI: http://dx . doi . org/10 . 7554/eLife . 27342 . 016 We first confirmed that each recorded neural population did not appear to contain any motorneurons with axons in P10 , which could make the decoding potentially trivial ( Figure 6—figure supplement 1 ) . To then decode P10 activity , we used a statistical model that predicts the firing rate of nerve P10 at each time point , by weighting and summing the recent history ( up to 100 ms ) of the trajectory in the low dimensional space , and using a non-linearity to convert this weighted sum into a firing rate . We controlled for over-fitting using cross-validation forecasting: we fit the model to a 40 s window of trajectory data , and predicted the next 10 s of P10 activity ( Figure 6B ) . By sliding the window over the data , we could assess the quality of the forecast over the entire recording ( Figure 6C ) . The model could accurately fit and forecast P10 activity from the low-dimensional trajectory in all nine population responses ( Figure 6D ) . Emphasising the quality of the model , in Figure 6D we plot example forecasts of the entire P10 recording based on fitting only to the first 40 s window , each example taken from the extremes we obtained for the fit-quality metrics . Notably , in one recording the population response shutdown half-way through; yet despite the model being fit only to the 40 s window containing strong oscillations , it correctly forecasts the collapse of P10 activity , and its slight rise in firing rate thereafter . Thus , the low dimensional trajectory of the periodic attractor appears to directly encode muscle commands for movement . To confirm this , we asked whether the encoding – as represented by the P10 activity – was truly low-dimensional . The successful decoding of future P10 activity was achieved despite needing only 3–5 embedding dimensions to account for 80% variance in the population activity for these nine recordings ( Figure 6—figure supplement 2 ) . Increasing the number of embedding dimensions to account for 90% variance , at least doubling the number of embedding dimensions , did not improve the forecasts of P10 activity ( Figure 6—figure supplement 2 ) . These results suggest that the low dimensional population trajectory is sufficient to encode the locomotion muscle commands . If the low-dimensional trajectory described by the joint activity of the population just is the motor program for locomotion , then how crucial to this program are the firing of individual neurons ( Katz et al . , 2004; Carmena et al . , 2005; Hill et al . , 2012; Huber et al . , 2012; Carroll and Ramirez , 2013; Hill et al . , 2015 ) ? Having quantified the motor program as the low-dimensional activity trajectory , we could uniquely ask how much each neuron participated in each evoked program . We quantified each neuron’s participation as the absolute sum of its weights on the principal axes ( eigenvectors ) : large total weights indicate a dominant contribution to the low-dimensional trajectory , and small weights indicate little contribution . So quantified , participation is a contextual measure , giving the contribution to the population trajectory of both a neuron’s firing rate and its synchrony with other neurons , relative to the rate and synchrony of all other neurons in the population ( Figure 7—figure supplement 1 ) . Every population response had a long-tailed distribution of participation ( Figure 7A ) , indicating that a minority of neurons dominated the dynamics of any given response . Nonetheless , these neurons were not fixed: many with high participation in one population response showed low participation in another ( Figure 7B , C ) . To rule out noise effects on the variability of participation ( for example , due to the finite duration of recording ) , we fitted a noise model to the change in participation separately for each preparation ( Figure 7D , E ) . Every preparation’s pedal ganglion contained neurons whose change in participation between responses well-exceeded that predicted by the noise model ( Figure 7F ) . Consequently , the contribution of single neurons was consistently and strongly variable between population responses in the same preparation . 10 . 7554/eLife . 27342 . 017Figure 7 . Single neuron participation varies within and between evoked locomotion bouts . ( A ) Distributions of single neuron participation per evoked population response . We plot the distribution of participation for all neurons in a population ( grey line ) , expressed as a percentage of the maximum participation in that population’s response . Black line gives the mean over all 30 population responses . ( B ) Change in participation between evoked locomotion bouts . Each dot plots one neuron’s maximum participation over all three evoked population responses , against its maximum change in participation between consecutive responses ( n=1131 neurons ) . ( C ) Two example neurons with variable participation between responses , from two different preparations . ( D ) Distribution of the change in participation between responses for one preparation . ( E ) Detecting strongly variable neurons . Gaussian fit ( red ) to the distribution of change in participation ( black ) from panel D . Neurons beyond thresholds ( grey lines ) of mean ±3SD of the fitted model were identified as strongly variable . ( F ) Proportion of identified strongly variable neurons per preparation . ( G ) Distance between pairs of population responses as a function of the total change in neuron participation between them . Each dot is a pair of responses from one preparation; the distance between them is given as a proportion of the mean distance between each response and a random projection ( <1: closer than random projections ) , allowing comparison between preparations ( Figure 3C ) . Black dots are excluded outliers , corresponding to the pairs containing response 1 in preparation 4 with apparent chaotic activity ( Figure 3—figure supplement 1 ) . ( H ) Distance between pairs of population responses as a function of the distance between the distributions of participation ( panel A ) . Conventions as for panel G . DOI: http://dx . doi . org/10 . 7554/eLife . 27342 . 01710 . 7554/eLife . 27342 . 018Figure 7—figure supplement 1 . Participation captures both rate and synchrony effects . We illustrate here that the participation of each neuron in the low-dimensional trajectory contains contributions from both its firing rate and its synchrony with other neurons . Hence changes to its participation correspond to changes in rate and/or synchrony . Note that as the participation quantifies each neuron’s contribution to multiple dominant patterns of activity co-variation in the population , so we expect only approximate correlations with the cruder , independent measures of rate and synchrony used here . ( A ) Example correlation between the evoked firing rates and participation scores of all neurons in one evoked population response . Participation is expressed as a percentage of the maximum score; the firing rate is calculated using all spikes from the offset of the stimulus to the end of the recording , and thus cannot capture contributions due to the decay or increase of rates . ( B ) Histogram of firing rate and participation correlations across all 30 programs . Firing rate explains on average ( R2 , red line ) 50% of the variation in participation within an evoked response . ( C ) Example correlation between the change in firing rates and the change in participation scores across one pair of evoked responses in the same preparation . One circle per neuron; red circles indicate neurons labelled as strongly variant for this preparation ( though not necessarily across this pair of responses ) . ( D ) Histogram of the correlations between change in rate and participation across all 30 pairs of programs ( pairs [1 , 2] , [1 , 3] , and [2 , 3] within each preparation ) . Firing rate changes explain on average 23% of the variation in the change of participation ( R2 , red line ) . ( E-H ) As A-D , for synchrony and changes in synchrony . The synchrony of each neuron is calculated as the total absolute correlation of that neuron with all other neurons , given by its corresponding column of the pairwise correlation matrix between all spike-density functions after stimulus offset . DOI: http://dx . doi . org/10 . 7554/eLife . 27342 . 01810 . 7554/eLife . 27342 . 019Figure 7—figure supplement 2 . Testing for an invariant central pattern generator . Here we test the possibility that hidden within the variation between programs is a small , invariant core of neurons that may form a classic central pattern generator . We do this by first determining the type of dynamical ensemble to which each neuron was assigned ( tonic , oscillator , burster , or pauser: see Figure 1C , main text ) , in order to find phasically-firing neurons expected of a pattern generator . ( A ) The proportion of neurons consistently assigned to the same type of dynamical ensemble across all three population responses . One circle per preparation . The inconsistent classification across programs is congruent with the wide variation in participation . Only the oscillator class of ensembles contained at least one consistently assigned neuron in all ten preparations , and so were candidates for containing an invariant core of neurons . ( B ) Scatter plot of participation against the maximum change in participation , restricted to neurons consistently assigned to the oscillator type of ensemble . One circle per neuron; colours indicate preparation . The same correlation between participation and variation as for the entire dataset is observed , with no clear evidence of an invariant group of strongly participating neurons across the preparations ( as would be indicated by neurons falling in the lower right quadrant ) . ( C ) The top 10% of consistent oscillator neurons by participation , and their corresponding maximum change in participation . Each circle plots the median across the top 10% for that preparation . Even restricted to just the top 10% , most preparations showed high variation in participation amongst the oscillator neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 27342 . 019 We also tested for the possibility that hidden within the variation between programs is a small core of neurons that are strongly participating , yet invariant across programs . Such a core of phasically active neurons may , for example , form the basis of a classical central pattern generator . However , in our observed portion of the ganglion we found no evidence for a core of strongly participating , invariant , and phasically active neurons across the preparations ( Figure 7—figure supplement 2 ) . These data show that a neuron’s role within the locomotion motor program is not fixed , but leave open the question of whether single neuron variability causes variation in the program itself . In our analysis , variation between sequentially-evoked population responses is quantified by the distance between their low-dimensional projections ( as in Figure 3C ) . We found that the distance between a pair of population responses did not correlate with either the total change in neuron participation between the two responses ( Figure 7G ) or the distance between their participation distributions ( Figure 7H ) . The execution of the motor program is thus robust to the participation of individual neurons . To get some insight into the physical substrate of the attractor , we plotted maps of the participation of each neuron in each preparation . We found that neurons with strong participation across the three evoked population responses were robustly located in the caudo-lateral quadrant of the ganglion ( Figure 8A , B ) . Maps of the right ganglion also indicated strong participation in the rostro-medial quadrant; due to the low numbers of maps for each side , it is unclear whether this is a true asymmetry of the ganglia or simply reflects sampling variation . Neurons with highly variable participation between population responses ( Figure 8C , D ) were similarly found in the caudo-lateral quadrants of both ganglia . Strongly participating neurons were thus confined to specific regions of the pedal ganglion’s network . 10 . 7554/eLife . 27342 . 020Figure 8 . Mapping of participation in the attractor across the ganglion network . Here we plot neuron location with respect to the photodiode array ( yellow hexagon ) . Each plot pools neurons from preparations of the left ( n=4 preparations ) or right ( n=4 ) ganglia . A , B Maps of maximum participation across the three evoked population responses for left ( A ) and right ( B ) ganglion recordings . The area of each marker is proportional to the neuron’s maximum participation . Neurons are colour coded ( light orange to dark red ) by the quintile of their participation across all preparations . C , D As for panels ( A , B ) , but plotting the range of participation across the three evoked population responses . DOI: http://dx . doi . org/10 . 7554/eLife . 27342 . 020 These data are consistent with a network-level distribution of the attractor , with a particularly strong contribution from the caudo-lateral quadrant . Encouragingly , from a different data-set we previously described this region as containing neural ensembles that generated a cyclical packet of neural activity , which moved in phase with activity from the neck-projecting P10 nerve ( Bruno et al . , 2015 ) . Consequently , both those data and our new data support our hypothesis that the pattern generator for locomotion is predominantly located in the caudo-lateral network . Testing the idea that high-dimensional population activity contains a low-dimensional signal has only been possible in the last decade or so , due to the necessary combination of large-scale multi-neuron recording and dimension reduction approaches ( Brown et al . , 2004; Briggman et al . , 2006; Cunningham and Yu , 2014; Kobak et al . , 2016 ) . Landmark studies have used this combination to project high-dimensional population activity into a more tractable low-dimensional space . In this space , studies have shown how activity trajectories are different between swimming and crawling ( Briggman et al . , 2005 ) ; distinguish olfactory ( Mazor and Laurent , 2005 ) , auditory ( Bartho et al . , 2009 ) , and visual ( Mante et al . , 2013 ) stimuli; and distinguish upcoming binary choices ( Harvey et al . , 2012 ) . Here we have gone a step further than previous studies by not only observing such low-dimensional signals , but explicitly testing for the first time the type of dynamical system that gives rise to the low-dimensional trajectories and its consistency between animals . Across all 30 evoked population responses examined here , there was a remarkable heterogeneity of spike-train patterns , from visually evident widespread oscillations to noisy , stuttering oscillations in a minority of neurons ( Figure 2—figure supplement 1 ) . Yet our analysis shows that underpinning this heterogeneity is the same dynamical system: a low-dimensional , decaying , periodic orbit . We found a remarkably consistent periodicity and rate of orbital decay across evoked responses within a preparation and between preparations . The stability of these dynamics , and the convergence of population activity to the same manifold , are all consistent with the expected behaviour of a true attractor . Our data thus suggest that only the low-dimensional system and not the high-dimensional population activity are consistent within and between nervous systems . We advance the hypothesis that the properties of the spiral attractor fully determine the parameters of the escape gallop: its frequency , physical distance per cycle , and duration . In this hypothesis , the orbital period of the attractor determines the period of the rhythmic gallop – the sequential activity of the neurons in each orbit thus driving the sequential contraction of the muscles driving the escape gallop ( Bruno et al . , 2015 ) . Further , the amplitude of the orbital period , corresponding to the spike rate of the neural population , could determine the strength of muscle contraction during the escape gallop , allowing control of the physical distance covered by each arching movement . Finally , the contraction rate of the attractor determines the duration of the escape: the faster the contraction rate , the shorter the escape gallop’s duration . The variation of these attractor properties between animals then determines the natural variability in the escape gallop . It follows that changes to parameters of the escape gallop caused by neuromodulation should correlate with changes to the orbital period and/or contraction rate of the attractor . For example , the reported increase in gallop duration by systemic injection of serotonin ( Marinesco et al . , 2004 ) should correlate with a decreased contraction rate of the attractor . Future work could test this hypothesis by determining the effects of neuromodulators on the spiral attractor’s properties and correlating those with read-outs of the escape gallop . Treating a neural circuit as a realisation of a dynamical system takes the emphasis away from the details of individual neurons - their neurotransmitters , their ion channel repertoire - and places it instead on their collective action . This allows us to take a Marr-ian perspective ( Marr , 1982 ) , which neatly separates the computational , algorithmic , and implementation levels of movement control . The computational problem here is of how to generate rhythmic locomotion for a finite duration; the algorithmic solution is a decaying periodic attractor - a spiral; and the implementation of that attractor is the particular configuration of neurons in the pedal ganglion - one of many possible implementations ( Kleinfeld and Sompolinsky , 1988; Pasemann , 1995; Eliasmith , 2005; Rokni and Sompolinsky , 2012 ) . Indeed , a spiral attractor is potentially a general solution to the problem of how to generate a finite rhythmic behaviour . We saw the separation of these levels most clearly in the variable participation of the individual neurons between evoked bouts of fictive locomotion . The projection of the pedal ganglion network’s joint activity into a low dimensional space captured the locomotion motor program independently of any single neuron’s activity . Even the most strongly participating neurons in a given population response could more than halve their participation in other evoked responses . These results suggest that the pedal ganglion’s pattern generator is not driven by neurons that are endogenous oscillators , as they would be expected to participate equally in every response . Rather , this variation supports the hypothesis that the periodic activity is an emergent property of the network . The adaptive function of having variably participating neurons is unknown . One possibility is that , by not relying on any core set of neurons to generate rhythmic activity , the pedal ganglion’s ability to generate locomotion is robust to the loss of neurons . A related possibility is that there is ‘sloppiness’ ( Panas et al . , 2015 ) in the pedal ganglion network , such that there are many possible configurations of neurons and their connections able to realise the spiral attractor that drives locomotion ( Marder et al . , 2015 ) . Such sloppiness allows for a far more compact specification of the developmental program than needing to genetically specify the type and wiring configuration of each specific neuron . The wide variation of single neuron participation between evoked bouts of fictive locomotion also raises new challenges for theories of neural network attractors ( Marder and Taylor , 2011 ) . While a variety of models present solutions for self-sustaining periodic activity in a network of neurons ( Kleinfeld and Sompolinsky , 1988; Eliasmith , 2005; Rokni and Sompolinsky , 2012 ) , it is unclear if they can account for the variable participation of single neurons . A further challenge is that while the variable participation of individual neurons does not affect the underlying program , clearly it takes a collective change in single neuron activity to transition between behaviours - as , for example , in the transition from galloping to crawling in Aplysia . What controls these transitions , and how they are realised by the population dynamics , is yet to be explored either experimentally or theoretically . Our results nonetheless argue against a number of hypotheses for the implementation of rhythmic locomotion by the pedal ganglion . As noted above , such single neuron variability between sequential locomotion bouts argues against the generation of rhythmic activity by one or more independent neurons that are endogenous oscillators . Our results also argue against the existence of many stable periodic states in this network ( Pasemann , 1995 ) . Such meta-stability would manifest as changes in periodicity following perturbation . Our results show that spontaneous divergences from the attractor overwhelmingly returned to the same attractor . How then might the pedal ganglion network implement a spiral attractor ? Our data were collected from an isolated central nervous system preparation , in which the modulatory influence of neurons outside the pedal ganglion cannot be discounted ( Jing et al . , 2008 ) . Nonetheless , as the pedal ganglion contains the central pattern generator for locomotion ( Jahan-Parwar and Fredman , 1980 ) , we can suggest how that generator is realised . Our results here support the hypothesis that the periodic activity is an emergent property of the ganglion’s network . We know the pedal ganglion contains a mix of interneurons and motorneurons ( Fredman and Jahan-Parwar , 1980 ) , and that the motorneurons are not synaptically coupled ( Hening et al . , 1979 ) , suggesting they read-out ( and potentially feedback to ) the dynamics of an interneuron network . An hypothesis consistent with our results here is that the ganglion contains a recurrent network of excitatory interneurons , centred on the caudo-lateral quadrant , which feed-forward to groups of motorneurons ( Bruno et al . , 2015 ) . This recurrent network embodies the attractor , in that stimulation of the network causes a self-sustained packet of activity to sweep around it ( Bruno et al . , 2015 ) . We see this as the periodic trajectory of joint population activity ( cf Figure 2A , Figure 3B ) . Our data further suggest that the pedal ganglion network supports at least two stable states , the spontaneous activity and the stable-spiral attractor . Reaching the stable-spiral attractor from the spontaneous activity required long-duration , high-voltage pedal nerve stimulation ( Figure 1; Bruno et al . , 2015 ) . In dynamical systems terms , this suggests that the spontaneous state’s basin of attraction is large: most perturbations return to that state , and it takes a large perturbation to move into a different basin of attraction . Multiple co-existing periodic attractors in a single network is also a challenge for current theories . While point attractor networks , such as Hopfield networks , can have vast number of stable states defined by different arrangements of the equilibrium activity of their neurons ( Miller , 2016 ) , a stable periodic attractor network typically has only two stable states: silence and periodic activity . The co-existence of stable spontaneous and periodic states in the same network suggests that something must reconfigure the network to sustain periodic activity ( Calin-Jageman et al . , 2007 ) ; otherwise , irrespective of the stimulation , the network would always return to the spontaneous state . One possibility in the pedal ganglion is that serotonin alters the effective connections between neurons: escape galloping is both dramatically extended by systemic injection of serotonin alongside tail stimulation ( Marinesco et al . , 2004 ) , and evoked by stimulating serotonergic command neurons CC9/CC10 in the cerebral ganglion ( Jing et al . , 2008 ) . Future experimental work should thus test the stability of the spontaneous state , and test how manipulating serotonin affects reaching and sustaining the stable-spiral attractor . There are potentially more stable states within the pedal ganglion’s network . The long-lasting crawl that follows the escape gallop is slower and omits the periodic arching of the body ( Flinn et al . , 2001 ) . We saw three perturbations of the attractor activity that were suggestive of a transition to a different , slower periodic orbit ( e . g . panel C in Figure 3—figure supplement 2 ) , consistent with a transition from galloping to crawling . Such crawling is also the animal’s normal mode of exploration ( Leonard and Lukowiak , 1986 ) , and so the ‘crawling’ attractor must be reachable from the spontaneous state too . Aplysia’s exploratory head-wave , moving its head side-to-side presumably to allow its tentacles and other head sensory organs to sample the environment ( Leonard and Lukowiak , 1986 ) , is also controlled by motorneurons in the pedal ganglion ( Kuenzi and Carew , 1994 ) . Previous studies of the Aplysia’s abdominal ganglion ( Wu et al . , 1994 ) , the leech segmental ganglion ( Briggman and Kristan , 2006 ) , and the crustacean stomatogastric ganglion ( reviewed in Marder and Bucher , 2007 ) have described multi-functional networks in which the same neurons are active in different motor behaviours . Our work here is consistent with the hypothesis that such multi-function is due to the neurons participating in different attractors realised by same network ( Briggman and Kristan , 2008 ) . Further work is needed to map the pedal ganglion network’s dynamics to the full range of Aplysia motor behaviour . Finding and quantifying the attractor required new analytical approaches . We introduce here the idea of using recurrence analysis to solve two problems: how to identify periodic activity in a high-dimensional space; and how to identify when the recorded system is and is not on the manifold of the attractor . By extracting the times when the population activity is on the manifold , we could then quantify and characterise the attractor , including identifying transient perturbations , and estimating changes in orbital period . Crucially , these manifold-times let us further introduce the idea of using linear models as a statistical estimator , to identify the type of attractor , and compare the detected attractor’s parameters within and between preparations . Our analysis approach thus offers a road-map for further understanding the dynamics of neural populations . There is rich potential for understanding spontaneous , evoked or learning-induced changes in the dynamics of populations for movement control . The dynamics of movement control populations transition between states either spontaneously or driven by external input ( Briggman et al . , 2005; Levi et al . , 2005 ) . Our recurrence approach allows both the detection of transitions away from the current state ( Figure 3 ) and the characterisation of the attractor in the new state . For learning , taking an attractor-view allows us to distinguish three distinct ways that short ( Stopfer and Carew , 1988; Katz et al . , 1994; Hill et al . , 2015 ) or long-term ( Hawkins et al . , 2006 ) plasticity could change the underlying attractor: by changing the shape of the manifold; by changing the rate of movement of the low-dimensional signal on the manifold; or by changing the read-out of the manifold by downstream targets . Such insights may contribute to the grand challenge of systems neuroscience , that of finding simplifying principles for neural systems in the face of overwhelming complexity ( Koch , 2012; Yuste , 2015 ) . Bandpassed optical data , spike-sorted data , and available P10 nerve recordings are hosted on CRCNS . org at: http://dx . doi . org/10 . 6080/K0SN074B All research code is available under a MIT License from ( Humphries , 2017 ) : https://github . com/mdhumphries/AplysiaAttractorAnalysis . A copy is archived at https://github . com/elifesciences-publications/AplysiaAttractorAnalysis . Full details of the Aplysia californica preparation are given in Bruno et al . ( 2015 ) . Briefly , the cerebral , pleural and pedal ganglia were dissected out , pinned to the bottom of a chamber , and maintained at 15-17∘C . Imaging of neural activity used the fast voltage sensitive absorbance dye RH-155 ( Anaspec ) , and a 464-element photodiode array ( NeuroPDA-III , RedShirtImaging ) sampled at 1600 Hz . Optical data from the 464 elements were bandpass filtered in Neuroplex ( 5 Hz high pass and 100 Hz low pass Butterworth filters ) , and then spike-sorted with independent component analysis in MATLAB to yield single neuron action potential traces ( the independent components ) , as detailed in ( Hill et al . , 2010 ) . Rhythmic locomotion motor programs were elicited using 8V 5 ms monophasic pulses delivered at 20 Hz for 2 . 5 s via suction electrode to pedal nerve 9 . A separate suction electrode was attached to pedal nerve 10 to continuously monitor the locomotion rhythm ( Xin et al . , 1996 ) . Evoked activity could last for many minutes; our system allowed us to capture a maximum of ≈125 s , divided between 30 s of spontaneous activity and 95 s of evoked activity . The stimulation protocol ( Figure 1B ) used short ( 15 min ) and long ( 60 min ) intervals between stimulations , as the original design also sought effects of sensitisation . Power spectra were computed using multi-taper spectra routines from the Chronux toolbox ( Bokil et al . , 2010 ) . We computed the power spectrum of each neuron’s spike-train post-stimulation , and plot means over all spectra within a recorded population , and the mean over all mean spectra . We computed the spike-density function f⁢ ( t ) for each neuron by convolving each spike at time ts with a Gaussian G:f ( t ) =∑t0<ts<t1G ( ts ) /∫t0t1G ( t∗ ) dt∗ , evaluated over some finite window between t0 and t1 ( see Szucs , 1998 ) . We set the window to be ±5⁢σ , and evaluated the convolution using a time-step of 10 ms . We defined the standard deviation σ of the Gaussian by the median inter-spike interval of the population: σ={median ISI in population}/12 ( see Humphries , 2011 ) . To visualise the entire population’s spiking activity ( Figure 1C ) , we cluster neurons by the similarity of their firing patterns using our modular deconstruction toolbox ( Bruno et al . , 2015 ) . Different dynamical types of ensembles were identified by the properties of their autocorrelograms: tonic , oscillator , burster , or pauser - see ( Bruno et al . , 2015 ) for details . We also assigned each neuron in the ensemble the same dynamical label , which we use in the analysis of Figure 7—figure supplement 2 . To demonstrate the firing rate change of each ensemble ( Figure 4 ) , we first counted the number of spikes emitted by that ensemble in 20 s windows , advanced in 5 s steps from the onset of stimulation . We then correlated ( Pearson’s R ) the time of each window against its spike count: ensembles were classified as decreasing rate if R < −0 . 2 , and increasing if R > 0 . 2 . We used a three-neuron network to demonstrate the dynamical properties of a periodic attractor as realised by neurons ( Figure 1—figure supplement 1 ) . Each neuron’s membrane dynamics were given by τa⁢a˙i=-ai⁢ ( t ) +ci⁢ ( t ) +∑j=13wj⁢i⁢rj⁢ ( t ) -γ⁢yi⁢ ( t ) , with adaptation dynamics τy⁢y˙i=-yi⁢ ( t ) +ri⁢ ( t ) , and output firing rate ri⁢ ( t ) =max⁡{0 , ai⁢ ( t ) } . Weights wj⁢i≤0 give the strength of inhibitory connections between the neurons , each of which receives a driving input ci . This model , due to Matsuoka ( Matsuoka , 1985 , 1987 ) , generates self-sustained oscillation of network firing rates given constant scalar inputs ci⁢ ( t ) =c , despite each neuron not being an endogenous oscillator: consequently the oscillations are an emergent property of the network . The time constants of membrane τa and adaptation τy dynamics , together with the strength of adaptation γ , determine the periodicity of the oscillations ( Matsuoka , 1985 , 1987 ) . Here we use τa=0 . 025 s , τy=0 . 2 s , and γ=2; input was ci=3 throughout except where noted . Low dimensional projections of the joint population activity were obtained for each program using standard principal components analysis , applied to the covariance matrix of the spike-density functions . The d leading eigenvectors Wi of the covariance matrix define the d principal dimensions , and the d corresponding eigenvalues are proportional to the variance accounted for by each dimension . The projection ( the ‘principal component’ ) onto each of the chosen dimensions is given by pi⁢ ( t ) =∑k=1nWik⁢fk⁢ ( t ) , where the sum is taken over all n neurons in the analyzed population . We used recurrence analysis ( Lathrop and Kostelich , 1989; Marwan et al . , 2007 ) to determine if the low-dimensional projection contained a stable periodic orbit . To do so , we checked if the low-dimensional projection P⁢ ( t ) = ( p1⁢ ( t ) , p2⁢ ( t ) , … , pd⁢ ( t ) ) at time t recurred at some time t+δ in the future . Recurrence was defined as the first point P⁢ ( t+δ ) = ( p1⁢ ( t+δ ) , p2⁢ ( t+δ ) , … , pd⁢ ( t+δ ) ) that was less than some Euclidean distance θ from P⁢ ( t ) . The recurrence time of point P⁢ ( t ) is thus δs . Contiguous regions of the projection’s trajectory from P⁢ ( t ) that remained within distance θ were excluded . Threshold θ was chosen based on the distribution of all distances between time-points , so that it was scaled to the activity levels in that particular program . Throughout we use the 10% value of that distribution as θ for robustness to noise; similar periodicity of recurrence was maintained at all tested thresholds from 2% upwards ( Figure 2—figure supplement 2 ) . We checked every time-point t between 5 s after stimulation until 10 s before the end of the recording ( around 7770 points per program ) , determining whether it was or was not recurrent . We then constructed a histogram of the recurrence times using 1 s bins to detect periodic orbits ( Figure 2E ) : a large peak in the histogram indicates a high frequency of the same delay between recurrent points , and thus a periodic orbit in the system . All delays less than 5 s were excluded to eliminate quasi-periodic activity due to noise in otherwise contiguous trajectories . Peaks were then defined as contiguous parts of the histogram between empty bins , and which contained more than 100 recurrent points . Programs had between one and four such periodic orbits . The peak containing the greatest number of recurrent points was considered the dominant periodic orbit of the program; the majority of programs had more than 50% of their recurrent points in this peak ( blue-scale vectors in Figure 2E ) . The mean orbit period of the program was then estimated from the mean value of all recurrence times in that peak . We measured the attractor’s stability as the percentage of all points that were in periodic orbits . Evolving dynamics of each program were analysed using 5 s sliding windows , advanced in steps of 1 s . We defined the ‘coalescence’ time of the attractor as the mid-point of the first window in which at least 90% of the points on the trajectory were recurrent . To determine if sequentially-evoked programs had the same manifold , we determined how closely the trajectories of each pair of programs overlapped in the low-dimensional space . We first projected all three programs from one preparation onto the principal axes of the first program , to define a common low-dimensional space . For each pair of programs ( A , B ) in this projection , we then computed the Haussdorf distance between their two sets of recurrent points , as this metric is suited to handling tests of closeness between irregularly shaped sets of points . Given the Euclidean distances {d⁢ ( A , B ) } from all recurrent points in A to those in B , and vice-versa {d ( B|A ) } , this is the maximum minimum distance needed to travel from a point in one program to a point in the other ( namely max{min{d ( A , B ) } , min{d ( B , A ) } ) . To understand if the resulting distances were close , we shuffled the assignment of time-series to neurons , then projected onto the same axes giving shuffled programs A* , B* . These give the trajectories in the low-dimensional space determined by just the firing patterns of neurons . We then computed the shuffled Haussdorf distance max{min{d ( A , B* ) } , min{d ( B , A* ) } ) . The shuffling was repeated 100 times . Mean ± 2SEM of the shuffled distances are plotted in ( Figure 3C ) ; the error bars are too small to see . To check the robustness of the convergence to the same manifold , we repeated this analysis starting from a common set of principal axes for the three programs , obtained using principal component analysis of their concatenated spike-density functions . We plot the results of this analysis in panel A of Figure 3—figure supplement 1 . As a further robustness control , we sought evidence of the manifold convergence independent of any low-dimensional projection . We made use of the idea that if neurons are part of sequential programs on a single manifold , then the firing of pairs of neurons should have a similar time-dependence between programs ( Yoon et al . , 2013; Peyrache et al . , 2015 ) . For each pair of programs ( A , B ) from the same preparation , we computed the similarity matrix S⁢ ( A ) between the spike-density functions of all neuron pairs in A , and similarly for B , giving S⁢ ( B ) . We then computed the correlation coefficient between S⁢ ( A ) and S⁢ ( B ) : if A and B are on the same manifold , so their pairwise correlations should themselves be strongly correlated . As a control we computed a null model where each neuron has same total amount of similarity as in the data , but its pairwise similarity with each neuron is randomly distributed ( Humphries , 2011 ) . The expected value of pairwise correlation between neurons i and j under this model is then Ei⁢j=si⁢sj/T , where ( si , sj ) are the total similarities for neurons i and j , and T is the total similarity in the data matrix . For comparison , we correlated S⁢ ( A ) with E , and plot these as the control correlations in Figure 3E . We detected divergences of the trajectory away from the putative manifold , indicating spontaneous perturbations of population dynamics . We first defined potential perturbations after coalescence as a contiguous set of 5 s windows when the density of recurrent points was below 90% and fell below 50% at least once . The window with the lowest recurrence density in this divergent period was labelled the divergent point . We removed all such divergent periods whose divergent point fell within two oscillation cycles of the end of the recording , to rule out a fall in recurrence due solely to the finite time horizon of the recording . For the remaining 19 divergent periods , we then determined if the population activity returned to a recurrent state after the divergent point; that is , whether the density of recurrence returned above 90% or not . The majority ( 17/19 ) did , indicating the perturbation returned to a manifold . For those 17 that did , we then determined if the recurrent state post-divergence was the same manifold , or a different manifold . For it to be the same manifold after the spontaneous perturbation , then the trajectory before the perturbation should recur after the maximum divergence . To check this , we took the final window before the divergent period , and counted the proportion of its recurrent delays that were beyond the end of the divergent period , so indicating that the dynamics were in the same trajectory before and after the divergence . We plot this in Figure 3H . We introduce here a statistical approach to analysing the dynamics of low-dimensional projections of neural activity time-series obtained from experiments . We first fitted a linear model around each point on the low-dimensional trajectory to capture the local dynamics . For each point P⁢ ( t ) , we took the time-series of points before and after P⁢ ( t ) that were contiguous in time and within 2 . 5×θ as its local neighbourhood; if less than 100 points met these criteria P⁢ ( t ) was discarded . We then fitted the dynamical model P*˙=A⁢P* that described the local evolution of the low-dimensional projection P* by using linear regression to find the Jacobian matrix A; to do so , we used the selected local neighbourhood time-series as P* , and their first-order difference as P*˙ . The maximum eigenvalue λ=a+i⁢b of A indicates the dominant local dynamics ( Strogatz , 1994 ) , whether contracting or expanding ( sign of the real part a of the eigenvalue ) , and whether oscillating or not ( existence of the complex part of the eigenvalue that is , b≠0 ) . The other eigenvalues , corresponding to the d-1 remaining dimensions , indicate other less-dominant dynamics; usually these were consistent across all dimensions ( Figure 4—figure supplement 1 ) . We fitted A to every point P⁢ ( t ) after the stimulation off-set , typically giving ≈5000 local estimates of dynamics from retained P⁢ ( t ) . The dominant dynamics for the whole program were estimated by averaging over the real a and the complex b parts of the maximum eigenvalues of the models fitted to all recurrent points in the dominant periodic orbit . The linear model’s estimate of the orbit rotation period was estimated from the complex part as ω=2⁢π⁢b⁢Δ⁢t , with the sampling time-step Δ⁢t=0 . 01 s here . The linear model’s estimate of the contraction rate is exp⁡ ( a/Δ⁢t ) , which we express as a percentage . We tracked changes in the oscillation period by first averaging the recurrence time of all recurrent points in a 5 s sliding window . We then correlated the mean time with the time-point of the window to look for sustained changes in the mean period over time , considering only windows between coalescence and the final window with 90% recurrent points . We used a weighted version of Spearman’s rank to weight the correlation in favour of time windows in which the trajectory was most clearly on the periodic orbit , namely those with a high proportion of recurrent points and low variation in recurrence time . The weighted rank correlation is: given vectors x and y of data rankings , and a vector of weights w , compute the weighted mean m=∑iwi⁢xi/∑iwi and standard deviation σx⁢y=∑iwi⁢ ( xi-mx ) ⁢ ( yi-my ) /∑iwi , and then the correlation ρ=σx⁢y/σx⁢x⁢σy⁢y . We used the weight vector: wi=si-1⁢Qi , where si is the standard deviation of recurrence times in window i , and Qi is the proportion of recurrent points in window i . P-values were obtained using a permutation test with 10000 permutations . We decoded P10 activity from the low-dimensional trajectory of population activity using a generalised linear model . We first ruled out that any simultaneously recorded neuron was a motorneuron with an axon in nerve P10 , by checking if any neurons had a high ratio of locking between their emitted spikes and spikes occurring at short latency in the P10 recording . Figure 6—figure supplement 1 shows that no neuron had a consistent , high ratio locking of its spikes with the P10 activity . We convolved the spikes of the P10 recording with a Gaussian of the same width as the spike-density functions of the simultaneously recorded program , to estimate its continuous firing rate f10 . We fitted the model f10⁢ ( t ) =exp⁡ ( β0+∑i=1d∑h=1mβi , h⁢Pi⁢ ( t-h ) ) to determine the P10 firing rate as a function of the past history of the population activity trajectory . Using a generalised linear model here allows us to transform the arbitrary co-ordinates of the d-dimensional projection P⁢ ( t ) into a strictly positive firing rate . Fitting used glmfit in MATLAB R2014 . To cross-validate the model , we found the coefficients β using a 40 s window of data , then forecast the P10 firing rate f10* using the next 10 s of population recording data as input to the model . Forecast error was measured as both the median absolute error and the correlation coefficient R between the actual and forecast P10 activity in the 10 s window . The fitting and forecasting were repeated using a 1 s step of the windows , until the final 40 s + 10 s pair of windows available in the recording . We tested activity histories between 50 and 200 ms duration , with time-steps of 10 ms , so that the largest model for a given program had d×20 coefficients . These short windows were chosen to rule out the contributions of other potential motorneurons in the population recording that would be phase offset from neck contraction ( as 200 ms is 2% of the typical period ) . All results were robust to the choice of history duration , so we plot results using history durations that had the smallest median absolute error in forecasting for that program . We quantified each neuron’s participation in the low-dimensional projection as the L1-norm: the absolute sum of its weights on the principal axes ( eigenvectors ) for program m:ρim=∑j=1d|λjmWjm ( i ) | , where the sum is over the d principal axes , Wjm⁢ ( i ) is the neuron’s weight on the jth axis , and λjm is the axis’ corresponding eigenvalue . Within a program , participation for each neuron was normalised to the maximum participation in that program . To fit a noise model for the variability in participation between programs , we first computed the change in participation for each neuron between all pairs of programs in the same preparation . We then fit a Gaussian model for the noise , using an iterative maximum likelihood approach to identify the likely outliers; here the outliers are the participation changes that are inconsistent with stochastic noise . In this approach , we compute the mean and variance of the Gaussian from the data , eliminate the data-point furthest from the estimate of the mean , re-estimate the mean and variance , and compute the new log likelihood of the Gaussian model without that data-point . We iterate elimination , re-estimation , and likelihood computation until the likelihood decreases . The final model ( mean and variance ) found before the decrease is then the best-fit Gaussian model to the bulk of the data . Neurons whose maximum change in participation exceeded a threshold of the mean ±3SD of that best-fit model were then considered ‘strongly variable’ neurons . We asked whether the variation in low-dimensional dynamics of sequentially-evoked programs was a consequence of the degree of variation in single neuron participation . Between a pair of consecutively evoked programs , we quantified the variation in their low dimensional dynamics as the Hausdorff distance between them , normalised by the mean distance between their random projections . This normalisation allowed us to put all programs on a single scale measuring the closeness relative to random projections , such that one indicates equivalence to a random projection , <1 indicates closer than random projections , and >1 indicates further apart then random projections . For a given pair of programs , we quantified the variability of individual neurons’ participation in two ways: by summing the change in participation of each neuron between the programs; and by computing the Hellinger distance between the two distributions of participation ( one distribution per program ) . Each neuron’s ( x , y ) location on the plane of the photodiode array could be estimated from the weight matrix from the independent component analysis of the original 464 photodiode time-series; see ( Bruno et al . , 2015 ) for full details . We were able to reconstruct locations for all neurons in 8 of the 10 recorded preparations; for the other two preparations , partial corruption of the original spike-sorting analysis data prevented reconstructions of some neuron locations in one; for the other , we could not determine on what side it was recorded . We merged all left or right ganglion recordings on to a common template of the photodiode array . The marker sizes and colour codes for each neuron were proportional to the normalised maximum participation of that neuron ( Figure 8A , C ) and to the range of normalised maximum participation across the three programs ( Figure 8B , D ) .

In all animals , neurons in the brain work together to generate movement . From a slug’s ability to crawl , to your ability to move your hand , movement is dependent on hundreds or thousands of neurons being active at the same time . Rhythmic movements such as crawling or swimming show this clearly: groups of neurons fire together and remain silent together in a repeating sequence , producing waves of muscle contraction . But do we need to understand the activity of each of the hundreds or thousands of individual neurons to know how they generate these movements ? Bruno et al . argue that we do not , and propose instead that brain circuits that generate movement show a few set patterns of activity . By recording the activity of a population of neurons , we can identify the pattern of activity that generates a particular movement . To illustrate this point , Bruno et al . examined the network of neurons that drives the rhythmic crawling movement of the sea slug Aplysia . The results show that the network of neurons seems to contain many different patterns of activity during crawling . Yet collectively these different patterns are reflections of a simpler hidden system , a spiral of ever-decreasing , oscillating activity . This pattern is referred to as a spiral attractor because whenever the network is activated , the overall pattern of activity is always pulled into this spiral regardless of its starting point . The same applies whenever the network is disturbed . The key thing to note , however , is that individual neurons within the network do not show the same activity each time the network is active . This means that only the spiral attractor itself , and not the contribution of the individual neurons , is constant every time the sea slug crawls . What do we need to know to understand the brain ? The results presented by Bruno et al . suggests that identifying the hidden systems that underlie seemingly complex and varying neural activity is the key to understanding how brains generate movement . This may also be true for how brains form memories , make decisions , and give rise to sight , hearing and touch .

2017

A spiral attractor network drives rhythmic locomotion

Cells employ regulated transport mechanisms to ensure that their plasma membranes ( PMs ) are optimally supplied with cholesterol derived from uptake of low-density lipoproteins ( LDL ) and synthesis . To date , all inhibitors of cholesterol transport block steps in lysosomes , limiting our understanding of post-lysosomal transport steps . Here , we establish the cholesterol-binding domain 4 of anthrolysin O ( ALOD4 ) as a reversible inhibitor of cholesterol transport from PM to endoplasmic reticulum ( ER ) . Using ALOD4 , we: ( 1 ) deplete ER cholesterol without altering PM or overall cellular cholesterol levels; ( 2 ) demonstrate that LDL-derived cholesterol travels from lysosomes first to PM to meet cholesterol needs , and subsequently from PM to regulatory domains of ER to suppress activation of SREBPs , halting cholesterol uptake and synthesis; and ( 3 ) determine that continuous PM-to-ER cholesterol transport allows ER to constantly monitor PM cholesterol levels , and respond rapidly to small declines in cellular cholesterol by activating SREBPs , increasing cholesterol uptake and synthesis . Animal cells carefully control both their overall content of cholesterol as well as its distribution among organelles . Although present in the membrane of every organelle , 60–90% of a cell’s cholesterol is concentrated in its plasma membrane ( PM ) ( De Duve , 1971; Lange et al . , 1989 ) . In the PM , the concentration of cholesterol is ~45 mole% of total lipids ( Das et al . , 2013 ) . How does PM acquire and maintain these high levels of cholesterol ? Much has been learned about the two sources through which cells acquire cholesterol – ( i ) receptor-mediated endocytosis of cholesterol-rich low-density lipoprotein ( LDL ) ( Brown and Goldstein , 1986 ) , and ( ii ) biosynthesis ( Brown and Goldstein , 1990 ) . However , neither of these two sources is located in the PM . LDL-derived cholesterol from lysosomes and synthesized cholesterol from endoplasmic reticulum ( ER ) must be moved to PM by transport mechanisms that are not understood . Moreover , the SREBP regulatory network that controls rates of LDL uptake and cholesterol biosynthesis is located not in the PM , but in the ER ( Brown and Goldstein , 2009 ) . Thus , PM cholesterol levels must be sensed and this information must be transmitted to ER to ensure optimal cholesterol levels in PM . These sensing mechanisms are also not fully understood . Recent studies have provided clues regarding how the organization of PM cholesterol into three distinct pools may play a role in regulating cholesterol levels ( Das et al . , 2014 ) . These studies showed that one pool of PM cholesterol , ~15 mole% of PM lipids , is sequestered by sphingomyelin . A second pool , ~12 mole% of PM lipids , is sequestered by other membrane factors . Cholesterol in excess of these two sequestered pools comprises a third pool that signals cholesterol excess to the regulatory machinery in the ER , thereby avoiding cholesterol overaccumulation while ensuring optimal cholesterol levels in PM . Defining these pools of cholesterol in PM was made possible by the use of mutant versions of Perfringolysin O ( PFO ) , a bacterial toxin that binds specifically to accessible cholesterol in membranes ( Das et al . , 2013; Flanagan et al . , 2009; Sokolov and Radhakrishnan , 2010 ) . Unfortunately , pore formation at 37°C by the PFO-derived probes prevented us from studying PMs of living cells , and restricted their use to PMs of cells that had been chilled to 4°C ( Das et al . , 2014 , 2013 ) . We overcame the undesired lytic properties of PFO probes by taking advantage of recent work where we showed that sub-domains of PFO and anthrolysin O ( ALO ) , another closely related toxin , bound membrane cholesterol but did not form large oligomeric pores in red blood cells at 37°C ( Gay et al . , 2015 ) . In the studies described here , we used the cholesterol-binding domain 4 of ALO , hereafter referred to as ALOD4 , due to the protein’s higher stability . When non-lytic ALOD4 was used to study cells at 37°C , we made the surprising observation that ALOD4 was not just a reporter of PM cholesterol accessibility; it also trapped cholesterol at PM and specifically inhibited PM-to-ER cholesterol transport . Using this new cholesterol trafficking inhibitor , we traced the route taken by LDL-derived cholesterol to show that it travels directly from lysosomes to PM . We also show that trapping as little as 1% of PM cholesterol triggers the activation of the SREBP regulatory network in ER . This activation leads to increased cholesterol production and uptake , thus enabling a rapid and switch-like response to restore PM cholesterol to optimal levels . To develop a tool to study PM cholesterol in cells growing at 37°C , we overexpressed His6-tagged ALOD4 in bacteria , and purified the resulting recombinant protein as described in Materials and methods . Gel-filtration chromatography of purified ALOD4 showed that it eluted as a single sharp peak ( Figure 1A , blue ) , and its homogeneity was confirmed by Coomassie staining ( Figure 1A inset ) . We also produced fluorescently-labeled versions of ALOD4 as described previously ( Gay et al . , 2015 ) . Gel-filtration chromatography and Coomassie staining combined with fluorescence gel imaging of ALOD4 labeled with Alexa Fluor 488 ( fALOD4-488 ) or Alexa Fluor 647 ( fALOD4-647 ) showed that their elution profiles and homogeneity were similar to those of unlabeled ALOD4 ( Figure 1A ) . 10 . 7554/eLife . 25466 . 003Figure 1 . Biochemical characterization and non-lytic properties of ALOD4 . ( A ) Gel filtration chromatography of purified proteins . Recombinant ALOD4 was purified and labeled with Alexa Fluor 488 ( fALOD4-488 ) or Alexa Fluor 647 ( fALOD4-647 ) fluorescent dyes as described in Materials and methods . Buffer B ( 1 ml ) containing 0 . 8 mg of ALOD4 , fALOD4-488 , or fALOD4-647 was loaded onto a Tricorn 10/300 Superdex 200 column and chromatographed at a flow rate of 0 . 5 ml/min . Absorbance at 280 nm ( A280 ) was monitored continuously to identify ALOD4 ( blue ) , fALOD4-488 ( green ) , or fALOD4-647 ( red ) proteins . Maximum A280 values for each protein ( ALOD4: 390 mAU , fALOD4-488: 231 mAU , and fALOD4-647: 279 mAU ) are normalized to one . ( Inset ) 3 µg of each protein was subjected to 15% SDS/PAGE and stained with Coomassie ( left ) or imaged with the 600 nm filter ( middle ) or the 700 nm filter ( right ) on a LICOR instrument . ( B ) Release of cytosolic proteins from CHO-K1 cells into media after incubation with ALOFL , but not ALOD4 . On day 0 , CHO-K1 cells were set up in medium B at a density of 6 × 104 cells/well of 48-well plates . On day 1 , media was removed , and cells were washed with 500 µl of PBS followed by addition of either 200 µl of medium C with the indicated concentration of ALOFL or ALOD4 , or with 200 µl of buffer C containing 1% SDS detergent . After incubation for 1 hr at 37°C , media was collected and cells were harvested as described in Materials and methods . Equal aliquots of cells and media ( 10% of total ) were subjected to immunoblot analysis as described in Materials and methods . Coom , Coomassie . DOI: http://dx . doi . org/10 . 7554/eLife . 25466 . 003 We next tested whether ALOD4 would form pores in CHO-K1 cells at 37°C . As a positive control for pore formation , we purified the full-length version of ALO ( ALOFL ) that forms large oligomeric pores in cells ( Bourdeau et al . , 2009; Gay et al . , 2015 ) . When added to CHO-K1 cells , ALOFL permeabilized the PM as revealed by immunoblotting of the medium for two cytosolic proteins , lactate dehydrogenase ( LDH ) and ubiquitin-activating enzyme ( E1 ) ( Figure 1B , lanes 2–4 ) . Similar release of cytosolic content was observed when the cells were lysed with SDS ( Figure 1B , lane 9 ) ; however , no such effect was observed when cells were treated with ALOD4 , even at the highest concentration of 10 µM ( Figure 1B , lanes 6–8 ) . This result is consistent with earlier studies where addition of the non-lytic domain 4 of PFO or ALO to the external medium resulted in labeling of the PM of intact cells , but not of internal organelle membranes ( He et al . , 2017; Ishitsuka et al . , 2011; Maekawa et al . , 2016; Shimada et al . , 2002 ) . Although we initially designed ALOD4 as a live-cell sensor of accessible cholesterol levels in PM , the question arose as to whether the binding of ALOD4 to PM cholesterol would perturb intracellular cholesterol distribution . To answer this question , we incubated CHO-K1 cells with ALOD4 for 1 hr at 37°C , and then processed the cells for immunoblot analysis . Not surprisingly , a dose-dependent increase in binding of ALOD4 to cells was observed ( Figure 2A , top panel ) . Most of the added ALOD4 was unbound and remained in the medium ( Figure 2A , two bottom panels ) . As described later , we used fluorescently-labeled ALOD4 to quantify the kinetics and stoichiometry of ALOD4 binding to PMs ( see Figure 4 ) . 10 . 7554/eLife . 25466 . 004Figure 2 . ALOD4 triggers activation of SREBP transcription factors in hamster and human cells . ( A–B ) On day 0 , CHO-K1 cells were set up in medium B at a density of 6 × 104 cells/well of 48-well plates ( A ) and SV-589 cells were set up in medium H at a density of 4 × 104 cells/well of 48-well plates ( B ) . On day 1 ( A ) or day 2 ( B ) , media was removed , cells were washed with 500 µl of PBS followed by addition of 200 µl of medium C ( A ) or medium G ( B ) with the indicated concentrations of ALOD4 . After incubation for 1 hr at 37°C , media was collected , and cells were harvested as described in Materials and methods . Equal aliquots of cells and media ( 10% of total ) were subjected to immunoblot analysis or Coomassie staining as described in Materials and methods . P = precursor form of SREBP1 or SREBP2; N = cleaved nuclear form of SREBP1 or SREBP2 . DOI: http://dx . doi . org/10 . 7554/eLife . 25466 . 004 To examine the consequence of ALOD4 binding to the PMs of these cells , we conducted immunoblot analysis of SREBP1 and SREBP2 , transcription factors that respond to declines in cellular cholesterol by activating genes encoding cholesterol biosynthetic enzymes and the LDL receptor that mediates uptake of cholesterol-rich LDL ( Horton et al . , 2003 ) . After being synthesized in the ER , both SREBPs bind to Scap , a cholesterol-sensing membrane protein that escorts SREBPs from ER to Golgi when ER cholesterol levels are below a threshold level of ~5 mole% of total ER lipids ( Brown and Goldstein , 2009 ) . In the Golgi , Site-1 protease and Site-2 protease sequentially cleave SREBPs , generating an active transcription factor fragment that travels to the nucleus to upregulate lipogenic genes , eventually raising cholesterol levels in cells and in ER . When ER cholesterol rises above the threshold concentration of ~5 mole% of total ER lipids , cholesterol binds to Scap and promotes Scap’s binding to Insigs , ER retention proteins . These interactions cause a conformational change in Scap , preventing its transport from ER to Golgi . Transport of SREBPs to Golgi is also blocked , and thus the proteolytic activation of SREBPs does not occur . As a result , cellular cholesterol levels decline and return to optimal levels . Activation of SREBPs is thus finely tuned to cellular cholesterol levels ( Brown and Goldstein , 2009; Goldstein and Brown , 2015 ) . As cells growing in lipoprotein-rich FCS were well supplied with cholesterol , almost all of their SREBP2 and about half of their SREBP1 were in their precursor ER forms ( Figure 2A , second and third panels , lane 1 ) . As increasing amounts of ALOD4 bound to these cells , we detected increasing amounts of the proteolyzed nuclear forms of both SREBP2 and SREBP1 , and a corresponding decline in their precursor forms ( Figure 2A , second and third panels , lanes 1–7 ) . Levels of Scap were not affected by ALOD4 binding ( Figure 2A , fourth panel , lanes 1–7 ) . ALOD4 also triggered the activation of SREBP1 and SREBP2 in human fibroblast cells ( SV-589 ) ( Figure 2B ) . To test whether ALOD4’s triggering of SREBP2 activation required its binding to PM cholesterol , we compared the effect of ALOD4 to that of a purified mutant version that can no longer bind membrane cholesterol ( Gay et al . , 2015 ) , designated as ALOD4 ( Mut ) . As shown in Figure 3A , when incubated with CHO-K1 cells , ALOD4 bound to cells ( top panel , lanes 1–4 ) and triggered activation of SREBP2 ( middle panel , lanes 1–4 ) , whereas ALOD4 ( Mut ) did not bind to cells ( top panel , lanes 5–8 ) or trigger activation of SREBP2 ( middle panel , lanes 5–8 ) . Coomassie staining of the medium confirmed that similar amounts of ALOD4 and ALOD4 ( Mut ) were added to cells ( Figure 3A , bottom panel , lanes 1–8 ) . 10 . 7554/eLife . 25466 . 005Figure 3 . ALOD4 binding to cells activates SREBP2 transcription factors without changing cellular cholesterol levels . ( A–B ) Immunoblot analysis of CHO-K1 cells after incubation with ALOD4 or ALOD4 ( Mut ) proteins in lipoprotein-rich or lipoprotein-poor serum . On day 0 , CHO-K1 cells were set up in medium B at a density of either 3 × 104 cells/well of 48-well plates ( A ) or 6 × 104 cells/well of 48-well plates ( B ) . On day 1 ( B ) or day 2 ( A ) , media was removed , cells were washed with 500 µl of PBS and then the following additions were made: 200 µl of medium C with indicated concentrations of ALOD4 or ALOD4 ( Mut ) ( A ) or 200 µl of lipoprotein-rich medium C or lipoprotein-poor medium F with indicated concentrations of ALOD4 ( B ) . After incubation for 1 hr at 37°C , media was collected , and cells were harvested as described in Materials and methods . Equal aliquots of cells and media ( 10% of total ) were subjected to immunoblot analysis or Coomassie staining as described in Materials and methods . ( C ) Cellular cholesterol levels decline after treatment with HPCD , but not after treatment with ALOD4 . On day 0 , CHO-K1 cells were set up in medium B at a density of 2 . 5 × 105 cells/60 mm dish . On day 3 , media was removed , followed by addition of 200 µl of medium C with the indicated concentration of HPCD or ALOD4 . After incubation for 1 hr at 37°C , media was removed and cells were harvested as described in Materials and methods . An aliquot of cells ( 5% of total ) was used for immunoblot analysis ( 25 µg/lane ) , and the remainder was used for quantification of cholesterol as described in Materials and methods . Each column represents the mean of cholesterol measurements from three independent experiments , and error bars show the standard error ( top panel ) . Asterisks denote level of statistical significance ( Student t test ) between cells treated without and with HPCD: *p<0 . 05 . Immunoblot analysis of the cells from one of the three experiments is shown in the bottom panel . ( D ) ALOD4 treatment causes increases in mRNA levels of HMG CoA Reductase and LDL receptor genes . On day 0 , CHO-K1 cells were set up in medium B at a density of 5 × 105 cells/100 mm dish . On day 2 , media was removed , followed by addition of 2 ml of medium C in the absence or presence of ALOD4 ( 5 µM ) or HPCD ( 1% w/v ) . After incubation for 4 hr at 37°C , media was removed , and cells were harvested for measurement of the indicated mRNA by quantitative RT-PCR as described in Materials and methods . For each gene , the amount of mRNA from untreated cells is set to 1 , and mRNA amounts from cells treated with ALOD4 or HPCD are expressed relative to this reference value . Each column represents the mean of relative mRNA values measured in three independent experiments , and error bars show the standard error . Asterisks denote level of statistical significance ( Student t test ) between cells treated without and with ALOD4 or HPCD: *p<0 . 05; **p<0 . 01; ***p<0 . 001 . The average Ct values for actin ( invariant control ) were 15 . 38 , 15 . 31 , and 15 . 18 for the untreated , ALOD4-treated , and HPCD-treated conditions , respectively . The average Ct values for HMG CoA Reductase and LDL receptor were 21 . 3 and 22 . 3 , respectively , for the untreated condition . P = precursor form of SREBP2; N = cleaved nuclear form of SREBP2 . DOI: http://dx . doi . org/10 . 7554/eLife . 25466 . 005 The results of Figures 2 and 3A were reminiscent of previous studies where SREBP activation was triggered by depleting cells of sterols , either by incubation in lipoprotein-poor serum ( Wang et al . , 1994 ) or by cholesterol extraction from PMs by cyclodextrin reagents ( Yang et al . , 2002 ) . If ALOD4 blocked receptor-mediated endocytosis of lipoproteins , then the net result would be the same as incubation of cells in lipoprotein-poor serum . To test this possibility , we incubated CHO-K1 cells with ALOD4 in lipoprotein-rich FCS as well as in lipoprotein-poor serum ( LPDS ) . As shown in Figure 3B , we observed similar binding of ALOD4 to cells ( top panel ) and similar triggering of SREBP2 activation ( bottom panel ) both in FCS ( lanes 1–5 ) and LPDS ( lanes 6–10 ) incubation conditions . See Figure 6 later for a more detailed analysis of LDL uptake and degradation by sterol-depleted cells in the presence of ALOD4 . We then tested whether ALOD4 triggered SREBP activation by removing cholesterol from cells . For this experiment , we compared ALOD4’s effects to that of HPCD , a commonly used cyclodextrin that extracts cholesterol from cells and lowers the overall cellular cholesterol content ( Ohtani et al . , 1989; Radhakrishnan et al . , 2008 ) . As shown in Figure 3C , incubation of CHO-K1 cells with increasing concentrations of HPCD or ALOD4 both triggered similar activation of SREBP2 ( compare lanes 1–3 to lanes 4–6 ) . The HPCD-induced increase in SREBP2 processing was accompanied by a decrease in total cellular cholesterol from ~30 µg/mg protein to ~18 µg/mg protein ( Figure 3C , lanes 1–3 ) . However , the ALOD4-induced increase in SREBP2 processing was observed even though there was no significant change in total cellular cholesterol ( Figure 3C , lanes 4–6 ) . Together , the experiments of Figure 3 suggest that ALOD4 triggers SREBP2 activation by binding to PM cholesterol and preventing its transport to ER , and not by blocking lipoprotein uptake or depleting cellular cholesterol . To determine whether ALOD4-induced activation of SREBPs led to upregulation of target genes in a manner similar to that induced by cholesterol depletion , we isolated total RNA from cells treated with ALOD4 or HPCD and measured the levels of mRNAs for HMG CoA reductase and LDL receptor by quantitative real-time PCR . As shown in Figure 3D , expression of mRNAs for both these SREBP target genes was increased by ALOD4 treatment , and these increases were similar to that observed after HPCD treatment . No change was observed in mRNA levels of actin , which is not a SREBP target gene . To further understand ALOD4’s effect on PM-to-ER cholesterol transport , we measured the kinetic parameters of this process . For more accurate quantification of binding of ALOD4 to PMs , we supplemented ALOD4 with tracer amounts ( ~5% of total protein ) of fALOD4-488 or fALOD4-647 , fluorescently-labeled versions of ALOD4 that trigger SREBP2 activation in CHO-K1 cells with the same concentration dependence as ALOD4 ( Figure 4A , compare lanes 1–4 to lanes 5–8 and lanes 9–12 ) . We first measured the rate of association of ALOD4 to PMs of CHO-K1 cells . We detected rapid binding of ALOD4 , both by immunoblot analysis ( Figure 4B , upper panel , lanes 1–7 ) and by fluorescence measurements ( Figure 4C , black filled circles ) . Half-maximal binding occurred at 10 min and saturation was reached after 40 min . Once ALOD4 bound to PMs , there was a lag time before triggering of SREBP2 activation at 60 min ( Figure 4B , lower panel , lanes 1–7 , and densitometry quantification in Figure 4C , red filled circles ) . As a control , ALOD4 ( Mut ) showed no significant binding to PMs ( Figure 4B , upper panel , lanes 8–14 , and Figure 4C , black open circles ) , and triggered minor SREBP2 activation only at the very longest incubation periods used ( Figure 4B , lower panel , lanes 8–14 , and densitometry quantification in Figure 4C , red open circles ) . 10 . 7554/eLife . 25466 . 006Figure 4 . Kinetics of ALOD4 binding to CHO-K1 cells . ( A ) Fluorescence labeling of ALOD4 does not affect its ability to trigger SREBP2 activation . Recombinant ALOD4 was purified and labeled with Alexa Fluor 488 ( fALOD4-488 ) or Alexa Fluor 647 ( fALOD4-647 ) fluorescent dyes as described in Materials and methods . On day 0 , CHO-K1 cells were set up in medium B at a density of 3 × 104 cells/well of 48-well plates . On day 2 , media was removed , and cells were washed with 500 µl of PBS followed by addition of 200 µl of medium C with the indicated concentrations of ALOD4 , fALOD4-488 , or fALOD4-647 . After incubation for 1 hr at 37°C , media was removed , and cells were harvested and subjected to immunoblot analysis as described in Materials and methods . ( B , C ) Association rate . On day 0 , CHO-K1 cells were set up in medium B at a density of 6 × 104 cells/well of 48-well plates . On day 1 , media was removed , and cells were washed with 500 µl of PBS followed by addition of 200 µl of medium C containing 3 µM of either ALOD4 or ALOD4 ( Mut ) . For quantification purposes , ~5% of ALOD4 or ALOD4 ( Mut ) proteins were labeled with Alexa Fluor 488 dyes . After incubation for the indicated times at 37°C , the cells were harvested and subjected to either immunoblot analysis for detection of SREBP2 processing and bound ALOD4 ( B ) or fluorescence analysis of bound ALOD4 ( C ) as described in Materials and methods . LICOR quantification of SREBP2 from ( B ) is expressed as the amount of nuclear form relative to the total ( precursor plus nuclear ) ( C , red ) . ( D , E ) Dissociation rate . On day 0 , CHO-K1 cells were set up in medium B at a density of 6 × 104 cells/well of 48-well plates . On day 1 , media was removed , and cells were washed with 500 µl of PBS followed by addition of 200 µl of lipoprotein-rich medium C or lipoprotein-poor medium F with 3 µM of ALOD4 . For quantification purposes , ~5% of ALOD4 or ALOD4 ( Mut ) proteins were labeled with Alexa Fluor 647 dyes . After incubation for 1 hr at 37°C , media was removed , and cells were washed twice with 500 µl of PBS followed by addition of 200 µl of medium C or medium F without ALOD4 . After incubation for the indicated times at 37°C , the cells were harvested and subjected to either immunoblot analysis for detection of SREBP2 processing and bound ALOD4 ( D ) or fluorescence analysis of bound ALOD4 ( E ) as described in Materials and methods . Curves are drawn merely to guide the eye and do not represent a fit . P = precursor form of SREBP2; N = cleaved nuclear form of SREBP2 . DOI: http://dx . doi . org/10 . 7554/eLife . 25466 . 006 Next , we measured the rate of dissociation of ALOD4 from PMs of CHO-K1 cells . We incubated cells with 3 µM ALOD4 for 1 hr , after which the ALOD4-containing medium was removed and replaced with medium without ALOD4 . Consistent with our previous observations , ALOD4 triggered SREBP2 activation in both lipoprotein-rich FCS and lipoprotein-poor LPDS ( Figure 4D , upper panel , compare lane 1 to 2 , and lane 7 to 8 ) . ALOD4 that had bound to cells during the 1 hr incubation was detected both by immunoblot analysis ( Figure 4D , lower panel , lanes 2 and 8 ) and by fluorescence measurements ( Figure 4E ) . Dissociation of this bound ALOD4 was rapid , with ~90% detaching from cells within 5 min ( Figure 4D , lower panel , lanes 2–6 and 8–12 , and Figure 4E ) . In addition to providing insights into the nature of ALOD4’s interaction with PM cholesterol , the rapid dissociation of bound ALOD4 from PMs also alleviated concerns that PM-bound ALOD4 may interfere with previously established methods to purify PM and ER membranes that we employ later for the analysis of Figure 5 ( Das et al . , 2013; Radhakrishnan et al . , 2008 ) . One might expect that dissociation of bound ALOD4 would restore PM-to-ER cholesterol transport and suppress the activation of SREBP2 transcription factors . However , we did not observe any such reversal during the short 30 min duration of this experiment . This is not surprising since previous studies have shown that suppression of SREBP2 activation by cholesterol delivered in complexes with MCD or in lipoproteins requires 1–2 hr ( Radhakrishnan et al . , 2008 ) . As shown later , suppression of SREBP2 activation after dissociation of bound ALOD4 was indeed observed after 4 hr ( see Figure 7C ) . 10 . 7554/eLife . 25466 . 007Figure 5 . ALOD4 triggers activation of SREBP2 by lowering ER cholesterol levels while leaving PM cholesterol levels unchanged . ( A–B ) On day 0 , CHO-K1 cells were set up in medium B at a density of 5 × 105 cells/100 mm dish ( 20 dishes/replicate/condition ) . On day 1 , media was removed and fresh medium B was added . On day 2 , media was removed , cells were washed with 5 mL PBS , followed by addition of 2 ml of medium B without or with ALOD4 ( 5 µM ) or HPCD ( 1% w/v ) . After incubation for 1 hr at 37°C , media was removed , and cells were washed twice with 5 ml of PBS . After the washes , 5 ml of PBS was added to each dish and cells were harvested . An aliquot of cells ( one dish/replicate/condition ) was used for immunoblot analysis and quantification of whole cell cholesterol as described in Materials and methods . Another aliquot of cells ( two dishes/replicate/condition ) was used for PM purification and quantification of PM cholesterol , as described in Materials and methods . The remainder of cells ( 17 dishes/replicate/condition ) was used for ER purification and quantification of ER cholesterol , as described in Materials and methods . Immunoblot of the cells from one of the three replicates for each treatment condition is shown in ( A ) , and whole cell , PM , and ER cholesterol levels are shown in ( B ) . Each column represents the mean of cholesterol measurements from triplicate assays , and error bars show the standard error . Asterisks denote level of statistical significance ( Student t test ) between cells treated without and with ALOD4 or HPCD: *p<0 . 05; **p<0 . 01 . P = precursor form of SREBP2; N = cleaved nuclear form of SREBP2 . DOI: http://dx . doi . org/10 . 7554/eLife . 25466 . 007 Our results so far suggest that the binding of ALOD4 to PMs blocks transport of PM cholesterol to ER , which leads to a decline in ER cholesterol levels even though overall cellular cholesterol levels do not change ( Figure 3C ) . The decline in ER cholesterol level is inferred from the triggering of SREBP activation , which has been shown to occur when ER cholesterol levels drop below a threshold concentration of ~5 mole% ( Radhakrishnan et al . , 2008 ) . We next sought to directly determine whether ALOD4 treatment lowers ER cholesterol levels . For this experiment , we set up three sets of CHO-K1 cells in lipoprotein-rich FCS . We then maintained one set of cells in FCS , and incubated the other two sets with either ALOD4 or cholesterol-depleting HPCD . After treatment for 1 hr , aliquots of cells from each set were subjected to immunoblot analysis for SREBP2 , whole cell cholesterol quantification , PM purification and quantification of PM cholesterol ( Das et al . , 2013 ) , and ER purification and quantification of ER cholesterol ( Radhakrishnan et al . , 2008 ) . Consistent with the results of Figure 3C , treatment of cells with ALOD4 or HPCD both triggered similar activation of SREBP2 ( Figure 5A , lanes 2 and 3 ) , however the treatments had distinct effects on cellular cholesterol distribution . Compared to the FCS-treated set , ALOD4 treatment did not significantly alter cholesterol levels in the whole cell or in PM , whereas HPCD treatment lowered total cellular cholesterol from ~33 µg/mg protein to ~20 µg/mg protein and PM cholesterol from 41 mole% of total PM lipids to ~31 mole% of total PM lipids ( Figure 5B ) . A different relationship was found when we quantified ER cholesterol levels . The ER cholesterol content of FCS-treated cells was ~8 mole% of total ER lipids and this value declined by a similar degree to ~4 mole% of total ER lipids when treated with ALOD4 or HPCD ( Figure 5B ) . The above experiments demonstrate that ALOD4 blocks the transport of cholesterol from PM to ER . To determine if this property of ALOD4 would allow us to clarify the intracellular trafficking route of LDL-derived cholesterol , we first depleted CHO-K1 cells of cholesterol by incubation with HPCD , and then added back increasing amounts of LDL . After cholesterol depletion , almost all of the SREBP2 was activated to its cleaved nuclear form ( Figure 6A , top panel , lane 1 ) . Incubation with increasing amounts of LDL for 3 hr suppressed SREBP2 activation , leading to a decrease in the cleaved nuclear form and a corresponding increase in the uncleaved precursor form ( Figure 6A , top panel lanes 2–4 ) . However , addition of 3 µM ALOD4 during the 3 hr LDL incubation period completely blocked the suppression of SREBP2 activation , and only the cleaved nuclear form was detected ( Figure 6A , top panel lanes 5–8 ) . We also measured the levels of ALOD4 that had bound to the PMs of LDL-treated cells . After cholesterol depletion , no binding of ALOD4 was detected ( Figure 6A , bottom panel , lane 5 ) , indicating levels of accessible PM cholesterol were low . As increasing amounts of LDL were added , a steady increase in PM-bound ALOD4 was observed , indicating a rise in levels of accessible PM cholesterol ( Figure 6A , bottom panel lanes 6–8 ) . Although LDL-derived cholesterol can reach the PM , its subsequent transport to ER is blocked due to sequestration at the PM by ALOD4 . 10 . 7554/eLife . 25466 . 008Figure 6 . ALOD4 blocks PM-to-ER transport of LDL-derived cholesterol in CHO-K1 cells . ( A ) Immunoblot analysis of sterol-depleted CHO-K1 cells incubated with LDL in the absence or presence of ALOD4 proteins . On day 0 , CHO-K1 cells were set up in medium B at a density of 3 × 104 cells/well of 48-well plates . On day 2 , media was removed , and cells were washed twice with 500 µl PBS followed by the addition of 200 µl of medium E with 2% HPCD ( sterol-depleting ) . After incubation for 1 hr at 37°C , media was removed , and sterol-depleted cells were washed twice with 500 µl of PBS followed by addition of 200 µl of medium E with the indicated concentration of human LDL in the absence or presence of 3 µM ALOD4 . After incubation for 3 hr at 37°C , the cells were harvested and subjected to immunoblot analysis for SREBP2 and cell surface-bound ALOD4 , as described in Materials and methods . ( B ) Cholesterol esterification in the presence of ALOD4 . On day 0 , CHO-K1 cells were set up in medium B at a density of 2 . 5 × 105 cells/60 mm dish . On day 2 , media was removed , and cells were washed twice with 2 ml of PBS followed by addition of 2 ml of medium D . On day 3 , media was removed , and cells were washed with 2 ml of PBS followed by addition of 1 ml of medium E with the indicated concentration of human LDL or 5 µg/ml of 25-HC ( in ethanol ) , all in the absence or presence of 3 µM ALOD4 . After incubation for 1 hr at 37°C , each monolayer was supplemented with 0 . 2 mM of sodium [14C]oleate ( 7759 dpm/nmol ) , and incubated for an additional 2 hr at 37°C . The cells were then harvested , and their levels of [14C]cholesteryl oleate and [14C]triglycerides were measured as described in Materials and methods . The levels of [14C]triglycerides formed at 0 and 100 µg/ml LDL treatment conditions were 38 . 6 and 39 . 3 nmol/mg/h , respectively , in the absence of ALOD4 , and 31 . 2 and 30 . 7 nmol/mg/h , respectively , in the presence of 3 µM ALOD4 . Each data point or column represents the mean of cholesterol esterification measurements from three independent experiments , and error bars show the standard error . ( C ) LDL uptake and degradation in the presence of ALOD4 . CHO-K1 cells were set up on day 0 and treated on day 2 exactly as described in ( B ) . On day 3 , media was removed , and cells were washed with 2 ml of PBS followed by addition of 1 ml of medium E containing 50 µg/ml of [125I]LDL ( 35 . 3 cpm/ng ) in the absence or presence of 1 or 3 µM of ALOD4 . The cells were incubated for 3 hr at either 4°C or 37°C , after which LDL uptake and degradation was determined as described in Materials and methods . Each column represents the mean of measurements from three independent experiments , and error bars show the standard error . P = precursor form of SREBP2; N = cleaved nuclear form of SREBP2 . DOI: http://dx . doi . org/10 . 7554/eLife . 25466 . 008 We were also able to show that ALOD4 blocks LDL-derived cholesterol from reaching the ER by using another measure of cholesterol delivery to ER , namely the activity of acyl-CoA acyltransferase ( ACAT ) ( Goldstein et al . , 1983 ) . ACAT is an ER enzyme that esterifies some of the LDL-derived cholesterol that reaches the ER . CHO-K1 cells were first depleted of cholesterol by incubation in lipoprotein-poor LPDS along with compactin , an inhibitor of cholesterol biosynthesis ( Brown et al . , 1978 ) . We then added increasing amounts of LDL together with [14C]oleate , and after 3 hr , processed the cells for measurement of cholesteryl [14C]oleate . As shown in Figure 6B , cholesteryl [14C]oleate formation increased in a dose-dependent manner as increasing amounts of LDL were added ( left panel , black filled circles ) , but was completely blocked when 3 µM ALOD4 was included during the incubation with LDL and [14C]oleate ( left panel , red filled circles ) . In contrast to the all-or-none effects observed with LDL , ALOD4 had little effect on 25-HC mediated cholesteryl [14C]oleate formation ( Figure 6B , right panel ) . This is likely because cholesterol-binding toxins like ALOD4 do not bind 25-HC ( Sokolov and Radhakrishnan , 2010 ) , and thus the oxysterol is free to enter cells and travel to the ER where it directly activates ACAT to esterify the cholesterol in that membrane ( Cheng et al . , 1995 ) . The increase in PM-bound ALOD4 detected after LDL addition ( Figure 6A ) argues against ALOD4 inhibiting the uptake or degradation of LDL; nevertheless the nature of ALOD4’s interaction with cholesterol in membranes gave us some concerns . ALOD4 could potentially bind to regions of the PM containing clathrin-coated pits and block LDL internalization . ALOD4 could also bind to cholesterol in the lipid monolayer coating the LDL surface and then block LDL’s binding to the LDL receptor or processing in lysosomes . To allay these concerns , we used previously described methods ( Goldstein and Brown , 1974 ) to directly measure the uptake and degradation of LDL in the presence of ALOD4 . CHO-K1 cells were depleted of cholesterol by incubation in lipoprotein-poor LPDS along with compactin . We then added 50 µg/ml [125I]LDL , a concentration at which SREBP2 activation was completely suppressed in the absence of ALOD4 and not suppressed at all in the presence of 3 µM ALOD4 ( Figure 6A , upper panel , compare lane 3 to lane 7 ) . After incubation for 3 hr with increasing concentrations of ALOD4 , we processed the cells for measurement of surface-bound plus internalized [125I]LDL ( a measure of LDL uptake ) and [125I]monoiodotyrosine released into medium ( a measure of degradation of LDL ) . As shown in Figure 6C , when incubated at 37°C , the uptake of [125I]LDL increased as increasing concentrations of ALOD4 were added ( left panel ) , a result that is consistent with increased expression of the LDL receptor induced by ALOD4 treatment ( Figure 3D ) . The degradation of [125I]LDL persisted even at the highest added concentration of 3 µM ALOD4 ( Figure 6C , right panel ) . As a negative control , incubation of the cells at a temperature that blocks endocytosis ( 4°C ) suppressed both [125I]LDL uptake and degradation by more than 75% ( Figure 6C , gray bars ) . We next sought to determine whether ALOD4 would affect PM-to-ER cholesterol transport in cells deficient in Niemann-Pick C1 ( NPC1 ) , a lysosomal membrane protein that is critical for transporting LDL-derived cholesterol out of lysosomes ( Liscum and Faust , 1987; Pentchev , 2004 ) . In NPC1-deficient cells , LDL-derived cholesterol does not reach the PM to fill up its pool of accessible cholesterol ( Das et al . , 2013 ) . The PMs of NPC1-deficient cells must thus rely on biosynthesis to meet their cholesterol requirements . For this experiment , we used a previously described mutant CHO-K1 cell line ( Wojtanik and Liscum , 2003 ) that had no detectable levels of NPC1 protein ( Figure 7A , top panel ) . Consistent with previous studies , we also show here that LDL was unable to stimulate cholesterol esterification in these NPC1-deficient cells ( Figure 7A , black bars ) . In contrast , stimulation of cholesterol esterification by 25-HC occurred in both control and NPC1-deficient cells ( Figure 7A , gray bars ) . We then set up control and NPC1-deficient cells in lipoprotein-rich FCS and incubated the cells with ALOD4 for 1 hr . We observed similar binding of ALOD4 to their PMs ( Figure 7B , bottom panel , compare lanes 1–4 to lanes 5–8 ) and identical concentration dependences for triggering SREBP2 activation ( Figure 7B , top panel , compare lanes 1–4 to lanes 5–8 ) . This result suggests that the pool of PM cholesterol that is transported to ER to suppress SREBP2 activation is of a similar magnitude in control and NPC1-deficient cells . 10 . 7554/eLife . 25466 . 009Figure 7 . ALOD4 blocks PM-to-ER transport of cholesterol in NPC1-deficient CHO-K1 cells . ( A ) Cholesterol esterification in control and NPC1-deficient CHO-K1 cells . On day 0 , cells were set up in medium B at a density of 2 . 5 × 105 cells/60 mm dish . On day 2 , media was removed , and cells were washed twice with 2 ml of PBS followed by addition of 2 ml of medium D . On day 3 , media was removed , and cells were washed with 2 ml of PBS followed by addition of 1 ml of medium E in the absence or presence of 50 µg/ml of human LDL or 4 µg/ml of 25-HC ( in ethanol ) . After incubation for 2 hr at 37°C , each monolayer was supplemented with 0 . 2 mM of sodium [14C]oleate ( 3913 dpm/nmol ) , and incubated for an additional 2 hr . The cells were then harvested , and their levels of [14C]cholesteryl oleate and [14C]triglycerides were measured as described in Materials and methods . The levels of [14C]triglycerides formed at 0 , 50 µg/ml LDL , and 4 µg/ml 25-HC treatment conditions were 21 . 8 , 22 . 1 and 18 . 7 nmol/mg/h , respectively , for control CHO-K1 cells , and 24 . 7 , 22 . 3 , and 19 . 7 nmol/mg/h , respectively , for NPC1-deficient CHO-K1 cells . Each column represents the mean of cholesterol esterification measurements from three independent experiments , and error bars show the standard error . ( Inset ) On day 0 , control and NPC1-deficient CHO-K1 cells were set up in medium B at a density of 3 × 104 cells/well of 48-well plates and 6 × 104 cells/well of 48-well plates , respectively . On day 2 , media was removed , and cells were harvested and subjected to immunoblot analysis of the indicated proteins as described in the Materials and methods . ( B ) Immunoblot analysis of control and NPC1-deficient CHO-K1 cells after incubation with ALOD4 . Cells were set up on day 0 in lipoprotein-rich FCS as described in A ( inset ) . On day 2 , media was removed , and cells were washed twice with 500 µl of PBS followed by addition of 200 µl of lipoprotein-rich medium C with the indicated concentrations of ALOD4 . After incubation for 1 hr at 37°C , the cells were harvested and subjected to immunoblot analysis as described in the Materials and methods . ( C ) Suppression of SREBP2 activation in control and NPC1-deficient CHO-K1 cells after induction with HPCD or ALOD4 . Cells were set up on day 0 as described in A ( inset ) . On day 2 , media was removed , and cells were washed twice with 500 µl of PBS followed by addition of 200 µl of medium E with 2% HPCD or 3 µM ALOD4 . After incubation for 1 hr at 37°C , media was removed , and cells were washed twice with 500 µl of PBS , followed by addition of 200 µl of medium C with the indicated amount of LDL , in the absence or presence of 3 µM ALOD4 . After incubation for 3 hr at 37°C , the cells were harvested and subjected to immunoblot analysis as described in the Materials and methods . P = precursor form of SREBP2; N = cleaved nuclear form of SREBP2 . DOI: http://dx . doi . org/10 . 7554/eLife . 25466 . 009 By blocking PM-to-ER cholesterol transport , ALOD4 provides a new way to rapidly deplete ER cholesterol without significantly altering total cellular cholesterol levels . In contrast , the commonly used method of HPCD treatment to rapidly deplete ER cholesterol also lowers total cellular cholesterol by 30–40% ( Figures 3C and 5 ) . Moreover , the use of HPCD to study cholesterol regulation in NPC1-deficient cells is confounded by HPCD’s ability to rescue the lysosomal cholesterol accumulation caused by NPC1-deficiency by as yet undefined mechanisms ( Abi-Mosleh et al . , 2009; Liu et al . , 2009 ) . We thus decided to compare the effects of these two methods of sterol depletion in NPC1-deficient cells . As shown in Figure 7C , sterol depletion by HPCD triggered the activation of SREBP2 in both control CHO-K1 cells ( upper panel , lane 1 ) and NPC1-deficient cells ( lower panel , lane 1 ) . Addition of LDL suppressed the activation of SREBP2 in control CHO-K1 cells ( upper panel , lanes 1–4 ) . LDL treatment also suppressed the activation of SREBP2 in NPC1-deficient cells ( lower panel , lanes 1–4 ) . This paradoxical result is consistent with an earlier study where HPCD treatment overcame the inability of NPC1-deficient cells to transport LDL-derived cholesterol out of lysosomes to ER ( Abi-Mosleh et al . , 2009 ) . A different result was obtained when ALOD4 was used to deplete ER cholesterol from NPC1-deficient cells . Consistent with the results of Figure 7B , incubation with 3 µM ALOD4 for 1 hr triggered the activation of SREBP2 in both control cells ( Figure 7C , upper panel , lane 5 ) and NPC1-deficient cells ( Figure 7C , lower panel , lane 5 ) . We then replaced the ALOD4-containing medium with medium containing increasing amounts of LDL . Rapid dissociation of ALOD4 from PMs ( Figure 4C–D ) allowed LDL-derived cholesterol to transit through the PM and reach the ER to suppress the activation of SREBP2 in control cells ( Figure 7C , upper panel , lanes 5–8 ) , but not in NPC1-deficient cells ( Figure 7C , lower panel , lanes 5–8 ) . Thus , unlike HPCD , ALOD4 does not overcome the post-lysosomal cholesterol trafficking defect caused by NPC1-deficiency . We next examined the effect of ALOD4 when cholesterol was added to cells in complexes with methyl-β-cyclodextrin ( MCD ) , which likely delivers cholesterol directly to PM without passing through lysosomes ( Abi-Mosleh et al . , 2009; Das et al . , 2014 ) . We first depleted cells of cholesterol by incubation with HPCD , triggering activation of SREBP2 ( Figure 8A , top and middle panels , lanes 1 and 5 ) . This activation of SREBP2 was suppressed when we added cholesterol/MCD complexes ( Figure 8A , top panel , lanes 1–4 ) , but suppression was blocked in the presence of 3 µM ALOD4 ( Figure 8A , middle panel , lanes 1–4 ) . After cholesterol depletion , no binding of ALOD4 was detected ( Figure 8A , bottom panel lane 1 ) , indicating levels of accessible PM cholesterol were low . As increasing amounts of cholesterol/MCD were added , we observed an increase in PM-bound ALOD4 ( Figure 8A , bottom panel lanes 1–4 ) , indicating that levels of accessible PM cholesterol had increased . This suggests that cholesterol/MCD complexes deliver their cholesterol to the PM , where it is immediately sequestered by ALOD4 , preventing its movement to the ER to suppress SREBP activation . In contrast , ALOD4 had no effect on the ability of 25-HC to suppress SREBP2 activation ( Figure 8A , top and middle panels , lanes 5–8 ) , and we detected no binding of ALOD4 to PMs after addition of 25-HC ( bottom panel , lanes 5–8 ) . This result is consistent with earlier studies showing that ( i ) cholesterol-binding toxins like ALOD4 bind cholesterol , but not oxysterols like 25-HC ( Sokolov and Radhakrishnan , 2010 ) ; ( ii ) addition of 25-HC does not increase cholesterol in the whole cell or in ER membranes ( Radhakrishnan et al . , 2008 ) , and ( iii ) 25-HC suppresses SREBP2 transport to Golgi by binding not to the cholesterol sensor Scap , but to Insigs , ER retention proteins ( Radhakrishnan et al . , 2007 ) . 10 . 7554/eLife . 25466 . 010Figure 8 . ALOD4 blocks PM-to-ER transport of cholesterol delivered directly to PM . ( A ) Immunoblot analysis of sterol-depleted CHO-K1 cells incubated with cholesterol/MCD complexes in the absence or presence of ALOD4 . On day 0 , CHO-K1 cells were set up in medium B at a density of 6 × 104 cells/well of 48-well plates in medium B . On day 1 , media was removed , and cells were washed twice with 500 µl of PBS followed by addition of 200 µl of medium E with 2% HPCD ( sterol-depleting ) . After incubation for 1 hr at 37°C , media was removed , and cells were washed twice with 500 µl of PBS followed by addition of 200 µl of medium E with the indicated amounts of cholesterol/MCD or 25-HC/ethanol , in the absence or presence of 3 µM ALOD4 . ( B ) Triggering of SREBP2 activation by ALOD4 is overcome by addition of excess cholesterol/MCD . On day 0 , CHO-K1 cells were set up in medium B at a density of 6 × 104 cells/well of 48-well plates . On day 1 , media was removed , and cells were washed with 500 µl of PBS followed by addition of 200 µl of medium C with the indicated concentration of cholesterol/MCD in the absence or presence of 3 µM ALOD4 . ( A–B ) After incubation for 4 hr at 37°C ( A ) or 1 hr at 37°C ( B ) , the cells were harvested and subjected to immunoblot analysis of SREBP2 and cell surface-bound ALOD4 as described in Materials and methods . ( C ) ALOD4 is not internalized by sterol-depleted cells . CHO-K1 cells were set up on day 0 and depleted of sterols on day 1 exactly as described in ( A ) . After sterol depletion , media was removed , and cells were washed twice with 500 µl of PBS followed by addition of 200 µl of medium E without or with 50 µM cholesterol/MCD , in the absence or presence of 3 µM ALOD4 . After incubation for 4 hr at 37°C , media was collected , cells were washed twice with 500 µl of PBS , and either harvested immediately or incubated with 200 µl of medium C without ALOD4 for 40 min at 37°C , and then harvested . Equal aliquots of cells and media ( 10% of total ) were subjected to immunoblot analysis as described in Materials and methods . P = precursor form of SREBP2; N = cleaved nuclear form of SREBP2 . DOI: http://dx . doi . org/10 . 7554/eLife . 25466 . 010 In Figure 8B , we tested whether ALOD4’s ability to sequester cholesterol delivered to PM in MCD complexes could be saturated . We incubated CHO-K1 cells that had been growing in FCS with increasing amounts of cholesterol/MCD in the absence or presence of 3 µM ALOD4 for 1 hr , and then processed the cells for immunoblot analysis . In the absence of ALOD4 , there was no triggering of SREBP2 activation since cellular sterol levels were high ( top panel , lanes 1–4 ) . When ALOD4 was added , it bound to PMs ( bottom panel , lane 5 ) and prevented the PM cholesterol from traveling to the ER , thereby triggering SREBP2 activation ( top panel , lane 5 ) . When further supplemented with 50 µM cholesterol/MCD , the cholesterol levels in PM increased as judged by an increase in bound ALOD4 ( bottom panel , lane 6 ) . However , SREBP2 activation was not suppressed since all of the added cholesterol was likely sequestered by ALOD4 at PM and prevented from traveling to ER ( top panel , lane 6 ) . Supplementation with 100 µM and 200 µM cholesterol/MCD delivered even more cholesterol to PM , as judged by further increases in bound ALOD4 ( bottom panel , lanes 7 and 8 ) . In these cases , the additional cholesterol in PM was in excess of ALOD4’s sequestering capacity , leading to transport of cholesterol to ER and suppression of SREBP2 activation ( top panel , lanes 7 and 8 ) . In a final experiment , we examined whether ALOD4 added to the extracellular medium of CHO-K1 cells was internalized during the treatment conditions of the experiments in the current study . We have already shown that when added to sterol replete cells for 1 hr , ALOD4 bound to PM ( Figure 4D , lower panel lane 2 ) but the PM-bound ALOD4 rapidly dissociated , with no detectable signal after 30 min ( Figure 4D–E ) . If ALOD4 had been internalized during the 1 hr incubation period , the immunoblot analysis and fluorescence assays would have revealed residual bound ALOD4 ( Figure 4D–E ) . We next measured whether ALOD4 could be internalized by sterol depleted cells ( Figure 8C ) . We first depleted cells of cholesterol by incubation with HPCD for 1 hr , after which we incubated the cells without or with 50 µM cholesterol/MCD complexes for 3 hr . As expected , SREBP2 activation was triggered by sterol depletion , and this activation was suppressed by addition of cholesterol ( top panel , lanes 1 and 2 ) . When we included 3 µM ALOD4 during the 3 hr incubation period ( bottom panel , lanes 3–6 ) , the added cholesterol was trapped at the PM by the binding of ALOD4 ( middle panel , lane 4 ) and was unable to reach the ER to suppress SREBP2 activation ( top panel , lane 4 ) . For a parallel set of cells , we removed the ALOD4-containing medium after the 3 hr incubation , and replaced it with medium without ALOD4 . After 40 min , a time period that is sufficient for complete dissociation of PM-bound ALOD4 ( Figure 4 ) , cells were harvested and subjected to immunoblot analysis . We observed no bound ALOD4 in the condition where cholesterol was added ( middle panel , lane 6 ) , suggesting that all of the PM-bound ALOD4 had dissociated during the 40 min period and that no detectable level of ALOD4 had been internalized during the 3 hr incubation period . Restoration of PM-to-ER cholesterol transport after ALOD4 dissociation was not sufficient to suppress the activation of SREBP2 in the short 40 min duration of this experiment , a result that is consistent with our previous observation in Figure 3D . Our current understanding of intracellular cholesterol trafficking has been largely made possible by tools that inhibit specific steps of this pathway ( see Figure 9 ) . These tools have been discovered through the study of genetic disorders or small molecule screens . Cells from individuals with familial hypercholesterolemia revealed how LDL receptors enable the uptake of cholesterol-rich LDL by a process called receptor-mediated endocytosis ( Brown and Goldstein , 1986 ) . Cells from individuals with Wolman disease or Cholesteryl Ester Storage disease revealed how lysosomal acid lipase converts cholesteryl esters liberated from the endocytosed LDL to unesterified cholesterol , a form that can be used by cells ( Brown et al . , 1976; Goldstein et al . , 1975; Sloan and Fredrickson , 1972; Wolman et al . , 1961 ) . The activity of lysosomal acid lipase , and other lysosomal degradative processes , are also blocked by chloroquine ( Goldstein et al . , 1975 ) . Cells from individuals with Niemann-Pick C disease revealed how two lysosomal proteins , NPC1 and NPC2 , move the unesterified cholesterol generated by lysosomal acid lipase out of lysosomes ( Carstea et al . , 1997; Kwon et al . , 2009; Liscum and Faust , 1987; Naureckiene et al . , 2000 ) . This last step is also blocked by U18666A , a cationic amphiphile that binds to NPC1 and inhibits cholesterol transport out of lysosomes ( Cenedella , 2009; Liscum and Faust , 1989; Lu et al . , 2015 ) . Despite these advances , little is known about how cholesterol moves to other organelles after it exits the lysosome . The current study introduces ALOD4 as a new inhibitor to understand steps in the post-lysosomal transport of cholesterol . 10 . 7554/eLife . 25466 . 011Figure 9 . Model for the intracellular itinerary of LDL-derived cholesterol . ( A ) Normal trafficking . LDL particles containing cholesteryl esters are internalized by sterol-depleted cells through receptor-mediated endocytosis . The internalized LDL is degraded in lysosomes , and its cholesteryl esters ( chol . esters ) are hydrolyzed by acid lipase to generate unesterified cholesterol ( yellow circles ) and fatty acids . Two lysosomal proteins , NPC1 and NPC2 , mediate the exit of LDL-derived cholesterol from lysosomes . Cholesterol is transported first to the sterol-depleted PM to replenish its cholesterol until optimal levels are reached . Excess PM cholesterol then traffics to regions of the ER where the SREBP regulatory network is located and where cholesterol levels are below a threshold concentration . After expanding ER cholesterol past the threshold level , excess cholesterol binds to SCAP , the ER cholesterol sensor , that is in complexes with SREBPs . Binding to cholesterol changes the conformation of Scap promoting its interaction with Insigs , ER retention proteins , and preventing recruitment of the Scap-SREBP complex into CopII vesicles for transport to Golgi . Proteolytic activation of SREBPs in Golgi does not occur , and genes encoding enzymes for cholesterol synthesis and uptake are shut down , thus maintaining cholesterol homeostasis . Excess ER cholesterol can be transported to the PM or can be converted by ACAT , an ER enzyme , to cholesteryl esters and stored in intracellular lipid droplets . ( B ) Genetic disorders or engineered tools that impair trafficking of LDL-cholesterol at various steps in its itinerary . See Discussion for a description of each of the impairments . DOI: http://dx . doi . org/10 . 7554/eLife . 25466 . 011 ALOD4 is a non-lytic peptide derived from a bacterial toxin that binds to accessible cholesterol in membranes ( Gay et al . , 2015 ) . Other versions of these toxins have been used in end-point assays to measure changes in accessible cholesterol in PM and ER membranes purified from cultured cells ( Das et al . , 2014; Sokolov and Radhakrishnan , 2010 ) . These earlier studies showed that the distribution of cholesterol between inaccessible and accessible pools in PM and ER membranes plays crucial roles in controlling cellular cholesterol levels . In the current study with live cells , we found that ALOD4 does much more than just serve as a passive reporter of accessibility of PM cholesterol . When added in the extracellular medium of cells that were replete with cholesterol , ALOD4 rapidly bound PM cholesterol and prevented its movement to ER . As a result , ER cholesterol levels dropped below a threshold concentration of ~5 mole% cholesterol , thereby triggering activation of SREBP transcription factors ( Figures 2–5 ) . ALOD4 did not bind to PMs of cells that were depleted of cholesterol ( Figures 6A and 8 ) . When we attempted to replenish sterol-depleted cells with MCD/cholesterol complexes that deliver cholesterol directly to PMs , ALOD4 immediately bound this newly delivered cholesterol in PMs , and prevented its movement to ER to suppress SREBP2 activation ( Figure 8 ) . A potentially useful feature of ALOD4 is that by sequestering PM cholesterol without extracting it from the membrane , ER cholesterol is depleted even though PM cholesterol levels and overall cellular cholesterol levels are not altered ( Figures 2D and 5 ) . This sequestration is reversible , as removing ALOD4 from the medium results in rapid dissociation of PM-bound ALOD4 ( Figure 4D–E ) , and restoration of PM-to-ER cholesterol transport ( Figure 7C ) . In contrast , other cholesterol modulators like cyclodextrins , statins , or lipoprotein-poor serum , all irreversibly deplete cholesterol from both PM and ER membranes ( Das et al . , 2014 , 2013; Radhakrishnan et al . , 2008 ) . ALOD4’s specific inhibition of PM-to-ER cholesterol transport allowed us to clarify the trafficking route of LDL-derived cholesterol after it exits lysosomes . The post-lysosomal fate of LDL-derived cholesterol has been a subject of debate with some studies suggesting that it moves from lysosomes to PM ( Das et al . , 2014; Lange and Steck , 1997; Xu and Tabas , 1991 ) , and other studies suggesting that it moves from lysosomes to ER ( Neufeld et al . , 1996; Underwood et al . , 1998 ) . In the presence of ALOD4 , sterol-depleted cells internalized LDL and degraded it in lysosomes ( Figure 6C ) , and the LDL-derived cholesterol exited the lysosomes and reached the PM ( Figure 6A ) . At the PM , ALOD4 sequestered the LDL-derived cholesterol , preventing it from traveling to the ER and suppressing SREBP2 activation ( Figure 6A ) , or stimulating the activity of ACAT ( Figure 6B ) . These data support the model that LDL-derived cholesterol moves first from lysosomes to PM ( Figure 9 ) . Only after PM’s cholesterol requirements are met does excess cholesterol travel from PM to ER . LDL-derived cholesterol from lysosomes cannot bypass the PM and travel directly to ER to shut down lipogenic genes , thus ensuring that the supply of cholesterol is not prematurely halted before the PM’s needs are satisfied . Our simplified model shows LDL-derived cholesterol traveling directly from lysosomes to PM ( Figure 9 ) , but we cannot rule out the possibility that this cholesterol makes intermediate stops at other organelles such as peroxisomes on its way to PM ( Chu et al . , 2015 ) , as long as these stops do not include the region of the ER housing the SREBP regulatory network or the ACAT enzyme . The PM to ER cholesterol transport route may also involve stops at other organelles . It should be noted that LDL-derived cholesterol is most likely delivered to the cytoplasmic leaflet of the PM , but must then flip to the outer leaflet of the PM to be accessible for binding to extracellularly added ALOD4 . This step is likely not rate-limiting , since previous studies have suggested that equilibration of cholesterol between the bilayer leaflets of red cell membranes is rapid ( t1/2 of ~1 s ) ( Steck et al . , 2002 ) . The kinetic studies of Figure 4 allowed us to estimate the number of cholesterol molecules that must be sequestered in PM to trigger SREBP transport from ER to Golgi and proteolytic activation . At saturation , ~150 ng ( ~10 pmol ) of ALOD4 bound to ~120 , 000 cells in a single well of a 48-well plate ( Figure 4C ) . If we assume that each molecule of ALOD4 sequesters one molecule of cholesterol , we calculate that ALOD4 binds ~0 . 08 fmol of cholesterol/cell . In an earlier study , we estimated that a CHO-K1 cell growing in lipoprotein-rich FCS contains ~10 fmol of cholesterol ( Radhakrishnan et al . , 2008 ) . The LIPID MAPS consortium quantified the lipidome of a mouse macrophage and reported that a macrophage cell contains 1000–2000 pmol cholesterol/µg DNA ( Dennis et al . , 2010 ) . Using this value and previous measurements of ~10 picograms DNA/mouse cell ( Collins , 1978; Collins et al . , 1980 ) , we calculate that a mouse macrophage contains ~10–20 fmol of cholesterol , similar to the estimated value for CHO-K1 cells . While being mindful of the many assumptions made in estimating these values , our results suggest that sequestering just a small fraction ( ~1% ) of cellular cholesterol at the PM and preventing its movement to ER , can trigger activation of SREBPs in an all-or-none manner . Treatment with HPCD , a commonly used tool to study cholesterol regulation , lowers cellular cholesterol by 30–40% , and also triggers activation of SREBPs ( Figures 3C and 5 ) . The results obtained with ALOD4 indicate that the cholesterol homeostatic machinery is sensitive to much more subtle changes in a small , labile fraction of PM cholesterol . It remains to be determined whether this small fraction represents a critical sub-pool of the previously determined accessible PM cholesterol pool ( ~15 mole% of PM lipids , Das et al . , 2014 ) . This metabolically active fraction of PM cholesterol can be maintained even in NPC1-deficient cells , which lack the ability to use LDL as a cholesterol source and must rely on cholesterol biosynthesis to stock its PM with optimal cholesterol levels . As shown in Figure 7B , ALOD4 triggered activation of SREBP2 in NPC1-deficient cells with identical concentration dependence as it did in control cells . In cells replete with cholesterol , robust PM-to-ER cholesterol transport ensures that the ER cholesterol concentration is above a threshold value of 5 mole% of ER lipids ( Radhakrishnan et al . , 2008 ) . At these high concentrations , cholesterol binds to the ER cholesterol sensor Scap , leading to formation of a complex with ER-resident Insigs , thus preventing Scap’s recruitment into CopII coated vesicles and transport of SREBPs to Golgi where they can be proteolytically activated ( Figure 9A ) . Incubating these cholesterol-replete cells with ALOD4 for 1 hr in the presence of compactin , an inhibitor of cholesterol synthesis , results in ER cholesterol dropping below the threshold concentration of 5 mole% ( Figure 5 ) . This drop in cholesterol concentration leads to cholesterol’s dissociation from Scap , disassembly of the Scap-Insig complex , and promotes Scap’s recruitment into CopII coated vesicles to transport and activate SREBPs ( Figure 9B ) . How does ALOD4 cause a drop in ER cholesterol levels without altering overall cellular cholesterol ? In the absence of synthesis , ER cholesterol is maintained at an equilibrium level by balancing the influx of cholesterol from the PM with the efflux of cholesterol out of ER to other organelles , including to the PM ( Lange et al . , 2004 , Das et al . , 2014 ) . ALOD4’s inhibition of the influx pathway , but not of the efflux pathway , disrupts this equilibrium and leads to ER cholesterol dropping below the threshold concentration of 5 mole% , even though cholesterol has not been removed from the PM ( Figure 5 ) . This suggests that when cells are replete with cholesterol , the ER continuously receives cholesterol from PM to offset losses of cholesterol due to efflux out of ER . This allows the ER to constantly sample the cholesterol content of PMs , and make adjustments via transcriptional regulation of cholesterol uptake and synthesis to ensure optimal PM cholesterol levels . How cholesterol moves from PM to ER is currently not understood , although PM proteins such as ABCA1 , and vesicular and non-vesicular trafficking pathways have been implicated ( Holthuis and Menon , 2014; Lahiri et al . , 2015; Mesmin and Maxfield , 2009; Yamauchi et al . , 2015 ) . In the experiments where we clarified the trafficking route of LDL-derived cholesterol , we purposefully included compactin to eliminate biosynthesis as a source for PM cholesterol . This allowed us to determine that LDL-derived cholesterol , once released from lysosomes , is directed first to the PM to ensure optimal cholesterol levels in that membrane before expanding the ER regulatory pool to downregulate cholesterol synthesis and uptake . A similar strategy is likely employed for newly synthesized cholesterol in ER , which must also be first directed to PM to meet the needs of that membrane before expanding the ER regulatory pool to shut down cholesterol synthesis and uptake . How proteins involved in cholesterol synthesis , regulation , and esterification are spatially organized in the ER , and how cholesterol is moved from ER to PM is currently not known . In addition to the uses outlined in the current study , ALOD4 could also help to reveal the transport pathways that originate from the ER and deliver cholesterol to PM . We obtained [14C]oleic acid ( 59 mCi/mmol ) and [3H]cholesteryl oleate ( 30–60 Ci/mmol ) from American Radiolabeled Chemicals , St . Louis , MO; Coomassie Brilliant Blue R-450 from BioRad , Inc . , Hercules , CA; methyl-β-cyclodextrin ( randomly methylated ) ( MCD ) and hydroxypropyl-β-cyclodextrin ( HPCD ) from Cyclodextrin Technologies Development , Inc . , Gainesville , FL; [125I]NaI ( 17 Ci/mg ) from PerkinElmer , Waltham , MA; protease inhibitor cocktail tablets ( cOmplete , EDTA-free ) from Roche , Indianapolis , IN; 25-hydroxycholesterol ( 25-HC ) from Steraloids , Inc . , Newport , RI; bovine serum albumin ( BSA ) , cholesterol , fetal bovine serum ( FCS ) , phenylmethanesulfonyl fluoride ( PMSF ) , and tris ( 2-carboxyethyl ) phosphine ( TCEP ) from Sigma , St . Louis , MO; and Alexa Fluor 488 C5-maleimide and Alexa Fluor 647 C2-maleimide from Thermo Fisher , Waltham , MA . Human low-density lipoprotein ( LDL , density <1 . 019 g/mL ( fixed rotor ) or density <1 . 063 g/mL ( zonal ) ) and newborn calf lipoprotein-deficient serum ( LPDS , density >1 . 3 g/mL ) were prepared by ultracentrifugation as described previously ( Goldstein et al . , 1983 ) . Solutions of compactin and sodium mevalonate were prepared as described previously ( Brown et al . , 1978 ) . Solutions of cholesterol/MCD complexes were prepared at a final cholesterol concentration of 2 . 5 mM and a cholesterol:MCD ratio of 1:10 as described previously ( Brown et al . , 2002 ) . We obtained monoclonal anti-His antibody from Millipore , Billerica , MA , monoclonal anti-lactate dehydrogenase ( LDH ) from Epitomics , Burlingame , CA , polyclonal anti-E1 and monoclonal anti-NPC1 from Abcam , Cambridge , MA , and polyclonal anti-calnexin from Novus Biologicals , Littleton , CO . Monoclonal antibody IgG-7D4 against hamster SREBP-2 ( amino acids 32–350 ) ( Yang et al . , 1995 ) , monoclonal antibody IgG-20B12 against hamster SREBP-1 ( amino acids 32–250 ) ( Rong et al . , 2017 ) , and monoclonal antibody IgG-4H4 against hamster Scap ( amino acids 1–767 ) ( Ikeda et al . , 2009 ) are described in the indicated references . Buffer A contained 50 mM Tris-HCl ( pH 7 . 5 ) and 1 mM TCEP . Buffer B is buffer A supplemented with 150 mM NaCl . Buffer C contains 10 mM Tris-HCl ( pH 6 . 8 ) , 100 mM NaCl , 1% ( w/v ) SDS , 1 mM EDTA , 1 mM EGTA , 20 µg/ml PMSF , and protease inhibitors ( one tablet/20 ml ) . Solution A contains 150 mM NaCl and 0 . 3 mM EDTA . Medium A is a 1:1 mixture of Ham’s F-12 and Dulbecco’s modified Eagle’s medium . Medium B is medium A supplemented with 100 units/ml penicillin , 100 μg/ml of streptomycin sulfate and 5% ( vol/vol ) FCS . Medium C is medium A supplemented with 5% ( vol/vol ) FCS . Medium D is medium A supplemented with 100 units/ml penicillin , 100 μg/ml of streptomycin sulfate , 50 μM compactin , 50 μM sodium mevalonate , and 5% ( vol/vol ) LPDS . Medium E is medium A supplemented with 50 μM compactin , 50 μM sodium mevalonate , and 5% ( vol/vol ) LPDS . Medium F is medium A supplemented with 5% ( vol/vol ) LPDS . Medium G is Dulbecco’s modified Eagle’s medium ( low glucose ) supplemented with 5% ( vol/vol ) FCS . Medium H is medium G supplemented with 100 units/ml penicillin and 100 µg/ml of streptomycin sulfate . Stock cultures of hamster CHO-K1 cells and 10–3 cells ( mutant CHO-K1 cells that lack detectable mRNA for Niemann-Pick C1 ) ( Wojtanik and Liscum , 2003 ) were maintained in monolayer culture at 37°C in 8 . 8% CO2 . Stock cultures of SV-589 cells were maintained in monolayer culture at 37°C in 5% CO2 . These environmental conditions were used for all experimental incubations , unless otherwise indicated . CHO-K1 , SV-589 , and 10–3 cells were obtained from American Type Culture Collection ( CCL 61 ) , Human Genetic Cell Repository ( GM 639 ) , and Laura Liscum ( Tufts University ) , respectively . Each cell line was propagated , aliquoted , and stored under liquid nitrogen . To guard against potential genomic instability , an aliquot of each cell line is passaged for only 4 weeks ( SV-589 and NPC 10–3 ) or 8 weeks ( CHO-K1 ) before a fresh batch of cells is thawed and propagated for experimental use . All the cell lines have been confirmed to be free of mycoplasma contamination using the MycoAlert Mycoplasma Detection Kit ( Lonza , Allendale , NJ ) . The following recombinant expression plasmids have been previously described: pALOFL encoding His6-tagged signal-peptide deficient ALO ( amino acids 35–512 ) ; pALOD4 encoding His6-tagged domain 4 ( amino acids 404–512 ) of ALO with two point mutations ( S404C and C472A ) ; and pALOD4 ( Mut ) encoding ALOD4 with six additional point mutations ( S404C , C472A , G501A , T502A , T503A , L504A , Y505A , and P506A ) ( Gay et al . , 2015 ) . Recombinant ALO proteins were overexpressed in E . Coli and purified by nickel chromatography as described previously ( Gay et al . , 2015 ) . ALOFL was further purified by gel filtration chromatography in buffer B , as described previously ( Gay et al . , 2015 ) , and stored at 4°C until use . ALOD4-rich elution fractions from nickel chromatography ( containing 150 mM NaCl ) were combined and concentrated to a final volume of 15 mL using a 10 , 000 molecular weight cut-off Amicon Ultra centrifugal filter ( Millipore ) . We then added 135 mL of NaCl-free buffer A to lower the NaCl concentration to ~15 mM , and loaded the mixture on a 1 ml anion-exchange chromatography column ( HiTrap Q , GE Healthcare , Pittsburgh , PA ) . After washing with 20 column volumes of buffer A containing 50 mM NaCl , bound ALOD4 was eluted with buffer A containing 500 mM NaCl into a single 2 ml fraction . After dilution with buffer A to lower the NaCl concentration to 150 mM , ALOD4 was further diluted in buffer B to reach a final ALOD4 protein concentration of 2 mg/ml . This material was either directly used in experiments or supplemented with 20% ( v/v ) glycerol , flash frozen in liquid nitrogen , and stored at −80°C for later use . In some cases , the lone engineered cysteine on ALOD4 ( at amino acid 404 ) was labeled with Alexa Fluor maleimide dyes as described previously ( Gay et al . , 2015 ) . Degree of labeling was greater than 0 . 5 in all cases . Protein concentrations were measured using a NanoDrop instrument ( Thermo Fisher ) or a bicinchoninic acid kit ( Thermo Fisher ) . The rate of incorporation of [14C]oleate into cholesteryl [14C]oleate and [14C]triglycerides in cultured CHO-K1 cells was measured as described previously ( Goldstein et al . , 1983 ) . Human LDL was iodinated using Pierce pre-coated iodination tubes ( Thermo Fisher ) according to the manufacturer’s instructions . Each tube , in a final volume of 500 µl of solution A , contained human LDL ( 5 mg ) and [125I]NaI ( 2 mCi ) . After incubation for 15 min at room temperature , 2 ml of solution A was added to the tube , and the entire mixture was loaded onto a PD-10 column ( GE Healthcare ) that had been pre-equilibrated with solution A . Elution fractions containing [125I]LDL were pooled and subjected to dialysis for 16 hr against 6 L of solution A to further eliminate unincorporated [125I]NaI . The dialyzed [125I]LDL had a specific activity of 70 . 6 cpm/ng and was stored at 4°C . The uptake and proteolytic degradation of [125I]LDL by cultured CHO-K1 cells was measured using previously described methods ( Goldstein and Brown , 1974 ) . After indicated treatments , media was removed from each well of 48-well plates , wells were washed twice with 500 µl PBS , and 200 µl of buffer C was added to each well . The plate was then placed on a shaker at room temperature for 20 min , after which the lysed cells were collected , mixed with 5x loading buffer , heated at 95°C for 10 min , and subjected to either 10% or 15% SDS/PAGE . The electrophoresed proteins were transferred to nitrocellulose filters using the Bio-Rad Trans Blot Turbo system , and subjected to immunoblot staining with the following primary antibodies: IgG-7D4 ( 10 µg/ml ) , anti-His ( 1:1000 dilution ) , IgG-20B12 ( 2 µg/ml ) , IgG-4H4 ( 0 . 2 µg/ml ) , anti-LDH ( 1:1000 dilution ) , anti-E1 ( 1:1000 dilution ) , anti-NPC1 ( 1:1000 dilution ) , and anti-calnexin ( 1:2000 dilution ) . Bound antibodies were visualized by chemiluminescence ( Super Signal Substrate; Thermo Fisher ) by using a 1:5000 dilution of donkey anti-mouse IgG ( Jackson ImmunoResearch , West Grove , PA ) or a 1:2000 dilution of anti-rabbit IgG ( GE Healthcare ) conjugated to horseradish peroxidase . Filters were exposed to Phoenix Blue X-Ray Film ( F-BX810; Phoenix Research Products , Pleasanton , CA ) at room temperature for 1–300 s or scanned using an Odyssey FC Imager ( Dual-Mode Imaging System; 2 min integration time ) and analyzed using Image Studio ver . 5 . 0 ( LI-COR , Lincoln , NE ) . After indicated treatments of 100 mm dishes , media was removed , and cells were washed twice with 5 ml of PBS . After addition of 1 ml PBS , cells were harvested by scraping and transferred to 1 . 7 ml centrifuge tube . An aliquot ( 100 µl ) was saved for immunoblot analysis . The remainder ( 900 µl ) was subjected to centrifugation at 2000 x g for 10 min , after which the pellets were resuspended in 600 µl of Buffer RLT ( RNeasy Mini Kit , Qiagen , Germantown , MD ) . Total RNA was prepared using the RNeasy Mini Kit ( Qiagen ) using the manufacturer’s directions , and subjected to real time PCR analysis . The primer sequences used for PCR were as follows: HMG CoA reductase ( Fwd: AGATACTGGAGAGTGCCGAGAAA; Rev: TTTGTAGGCTGGGATGTGCTT ) , LDL receptor ( Fwd: AGACACATGCGACAGGAATGAG; Rev: GACCCACTTGCTGGCGATA ) , and Actin ( Fwd: GGCTCCCAGCACCATGAA; Rev GCCACCGATCCACACAGAGT ) . After indicated treatments of 60 mm dishes , media was removed , and cells were washed twice with 2 ml PBS . After addition of 1 ml PBS , cells were harvested by scraping and transferred to 1 . 7 ml centrifuge tubes . An aliquot ( 50 µl ) was centrifuged at 4000 rpm for 10 min , followed by resuspension of the pellet in 100 µl of buffer C , lysis using a 22-gauge needle , and vigorous disruption on a shaker for 30 min . After using 20 µl of the resuspended pellet for quantifying protein content using a bicinchoninic acid assay kit , the remainder of the resuspended pellet ( 80 µl ) was mixed with 5x loading buffer , heated at 95°C for 10 min , and then subjected to SDS/PAGE and immunoblot analysis as described above . The remaining 950 µl of the original cell sample was used for cholesterol measurements . Lipids were extracted using a chloroform/methanol ( 1:1 , v/v ) mixture , the organic solvent was evaporated under a gentle stream of nitrogen , and unesterified cholesterol was measured using the Amplex Red Cholesterol Assay Kit ( Thermo Fisher ) . PM and ER membranes were purified and their cholesterol content was quantified as described previously ( Das et al . , 2013; Radhakrishnan et al . , 2008 ) . After indicated treatments , media was removed from each well of 48-well plates , wells were washed twice with 500 µl PBS to eliminate unbound fALOD4 , and 200 µl of buffer C was added to each well . The plate was then placed on a shaker at room temperature for 20 min , after which the lysed cells including surface-bound fALOD4 were collected , and transferred to a 96-well plate ( Greiner Bio-One , Monroe , NC ) . Solubilization of bound fALOD4 in SDS-containing buffer C eliminated fluorescence quenching effects that occur after binding of fALOD4 to cholesterol-containing membranes ( Gay et al . , 2015 ) . The 96-well plate was stored in a −20°C freezer for at least 2 hr to eliminate bubbles , after which bound fALOD4 fluorescence was measured using a Tecan microplate reader ( fALOD4-488: excitation wavelength: 495 nm , emission wavelength: 517 nm; fALOD4-647: excitation wavelength: 651 nm , emission wavelength: 672 nm ) . By measuring the fluorescence ( at identical gain settings ) from wells containing known amounts of fALOD4 in buffer C , we were able to quantify the amount of bound fALOD4 . All results were confirmed in at least three independent experiments conducted on different days using different batches of cells and different batches of purified ALOD4 . The only exception was the large-scale study described in Figure 5 which was conducted in triplicate on one occasion .

Cells are surrounded by a plasma membrane made mostly from oily molecules known as lipids . One of these lipids , called cholesterol , is essential for keeping this membrane stable . Cholesterol is partly produced within the cells at a specialized structure called the endoplasmic reticulum , and partly imported from the blood surrounding the cell . In the blood , cholesterol is shielded inside particles called low-density lipoprotein ( or LDL for short ) , which is taken into the cell and then sent to another structure called the lysosome . Inside the cell , cholesterol that is freshly produced in the endoplasmic reticulum or freshly imported into the lysosome , must be moved to the plasma membrane , where most of the cholesterol is located . Cholesterol levels are regulated by a ‘control machinery’ of proteins located in the endoplasmic reticulum . To keep the cholesterol levels constant , the endoplasmic reticulum needs to be in continual communication with the plasma membrane . However , the mechanisms by which cholesterol is transported between membranes are still poorly understood . Here , Infante and Radhakrishnan report a new tool to study how cholesterol is transported in human and hamster cells . The tool , which is based on part of a bacterial protein , traps cholesterol in the plasma membrane and prevents it from moving to the endoplasmic reticulum , and thus from updating the control machinery about cholesterol levels . From this inhibition , it is inferred that a stream of cholesterol constantly travels from the plasma membrane back to endoplasmic reticulum . This way , proteins in the endoplasmic reticulum can monitor the cholesterol levels in the plasma membrane in real-time . The endoplasmic reticulum responded rapidly even to small declines in cholesterol levels by activating genes that increase cholesterol production or the amount of cholesterol imported via the LDL pathway . Further work showed that cholesterol derived from LDL travels from the lysosome directly to the plasma membrane to maintain optimal cholesterol levels . It then moves to the endoplasmic reticulum to signal that cholesterol levels in the cell have been satisfied . The findings and tools described in this study will help to further investigate the mechanisms underlying the transport of cholesterol between the different membranes and structures in a cell . A next step will be to see if the mechanisms that apply to distribution of imported cholesterol from lysosomes , also apply to the cholesterol produced in the endoplasmic reticulum .

2017

Continuous transport of a small fraction of plasma membrane cholesterol to endoplasmic reticulum regulates total cellular cholesterol

Nitrogen-containing-bisphosphonates ( N-BPs ) are a class of drugs widely prescribed to treat osteoporosis and other bone-related diseases . Although previous studies have established that N-BPs function by inhibiting the mevalonate pathway in osteoclasts , the mechanism by which N-BPs enter the cytosol from the extracellular space to reach their molecular target is not understood . Here , we implemented a CRISPRi-mediated genome-wide screen and identified SLC37A3 ( solute carrier family 37 member A3 ) as a gene required for the action of N-BPs in mammalian cells . We observed that SLC37A3 forms a complex with ATRAID ( all-trans retinoic acid-induced differentiation factor ) , a previously identified genetic target of N-BPs . SLC37A3 and ATRAID localize to lysosomes and are required for releasing N-BP molecules that have trafficked to lysosomes through fluid-phase endocytosis into the cytosol . Our results elucidate the route by which N-BPs are delivered to their molecular target , addressing a key aspect of the mechanism of action of N-BPs that may have significant clinical relevance . N-BPs are the most commonly prescribed drugs used to treat osteoporosis ( Drake et al . , 2008 ) . They have two negatively charged phosphonate groups that bind to hydroxyapatite crystals with high affinity and enable efficient accumulation of N-BPs on the bone surface ( Drake et al . , 2008 ) . Osteoclasts , the major cell type responsible for bone resorption , release N-BPs from the bone matrix by dissolving bone mineral and then take up N-BPs through fluid-phase endocytosis ( Drake et al . , 2008; Thompson et al . , 2006 ) . N-BPs subsequently inhibit farnesyl diphosphate synthase ( FDPS ) in the mevalonate pathway and reduce protein prenylation , an essential post-translational lipid modification required for the function of numerous proteins such as Ras , Rab and Rho , thereby inducing apoptosis in osteoclasts and diminishing their bone-resorption activities ( Drake et al . , 2008; Dunford et al . , 2001; Fisher et al . , 2000; Hughes et al . , 1995; Kavanagh et al . , 2006; Luckman et al . , 1998a; Luckman et al . , 1998b; van Beek et al . , 1999 ) . However , it is not known how highly charged N-BPs exit the endocytic pathway to target FDPS , which is localized to the cytosol and peroxisomes ( Martín et al . , 2007 ) . It has been proposed that the acidification of endocytic compartments might neutralize the negative charges on the phosphonate groups and allow N-BPs to diffuse across the vesicle membrane ( Thompson et al . , 2006 ) , but this model does not address the issue that the amine groups in N-BPs become positively charged in acidic environments . An alternative model is that a transporter exists that facilitates the exit of N-BPs from endocytic vesicles . To gain further insight into the mechanism of action of N-BPs , including the mechanism by which N-BPs are delivered to their molecular target , we implemented an unbiased genome-wide screening approach based on CRISPR-mediated interference ( CRISPRi ) ( Figure 1A ) ( Gilbert et al . , 2014 ) . We transduced a genome-scale CRISPRi single-guide RNA ( sgRNA ) library into K562 human myeloid leukemia cells that stably express a dCas9-KRAB fusion protein , which functions as an sgRNA-guided transcription inhibitor . The cells were split into a population treated with alendronate ( ALN ) , a representative N-BP , and an untreated control population . Through deep sequencing , we quantified the enrichment/depletion of each sgRNA in the treated population compared to the control population ( Figure 1B , Figure 1—figure supplement 1A ) , and designated the target genes of those sgRNAs enriched in the treated population as resistance hits and those depleted as sensitizing hits ( Figure 1B , Figure 1—figure supplement 1B and Supplementary file 1 ) . Consistent with the current model for the action of N-BPs , enzymes , co-factors , and regulators of the mevalonate pathway are enriched in top hits ( Figure 1C–D and Figure 1—figure supplement 1C–D ) . Particularly , in accordance with the model that N-BPs induce cell death through inhibiting the enzymatic activities of FDPS and GGPPS1 ( Drake et al . , 2008 ) , silencing of FDPS and GGPPS1 strongly sensitized cells to ALN ( Figure 1C–D ) . However , in contradiction with the current model of N-BP action , we observed that silencing of numerous enzymes in the pathway upstream of FDPS in fact conferred strong resistance to ALN ( Figure 1D ) . A recent genome-wide genetic interaction study may resolve this paradox ( Horlbeck et al . , unpublished ) . That work demonstrated that isopentenyl-5-pyrophosphate ( IPP ) , the substrate of FDPS , is a toxic intermediate that interferes with DNA synthesis and causes DNA damage , suggesting that inhibition of enzymes upstream of FDPS protects cells from ALN by preventing ALN-induced accumulation of IPP . Amongst the resistance hits not known to be involved in the mevalonate pathway , the gene that conferred the strongest resistance was SLC37A3 ( Figure 1C ) , which is predicted to encode a membrane protein with 12 transmembrane segments ( Chou et al . , 2013 ) . SLC37A3 also appeared as a top resistance hit in a second CRISPRi screen using zoledronate , another representative N-BP , as the selection agent ( Figure 1—figure supplement 1E and Supplementary file 2 ) , further supporting its role in the mechanism of action of N-BPs . SLC37A3 is predicted based on sequence homology to be a glucose-6-phosphate/phosphate antiporter ( Chou et al . , 2013 ) . However , it has been demonstrated that SLC37A3 in fact lacks this predicted activity ( Pan et al . , 2011 ) . To the best of our knowledge , the physiological function of SLC37A3 has remained elusive . Intriguingly , a recent human protein interactome study reported an interaction between SLC37A3 and ATRAID ( Huttlin et al . , 2017 ) , a type I transmembrane protein that was identified as an N-BP target in a previous work ( Surface et al . , unpublished ) and our zoledronate CRISPRi screen ( Figure 1—figure supplement 1E ) , suggesting SLC37A3 and ATRAID might be functionally related . To investigate the roles of SLC37A3 and ATRAID in the action of N-BPs , we generated SLC37A3-knockout ( SLC37A3KO ) and ATRAID-knockout ( ATRAIDKO ) cells in K562 cells , human embryo kidney ( HEK ) 293 T cells and murine macrophage-like RAW 264 . 7 cells ( Figure 2—figure supplement 1A–E ) ( Surface et al . , unpublished; Ran et al . , 2013 ) , with K562 and HEK 293 T cells serving as human cell models that represent distinct lineages , and RAW 264 . 7 macrophages as a mouse cell model that can be differentiated into mature osteoclasts ( Collin-Osdoby and Osdoby , 2012 ) . Consistent with the CRISPRi screen and our previous results , knockout of SLC37A3 and ATRAID in all cell types conferred resistance to ALN ( Figure 2A–C ) ( Surface et al . , unpublished ) . Similarly , mature osteoclasts differentiated from knockout RAW cells are more resistant to ALN compared to those differentiated from wild-type cells ( Figure 2D and Figure 2—figure supplement 2A ) ( Surface et al . , unpublished ) . We also measured the reduction in protein prenylation as a readout of N-BP toxicity and verified that ALN treatment had significantly less of an effect on protein prenylation in SLC37A3KO and ATRAIDKO cells compared to wild-type cells ( Figure 2E ) . Complementation with epitope-tagged SLC37A3 or either isoform of ATRAID ( a short isoform , UniProt Q6UW56-1 , and a long isoform , UniProt Q6UW56-3 ) in SLC37A3KO or ATRAIDKO HEK 293 T cells , respectively , restored sensitivity to ALN ( Figure 2—figure supplement 2B–D ) , confirming that the resistance to ALN observed in the knockout cells is indeed caused by the lack of SLC37A3 or ATRAID expression , and that the epitope-tagged versions of the two proteins are functional . The knockout cells are also resistant to N-BPs other than ALN ( Figure 2—figure supplement 2E ) but not to non-nitrogen-containing bisphosphonates ( non-N-BPs ) , which do not target FDPS ( Figure 2—figure supplement 2F ) ( Drake et al . , 2008 ) . Interestingly , knockout of SLC37A3 and ATRAID did not protect cells from lovastatin ( LOV ) , a statin drug that also targets the mevalonate pathway ( Figure 2—figure supplement 2G–I ) ( Tiwari and Khokhar , 2014 ) . The distinctive responses of SLC37A3KO and ATRAIDKO cells to N-BPs , non-N-BPs , and LOV indicate that the roles of SLC37A3 and ATRAID in the mechanism of action of N-BPs are not related to the mevalonate pathway but are instead specific to N-BPs . As knockout of SLC37A3 and ATRAID in different cell types conferred similar responses , we focused on HEK 293 T cells for further studies as they host various tools for molecular biology . To probe the epistatic relationship between the two genes , we generated double knockout ( ATRAIDKO; SLC37A3KO , abbreviated as KO2 ) cells ( Figure 2—figure supplement 1F ) . We observed that SLC37A3KO cells are more resistant to ALN compared to ATRAIDKO cells , and the knockout of ATRAID in addition to SLC37A3 did not further protect cells from the drug ( Figure 2F and Figure 2—figure supplement 2J ) , indicating that ATRAID and SLC37A3 are indeed functionally related and that SLC37A3 is epistatic to ATRAID . Next , we investigated the mechanism underlying the observed functional relationship between ATRAID and SLC37A3 . As protein localization can often provide clues to protein function and functionally linked proteins frequently share subcellular distribution patterns , we expressed functional , epitope-tagged SLC37A3 and ATRAID ( Figure 2—figure supplement 2B–C ) and characterized their localization with immunofluorescence ( IF ) . We confirmed that epitope-tagged SLC37A3 and ATRAID are not over-expressed ( Figure 3—figure supplement 1A ) . We observed that SLC37A3 co-localizes with LAMP2 , a lysosomal marker , but not with Na+/K+-ATPase or EEA1 , which mark the plasma membrane and early endosomes , respectively ( Figure 3A , B and Figure 3—figure supplement 1C ) . Consistent with a previous report ( Ding et al . , 2015 ) , both isoforms of ATRAID also predominantly localize to lysosomes but not to the plasma membrane or early endosomes ( Figure 3C–D and Figure 3—figure supplement 1D , F–H ) . When we co-expressed functional and epitope-tagged SLC37A3 and ATRAID ( Figure 3—figure supplement 1A–B ) , we observed that both isoforms of ATRAID predominantly co-localize with SLC37A3 ( Figure 3E and Figure 3—figure supplement 1E ) . To investigate whether the observed co-localization between SLC37A3 and ATRAID represents a physical interaction , we performed reciprocal co-immunoprecipitation ( co-IP ) experiments in HEK 293 T cells over-expressing the two proteins . We detected an interaction between SLC37A3 and ATRAID in both pull-down directions ( Figure 3F ) , confirming that SLC37A3 and ATRAID physically interact , likely forming a lysosomal complex . As it has been reported that the expression levels of certain solute carriers depend on the presence of accessory proteins ( Makrides et al . , 2014 ) , we investigated the possibility that the functional relationship between ATRAID and SLC37A3 is due to the impaired expression of SLC37A3 in the absence of ATRAID . Indeed , when we expressed SLC37A3 in the KO2 cells , we observed a substantial decrease in the protein level of SLC37A3 compared to that when expressed in the SLC37A3KO background ( Figure 3G and Figure 3—figure supplement 2A ) , even though the mRNA levels of SLC37A3 in both backgrounds are similar ( Figure 3—figure supplement 1A ) . The protein level of SLC37A3 was restored by complementation with either isoform of ATRAID ( Figure 3G ) . ( Note that the reduction in SLC37A3 expression in the absence of ATRAID is not clearly observed in Figure 3F due to the over-expression of SLC37A3 . Indeed , in samples analyzed in Figure 3F , SLC37A3 exists predominantly in an un-glycosylated form , suggesting saturation of machineries required for the post-translational processing of SLC37A3 ( compare Figure 3F and G , also see discussion below ) . ) In reciprocity , we also observed reduced protein levels of both isoforms of ATRAID in the KO2 background that cannot be explained by changes in transcript levels ( Figure 3H , Figure 3—figure supplement 2B–C and Figure 3—figure supplement 1A ) . As the translation efficiency of SLC37A3 transcripts is not significantly altered in the absence of ATRAID ( Figure 3—figure supplement 2D–E ) , the decreased SLC37A3 protein level in the KO2 background is likely caused by shortened protein half-life , suggesting that ATRAID and SLC37A3 are mutually dependent for their stability . Intriguingly , the deletion of ATRAID also altered the glycosylation pattern of SLC37A3 . In the absence of ATRAID , the mature , glycosylated population of SLC37A3 ( around 50kD ) became undetectable , whereas a population of un-glycosylated SLC37A3 ( around 40kD ) emerged ( Figure 3G and Figure 3—figure supplement 2F ) . Moreover , in cells overexpressing SLC37A3 , although a significant proportion of SLC37A3 remained un-glycosylated , only the glycosylated population of SLC37A3 interacted with ATRAID ( Figure 3—figure supplement 2G ) , suggesting that the interaction with ATRAID is crucial for the expression of correctly modified SLC37A3 . Finally , we explored the mechanism by which the knockout of SLC37A3 and ATRAID conferred resistance to N-BPs . Given the predicted function of SLC37A3 as a transporter , we hypothesized that ATRAID and SLC37A3 transport N-BP molecules across the lipid bilayer to inhibit FDPS . Indeed , we observed that SLC37A3 resides on the membrane of vesicles in which internalized fluorescently-labeled zoledronate ( AF647-ZLN ) accumulates ( Figure 4A ) , supporting its role as a transporter of N-BPs . This hypothesis is also consistent with our finding that the role of SLC37A3 and ATRAID in the mechanism of action of N-BPs is specific to the chemical properties of N-BPs . We implemented radioactive uptake assays to test this hypothesis . When we incubated wild-type , ATRAIDKO and SLC37A3KO cells with radioactive ALN ( 3H-ALN ) and measured total intracellular radioactivity , we observed no significant difference in whole-cell accumulation of radioactivity between the knockouts and wild-type cells ( Figure 4B ) . As N-BP molecules accumulate in SLC37A3-positive vesicles and SLC37A3 localizes to lysosomes , we further hypothesized that N-BP molecules traffic to lysosomes and that SLC37A3 and ATRAID , together as a lysosomal complex , might function to release N-BP molecules from the lumen of lysosomes into the cytosol . This model predicts that the total amount of intracellular 3H-ALN will remain the same in wild-type , SLC37A3KO and ATRAIDKO cells , but 3H-ALN will not be able to exit lysosomes in knockout cells . To detect this potential shift in the subcellular distribution of 3H-ALN in knockout cells , we used digitonin to selectively permeabilize the plasma membrane of 3H-ALN treated cells and generated a cytosolic fraction and a membranous fraction , which contained intact membrane-bound organelles ( Figure 4—figure supplement 1A ) ( Liu and Fagotto , 2011 ) . We observed that the distribution of radioactive signal changed from being primarily in the cytosolic fraction in wild-type cells to being predominantly in membranous fractions in SLC37A3KO and ATRAIDKO cells ( Figure 4C ) , suggesting that 3H-ALN cannot be released from membrane-bound organelles in knockout cells . As our model specifically predicts that 3H-ALN will be trapped in lysosomes in the absence of SLC37A3 or ATRAID , we affinity-purified lysosomes from 3H-ALN treated cells ( Figure 4—figure supplement 1B–D ) ( Abu-Remaileh et al . , 2017; Wyant et al . , 2017 ) to assess the lysosomal accumulation of 3H-ALN . As our model predicted , we observed a significant enrichment of 3H-ALN in lysosomes purified from knockout cells compared to those from wild-type cells ( Figure 4D ) . Additionally , consistent with the observation that ATRAID is required for the stable expression of SLC37A3 , ATRAIDKO cells phenocopied SLC37A3KO cells in these uptake assays . Taken together , our results suggest that N-BPs traffic to lysosomes after internalization through endocytosis , and SLC37A3 and ATRAID form a lysosomal transporter complex that releases N-BP molecules from the lumen of lysosomes into the cytosol . In summary , this study elucidates the route by which N-BPs enter the cytosol and inhibit their molecular target . As a recent study has proposed that patients who harbor a genetic variant of GGPS1 might be more prone to the side-effects of N-BP treatment ( Roca-Ayats et al . , 2017 ) , it is possible that patients with variants of SLC37A3 or ATRAID , which are genes crucial for the action of N-BPs , might also exhibit non-canonical responsiveness to the drugs . Therefore , our results may bear significant relevance to the clinical applications of N-BPs . Reagents were obtained from the following sources: antibodies against LAMP2 ( mouse , sc-18822 ) , Ran BP3 ( mouse , sc-373678 ) and Rap 1A ( goat , sc-1482 ) were from Santa Cruz Biotechnology ( Dallas , Texas ) ; antibodies against GAPDH ( rabbit , 2118 ) , V5-tag ( rabbit , 13202 ) and EEA1 ( rabbit , 2411 ) were from Cell Signaling Technology ( Danvers , Massachusetts ) ; antibodies against Na+/K+-ATPase ( mouse , ab7671 ) and Lamin B1 ( rabbit , ab16048 ) were from Abcam ( Cambridge , Massachusetts ) ; antibodies against V5-tag ( mouse , R960-25 ) and HDJ2 ( mouse , MS-225-P0 ) were from ThermoFisher Scientific ( Waltham , Massachusetts ) ; antibodies against Caveolin-1 ( rabbit , C3237 ) and HA-tag ( rat , 11867423001 ) were from Sigma-Aldrich ( Burlington , Massachusetts ) ; antibody against EEA1 ( mouse , 610456 ) was from BD Biosciences ( San Jose , California ) ; antibody against LMAP1 ( mouse , H4A3 ) was from DSYB ( Developmental Studies Hybridoma Bank , Iowa City , Iowa ) ; antibody against Ubiquitin ( mouse , 05–944 ) was from Millipore-Sigma ( Burlington , Massachusetts ) ; alendronate ( A4978 ) , lovastatin ( PHR1285 ) , zoledronate ( SML0223 ) , ibandronate ( I5784 ) , etidronate ( P5248 ) , tiludronate ( T4580 ) , chloroquine ( C6628 ) , puromycin ( P8833 ) , polybrene ( H9268 ) , anti-HA ( A2095 ) and anti-V5 ( A7345 ) agarose affinity matrix , poly-L-lysine solution ( P4707 ) , fibronectin solution ( F0895 ) , Triton X-100 ( T8787 ) , saponin ( 47036 ) , Bovine Serum Albumin ( BSA , A9647 ) and complete protease inhibitor cocktail ( 11836170001 ) were from Sigma-Aldrich; HRP-conjugated anti-rat secondary antibody ( 31470 ) , Alexa 488 and Alexa 647-conjugated secondary antibodies ( A21208 , A32728 , A11034 and A32733 ) , 0 . 1 µm TetraSpeck microspheres ( T7279 ) , SlowFade Diamond mounting medium ( S36968 ) , Halt protease-phosphatase inhibitor cocktail ( 78443 ) , BCA protein assay kit ( 23225 ) , SuperSignal west femto substrate ( 34095 ) , TOPO TA cloning kit ( 450030 ) , TRIzol and TRIzol LS reagents ( 15596018 and 10296028 ) , SuperScript IV ( 18090050 ) , RNase-free Turbo DNase ( AM2238 ) , SUPERase . In RNase Inhibitor ( AM2694 ) , SYBR Green qPCR master mix ( A25742 ) , Hygromycin B ( 10687010 ) , DMEM ( 11965118 ) , RPMI ( 11875093 ) , Fetal Bovine Serum ( FBS , 16000044 ) and Lipofectamine 3000 reagent ( L3000008 ) were from ThermoFisher Scientific; HRP-conjugated anti-mouse and anti-rabbit secondary antibodies ( 1706515 and 1706516 ) were from Bio-Rad ( Hercules , California ) ; PNGase F ( P0704 ) and Endo Hf ( P0703 ) were from New England Biolabs ( NEB , Ipswich , Massachusetts ) ; cell line Nucleofector kit V ( VACA-1003 ) was from Lonza ( Walkersville , Maryland ) ; CellTiter-Glo kit ( G7572 ) was from Promega ( Fitchburg , Wisconsin ) ; mouse RANK ligand ( RANKL , 462-TEC-010 ) was from R and D systems ( Minneapolis , Minnesota ) ; IMDM ( 30–2005 ) was from ATCC ( American Type Culture Collection , Manassas , Virginia ) ; digitonin ( 300410 ) was from Millipore-Sigma; QuickExtract DNA extraction solution ( QE0905T ) was from Epicentre ( Madison , Wisconsin ) ; tritium-labeled alendronate ( MT-1727 ) was from Moravek ( Brea , California ) ; Alexa flour 647 labeled zoledronate ( AF647-ZOL ) was from BioVinc ( Pasadena , California ) ; Janelia flour 549 was a gift from Luke Lavis; pSpCas9 ( BB ) −2A-GFP ( PX458 ) was a gift from Feng Zhang ( Addgene plasmid # 48138 ) ; AAVS1-Puro-PGK1 −3 × FLAG TwinStrep was a gift from Yannick Doyon ( Addgene plasmid # 68375 ) ; pLenti PGK Hygro DEST ( w530-1 ) was a gift from Eric Campeau and Paul Kaufman ( Addgene plasmid # 19066 ) ; psPAX2 and pMD2 . G were gifts from Didier Trono ( Addgene plasmid # 12260 and # 12259 ) . K562 human myeloid leukemia cells , Human Embryo Kidney ( HEK ) 293 T cells and RAW 264 . 7 murine macrophage-like cells were obtained from ATCC . Cell line identities were verified by determining species identity and examining morphology . All cell lines were tested for mycoplasma contamination using ATCC universal mycoplasma detection kit and/or DAPI staining . All cell lines were free of mycoplasma contamination . K562 cells were cultured in RPMI supplemented with 25 mM HEPES , 2 mM L-glutamine , 2 g/L NaHCO3 , 10% FBS and penicillin/streptomycin; HEK 293 T cells and RAW 264 . 7 cells were cultured in DMEM supplemented with 10% FBS and penicillin/streptomycin . All cultures were maintained at 37°C and 5% CO2 . K562 cell line generation , genome-scale library design and cloning , virus production , and bioinformatic analysis were conducted as previously described ( Gilbert et al . , 2014; Jost et al . , 2017 ) . In summary , K562 cells stably expressing dCas9-KRAB were transduced in duplicate with the v1 CRISPRi sgRNA library to achieve ~30% infection to ensure no more than one viral integration event per cell . Two days after transduction , cells were selected with 0 . 75 µg/mL of puromycin for 2 days and then kept with fresh puromycin-free medium for 2 days for recovery . At this point ( t0 ) , 250 million cells ( ensuring a minimum of 1000 × library coverage ) were harvested from each replicate and the remaining cells in each replicate were split into two populations for untreated growth and alendronate-treated growth . For alendronate treatment , cells were cultured in medium containing 250 µM alendronate for 24 hr , spun down to remove the drug and re-suspended in fresh medium . Cells were cultured for another 13 days to allow the untreated population to double seven more times than the alendronate-treated population . 250 million cells were then harvest from each group ( two replicates for each condition , four groups in total ) . Cells were maintained at a density of 500 , 000 to 1 , 000 , 000 cells/mL in 2-liter cultures to ensure a library coverage of at least 1000 cells per sgRNA during the entire screening period . Genomic DNA was collected from all harvested samples and the genomic regions containing the inserted sgRNAs were amplified for 20 cycles by PCR and sequenced at 800 × coverage on Ilumina HiSeq-2500 using custom primers as previously described ( Kampmann et al . , 2013 ) . For data analysis , sequencing reads were aligned to the v1 CRISPRi library sequences , counted , and quantified using the Python-based ScreenProcessing pipeline ( Horlbeck et al . , 2016 ) . Sensitivity phenotypes ( ρ ) were calculated by computing the log2 difference in enrichment of each sgRNA between the treated and untreated samples , subtracting the equivalent median value for all non-targeting sgRNAs , and dividing by the number of population doubling differences between the treated and untreated populations ( Gilbert et al . , 2014; Jost et al . , 2017 ) . Similarly , untreated growth phenotypes ( γ ) were calculated from the untreated endpoint samples and t0 samples , dividing by the total number of doublings of the untreated population . Phenotypes from sgRNAs targeting the same gene were collapsed into a single sensitivity phenotype for each gene using the average of the top three scoring sgRNAs ( by absolute value ) and assigned a P-value using the Mann-Whitney test of all sgRNAs targeting the same gene compared to the non-targeting controls ( Supplementary file 1 ) . For genes with multiple independent transcription start sites ( TSSs ) targeted by the sgRNA libraries , phenotypes and P-values were calculated independently for each TSS and then collapsed to a single score by selecting the TSS with the lowest Mann-Whitney P-value . Replicate-averaged sensitivity phenotype and P-value for each gene were obtained by performing the above computations on the average of sgRNA phenotype values calculated from both replicates and used for illustration . The CRISPRi screen was performed only once . Genome editing experiments were designed based on an established protocol ( Ran et al . , 2013 ) . For the human ATRAID locus , one sgRNA targeting exon three and another targeting exon five were used to act simultaneously and remove part of exon 3 , the entire exon four and part of exon 5 . For the human SLC37A3 locus , one sgRNA targeting intron five and another targeting intron six were used to act simultaneously and remove the entire exon 6 , which contains 146 bp of CDS . For murine SLC37A3 locus , two sgRNAs targeting exon two were designed to cause microdeletions and frameshifts in the CDS . sgRNAs were cloned into PX458 for co-expression with Cas9 . sgRNA_human_ATRAID_exon3_1: GCCTGATGAAAGTTTGGACC sgRNA_human_ATRAID_exon3_2: CCCTGGTCCAAACTTTCATC sgRNA_human_ATRAID_exon5: GTCCTGGAGGAATTAATGCC sgRNA_human_SLC37A3_intron5_1: GTGTGAGTGTATCCTTCACG sgRNA_human_SLC37A3_intron5_2: GCCAGTGCCTGTAAGTCACG sgRNA_human_SLC37A3_intron6: GTAGCAAGTCAGAGTTGTTCA sgRNA_mouse_ SLC37A3_exon2_1: TCTCTGCAAAAATCGTGGCC sgRNA_mouse_ SLC37A3_exon2_2: TGTTCCTGCTCACGTTCTTC For K562 and HEK 293 T cells , on day one , 500 , 000 cells were seeded onto a 6 cm dish . 24 hr later , cells were transfected with 1 . 25 µg of PX458 construct for each sgRNA ( total of 2 . 5 µg DNA ) using the Lipofectamine 3000 reagent according to manufacturer instructions . Medium was replenished on the following day . 48 hr after transfection , cells were trypsinized , filtered through a 50 µm strainer into ice-cold FACS buffer ( PBS containing 1% FBS ) and sorted with a flow cytometer for GFP-positive cells . Single GFP-positive cells were seeded into the wells of a 96-well plate containing 150 µL of DMEM in each well . 12 to 14 days later , each surviving clone was split into two wells , with one well saved for expansion and the other for genotyping . For genotyping , genomic DNA was extracted from each confluent clone with QuickExtract solution and used to perform genomic PCR using a pair of primers flanking the target region . Successful deletion events were identified by a significant decrease in the size of PCR products . Clones with deletions on all alleles of the target gene were further expanded and stored . Genomic PCR products from clones with homozygous deletions were inserted into TOPO-TA cloning vectors and sequenced to identify clones that have frameshifts on all alleles of the target gene . ATRAIDKO; SLC37A3KO cells are generated by knocking out ATRAID in SLC37A3KO cells . human_ATRAID_KO_verification_forward: CTGAAAAGGGGGTTGTGTAGTCAA human_ATRAID_KO_verification_reverse: GGGTTATAGCCCCAGAACTCTGAA human_SLC37A3_KO_verification_forward: GTTGGAGGGCTGATAGCTTAATG human_SLC37A3_KO_verification_reverse: AAAAATTGAGACCTCCTGCCTTG For RAW 264 . 7 cells , 2 million cells were electroporated with 2 µg of PX458 harboring the sgRNAs of interest ( 1 µg per sgRNA construct ) in 100 µL of Nucleofector solution V on a Nucleofector device ( Lonza ) using program D-023 . Cells were allowed to recover for two days before they were single-cell sorted into a 96-well plate . Clonal expansion and genotyping were then performed as described above for HEK 293 T cells . Mouse_SLC37A3_KO_verification_forward: CCCACAGGCAGAAGACAAGA Mouse_ SLC37A3_KO_verification_reverse: TGTAACTCAGTCACTGGGAGGA Differentiation of RAW cells to osteoclasts was achieved following an established protocol ( Collin-Osdoby and Osdoby , 2012 ) . Briefly , RAW 264 . 7 cells were seeded into a 24-well plate and treated with 35 ng/mL RANKL for 6 days to reach a large , multi-nucleated morphology that is characteristic of osteoclasts . For experiments with alendronate , the drug was added at the indicated concentrations 48 hr prior to harvesting . RAW cell differentiation was repeated independently for three times . On day one , cells were seeded at 8000 cells per well for K562 and HEK 293 T cells or 4000 cells per well for RAW 264 . 7 cells in a 96-well plate and treated with a series of doses of the desired drug . Three wells were prepared for each combination of cell line and drug concentration . Forty-eight hours later , the total cellular ATP level in each well was measured using the CellTiter-Glo luminescent assay following manufacturer instructions . Relative ATP levels were then plotted as percentages of ATP levels in the untreated samples and interpreted as a proxy for cell viability under drug treatments . Each viability curve was repeated independently for at least two times . V5-tagged ATRAID and HA-tagged SLC37A3 were constructed by appending codon-optimized sequences of V5 ( sequence: GGA AAG CCC ATA CCG AAT CCT CTC CTT GGG TTG GAT AGC ACT ) and HA tags ( sequence: TAC CCC TAT GAT GTT CCT GAT TAC GCG ) to the C-termini of ATRAID ( both the short isoform , 229 a . a . , and the long isoform , 284 a . a . ) and SLC37A3 CDSs , respectively . A GGGGSGGGGS flexible linker ( sequence: GGT GGA GGG GGA AGT GGC GGA GGA GGT TCA ) was added between each CDS and its epitope tag . To generate HEK 293 T cells that stably express sATRAID-V5 , lATRAID-V5 or SLC37A3-HA at sub-endogenous levels , CDSs were cloned into pPGK-AAVS-Puro ( derived from Addgene # 68375 ) and inserted into human AAVS1 locus via CRISPR-Cas9 mediated homologous recombination . For genome integration , a 40% confluent dish of ATRAIDKO , SLC37A3KO or KO2 HEK 293 T cells was transfected with 1 . 5 µg of pPGK-AAVS-Puro carrying the desired CDS and 1 µg of PX458 expressing sgRNA_AAVS1 ( see below ) using Lipofectamine 3000 reagent . Culture medium was replenished on the next day . 48 hr after transfection , cells were re-plated in a 10 cm dish in DMEM containing 2 µg/mL puromycin to select for successfully edited cells . Puromycin selection was continued for a week , with medium replenished every other day . After selection , expression levels of epitope-tagged proteins were analyzed by immunoblotting and RT-qPCR as described below . sgRNA_AAVS1: GGGGCCACTAGGGACAGGAT To generate HKE 293 T cells that co-express ATRAID-V5 and SLC37A3-HA at near-endogenous levels , the CDS of sATRAID-V5 or lATRAID-V5 was cloned into w530-1 and transduced into KO2 +SLC37A3-HA cells . To generate the lentiviruses , on day one 150 , 000 HEK 293 T cells were seeded into a well of a 6-well plate . 24 hr later , cells were transfected with 2 µg w530-1 construct , 0 . 8 µg psPAX2 and 0 . 4 µg pMD2 . G using Lipofectamine 3000 reagent . On the next day culture medium was replenished . 48 hr after transfection , supernatant from the culture was collected and filtered through a 0 . 45 µm filter . 150 µL of the viral medium was added to a 40% confluent well of KO2 +SLC37A3-HA cells in a 6-well plate in the presence of 8 µg/mL Polybrene to a total volume of 2 mL . 48 hr later cells were re-plated into a 6 cm dish in DMEM containing 200 µg/mL hygromycin . Selection was continued for one week , with medium replenished every other day . After selection , expression levels of epitope-tagged proteins were analyzed by immunoblotting and RT-qPCR as described below . Cells were washed once with PBS and lysed on ice by scraping into ice-cold RIPA buffer supplemented with Halt protease-phosphatase inhibitor . The lysate was then cleared by centrifuging at 20 , 000 × g , 4°C for 15 min . Protein concentration in the lysate was determined with BCA protein assays . Loading samples were prepared by mixing lysates with SDS loading buffer and incubating at 37°C for 15 min . ( Higher denaturing temperatures may cause SLC37A3 to aggregate and prevent it from entering the gel . ) SDS-PAGE electrophoresis and protein transfer onto nitrocellulose membranes were performed according to standard protocols . Membranes were blocked in TBST containing 2 . 5% BSA and 2 . 5% skim milk for 1 hr and incubated with primary antibodies overnight in TBST containing 5% BSA . ( Primary antibody concentrations: α-HA , 100 ng/mL; α-GAPDH and α-HDJ-2 , 250 ng/mL; α-V5 ( mouse ) , α-LAMP1 , and α-Lamin B1 , 500 ng/mL; α-Rap 1A , α-Ran BP3 , α-EEA1 ( rabbit ) and α-Caveolin-1 , 1 µg/mL . ) Membranes were then washed 3 × 5 min in TBST and incubated with either HRP-conjugated or fluorophore-conjugated secondary antibodies ( 1:2000 dilution for all secondary antibodies ) in TBST containing 5% skim milk for 1 hr . Membranes were then washed again for 3 × 5 min in TBST and visualized with either SuperSignal substrate or a Typhoon scanner . Each blot was repeated independently for two times . RNA was extracted from near-confluent 3 cm dishes using TRIzol reagent following manufacturer instructions . Purified RNA was reverse transcribed using SuperScript IV and oligo d ( T ) 20 following manufacturer instructions . qPCR reactions were performed with SYBR Green qPCR master mix and primers listed below using a CFX96 or CFX384 Real-Time PCR machine ( Bio-Rad ) . Ct values were calculated for each transcript using triplicate measurements , and relative mRNA levels were determined for each gene using TBP ( TATA binding protein ) and RPLP1 ( 60S acidic ribosomal protein P1 ) as loading references for human transcripts and Actb ( β actin ) and Rplp0 ( 60S acidic ribosomal protein P0 ) as loading references for mouse transcripts . Each qPCR measurement was repeated independently for two times . human_TBP_forward: ATAAGAGAGCCACGAACCACG human_TBP_reverse: TGCCAGTCTGGACTGTTCTTC human_RPLP1_forward: AGCCTCATCTGCAATGTAGGG human_ RPLP1_reverse: TCAGACTCCTCGGATTCTTCTTT human_ATRAID_forward: CAGAAGGGCACCATCTTGGG human_ATRAID_reverse: ACCTTTGAGGGGGTTTGCTT human_SLC37A3_forward: GCTGCCTGTGGATTGTGAAC human_SLC37A3_reverse: AAAATGTTGCCCACCGAAGC mouse_Actb_forward: TGTCGAGTCGCGTCCA mouse_Actb_reverse: ATGCCGGAGCCGTTGTC mouse_Rplp0_forward: TGCTCGACATCACAGAGCAG mouse_Rplp0_reverse: ACGCGCTTGTACCCATTGAT mouse_Ctsk_forward: CCTTCCAATACGTGCAGCAG mouse_Ctsk_reverse: CATTTAGCTGCCTTTGCCGT mouse_Rank_forward: GCAGCTCAACAAGGATACGG mouse_Rank_reverse: TAGCTTTCCAAGGAGGGTGC mouse_Trap_forward: AAGAGATCGCCAGAACCGTG mouse_Trap_reverse: CGTCCTCAAAGGTCTCCTGG On day one , coverslips are placed into wells of 6-well plates and coated for 1 hr at 37°C with 0 . 01% Poly-L-lysine solution supplemented with 10 µg/mL fibronectin . Coating solution was aspirated and 120 , 000 HEK 293 T cells were plated into each well . Twenty-four hours later , the coverslips were rinsed two times with PBS++ ( PBS containing calcium and magnesium ) and fixed in 4% formaldehyde in PBS++ for 15 min at room temperature . The coverslips were rinsed three times with PBS++ and cells were permeabilized/blocked with 0 . 1% Saponin and 2% BSA in PBS++ for 30 min . After rinsing briefly with PBS++ , the coverslips were transferred to a humidity chamber and incubated overnight at 4°C in PBS++ containing 2% BSA and desired primary antibodies . ( Antibody concentrations: α-HA and α-EEA1 ( mouse ) , 500 ng/mL; α-LAMP2 , 1 µg/mL; α-V5 ( rabbit ) , 2 µg/mL; α-Na+/K+-ATPase , 5 µg/mL . ) On the next day , the coverslips were washed 3 × 5 min in PBS++ and incubated for 1 hr with Alexa-488 and Alexa-647-conjugated secondary antibodies diluted in PBS++ containing 2% BSA . The coverslips were again washed 4 × 5 min in PBS++ and mounted onto slides in SlowFade Diamond anti-fade mountant supplemented with DAPI , and the edges of the coverslips were sealed with nail polish . ( We do not recommend curing mountants such as ProLong Gold , as they distort/flatten the samples . We also do not recommend Vectashield if Alexa 647 is chosen as a fluorophore . ) Images were acquired on a Zeiss AxiObserver Z . 1 microscope equipped with a Zeiss α Plan-APO 100× ( NA1 . 46 ) oil-immersion objective , a 561 nm Hamamatsu Gemini 2C beam splitter , dual Hamamatsu Image-EM 1K EM-CCD cameras and CoolLED P4000 light sources . The ZEN blue software package ( Zeiss ) was used to control the hardware and image acquisition conditions . An LED-DA/FI/TR/Cy5-A quadruple band-pass filter set was installed in the microscope turret and used for all channels . Excitation wavelength ( EW ) , excitation filters ( ExF ) and emission filters ( EmF ) used for each channel are: DAPI channel , EW 385 nm , ExF 387/11 nm , EmF 525/45 nm; Alexa 488 channel , EW 470 nm , ExF 473/10 nm , EmF 525/45 nm; Alexa 647 channel , EW 635 nm , ExF 635/6 nm , EmF 682/40 nm . Images were sampled above the Nyquist limit with a voxel size of 81 . 25 nm ×81 . 25 nm×220 nm . Each IF sample was independently prepared and imaged for two times . With each batch of samples , one slide of 0 . 1 µm TetraSpeck microspheres immersed in SlowFade Diamond was also prepared and imaged with the same settings . The point-spread function ( PSF ) of each channel was distilled from the microsphere images and used to deconvolute sample images using Huygens Professional deconvolution software ( Scientific Volume Imaging ) . Chromatic aberration ( shift , rotation and scaling ) were estimated by correlating different channels of the microsphere images . The estimated aberration parameters were then used to align different channels of deconvoluted images using Huygens Profession software . Only images that have been deconvoluted and aligned were used for analysis . Images displayed in figures are representative single Z-slices . 150 , 000 HEK 293 T cells expressing SLC37A3-Halo were seeded into a 35 mm glass bottom tissue culture dish ( MatTek ) coated with poly-L-lysine and fibronectin ( see the previous section ) . Cells were cultured in DMEM containing 500 nM AF647-ZLN for 16 hr . Cells were washed once in PBS and stained with DMEM containing 100 nM JF549 for 30 min , washed three times with PBS , incubated in fresh DMEM for 30 min , and washed again in PBS for three times and finally cultured in DMEM without phenol Red to facilitate live cell imaging . The glass bottom dish was then placed in an incubation chamber mounted onto a microscope , and images were taken and analyzed as described in the previous section . 10 cm dishes of near-confluent KO2 HEK 293 T cells over-expressing the proteins indicated in Figure 3Ff were lysed by scraping on ice into 1 mL of ice-cold lysis buffer ( 1% Trition X-100 , 20 mM Tris-HCl pH 8 . 0 , 150 mM NaCl and 2 mM EDTA in ddH2O ) supplemented with cOmplete protease inhibitors . The lysates were cleared as described above . 10 µL of each cleared lysate was saved for input analysis . The rest of the lysates were transferred to Eppendorf tubes containing 20 µL ( settled volume ) of anti-HA or anti-V5 agarose beads that had been blocked overnight in lysis buffer containing 2% BSA . Lysates were incubated with the beads for 90 min at 4°C , washed three times in low-salt wash buffer ( 0 . 1% Triton X-100 , 10 mM Tris-HCl pH 7 . 5 , 150 mM NaCl and 1 mM EDTA in ddH2O ) supplemented with protease inhibitors and three times in high-salt wash buffer ( 0 . 1% Triton X-100 , 10 mM Tris-HCl pH 7 . 5 , 300 mM NaCl and 1 mM EDTA in ddH2O ) supplemented with protease inhibitors , and eluted in 100 µL elution buffer ( 1% Triton X-100 , 10 mM Tris-HCl pH 7 , 5 and 150 mM NaCl ) containing 2 mg/mL HA or V5 peptide . Inputs and eluates were then analyzed by immunoblotting as described . The co-IP experiment was repeated independently for two times . Protein lysates and IP eluates were prepared as described above . Once obtained , samples ( lysates containing 20–40 µg of total protein or IP eluates that correspond to 60 µg of total protein input ) were either left untreated , or treated with peptide-N-Glycosidase F ( PNGase F ) or endo-glycosidase H ( Endo H ) following manufacturer instructions ( NEB ) using non-denaturing conditions ( without adding SDS or boiling for denaturation ) . The reactions were incubated at 37°C overnight and subsequently analyzed by SDS-PAGE electrophoresis and immunboblotting . For each cell line of interest , one near-confluent 15 cm dish of cells was lysed on ice by scraping into 300 µL of ice-cold lysis buffer ( 1% Triton X-100 , 20 mM Tris-HCl pH 7 . 0 , 20 mM Tris-HCl pH 8 . 0 , 15 mM MgCl2 , 150 mM NaCl , 5 mM CaCl2 , 25 U/mL RNase-free Turbo DNase , 500 U/mL SUPERase . In RNase Inhibitor and 0 . 1 mg/mL cycloheximide in RNase-free water ) supplemented with protease inhibitors . Lysates were homogenized by passing through a 22-gauge needle for 10 times and then cleared as described above . 10–50% Sucrose gradients were composed by mixing 6 mL of 10% Sucrose solution ( 20 mM Tris-HCl pH 7 . 0 , 20 mM Tris-HCl pH 8 . 0 , 15 mM MgCl2 , 150 mM NaCl , 0 . 1 mg/mL cycloheximide and 10% ( w/v ) sucrose in RNase-free water ) with 6 mL of 50% sucrose solution ( 20 mM Tris-HCl pH 7 . 0 , 20 mM Tris-HCl pH 8 . 0 , 15 mM MgCl2 , 150 mM NaCl , 0 . 1 mg/mL cycloheximide and 50% ( w/v ) sucrose in RNase-free water ) on a Gradient Master device ( BioComp ) . 450 µL of each cleared lysate was loaded onto a 10–50% sucrose gradient and centrifuged at 35 , 000 rpm , 4°C for 2 . 5 hr using a SW 41 Ti rotor ( Beckman ) . At the same time , 50 µL of each cleared lysate was used to extract total RNA . The density-separated RNA samples were loaded onto a Gradient Master device , collected from the top ( low density ) to the bottom ( high density ) and analyzed by measuring A254 . Fractionation of samples was carried out manually by visually determining the boundaries between desired fractions during sample analysis . RNA was then extracted from each fraction using TRIzol LS reagent following manufacturer instructions . Levels of SLC37A3 transcripts in collected fractions were then measured by RT-qPCR as described above and the distribution of SLC37A3 transcript in the differentially translated fractions was plotted and interpreted as a proxy for translation efficiency . The polysome profiling experiment was repeated independently for two times . For whole-cell uptake assays , on day one 1 . 5 million WT , ATRAIDKO or SLC37A3KO HEK 293 T cells were seeded into each poly-L-lysine coated 6 cm dish in DMEM containing 1 µCi/mL 3H-alendronate . 24 hr later , cells were washed three times with PBS++ , each time thoroughly removing the PBS . Cells were then trypsinized , pelleted and re-suspended in PBS . Cell density in each sample was measured on a Multisizer 3 Coulter Counter ( Beckman Coulter ) . Total intracellular radioactivity in 2 million cells from each sample was measured by scintillation counting on a LS6500 Liquid Scintillation Counter ( Beckman Coulter ) . Three plates were used as triplicates for each cell line . Fractionation-based uptake assays were designed based on an established protocol ( Liu and Fagotto , 2011 ) . On day one 6 million WT , ATRAIDKO or SLC37A3KO HEK 293 T cells were seeded into each Poly-L-lysine coated 10 cm dish in DMEM containing 1 µCi/mL 3H-alendronate . 24 hr later , dishes were transferred to a cold room and washed three times with 5 mL of PBS++ , each time thoroughly removing the PBS . 3 mL of permeabilization buffer ( 20 mM HEPES pH 7 . 4 , 150 mM NaCl , 0 . 2 mM EDTA , 2 mM MgCl2 , 2 mM DTT and 42 µg/mL digitonin in ddH2O ) supplemented with protease inhibitors was then added to each dish , and the dishes were incubated on an orbital shaker ( 100 rpm ) at 4°C for 10 min . 2 . 5 mL supernatant was collected from each dish as the cytosolic fraction , the rest of the supernatant was thoroughly aspirated . The permeabilized cells were then washed again in 5 mL of PBS++ and lysed in 500 µL of RIPA buffer supplemented with protease inhibitors . The RIPA lysates were cleared as described above and collected as membranous fractions . Radioactivity in cytosolic and membranous fractions was measured by scintillation counting and normalized to protein concentrations measured by BCA protein assays . Two plates were used as duplicates for each cell line . Lysosome purification-based uptake assays were adapted from established protocols ( Abu-Remaileh et al . , 2017; Wyant et al . , 2017 ) . Three 15 cm dishes of ~35 million WT , ATRAIDKO or SLC37A3KO HEK 293 T cells expressing Tmem192−3 × HA were used for each experiment . Cells were incubated in RPMI containing 1 µCi/mL 3H-alendronate and incubated for 3 hr , quickly rinsed twice with PBS , scraped in 1 mL of KPBS ( 136 mM KCl , 10 mM KH2PO4 , pH 7 . 25 was adjusted with KOH ) , pelleted by centrifuging at 1000 × g , 4°C for 2 min and re-suspended in 950 µL KPBS . The cell suspension was then homogenized with 20 strokes of a 2 mL dounce homogenizer and centrifuged at 1000 × g , 4°C for 2 min . The supernatant was then incubated with 150 µL of KPBS-prewashed anti-HA magnetic beads for 3 min . The beads were then gently washed three times with KPBS and then re-suspended in 50 µL ice-chilled extraction buffer ( 80% methanol and 20% ddH2O ) and incubated for 5 min on ice . The lysosome extract was then centrifuged at 1000 × g , 4°C for 2 min . Radioactivity in the supernatant was then measured by scintillation counting . All uptake assays were repeated independently for two times . P-values assigned to individual genes in the CRISPRi screen were calculated using non-parametric Mann-Whitney U test ( 11 , 23 ) . P-values in Figure 4b were calculated using unpaired two-way ANOVA test . P-values in Figure 4c were calculated using two-tailed unpaired t-test assuming equal variance . Degrees of freedom , F values and t values are reported in figure legends . The data that support the findings in this study are available within the paper and its supplementary files .

As some people age , their bones may become weak , brittle , and break easily . This condition is called osteoporosis . To treat osteoporosis , doctors often prescribe drugs called nitrogen-containing bisphosphonates ( NBPs ) . These drugs destroy cells called osteoclasts , which break down bone . This helps restore bone mass . To kill osteoclasts , the drugs must enter these cells . First , they must pass through an oily protective layer called a membrane . It is not completely clear how NBPs , which prefer to stay in water-like environments , can cross this oily membrane and enter osteoclasts . Understanding how NBPs cross the membrane is important to ensure the drugs work effectively . If NBPs do not efficiently cross the membrane , they will not work properly and may cause harmful side effects . Many patients who take NBPs suffer from side effects such as abnormal fractures . Now , Yu et al . show that two proteins help NBPs cross the membrane . In the experiments , proteins were removed from human cancer cells one at a time using a technique called CRISPRi . CRISPRi enabled the researchers to systematically turn off the genes for each protein and track what affect this had on the NBPs’ ability to cross the membrane . When one of the two genes called SLC37A3 and ATRAID was turned off , NBPs could not get into cells . The protein produced by the SLC37A3 gene opens a gate in the cell membrane allowing NBPs to enter osteoclasts . The protein made by the ATRAID gene helps this gate protein , and without it , the SLC37A3 proteins are unstable and NBPs cannot enter . Some people have variations of the SLC37A3 and ATRAID genes . Testing whether these genetic variations may alter NBPs’ ability to cross the membrane of osteoclasts in mice , might one day help physicians predict which patients with have side effects .

2018

Identification of a transporter complex responsible for the cytosolic entry of nitrogen-containing bisphosphonates

Mammalian heart development requires precise allocation of cardiac progenitors . The existence of a multipotent progenitor for all anatomic and cellular components of the heart has been predicted but its identity and contribution to the two cardiac progenitor ‘fields’ has remained undefined . Here we show , using clonal genetic fate mapping , that Mesp1+ cells in gastrulating mesoderm are rapidly specified into committed cardiac precursors fated for distinct anatomic regions of the heart . We identify Smarcd3 as a marker of early specified cardiac precursors and identify within these precursors a compartment boundary at the future junction of the left and right ventricles that arises prior to morphogenesis . Our studies define the timing and hierarchy of cardiac progenitor specification and demonstrate that the cellular and anatomical fate of mesoderm-derived cardiac cells is specified very early . These findings will be important to understand the basis of congenital heart defects and to derive cardiac regeneration strategies . Mammalian heart development involves the allocation of cardiac progenitors in a discrete spatial and temporal order ( Evans et al . , 2010; Bruneau , 2013 ) . Understanding the identity and regulation of these progenitors is critical to understanding the origins of congenital heart defects and may lead to novel cell-based regenerative therapies for heart disease ( Bruneau , 2008; Xin et al . , 2013 ) . The existence of an early and specific multipotent progenitor for all anatomic and cellular components of the heart has been predicted ( Parameswaran and Tam 1995; Tam et al . , 1997; Meilhac et al . , 2004b; Kinder et al . , 2001 ) , but the identity of this progenitor and when it arises in embryonic development has remained undefined . At least two sets of molecularly and morphologically distinct cardiac precursors have been identified in the mammalian embryo , referred to as the first and second heart fields; these populations contribute to distinct anatomical structures within the heart ( Evans et al . , 2010; Buckingham et al . , 2005 ) . Separation of the left and right ventricles is dependent on a single structure , the interventricular septum ( IVS ) and it has been postulated that the IVS myocardium has a dual contribution from these two heart fields ( Bruneau et al . , 1999; Takeuchi et al . , 2003 ) . The existence of these two cardiac progenitor ‘fields’ raises the question of when cardiac precursors are allocated to these populations and their contributions to mature structures in the heart , such as the IVS . Previous studies have suggested that a population of early mesoderm expressing the transcription factor Mesp1 precedes the establishment of the anatomically and molecularly distinct ‘heart fields’ ( Saga et al . , 1996 , 1999; Bondue et al . , 2008; Lindsley et al . , 2008 ) that ultimately will differentially populate the great vessels , RV and atria or the LV and atria . However it is clear that Mesp1+ cells can also contribute to a broad range of mesodermal derivatives that include , but are not restricted to , the developing heart ( Chan et al . , 2013; Yoshida et al . , 2008 ) . A population of mesoderm labeled by Eomesodermin has also been shown to contribute to the developing heart , but again these cells have broad contributions in the embryo ( Arnold and Robertson , 2009 ) . Retrospective lineage analysis supports the distinct origins of segments of the heart from individual precursor pools ( Meilhac et al . , 2003; Buckingham et al . , 2005; Meilhac et al . , 2004b ) , but several questions remain regarding the timing and molecular progression of cardiac specification ( Meilhac et al . , 2004b ) . For example , do early mesodermal cells become ‘locked into’ a cardiac fate early on and when do they become ‘assigned’ to an anatomical location ? Is there a multipotent , specified cardiac progenitor that anticipates the currently understood heart fields ? Here we show that early cardiac progenitors are assigned to a specific developmental path prior to or shortly after the initiation of gastrulation . We identify a population of specified cardiac precursors arising from these mesodermal progenitors that express the chromatin remodeling factor Smarcd3 prior to the onset of expression of known cardiac progenitor markers ( Nkx2-5 , Isl1 , and Tbx5 ) . Clonal labeling of early cardiac precursors highlights the heterogeneity among Mesp1+ progenitors , including the existence of precursors that are restricted in their anatomical contribution , especially to distinct ventricular chambers . Finally , inducible genetic marking of early Tbx5+ and Mef2c-AHF + populations highlights this early segregation of cardiac progenitors and suggests that the compartment boundary that exists between the right and left ventricles arises from an early clonal boundary , prior to the onset of septum morphogenesis . Overall our findings delineate the progression and molecular identity of cardiac precursors in the early mouse embryo . In reassessing the in vivo differentiation potential of Mesp1+ cells , we find that this population contributes broadly to several mesodermal derivatives , ( Figure 1A ) , consistent with other reports ( Yoshida et al . , 2008 ) . We reasoned that among this diverse mesodermal population , a more specific population destined for the cardiac lineage exists . To test this model , we performed in vivo clonal analysis by generating mosaic mice in which very few Mesp1+ cells were labeled at isolated clonal density via the mosaic analysis with double markers ( MADM ) system ( Zong et al . , 2005; Hippenmeyer et al . , 2010 ) ( Figure 1B–C ) . This approach is particularly advantageous because labeling events are rare , labeling is permanent , and one can identify labeled daughter cells ( twin spots ) based on color ( Figure 1—figure supplement 1A ) . We analyzed in fetuses ( E12 . 5-E14 . 5 ) the anatomic distribution and cellular constituents of clones induced by Mesp1Cre ( which is active in mesoderm from ∼E6 . 0 to E7 . 5 ) ( Saga et al . , 1999 ) . While we did not use a conditional Cre allele to control the timing of Cre activity , we confirmed the timing of Cre expression by in situ hybridization ( Figure 1—figure supplement 1B ) . By the late head fold stage ( LHF ) , we see a downregulation of Cre mRNA and localization to the base of the allantois . We see no expression in the area of forming cardiogenic mesoderm . In addition , we counted the number of labeling events in embryos at E8 . 5 and E14 . 5 ( Figure 1—figure supplement 1D–E and Statistical Analysis ) and saw no change in the distribution of labeled clusters , suggesting that no additional recombination events have occurred over this time interval . Finally , a complementary lineage labeling approach using a Mesp1-rtTA transgenic allele ( Lescroart et al . , 2014 ) defines a functional window of Mesp1 activity based on the timing of doxycycline administration between E6 . 25-E7 . 5 , again supporting the narrow timing of Mesp1 activity . 10 . 7554/eLife . 03848 . 003Figure 1 . The first and second heart fields diverge early in gastrulating mesoderm . ( A ) Genetic lineage tracing of Mesp1Cre;ROSAR26R mice reveals widespread labeling of mesodermal derivatives at E10 . 5 , including forelimb ( dotted outline ) . ( B ) Schematic of experimental protocol . Single cells are labeled early in gastrulating mesoderm and progeny of labeled cells observed later in development . ( C ) Example of clonal labeling in E9 . 5 embryo . Four distinct , scattered , clusters of labeled cells are present throughout trunk and neck , including a single yellow clone in ventricle ( 1 , arrowhead ) . ( D ) Ventral view of a second heart field progenitor clone ( red and green twin spots ) with an additional yellow clone in the septal region ( embryo ID MM2 ) . ( E ) Whole mount view revealing an absence of non-cardiac clones elsewhere in the same embryo . ( F ) Ventral view of left ventricle clone ( embryo ID MM15 ) . Red and green twin spots are adjacent to each other . ( G ) Section through red-boxed area of embryo MM15 showing intermingling of red and green twin spots . ( H ) Ventral view of large , yellow septal clone ( embryo ID MM34 ) . ( I ) Whole-mount ventral view of red and green twin-spots in right ventricle ( embryo ID MM27 ) . Boxed regions indicate areas shown in higher magnification sections . ( J ) Section through clone in embryo MM27 reveals green labeled cardiomyocyte ( asterisk ) and red endocardial twin spot ( arrows ) . Note overlap of red clonal labeling with blue PECAM staining . ( K ) Additional section through clone in embryo MM27 . Green twin spot contributes to both cardiomyocytes ( asterisk ) as well as PECAM stained endocardial cells ( arrows ) . ( L ) Whole-mount ventral view and ( M ) section of heart at E14 . 5 with a left ventricle-atria clone ( embryo ID MM26 ) . Note red and green twin spots ( in LV and RA ) in whole-mount view . Sectioning reveals a subset of the green twin spot has remained in the top of the left ventricle ( asterisk ) . ( N ) Whole-mount ventral view of red and green twin spots in out-flow tract from embryo MM24 . ( O ) Section through outflow-tract region reveals red twin spot contributing predominantly to pulmonary artery . Green twin spot contributes to both pulmonary artery and aorta . In addition , green twin spot appears to contribute to both endothelial lining of aorta ( arrows ) as well as cardiomyocytes ( asterisk ) at base of aorta . In ( G , J–K ) white: DAPI stained nuclei . In ( J–K ) blue: PECAM stained endothelial cells . In ( M and O ) blue: phalloidin stained actin . A , anterior; LA , left atrium; LV , left ventricle; OFT , out-flow tract; P , posterior; RA , right atrium; RV , right ventricle . Scale bars: ( G and J–K ) , 100 µm ( D , F , H , I , L , N ) , 200 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03848 . 00310 . 7554/eLife . 03848 . 004Figure 1—figure supplement 1 . Overview of MADM clonal analysis and twin spot labeling . ( A ) Prior to expression of Cre-recombinase , GFP and Tomato fluorophores are inactive . Interchromosomal recombination is induced by Cre recominase , thus restoring activity of GFP and Tomato . Recombination occurring in post-mitotic cells ( G0 ) or in G1 phase of cell cycle generates red and green positive ( yellow ) cells . Recombination occurring after DNA replication ( G2 phase of cell cycle ) generates daughter cells that are uniquely labeled . So-called X-segregation generates daughter cells that are Red and Green and Z-segregation generates daughter cells that are colorless or yellow . ( B ) In situ hybridization for Cre mRNA in a Mesp1Cre embryo at the late head fold stage ( LHF ) . Note expression in allantois ( asterisk ) as well as the allantoic membrane ( arrowheads ) . Expression in the area of the forming cardiogenic mesoderm ( dotted circle ) is largely absent . ( C ) The total number of Mesp1Cre-derived cells was empirically determined by FACS analysis . Five independent Mesp1Cre; RosaTdTomato embryos at E7 . 5 were collected , dissociated , and stained for cardiac-Troponin ( cTNT ) and DAPI . The total number of cells as well as TdTomato and cTNT positive cells were counted and plotted . On average , 1/3 of the total number of cells are Mesp1Cre-derived . No cTNT positive cells were seen at this time point . ( D–E ) Counting the number of labeled clones at two different developmental time-points ( E8 . 5 and E14 . 5 ) reveals a similar distribution of labeling frequency . A stable distribution of clonal labeling between the early and late time points argues against loss of twin spots over time due to apoptosis or ectopic induction after the initial Mesp1Cre induced clonal labeling . ( F ) Ventral view of heart with no clones ( embryo ID MM21 ) . ( G ) Lung and attached esophagus from same specimen with clones in mesenchyme surrounding esophagus and trachea ( arrow ) in mesenchyme of lung parenchyma ( arrowhead ) . E , esophagus; LA , left atrium; LV , left ventricle; OFT , out-flow tract; RA , right atrium; RV , right ventricle . Scale bar: ( B ) 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03848 . 00410 . 7554/eLife . 03848 . 005Figure 1—figure supplement 2 . Epiblast-specific induction of MADM clones . ( A–B ) The epiblast-specific Cre line , Sox2::Cre , is active in epiblast cells at E6 . 5 with little or no activity in extraembryonic tissues when passed through the male germline . Crossing male Cre-containing mice with ROSAAi14 ( TdTomato ) results in broad reporter activity throughout the embryo proper at E10 . 5 . ( C–D ) Left lateral views of E9 . 5 embryos of the indicated genotype reveals many yellow clones through the embryo , including the heart ( asterisk ) ( E–G ) Ventral view of isolated heart from an epiblast-induced MADM clone . A large , dispersed , yellow clone consisting of ∼75–100 cells extends from the OFT through the RV into the LV . Because of the dispersive nature and single color labeling of this particular example , it is unclear if this represents a single recombination event or multiple events . LV , left ventricle; OFT , outflow tract; RV , right ventricle . DOI: http://dx . doi . org/10 . 7554/eLife . 03848 . 00510 . 7554/eLife . 03848 . 006Figure 1—figure supplement 3 . Additional examples of early cardiac progenitor clones . ( A ) Dorsal view of left ventricle progenitor clone in embryo MM15 with both twin spots in the left ventricle . ( B ) Dorsal view of heart from embryo MM35 . Note red twin spot in left ventricle and corresponding green twin spot in right atrium . ( C ) Dorsal view of heart from embryo MM36 . Note red and green twin spots in right ventricle . Additional green labeling is present in the right atrium . ( D ) Ventral view of heart from embryo MM11 reveals red and green twin spots within the right ventricle . Because multiple twin spots are adjacent to one another , it is unclear if clones represent a single progenitor labeling event or more than one progenitor labeling events . ( E ) Ventral view of heart from MM3 showing an isolated yellow clone in ventral part of outflow-tract . ( F ) Ventral view of heart from MM4 showing red and green twin spots on outflow-tract . ( G–H ) Ventral views of hearts from MM38 ( G ) and MM6 ( H ) show red a green twin spots in left and right atria . ( I ) Ventral view of heart from MM1 . Two separate clones are seen in this particular example . The first clone is clearly identifiable as an out flow tract clone and consists of a band of green cells around the ventral part of the outflow-tract and a red twin spot ( not shown ) on dorsal surface of outflow-tract . In addition , a large yellow clone is seen in the left atrium . ( J ) Ventral view of right ventricle clone ( embryo ID MM9 ) . Red and green twin spots are adjacent to each other . ( K ) Section through red-boxed area of embryo MM9 reveals red labeled cardiomyocytes ( asterisks ) and green labeled endocardial cells arrows ) . In ( K ) blue: phalloidin stained actin . ( L ) Section through right ventricle of embryo MM33 . A single green twin spot is present within the right ventricle ( stained with GFP ) . Endothelial cells have been labeled with anti-PECAM ( red ) and cardiomyoctes have been labeled with anti-tropomyosin ( blue ) . Note that cells from this twin spot contribute to both PECAM-positive endocardial cells ( arrows ) as well as tropomyosin positive cardiomyocytes ( asterisks ) . LA , left atrium; LV , left ventricle; OFT , out-flow tract; RA , right atrium; RV , right ventricle . DOI: http://dx . doi . org/10 . 7554/eLife . 03848 . 006 In order to ensure an accurate description of clone locations throughout the embryo , a thorough external examination of embryos was performed followed by removal and , in many instances , immunostaining of dissected hearts for labeled twin spots . Coherent clusters of uniquely colored cells separated by > 100 µm were classified as a twin spot derived from a single labeling event . Because the majority of hearts ( 32 of 38 embryos ) contained three or fewer uniquely colored clusters , determining lineage relationships between and among clusters was straightforward . A subset of clones uniquely label the heart ( Figure 1D–E and Figure 2 ) , demonstrating the existence of an early , cardiac-specific progenitor . We also found specimens with clones of cells in other mesodermal derivatives but with no apparent clones within the heart ( Figure 1—figure supplement 1F–G and Figure 2 ) , conclusively demonstrating that within the population of Mesp1+ mesoderm a dedicated population of cardiac progenitors exists . 10 . 7554/eLife . 03848 . 007Figure 2 . Complete description of all Mesp1Cre-MADM clones examined . All observed clones are detailed here , including cardiac as well as extra-cardiac clones . An exhaustive description of extra-cardiac clones is beyond the scope of the current study and thus only a simple description of the tissue or organ containing a labeled clone is included . Red triangles correspond to red twin spots , green triangles correspond to green twin spots , and yellow triangles correspond to yellow twin spots . DOI: http://dx . doi . org/10 . 7554/eLife . 03848 . 007 To determine if these early cardiac progenitors represent a common precursor for the two heart fields , we analyzed the anatomic distribution of Mesp1Cre-MADM twin spot clones within the heart . In contrast to previous retrospective clonal analysis ( Meilhac et al . , 2004b ) , we did not observe twin spots that populated anatomic structures classically thought to derive from first and second heart field progenitors , for example spanning the left and right ventricles , or contributing to the left ventricle and outflow tract . Rather , we saw twin spots that populated discreet anatomic locations including the left ventricle ( Figure 1F–G and Figure 1—figure supplement 3A–B ) , right ventricle ( Figure 1D and Figure 1—figure supplement 3C–D ) , outflow tract ( Figure 1N–O and Figure 1—figure supplement 3E–F ) , atria-left ventricle ( Figure 1L–M ) , and interventricular septum ( Figure 1H ) . Notably absent from our clonal analysis were twin spots that spanned the right and left ventricles . Based on the total number of clonal observations made ( n = 96 ) , the probability that a common progenitor for the left and right ventricles does not exist , assuming a binomial distribution ( success , or 1 , = the progenitor is observed; failure , or 0 = the progenitor is not observed ) , can be calculated using Jeffreys interval . The upper and lower limits of a 95% confidence interval was calculated such that the upper limit is 0 . 019 ( See Statistical Methods ) . Thus , we can be quite confident given the number of observations made that such a common progenitor does not exist within the Mesp1+ population . An explanation for the discrepancy between our Mesp1Cre-MADM clonal analysis and prior retrospective clonal analyses lies with the timing of the labeling events . Mesp1 is transiently expressed ( ( Saga et al . , 1999 ) and Figure 1—figure supplement 1B ) , thus all of our clones are induced over a narrow window of time during gastrulation . The retrospective clonal analysis employed a cardiomyocyte-specific promoter , so an early labeling event ( in the epiblast for example ) that would contribute broadly throughout the embryo would only be visualized in the heart , giving the impression of a specific common progenitor cell . Indeed , when we perform MADM clonal analysis using the epiblast-specific Cre line Sox2::Cre , we see many clones in the embryos , including large , dispersed clones throughout the looping heart tube ( Figure 1—figure supplement 2 ) that could be interpreted as a labeling event in a single , common cardiac progenitor . We cannot definitively conclude , however , that these labeled cells within the heart are all clonally related because of the large number of labeling events outside of the heart . In summary , our results suggest that soon after a heart field progenitor is specified , shortly after the initiation of gastrulation , the anatomic destiny of daughter cells quickly diverge , especially those destined to occupy /contribute to the left or right ventricle , and these fates remain fixed . In analyzing the cellular composition of Mesp1Cre-MADM clones , we noticed that while most twin spots give rise to homogenous cellular progeny ( cardiomyocytes or endothelial cells ) , individual cardiac progenitors ( <5% , see Figure 2 ) can contribute to multiple cell types , for example myocardium and endocardium ( Figure 1J–K and Figure 1—figure supplement 3J–L ) , or endothelium and smooth muscle ( Figure 1O ) . While a multipotent progenitor had been predicted from in vitro ES cell-based differentiation models ( Wu et al . , 2006; Moretti et al . , 2006 ) this is the first evidence of the existence of such a multipotent progenitor in vivo . The existence of a dedicated population of cardiac progenitors within gastrulating mesoderm suggests that these cells may have a unique molecular signature . To define the order and timing of early cardiac gene expression in vivo , we examined by in situ hybridization and by ß-galactosidase reporter activity early markers of cardiac differentiation ( Mef2c , Tbx5 , Isl1 , Mesp1 , and Smarcd3 ) at the late streak ( LS ) and early head fold ( EHF ) stages . We find that Smarcd3 expression precedes Isl1 and Tbx5 in a domain that lies at the anterior-proximal region of the embryo and extends into the extraembryonic tissues ( Figure 3B , D–E ) . This expression domain appears coincident temporally but non-overlapping with Mesp1 expression ( Figure 3A–B and Figure 3—figure supplement 1 ) , but is within the Mesp1Cre-derived lineage ( Figure 4O–P ) . Activity of the Mef2cAHF enhancer ( Verzi et al . , 2005 ) is also detected at the LS stage ( Figure 3C ) prior to the expression of Tbx5 and Isl1 . Previous studies have shown that activity of this enhancer is dependent on Isl1 ( Dodou et al . , 2004 ) . In order to confirm an absence of Isl1 expression at the LS stage , we used a reporter allele where a nuclear lacZ ( Isl1nLacZ ) has replaced a short segment of the coding sequence , including the endogenous start codon ( Sun et al . , 2007 ) . While there was robust ß-galactosidase reporter activity at the cardiac crescent stage , no detectable staining was seen in LS stage embryos ( Figure 3—figure supplement 2A–B ) , suggesting that initiation of Mef2cAHF enhancer expression precedes Isl1 expression and its initiation may be independent of Isl1 . Several hours later ( EHF stage ) , weak Isl1 and Tbx5 expression is detectable , Mef2cAHF activity remains , and Smarcd3 expression restricts to the embryo proper ( Figure 3G–J ) . The VEGF-A receptor Flk1 labels multipotent progenitors that can differentiate into hematopoietic , endothelial , smooth muscle and cardiac lineages ( Ishitobi et al . , 2011; Kattman et al . , 2006 ) , thus we looked at co-localization of Mesp1 , Flk1 , and Smarcd3 in late streak embryos . While we find minimal overlap between Mesp1 and Flk1 and Mesp1 and a reporter of early Smarcd3 expression ( Figure 3—figure supplement 1A , C–D ) , there is significant overlap of Flk1 with this same reporter of early Smarcd3 expression ( see below ) in late streak stage embryos . Taken together , we see an orderly progression of gene expression that follows the progressive commitment of nascent mesoderm , from early expression of Mesp1 to the intermediate expression of Smarcd3 , Flk1 , and the Mef2cAHF enhancer , and finally the later cardiac lineage-specific expression of Tbx5 and Isl1 ( Figure 3K ) . 10 . 7554/eLife . 03848 . 008Figure 3 . Smarcd3 expression initiates in gastrulating mesoderm and precedes expression of Isl1 and Tbx5 . ( A–B and D–E ) In situ hybridization at late-streak ( LS ) stage for Mesp1 , Smarcd3 , Isl1 , and Tbx5 . ( C ) X-gal staining at late-streak stage for Mef2cAHF-lacZ . Smarcd3 mRNA is expressed anterior to Mesp1 mRNA in embryonic and extraembryonic tissues . Tbx5 and Isl1 are undetectable by in situ hybridization at this stage . Activity of the Mef2cAHF enhancer is detectable around the node ( asterisk ) and in the anterior embryonic tissues ( white arrowheads ) . ( F–G and I–J ) In situ hybridization at early-head-fold ( EHF ) stage for Mesp1 , Smarcd3 , Isl1 , and Tbx5 . ( H ) X-gal staining at early-head-fold ( EHF ) stage for Mef2cAHF-lacZ enhancer . Isl1 and Tbx5 expression is now detectable ( arrowheads ) . ( K ) Summary of gene expression . Mesp1 expression ( blue ) precedes all other genes . Flk1 ( gray ) , Smarcd3 ( red ) , and Mef2cAHF ( yellow ) expression follows , beginning at the mid-streak stage , in overlapping domains . Isl1 ( orange ) and Tbx5 ( magenta ) expression begins at the late-streak/early head fold stage and overlaps with the stripe of Smarcd3 expression . Mesp1 expression at this stage is restricted to a small domain at the posterior of the embryo . Scale bars: ( A–J ) ( 100 µm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03848 . 00810 . 7554/eLife . 03848 . 009Figure 3—figure supplement 1 . Additional characterization of Smarcd3-F1-LacZ expression . ( A ) In situ hybridization for Mesp1 mRNA and antibody staining for GFP in Flk1GFP/+ LS stage embryo reveals minimal overlap between active expression of Mesp1 and Flk1 . ( B ) Antibody staining for GFP and salmon-gal staining in Smarcd3-F1-lacZ; Flk1GFP/+ late-streak stage embryo shows significant overlap between Flk1 and Smarcd3 expression . ( C ) In situ hybridization for Mesp1 mRNA and salmon-gal staining of Smarcd3-F1-lacZ in LS and EHF ( D ) stage embryos . Mesp1 mRNA is localized predominantly at the base of the forming allantois ( asterisks ) and is largely non-overlapping with Salmon-gal staining of Smarcd3-F1-lacZ . Scale bars: ( A–D ) , 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03848 . 00910 . 7554/eLife . 03848 . 010Figure 3—figure supplement 2 . Characterization of early Isl1nlacZ expression . ( A ) X-gal staining of Isl1nlacZ/+ late streak ( LS ) stage embryos reveals no detectable staining even after over-night incubation at 37°C in staining solution . ( B ) One day later at E7 . 75 , X-gal staining is seen within the developing cardiac crescent . DOI: http://dx . doi . org/10 . 7554/eLife . 03848 . 010 In comparing human and mouse sequences at the Smarcd3 locus , we identified several regions of conservation in noncoding sequences upstream of the Smarcd3 transcriptional start site . We tested an ∼9 kb genomic element we called Smarcd3-F1 in a transgenic mouse reporter assay and found that it was sufficient to recapitulate early endogenous Smarcd3 expression ( Figure 4A–C ) . Expression was maintained in the cardiac crescent and looping heart at later stages of development ( Figure 4E–I ) , however expression in extraembryonic tissues was noted . Given the broad expression domain of Smarcd3 mRNA and our Smarcd3-F1::lacZ reporter lines , we sought to define an enhancer fragment that might uniquely label early cardiac progenitors . Through deletion analysis of the 9 kb enhancer/promoter region ( Smarcd3-F1 ) , we defined an ∼2 . 5 kb genomic element ( Smarcd3-F6 ) that is expressed only in the embryo proper , in a more restricted pattern than Smarcd3-F1 ( Figure 4D and J–N and Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 03848 . 011Figure 4 . Identification of an early cardiac specific enhancer of Smarcd3 . ( A ) Genomic region upstream of Smarcd3 translational start site ( black arrow ) . Grey boxes: regions of non-coding sequence conservation between human and mouse . Indicated regions were used to generate Smarcd3-F1-lacZ and Smarcd3-F6-nlacZ alleles . ( B–C ) Salmon-gal staining of Smarcd3-F1-lacZ allele ( red ) closely mimics endogenous expression of Smarcd3 mRNA ( dark blue ) . ( D ) X-gal staining of Smarcd3-F6-nlacZ allele ( light blue ) labels a fraction of total Smarcd3 mRNA . ( E–I ) lateral and frontal views of salmon-gal stained Smarcd3-F1-LacZ embryos at ( E ) early head fold ( EHF ) , ( F–G ) cardiac crescent , ( H ) E8 . 5 , and ( I ) E9 . 5 stages . Note the staining in extraembryonic tissues at EHF and cardiac crescent stages ( asterisks ) . Also note salmon-gal staining in lateral mesoderm of E8 . 5 and E9 . 5 embryos ( arrowheads ) . ( J–N ) A single copy of the F6 enhancer along with Hsp68 minimal promoter and nls-LacZ coding sequence were targeted to the Hipp11 locus on chromosome 11 ( see extended methods for details ) . Lateral and frontal views of X-gal stained Smarcd3-F6-Hsp68-nLacZ embryos at ( J ) early head fold ( EHF ) , ( K–L ) cardiac crescent , ( M ) E8 . 5 , and ( N ) E9 . 5 stages . Note absence of staining in extraembryonic tissues at EHF and cardiac crescent stages as well as restricted cardiac expression at E8 . 5 and E9 . 5 ( arrowheads ) . ( O ) Lateral view of late streak stage Mesp1Cre; ROSAmTmG; Smarcd3-F1-lacZ embryo showing partial overlap of Smarcd3 expression with the Mesp1-derived lineage . ( P ) Additional anterior view . Blue: DAPI stained nuclei , Green: GFP staining , Red: Beta-galactosidase . Scale bars: ( B–D , E–G , J–L ) , 100 µm , ( H and M ) , 100 µm , ( I and N ) , 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03848 . 01110 . 7554/eLife . 03848 . 012Figure 4—figure supplement 1 . Generation of Smarcd3-F6nLacZ reporter mice . ( A ) A single copy of the F6 enhancer along with the Hsp68 minimal promoter and nLacZ coding sequence were targeted to the Hipp11 locus on chromosome 11 ( see extended methods for details ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03848 . 012 Based on the temporal and spatial expression of Smarcd3 , prior to expression of the canonical heart field markers Tbx5 and Isl1 , and within a small subset of Mesp1Cre-derived cells , we hypothesized that Smarcd3+ cells may represent cardiac precursors . To define the lineage potential of Smarcd3+ cells and to compare the lineage potential of the cell populations marked by our two enhancer elements , we performed temporally regulated genetic fate-mapping using mice expressing a tamoxifen inducible CreERT2 under control of the Smarcd3-F1 and Smarcd3-F6 sequences ( Smarcd3-F1CreERT2 and Smarcd3-F6CreERT2 mice; Figure 5—figure supplement 1A–H ) . Lineage labeling of Smarcd3-F1CreERT2; RosaR26R at E5 . 5 and observation at E10 . 5 marked cells that contribute predominantly to the heart and anterior forelimb ( Figure 5A–B ) , with scattered cells in the trunk and cranial mesoderm ( Figure 5A and data not shown ) ; observation at earlier time points revealed significant labeling in extraembryonic tissues ( Figure 5—figure supplement 2A–C ) . Within the heart , labeled cells contribute to cardiomyocyte , endocardial , and pericardial layers ( Figure 5C ) . Labeling one day earlier ( E4 . 5 ) marks a similar distribution of cells but with reduced labeling ( Figure 5—figure supplement 2D–F ) . Earlier induction ( E3 . 5 ) and non-injected animals have minimal to no detectable labeling ( Figure 5—figure supplement 2G–K and data not shown ) . Given the restricted expression pattern of the Smarcd-F6 enhancer , we hypothesized that marking these cells early on would label a more restricted population of cells later in the embryo . Indeed , lineage tracing with Smarcd3-F6CreERT2 labeled a much smaller number of more spatially defined cells , including the heart and a small group of cells in the anterior forelimb ( Figure 5D–E ) . There was minimal to no labeling in head/trunk mesoderm or in extraembryonic tissues , and within the heart Smarcd3-F6CreERT2 labeled primarily the myocardial layer ( Figure 5F and Figure 5—figure supplement 3A–C ) , consistent with this enhancer labeling a more restricted population of cells . We addressed the clonal potential of early Smarcd3+ cells with limiting doses of tamoxifen to induce small numbers of well-isolated labeled cells in Smarcd3-F1CreERT2;RosaR26R embryos ( Figure 5G–H ) . The shape and distribution of the labeled patches is reminiscent of previous work describing the oriented clonal cell growth throughout the myocardium ( Meilhac et al . , 2004a ) . In order to confirm the clonal nature of these patches induced with our Smarcd3-F1CreERT2 allele , we attempted to perform MADM analysis with this particular allele . Unfortunately , the recombination frequency was too low to identify any clones in several litters of embryos and we proceeded with using the RosaConfetti multicolor reporter ( Snippert et al . , 2010 ) . We identified 6 Smarcd3-F1CreERT2;RosaConfetti embryos from several litters where tamoxifen was administered at E5 . 5 , which contained rare or infrequent labeling events; none of these examples showed a labeling pattern that would support a common progenitor contributing to both right and left ventricles . Instead , we saw isolated , single-color clones ( Figure 5I and Figure 5—figure supplement 3D ) that contributed to a single chamber . While the number of embryos we examined is not sufficient to reach statistical significance , the results are highly supportive of our conclusions from the Mesp1Cre-MADM analysis . We conclude that Smarcd3 expression in the late gastrulating embryo labels a defined population of specified cardiac precursors that are fated to occupy unique anatomic structures within the mature heart . 10 . 7554/eLife . 03848 . 013Figure 5 . A Smarcd3 enhancer in the late gastrulating embryo labels a population of specified cardiac precursors . ( A ) Smarcd3-F1+ cells labeled at E5 . 5 and observed at E10 . 5 contribute to the heart and anterior forelimb ( arrows ) . In addition , scattered cells are observed in the trunk and neck ( not shown ) . ( B ) Labeled cells are present in all chambers of the heart , including the RV , LV , OFT , and RA and LA ( not shown ) . ( C ) Within the heart , labeled cells contribute to the pericardial layer as well as the cardiomyocte and endocardial cell layers . ( D ) Smarcd3-F6+ cells labeled at E6 . 5 and observed at E10 . 5 contribute to the heart and anterior forelimb . No additional labeling in the trunk or neck is observed . ( E ) Labeled cells are also present in all chambers of the heart . The number of labeled cells , however , appears reduced . ( F ) Within the heart , myocardial and pericardial ( not shown ) cells are labeled . ( G–H ) Limiting doses of tamoxifen administered to Smarcd3-F1CreERT2;RosaR26R embryos at E5 . 5 label scattered clusters of cells throughout the heart at E10 . 5 . ( I ) Clonal analysis with the Smarcd3-F1CreERT2;RosaConfetti line ( E5 . 5 label , harvested at E10 . 5 ) shows that a single Smarcd3-F1+ progenitor can populate the left ventricle ( YFP , arrow heads ) or the left atrium ( red fluorescent protein , asterisk ) . Scale bars: ( A and D ) , 500 µm ( B and E ) , 200 µm ( I ) , 50 µm . en , endocardium; LV , left ventricle; LA , left atrium; m , myocardium; OFT , out flow tract; p , pericardium; RA , right atrium; RV , right ventricle . DOI: http://dx . doi . org/10 . 7554/eLife . 03848 . 01310 . 7554/eLife . 03848 . 014Figure 5—figure supplement 1 . Generation and characterization of Smarcd3-F1CreERT2 and Smarcd3-F6CreERT2 mice . ( A ) Schematic of Smarcd3-F1CreERT2 construction . The Smarcd3-F1 enhancer/promoter was cloned upstream of the CreERT2 coding sequence . Flanking H19 insulator sequences were added to minimize position effects ( see extended methods for details ) . ( B–G ) In situ hybridization for CreERT2 in Smarcd3-F1CreERT2 embryos at MS ( B and E ) LS ( C and F ) and EHF ( D and G ) stages reveals an identical pattern of expression to the Smarcd3-F1-lacZ allele and endogenous Smarcd3 early in development . ( H ) Schematic of Smarcd3-F6CreERT2 construction . A single copy of the F6 enhancer along with the Hsp68 minimal promoter and CreERT2 coding sequence were targeted to the Hipp11 locus on chromosome 11 ( see extended methods for details ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03848 . 01410 . 7554/eLife . 03848 . 015Figure 5—figure supplement 2 . Additional characterization of Smarcd3-F1CreERT2 mice . ( A ) Early labeling using Smarcd3-F1CreERT2 at E5 . 5 followed by observation at E7 . 5 reveals robust labeling within the cardiac crescent ( arrows ) as well as lateral plate mesoderm ( arrowheads ) and allantois ( asterisk ) . ( B ) Observation one day later at E8 . 5 reveals a similar distribution of cells within the looping heart ( asterisk ) and lateral plate mesoderm ( arrowheads ) . ( C ) At E9 . 5 , labeled cells are present within the heart ( asterisk ) and scattered along the lateral plate mesoderm ( arrowheads ) and cranial mesoderm ( asterisks ) . ( D ) Smarcd3-F1CreERT2 cells labeled at E4 . 5 and observed at E10 . 5 contribute to the heart and anterior forelimb ( arrows ) . In addition , labeled cells within trunk and cranial mesoderm are present ( asterisks ) . ( E–F ) Labeled cells are present in all chambers of the heart , including the RV , LV , OFT , and RA and LA ( not shown ) and contribute to the myocardium and pericardium . ( G–H ) Injection of tamoxifen into Smarcd3-F1CreERT2 animals at E3 . 5 and observation at E9 . 5 reveals no labeling in embryos . ( I–K ) Un-injected Smarcd3-F1CreERT2 animals have no labeled cells in the head , trunk , or limbs , and rare un-injected embryos have a few scattered cells in the heart ( K , asterisks ) at E13 . 5 . A total of 16 mock injected embryos were examined and the embryo shown is an extreme example that represents the most labeling that we saw throughout the 16 mock-injected embryos . LA , left atrium; LV , left ventricle; m , myocardium; OFT , out-flow tract; p , pericardium; RA , right atrium; RV , right ventricle . DOI: http://dx . doi . org/10 . 7554/eLife . 03848 . 01510 . 7554/eLife . 03848 . 016Figure 5—figure supplement 3 . Additional characterization of Smarcd3-F6CreERT2 mice . ( A–C ) Early labeling using Smarcd3-F6CreERT2 at E6 . 5 and observation at E10 . 5 reveals scattered cells ( green ) throughout the heart ( red ) without significant contribution to trunk or cranial mesoderm . ( D ) limiting doses of tamoxifen were used to induce small numbers of well-isolated labeled cells at E6 . 5 in Smarcd3-F1CreERT2;RosaConfetti embryos . Hearts were observed at E10 . 5 . Note isolated clones of YFP labeled cells adjacent to interventricular septum and top of left ventricle ( asterisks ) . LA , left atrium; LV , left ventricle; m , myocardium; OFT , out-flow tract; p , pericardium; RA , right atrium; RV , right ventricle . DOI: http://dx . doi . org/10 . 7554/eLife . 03848 . 016 Given the restricted expression and lineage potential of cells marked by the Smarcd3-F6 enhancer , we sought to characterize the global molecular signature that uniquely identifies this population . We generated a mouse ES cell line with a targeted insertion of the Smarcd3-F6 enhancer driving expression of a fluorescent reporter and we differentiated these cells into cardiomyocytes using an established directed differentiation protocol ( Figure 6A and Figure 6—figure supplement 1 ) ( Wamstad et al . , 2012 ) . In this system , we find expression of Mesp1 is rapidly induced at early mesoderm stage ( Day 3-Day 4 ) while expression of the cardiac precursor markers Nkx2-5 and Tbx5 initiate about a day later ( Figure 6C ) . Expression of Smarcd3 as well as the Smarcd3-F6 reporter are detectable in ES cells and gradually decline until both are sharply induced at Day 4 . At the same time , Mesp1 expression is rapidly downregulated ( Figure 6B ) . As a result , Mesp1 and Smarcd3 mRNA are largely non-overlapping , temporally . We isolated total RNA from three biological replicates of sorted cells shortly after the initiation of reporter activity at the mesoderm stage ( D4 . 75 ) and compared the gene expression profiles of GFP + to GFP- cells by RNA-seq ( Figure 6D ) . GFP + cells expressed genes associated with early cardiovascular progenitors ( e . g . Hand2 , Gata4 , and Meis1 ) , while GFP- cells expressed genes associated with hematopoietic and other lineages . Among the GFP + population , we attempted to identify markers that were unique to either cardiomyocyte or endocardial progenitors . While cell type-specific markers for differentiated cells are well documented ( eg . contractile proteins for cardiomyocytes vs Nfatc1 for endocardial cells ) such markers are less well defined at the early stages when we are isolating mRNA . In addition , our differentiation protocol pushes cells towards a cardiomyocyte fate , thus we may not expect to see any endocardial cell markers . One gene of interest that appears to be depleted in our GFP + population/enriched in our GFP- population is Tal1/Scl . This gene had been implicated as a critical component of endocardial morphogenesis ( Bussmann et al . , 2007 ) and more recently has been shown to repress cardiomyogenesis in yolk sac vasculature and endocardium ( Van Handel et al . , 2012 ) . A complete list of differentially expressed genes meeting a strict false discovery rate ( FDR ) of 0 . 02 ( 2% ) is presented in Figure 6—Source data 1 . Overall , these results indicate that cells labeled by the Smarcd3-F6 enhancer , both in vivo and in vitro , represent an early cardiac progenitor population . 10 . 7554/eLife . 03848 . 017Figure 6 . Smarcd3-F6 enhancer labels an early cardiac progenitor in differentiating ES cells . ( A ) Schematic of cardiac differentiation protocol and representative images at indicated time points during differentiation of Smarcd3-F6nlsEGFP mESCs . ( B ) Time course of gene expression during cardiac differentiation . Low expression of Smarcd3 as well as GFP is detectable in ES cells . Expression decreases over the course of the differential protocol but is rapidly induced shortly after day 4 . ( C ) Expression of Mesp1 peaks between day 3-day 4 of the differentiation protocol . Expression of Nkx2-5 and Tbx5 begins later at day 5 . Values shown are the mean plus SEM for 4 independent experiments , each performed in triplicate ( D ) Heatmap showing differential gene expression in GFP + compared to GFP- sorted cells . Many genes involved in cardiac progenitor development are enriched in the GFP + population . Markers of primitive mesoderm and of hematopoietic and other cell lineage development are enriched in the GFP- population . Values are log2 fold change and are clipped at 1 . 2 . Analysis is based on three biological replicates . Yellow = higher in GFP positive population , blue = higher in GFP negative population . DOI: http://dx . doi . org/10 . 7554/eLife . 03848 . 01710 . 7554/eLife . 03848 . 018Figure 6—Source data 1 . Complete list of differentially expressed genes meeting strict FDR of 0 . 02 . A mouse ESC line with the Smarcd3-F6 nlsEGFP reporter was differentiated toward the cardiac lineage ( see methods for complete details ) . 18 hours after replating , GFP + cells were sorted from GFP- cells . Total RNA was extracted and analyzed by RNA-sequencing . Following bioinformatics analysis ( see methods ) a list of differentially expressed transcripts meeting a strict FDR of 0 . 02 was selected . This list of transcripts was used to generate the heatmap in Figure 3K . DOI: http://dx . doi . org/10 . 7554/eLife . 03848 . 01810 . 7554/eLife . 03848 . 019Figure 6—figure supplement 1 . Generation and characterization of Smarcd3-F6nlsEGFP mESC line . ( A ) Schematic of Smarcd3-F6nlsEGFP mESC construction . The Smarcd3-F6 enhancer was cloned upstream of the Hsp68 minimal promoter and nlsEGFP coding sequences . The reporter construct was flanked by H19 insulator sequences . The entire construct was targeted to the Hipp11 locus on chromosome 11 and successfully targeted ESC clones were identified by PCR ( not shown ) and Southern blotting ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03848 . 019 One class of clones identified in our Mesp1Cre-MADM clonal analysis included twin spots within the interventricular septum ( IVS ) between the left and right ventricles ( Figure 7A–D and Figure 2 ) . These twin spots formed a sharp boundary within the IVS , reminiscent of compartment boundaries classically defined in the Drosophila wing imaginal disc and mammalian midbrain-hindbrain boundary ( Lawrence and Struhl , 1996; Dahmann and Basler , 1999 ) . To determine if and when this boundary between the right ventricle and left ventricle is established , we performed temporally regulated genetic fate-mapping to mark early cardiac progenitors and followed their contribution later in the mature IVS . In order to mark cells that would contribute to the right ventricle , we used the Mef2c-AHF enhancer ( Verzi et al . , 2005 ) to drive expression of a fusion of the Dre recombinase ( Anastassiadis et al . , 2009 ) to the tamoxifen inducible ERT2 protein ( Mef2cAHFDreERT2 ) . We found that this enhancer element is expressed very early in the embryo , in a domain that appeared to partially overlap with that of Smarcd3 ( Figure 3C , H ) . Previous work using a constitutive Cre recombinase expressed under the control of the Mef2cAHF enhancer suggested that endothelial and myocardial components of the outflow tract , right ventricle , and IVS are derived from this population ( Verzi et al . , 2005 ) . To label the complementary population of progenitors that contribute predominantly to the left ventricle , we targeted a tamoxifen inducible Cre recombinase to the Tbx5 locus ( Tbx5CreERT2 ) ( Figure 7—figure supplement 1A–B ) . In the looped heart , expression of Tbx5 is restricted to the left ventricle and atria with very little expression in the right ventricle or outflow tract ( Bruneau et al . , 1999 ) . Early labeling of Tbx5+ cells marked cells that were predominantly restricted to the left ventricle and atria ( Figure 7E–F and 7I–L ) ; scattered surface labeling on the right ventricle ( Figure 7—figure supplement 1I–J ) was also seen with early tamoxifen injection , but myocardial labeling was limited to the LV up to the junction with the RV at the IVS . Earlier observation at E8 . 5 following a pulse of tamoxifen at E6 . 5 revealed labeling in a restricted population of cells within the presumptive left ventricle ( Figure 7—figure supplement 2 ) , suggesting that the Tbx5 lineage restriction arises early and is not a consequence of later sorting out of cells . Labeling Mef2cAHF + cells at E6 . 5 marked a complementary population of cells that were largely restricted to the right ventricle and outflow tract ( Figure 7G–H and Figure 7—figure supplement 3A ) . Using an intersectional reporter that responds to the combined activity of Cre and Dre we confirmed that these early Mef2cAHF + cells are also within the Smarcd3-F1 lineage ( Figure 7M and Figure 7—figure supplement 3B–C ) . The expression of Tbx5 and the Mef2AHF enhancer appears to label complementary populations that are established prior to morphogenesis , which correspond , in part , to left and right ventricular precursors , respectively . 10 . 7554/eLife . 03848 . 020Figure 7 . Early establishment of a boundary between the right and left ventricle at the interventricular septum . ( A ) Ventral view of yellow septal clone ( embryo ID MM8 ) with additional red and green twin-spots in outflow tract ( arrow heads ) . ( B ) Section through heart reveals sharp septal boundary of clone ( arrows ) with an extension of cells at top of septum into RV ( asterisk ) . ( C ) Ventral view of large septal clone ( embryo ID MM25 ) originating from left ventricle . ( D ) Section through red-boxed area reveals large yellow clone ( arrows ) extending from apex of left ventricle towards the top of the interventricular septum . An additional clone of red cells ( asterisks ) is directly adjacent to the yellow clone . The green twin spot is located just medial to the red twin spot ( arrowheads ) . All clones appear to be originating from the apex of the left ventricle . ( E–F ) Tbx5+ cells labeled at E6 . 5 and observed at E10 . 5 contribute to left ventricle and atria . ( G–H ) Mef2cAHF + cells labeled at E6 . 5 and observed at E10 . 5 contribute to specific anterior heart field structures , including the right ventricle and outflow tract . Note sharp boundary at future site of interventricular septum ( IVS , asterisks ) . ( I–J ) Tbx5+ cells labeled at E7 . 5 and observed at E14 . 5 for mTmG . ( K–L ) Tbx5+ cells labeled at E6 . 5 and observed at E14 . 5 for R26R . A sharp boundary at IVS between the left and right ventricles is present following early labeling of Tbx5+ cells . ( M ) Smarcd3-F1/Mef2cAHF double positive cells were labeled at E6 . 5 and observed at E10 . 5 using the intersectional reporter , RosaAi66 . Labeled cells contribute to right ventricle and outflow tract with a minor population of cells extending into the left ventricle . ( N ) Tbx5+ cells were labeled at E6 . 5 and their lineage followed using the RosamTmG Cre-reporter ( green ) . Mef2cAHF + cells were also labeled at E6 . 5 and their lineage followed using the RosanKmB Dre-reporter ( blue ) . Note largely non-overlapping Tbx5 and Mef2cAHF derived lineages in left and right ventricles , respectively , except perhaps a small area of overlap at the forming interventricular septum ( asterisks ) . ( O ) Tbx5/Mef2cAHF double positive cells were labeled at E6 . 5 and observed at E10 . 5 using the intersectional reporter , RosaAi66 . A narrow ring of labeled cells is present between the left and right ventricles . ( P ) Sections confirm a restricted population of labeled cells within the interventricular septum and superior aspect of the ventricular chamber ( asterisk ) . Red , TdTomato; Green , Tropomyosin . DOI: http://dx . doi . org/10 . 7554/eLife . 03848 . 02010 . 7554/eLife . 03848 . 021Figure 7—figure supplement 1 . Generation of a multi-use Tbx5 allele . ( A ) Tbx5 targeting strategy . Top shows the wild-type ( WT ) allele of Tbx5: exons are indicated by roman numerals . Open boxes are untranslated regions , filled boxes are coding regions , and red box shows T-box encoding sequence . The targeting vector has a CreERT2-IRES-2X-FLAG inserted in frame with the endogenous Tbx5 coding region and a neomycin resistance cassette ( neoR ) flanked by frt sites ( triangles ) in the second intron . Flpase was used to remove the neoR cassette . ( B ) Southern analysis shows proper targeting of multiple independent ES cell lines . ( C–N ) Tbx5CreERT2 mice were crossed to RosaR26R reporter mice . Cre activity was induced at the indicated timepoints by tamoxifen injection and analyzed at E14 . 5 . ( C–H ) Early induction labeled few reporter cells in the retina of double transgenic embryos , however later induction robustly labeled cells within the retina . ( I–N ) Early induction robustly labeled cells in the left ventricle and atria of double transgenic embryos . In addition , scattered cells are noted on the surface of the right ventricle and outflow-tract ( arrows ) . Induction after E8 . 5 ( L–N ) continued to label the left ventricle and atria as well as the trabeculae of the right ventricle ( asterisks M–N ) . Surface labeling on the right ventricle was no longer present after induction at E8 . 5 . LV , left ventricle; RV , right ventricle; IVS , interventricular septum . DOI: http://dx . doi . org/10 . 7554/eLife . 03848 . 02110 . 7554/eLife . 03848 . 022Figure 7—figure supplement 2 . Early labeling of Tbx5CreERT2 lineage . ( A–F ) Tbx5CreERT2 mice were crossed to RosaR26R reporter mice . Cre activity was induced at E6 . 5 and embryos were evaluated at E8 . 5 . Early labeling marks a restricted population of cells that is predominantly localized to the presumptive left ventricle . A few scattered cells are noted on the surface of the forming right ventricle ( asterisks , F ) . In situ hybridization for CreERT2 ( G–H ) or Tbx5 ( I–J ) along with salmon gal staining for lacZ in embryos labeled at E6 . 25 and analyzed at E8 . 25 . Note significant overlap of salmon gal staining with CreERT2 and Tbx5 mRNA . A small domain of CreERT2 and Tbx5 mRNA expression at the caudal end of the sinus horns ( asterisks ) is not double labeled , consistent with the dynamic nature of Tbx5 expression . DOI: http://dx . doi . org/10 . 7554/eLife . 03848 . 02210 . 7554/eLife . 03848 . 023Figure 7—figure supplement 3 . Additional characterization of early Mef2cAHF and Tbx5 lineages . ( A ) Mef2cAHFDreERT2 mice were crossed to RosaRoxlacZ reporter mice . Dre activity was induced at E6 . 5 and embryos were evaluated at E14 . 5 . Early labeling marks a restricted population of cells that is predominantly localized to the right ventricle and outflow tract . A few scattered cells are noted extending into the left ventricle ( asterisks , and data not shown ) . ( B–C ) Smarcd3-F1/Mef2cAHF double positive cells were labeled at E6 . 5 and observed at E14 . 5 using the intersectional reporter , RosaAi66 . Labeled cells contribute to right ventricle and outflow tract . A sharp boundary is present within the interventricular septum at the apex of the heart ( arrowheads ) , however a minor population of cells extend into the left side near the superior portion of the septum ( asterisk ) . Note an absence of contribution to the cushion at the top of the interventricular septum ( arrow ) . ( D ) Tbx5/Mef2cAHF double positive cells were labeled at E6 . 5 and observed at E12 . 5 using the intersectional reporter , RosaAi66 . A narrow ring of labeled cells is present between the left and right ventricles . A few , scattered double positive cells also populate the left ventricular chamber ( asterisk ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03848 . 023 Given the seemingly complementary lineage contributions of early Tbx5 and Mef2cAHF expressing progenitors to the mature heart , we sought to map their contributions and overlap within the interventricular septum . We performed simultaneously lineage tracing of Tbx5+ and Mef2cAHF + cells using separate reporters for Cre ( RosamTmG ) and Dre ( RosanKmB ) activity and found that the two populations of labeled cells appeared largely mutually exclusive , except perhaps a narrow stripe of cells at the border between the forming left and right ventricles ( Figure 7N ) . We confirmed the existence of a small population of double positive cells early in development using the intersectional Cre/Dre reporter described previously . Labeling at E6 . 5 marked a small population of cells that would eventually go on to form a narrow ring of cells at the border between the left and right ventricles in the forming IVS ( Figure 7O–P and Figure 7—figure supplement 3D ) . In summary , genetic lineage labeling of the earlyTbx5+ and Mef2cAHF + precursors along with our clonal analysis of early Mesp1+ progenitors that contribute to the IVS suggest that a compartment boundary between the future left and right ventricles is established early , prior to cardiac morphogenesis , and that early progenitors positive for both Tbx5 and Mef2cAHF will go on to contribute specifically to the forming IVS . The existence of a multipotent cardiac progenitor that can contribute to all anatomic and cellular components of the mature heart has been predicted , but the identity and origins of such a progenitor has remained undefined . Our findings , summarized in Figure 8 , pinpoint the existence of specified cardiac precursors in gastrulating mesoderm , and also highlight an unanticipated early segregation of first and second heart field progenitors in their contribution to distinct chambers of the developing heart . Further , we define the orderly progression of gene expression that parallels the commitment of nascent mesoderm to a cardiovascular fate . In particular , the expression of Smarcd3 labels a population primarily comprised of the earliest specified precursors , which can be identified in vivo and in ES cell-derived differentiating cells . 10 . 7554/eLife . 03848 . 024Figure 8 . Summary of patterning and specification of early gastrulating mesoderm . Clonal analysis reveals early patterning of gastrulating mesoderm including the segregation of cardiac vs non-cardiac mesoderm . Among cardiac mesoderm , progenitors for the two heart fields diverge soon after the initiation of gastrulation and rapidly become specified into discrete populations of committed precursors . Expression of Smarcd3 begins prior to that of other known markers of cardiac progenitors ( Nkx2-5 , Tbx5 , and Isl1 ) and an enhancer of Smarcd3 ( Smarcd3-F6 ) marks the earliest cardiac-specific progenitor population . Expression of the Mef2AHF enhancer and Tbx5 further subdivides this early population into first vs second heart field progenitors . Inducible genetic lineage tracing along with clonal analysis predicts the existence of a compartment boundary between the future left and right ventricles that is established prior to cardiac morphogenesis . DOI: http://dx . doi . org/10 . 7554/eLife . 03848 . 024 Our results show that mesoderm is rapidly specified at or before E6 . 0 to E7 . 5 into discrete fates that anticipate anatomical localization . Retrospective lineage tracing using a cardiac marker suggested the existence of a precursor that could contribute to all chambers of the heart ( Meilhac et al . , 2004b ) . Our results , and those of Lescroart et al . ( 2014 ) , suggest that such a precursor could only exist prior to gastrulation , likely in the epiblast or shortly thereafter , and that early mesoderm is rapidly patterned such that specific cells are shunted towards a unique path that assigns them to a specific cardiac compartment . Whether these mesodermal cells retain plasticity and could adopt different fates if transplanted elsewhere in the embryo is not clear . We also cannot at this point distinguish between a scenario where Mesp1+ cells are pre-patterned or the gradual restriction of Mesp1+ cells to different lineage fates during gastrulation in response to signaling cues . Understanding which molecular cues induce the emergence and patterning of this population of early cardiac precursors will be of considerable interest , particularly in the context of regenerative cell therapy , and especially given the apparent heterogeneity of early cardiac progenitors . The Smarcd3-expressing cardiac precursor population is already specified into separate populations that will contribute to the left ventricle and atria , as marked by Tbx5 , and the right ventricle and outflow tract , as labeled by the Mef2cAHF enhancer . These results are consistent with genetic tracing of cardiac precursors expressing Hcn4 , which contribute to the left ventricle and conduction system ( Später et al . , 2013 ) , and we additionally show , at both a population and single cell level , that both left and right ventricular progenitors are independently established early in development and are separated by a compartment boundary . This compartment boundary is derived from a unique population of early progenitors that express Tbx5 as well as a subpopulation of cells co-expressing the Mef2cAHF enhancer , suggesting that a relatively small number of cells may be important in establishing and/or maintaining this border between the left and right ventricles . Understanding the molecular and cellular basis of this compartment boundary , and its importance in ventricular septation , will be of great relevance to congenital heart defects . This knowledge will also be relevant for disease modeling from stem cells , or for regenerative strategies , since knowing whether a progenitor cell is committed to become a right ventricular cardiomyocyte vs an atrial cardiomyocyte , for example , could be critical depending upon the anatomic or cellular defect one is attempting to correct . In organisms with a simpler heart , such as zebrafish or Ciona ( Stolfi et al . , 2010; Tolkin and Christiaen , 2012; Stainier et al . , 1993; Keegan et al . , 2004 ) , there is also a very early specification of cardiac precursors . Indeed , fate mapping in the chick embryo suggests distinct origins of the outflow tract vs the remainder of the heart , but does not indicate a compartmentalization of left and right ventricular progenitors , nor what molecular identity early cardiac precursors adopt ( Camp et al . , 2012 ) . Our data suggest that the mouse embryo , despite the complexity of its developing heart , has a very early patterning and allocation of mesodermal cells to specific cardiac fates and anatomical derivatives . In addition , it has been hypothesized that simpler organisms have integrated cardiac and pharyngeal mesoderm into a single lineage , the so-called cardiopharyngeal mesoderm ( Stolfi et al . , 2010 ) . Retrospective clonal analysis in the mouse ( Lescroart et al . , 2010 ) suggests a conservation of shared cardiac and pharyngeal lineages; caveats applicable to chamber-specific lineage allocation , including the timing of clone induction and the promoter used to visualize clones , remain . Lescroart and colleagues more recently reported rare instances of Mesp1+ clones co-labeling head muscles and the heart ( Lescroart et al . , 2014 ) . In our lineage labeling experiments , using the MADM system , we do not see convincing evidence for a common heart and head muscle progenitor . Additionally , Smarcd3+ precursors contribute to heart but not facial muscle . While our collective data do not support a common precursor for heart and facial muscle , we cannot definitively rule out a rare population of common progenitors . Having a well-defined understanding of the developmental origins of patterned organs is critical for the development of regenerative strategies , and for our understanding of the basis of congenital malformations . For the future development of cellular therapies , the knowledge that molecularly distinct populations of progenitors contribute to distinct anatomic regions of the heart will guide the selection of the most appropriate cellular and molecular signature . More broadly , our results suggest a rapid and precise patterning of progenitor populations during gastrulation . The Mesp1Cre knock-in mice ( Saga et al . , 1999 ) were obtained from Yumiko Saga . Mef2cAHF-lacZ mice ( Dodou et al . , 2004 ) were obtained from Brian Black . Isl1nLacZ knock-in mice ( Sun et al . , 2007 ) were obtained from Sylvia Evans . MADM11TG/TG , MADM11GT/GT mice ( Hippenmeyer et al . , 2010 ) , ROSAR26R mice ( Soriano , 1999 ) , ROSAmTmG mice ( Muzumdar et al . , 2007 ) , ROSAAi66 mice ( Allen Brain Institute strain B6;129S-Gt ( ROSA ) 26Sortm66 . 1 ( CAG-tdTomato ) Hze/J; JAX stock number 021876 ) , Flk1GFP knock-in mice ( Ema et al . , 2006 ) , and Sox2::Cre transgenic mice were obtained from Jackson Laboratory . ROSARoxlacZ mice ( Anastassiadis et al . , 2009 ) were generated from cryopreserved embryos purchased from MMRRC . Standard tamoxifen induction was done by injecting 3 mg/40 gram of body weight of tamoxifen dissolved in sesame oil intraperitoneally . Low dose tamoxifen induction was done by injecting 1/10th of this concentration ( 0 . 3 mg/40 gram body weight ) . Clonal analysis of Mesp1 mesodermal progenitors was performed by crossing Mesp1Cre;MADM11TG/TG with MADM11GT/GT mice . Lineage analysis of Smarcd3 cardiac progenitors was performed by crossing Smarcd3-F1CreERT2 mice with ROSAR26R , ROSAmTmG , or ROSAConfetti mice followed by administration of tamoxifen ( 3 mg/40 grams of pregnant dam's body weight ( Hayashi and McMahon , 2002 ) ) at E5 . 5 or E6 . 5 . Lineage analysis of Mef2cAHF-expressing progenitors was performed by crossing ROSARox-lacZ and Mef2cAHFDreERT2 transgenic mice followed by administration of tamoxifen at E5 . 5 or E6 . 5 . Lineage analysis of Tbx5-expressing progenitors was performed by crossing Tbx5CreERT2 knock-in mice and ROSAR26R or ROSAmTmG mice followed by administration of tamoxifen at indicated times . The Smarcd3-F1 fragment spanned 8796bp upstream of the start codon of Smarcd3 ( Chr5:24 , 107 , 677-24 , 116 , 473 ) and was cloned using bacterial recombineering from BAC bMQ133n21 into pENTR1A ( Invitrogen , Carlsbad , CA ) . The pWHERE plasmid ( InvivoGen ) was digested using a blunt cutter ( SmaI ) and an RFA ‘C’ Gateway cassette ( Invitrogen ) was inserted 5′ to the promoterless nls-LacZ reporter gene to make a destination vector ( pWHERE-DV ) . For construction of pWHERE-DV-CreERT2 , a Gateway RFA ‘B’ cassette was amplified by PCR and cloned into the AvrII site of pWHERE using Cold Fusion cloning ( System Biosciences Inc . ) . The resulting plasmid was digested with XhoI and NheI to remove the nls-LacZ reporter and a cassette encoding CreERT2 ( Feil et al . , 1997 ) was amplified and inserted using Cold Fusion Cloning . Construction of the Mef2cAHFDreERT2 allele will be described in detail elsewhere . Briefly , the CAGEN-DV plasmid ( Devine and Bruneau , unpublished ) was digested with SpeI and XbaI to remove the CAGGS ( Chicken Beta-Actin promoter with CMV enhancer ) promoter and the resulting ends blunted . The Mef2c-F6/frag3 plasmid ( Dodou et al . , 2004 ) was digested with XhoI and SalI to isolate the 3970 bp cardiac enhancer fragment . The resulting ends were blunted and the two blunt-ended fragments ligated together to make the destination vector ( Mef2c-F6-DV ) . A Gateway compatible entry clone for DreERT2 was constructed by PCR stitching of Dre ( minus the nls ) from plasmid DNA containing a codon-improved version of Dre ( a generous gift of Francis Stewart , Biotechnology Center TU , Dresden , Germany ) and ERT2 ( Feil et al . , 1997 ) separated by a short linker se quence ( Hunter et al . , 2005 ) to generate DreERT2 . Upon LR recombination with entry clones and destination reporter plasmids , the final vectors were restriction mapped and verified by DNA sequencing . Multiple founders were examined for each transgene . All experiments using mice were reviewed and approved by the UCSF Institutional Animal Care and Use Committee and complied with all institutional and federal guidelines . For construction of the RosanKmB fluorescent Dre-reporter allele , a nuclear localized mKateV5 ( nlsmKateV5 ) followed by a rabbit globin polyA sequence was cloned between ROX recombination sites downstream of a CAGGS promoter . A Gateway RFA destination cassette ( including a rabbit globin polyA sequence ) was subsequently cloned downstream to make pCAGGS-nK-DV . A SacI-SalI fragment containing an FRT-pGK-FRT cassette from pK11 ( plasmid courtesy of Gail Martin ) was cloned downstream of the Gateway RFA-rabbit globin polyA sequence to make pCAGGS-nK-DV-FRT-pGK-FRT . A Gateway entry clone containing a membrane localized TagBFP-FLAG was inserted into pCAGGS-nK-DV-FRT-pGK-FRT using LR recombination . The entire construct was excised using AscI and PacI and cloned into Rosa26PAm1 ( Addgene plasmid# 15 , 036 ) for targeting to the Rosa26 locus . The construct was linearized with SgfI and electroporated into E14 mouse ES cells . Drug resistant clones were screened by long range PCR using the following primers:5′ homology arm:ROSA1 5′-CCACTGACCGCACGGGGATTC-3′ ( in genomic ) ROSA7 5′-GGGGAACTTCCTGACTAGGG-3′ ( in FRT ) 3′ homology arm:ROSA2 5′-TCAATGGGCGGGGGTCGTT-3′ ( in CAGGS ) ROSA5 5′-GGGGAAAATTTTTAATATAAC-3′ ( in genomic ) Positive clones were subsequently screened by qPCR for single copy insertions of the pGK-Neo cassette . Following verification of correct targeting and karyotyping , two positive ES cell clones were expanded and injected into blastocysts for generation of mice . Chimeric founders were crossed to C57B6 lines to confirm germline transmission . The pGK-Neo cassette was eventually removed by breeding mice to ActFlpe expressing mice . The Smarcd3-F6 fragment , including approximately 2 . 5 kb of the 5′ end of the Smarcd3-F1 promoter fragment was cloned by PCR into pENTR1A . Modified pWHERE-DV and pWHERE-DV-CreERT2 plasmids were generated by inserting an Hsp68 minimal promoter into the XhoI restriction site using Cold Fusion Cloning . Upon LR recombination with entry clones and destination reporter plasmids , the final vectors were restriction mapped and verified by DNA sequencing . For generating TARGATT constructs , the PacI fragment from the final construct was subcloned into a PacI digested pBT346 . 3 plasmid ( Applied Stem Cells ) . DNA was purified and injected along with mRNA for the Phi31o transposase according to manufacturer's protocol . A 129 BAC clone ( 67H11 , RPCI-22 mouse BAC library ) containing the entire mouse Tbx5 locus was obtained from GeneService ( UK ) . Briefly , the cDNA for CreERT2 ( Feil et al . , 1997 ) , followed by an IRES element ( BamHI to NcoI , from pIRES2-EGFP ) , a Kozak consensus sequence and a 2X FLAG epitope sequence in frame with the translational start of endogenous Tbx5 was inserted between the FspI and NcoI sites of exon 2 of Tbx5 , upstream of the endogenous translation start site . At 40bp downstream of exon 2 , cloning sites ( NotI and AflII ) were added and a PGK-EM7-Neo-polyA cassette , flanked by Frt sites ( from PL451 ) , was inserted for positive selection . The entire cassette , as well as 5 kb upstream of exon 2 ( ClaI to FspI ) , and 6 kb downstream of exon 2 ( 40 bp downstream of exon 2 to BclI ) were cloned into a modified pBS containing a 5′ DTA negative selection cassette ( from pRosa-26-1 ) . The targeting vector was linearized by SalI digestion and electroporated into embryonic stem cells , and G418-resistant clones were tested for correct gene targeting by Southern analysis using 5′ and 3′ ( not shown ) probes external to the targeting vector . The following primers were used for the 5′ probe: probeA-F1: 5′-GGCCACTGATGGTGTAGAAGCAAC-3′ . probeA-R1: 5′- GTAGAGAGAAAGGCCATTCGGTCTG -3′ . The following primers were used for the 3′ probe: probeB-F1: 5′-GGGCCATTAGATCACCCTCATTCTG-3′ . probeB-R1: 5′- AACTCTGTGTATAAGGGCACTTCCC -3′ . Following verification of correct targeting and karyotyping , positive ES cells were expanded and injected into blastocysts for generation of mice . Chimeric founders were crossed to C57B6 lines to confirm germline transmission . The following primers were used for initial genotyping:5′ end of targeted locus:Tbx5CreF1: TATGTCGCTAGACACTCTCCTbx5CreR1:CCGGCAAACGGACAGAAGCAKnock-in = 226 bp , WT = no band3′ end of the targeted locus:Tbx5CreNeoFor1: ACTGTGCCTTCTAGTTGCCAGC . Tbx5CreWT: Rev: AAAGTGGATTGGGATAGAGTGGKnock-in = 470 bp , WT = no bandAfter mating to B-actin-FlpE mice to remove the Neo cassette:Tbx5NeoFlp’DF1: ACAACCATGGACTACAAGGACGTbx5CreWT Rev: AAAGTGGATTGGGATAGAGTGGKnock-in = 420 bp , WT = no bandMice were maintained on a C57B6 background after crossing to various reporter mice . Inheritance of the allele was confirmed by PCRTbx5Exon2WTFor2: ATACAGATGAGGGCTTTGGCCTGGTbx5CreWTRev: AAAGTGGATTGGGATAGAGTGGWT band = 290 bp , CreRT2 band = 360 bp The Smarca4FLAG knock-in ES cell line ( Attanasio et al . , 2014 ) was used for targeting of the Smarcd3-F6-Hsp68-nlsEGFP construct to the Hipp11 locus . Briefly , a modified shuttle vector containing a polylinker including PacI , XhoI , SacII , and flanking AscI sites was purchased from IDT . A pGKNeo selection cassette was subcloned from the pL451 plasmid using XhoI and SacII into the modified shuttle vector . A PacI fragment including flanking H19 insulator sequences , the Smarcd3-F6 enhancer , an Hsp68 minimal promoter , nlsEGFP coding sequence , WPRE mRNA stablilization sequence , and EF1alpha poly A sequence was subcloned into the modified shuttle vector . The entire reporter-selection construct was cloned into the Hipp11 targeting vector ( Hippenmeyer et al . , 2010 ) using AscI . The targeting vector was linearized using ApaI and electroporated into ES cells . Following G418 selection , correctly targeted clones were screened by PCR and Southern blotting . For culturing , ES cells were maintained in 2i + LIF media . Directed cardiomyocyte differentiations were performed as previously described ( Wamstad et al . , 2012 ) using the Smarcd3-F6nlsEGFP mESC line with minor modifications . 18 hours after plating and cardiac induction ( with VEGF , Fgf10 , and Fgf2 ) , supernatant was collected and 0 . 22 μm filtered . Cells were then washed with PBS ( w/o Ca2+/Mg2+ ) , dissociated from plates using TrypLE ( Gibco ) , resuspended in filtered supernatant , and placed on ice . GFP+ and GFP− populations were subsequently sorted into RNAprotect cell reagent ( Qiagen ) using a BD FACSAria II flow cytometer . RNA was then purified from each population using the RNeasy Mini kit ( Qiagen ) . Stranded RNA-seq libraries were then prepared using the Ovation Mouse FFPE RNA-Seq Multiplex System ( NuGEN ) and sequenced on an Illumina HiSeq 2000 . Three biological replicates for each population ( GFP+ and GFP- ) were obtained and analyzed by RNA-sequencing . Sequence reads were aligned to the mm9 ( mouse ) assembly with Tophat 2 ( Kim and Salzberg , 2011 ) , using Ensembl version 67 exon annotation . Differential expression and variance-corrected log fold change was calculated using the program ‘DefinedRegionDifferentialSeq’ in USeq version 8 . 6 . 4 ( http://useq . sourceforge . net/ ) . In order to report gene-level counts , the highest-total-read-count transcript was reported for each gene , resulting in gene level annotation only . The final heatmap reports all genes/nonCode elements where the Benjamini-Hochberg FDR-corrected value ( Benjamini and Hochberg , 1995; Nix et al . , 2008 ) ( false discovery rate ) was less than 0 . 02 ( i . e . 2% false discovery ) in the comparison of ( GFP+ ) vs ( GFP- ) . Embryos were processed as previously described ( Wythe et al . , 2013 ) . For lacZ embryos , beta galactosidase activity was detected using Salmon Gal ( Sigma ) prior to processing for in situ hybridization . Immunostaining of MADM samples for GFP and Myc was as previously described ( Zong et al . , 2005 ) . Antibodies used on cryosections include: Rabbit anti-GFP ( Invitrogen , 1:1000 ) , Goat anti-Myc ( Novus , 1:200 ) , Rat anti-CD31 ( BD Pharmingen , 1:100 ) , Mouse anti-tropomyosin ( DSHB clone CH-1 , 1:50 ) . Whole-mount lacZ and indirect immunofluorescent images were obtained using a Leica dissecting microscope and camera with the Leica LAS Montage extended focus function . Confocal images were obtained on a Nikon ECLIPSE Ti 2000 confocal microscope with a Yokogawa CSU-X1 spinning disk and Hamamatsu ImagEM CCD camera . Images were processed using Volocity software ( Perkin Elmer ) . All images , including immunofluorescent , in situ hybridization , and LacZ images are representative images . At least 5 embryos ( in situ hybridization and LacZ ) or 3 independent sections ( immunofluorescence ) were examined for each experiment . Images shown represent average or representative expression levels . All mouse protocols were approved by the Institutional Animal Care and Use Committee at UCSF . In the Mesp1Cre-MADM mice , heart labeling results from a Cre-mediated chromosomal translocation that occurs within a narrow developmental time window of Cre expression , between the initiation of gastrulation and shortly there after ( E6 . 0 and E7 . 0 ) . Importantly , this translocation event does not happen in the absence of Cre-recombinase ( our observations and previously published ( Tasic et al . , 2012 ) ) and there is no visible labeling prior to translocation . We have empirically defined the frequency of this event by measuring two different variables: ( 1 ) the total number of cells that can undergo this translocation and ( 2 ) the observed frequency of labeled clones . The total number of cells that could undergo translocation was determined by counting the number of cells that had recombined a Cre-dependent reporter ( Rosatd-Tomato ) at the end of gastrulation ( E7 . 5 ) . FACS analysis determined that this number was ∼850 cells ( ∼1/3 of the total number of cells in the embryo at this time ( Figure 1—figure supplement 1C ) ) . Given the fact that we observed a total of 96 clones across 38 embryos , our clonal sampling represents 11% ( or 96 clones / 850 Mesp1+ cells ) of the total Mesp1+ population ( assuming a random sampling ) . Thus a rare subpopulation , such as a common progenitor for both the right and left ventricles , would be missed only if it represented less than 10% of the total Mesp1 population . We also measured the observed frequency of labeled clones at two different time points ( E8 . 5 and E14 . 5 ) to determine if the frequency of labeled clones changed over time ( either increased or decreased ) as might happen with additional recombination outside of our time window ( E6 . 0–E7 . 0 ) or with selective loss of a twin spot via apoptosis ( Figure 1—figure supplement 1D–E ) . Although the number of samples we have analyzed at E8 . 5 is small , we detected no significant change in the observed frequency of labeled clones at the two time points analyzed and thus conclude that recombination frequency is stable over developmental time and there is no gain or loss of clones outside our narrow labeling time window . Based on the number of observations made ( n = 96 ) and the fact that we did not observe a common progenitor ( number of successes = 0 ) , we calculated the upper and lower bounds of a 95% and 99% confidence interval ( CI ) that a common progenitor does not exist using a binomial probability ( Jeffreys interval ) appropriate for instances when the number of successes is either very close to 0 or very close to 1 . from {I1−c2−1 ( x+12 , n−x+12 ) 0 1 } 0<1−c2<11−c2≤0 for−2≤c−1≤0to {Ic−12+1−1 ( x+12 , n−x+12 ) 0 1 } 0<c−12+1<1c−12+1≤0 for−2≤c−1≤0 x = number of successes , n = sample size , c = confidence interval Ix−1 ( a , b ) is the inverse regularized beta function . We used the R Package ‘binom’ , version 1 . 1-1 for calculating these values . The calculated values for a 95% CI were: 0 ( lower ) and 0 . 01975768 ( upper ) and the calculated values for a 99% CI were: 0 ( lower ) and 0 . 03387941 ( upper ) . Thus , given the number of observations we made , we can be quite confident that a common progenitor does not exist . An additional level of confidence regarding the lineage relationship of clones in different anatomical regions of the heart can be gained by comparing the color combinations observed . For example , clone MM2 could be interpreted as one recombination event that gave rise to two clusters: one in the right ventricle and one in the left ventricle . However the color combinations seen ( red/green twin spots in right ventricle and yellow/blank twin spots in left ventricle ) exclude the possibility that these two clusters are derived from the same event . Likewise , for clone MM26 , only two labeled patches are seen in the entire embryo ( red twin spot in left ventricle and green twin spot in right atrium ) . Assuming there is no selective loss of twin spots ( as we have determined above by measuring the clonal frequency at two different time points ) we can conclude that these two clusters are derived from a single event .

Most internal organs in the body are made up of several different kinds of cells . Understanding where these cells come from and how these different cells develop from a single cell in an embryo could help to guide regenerative therapies , where tissues grown in the laboratory are used to repair damage that the body cannot repair itself . The existence of a single heart progenitor cell that can produce all of the heart's structures has long been predicted , but has so far escaped discovery . Currently , it is known that two distinct sets of heart precursor cells exist in mammals , which each produce cells for different parts of the heart . Work performed in mouse embryos has hinted that both sets of cells develop from cells that produce a protein called Mesp1 . This protein controls when many genes—including those involved in heart development—are activated . Devine et al . marked a small number of Mesp1-producing cells and followed the fate of these cells through development to see where their descendants would end up within the embryo—and specifically within the mature heart . Labeling occurred at a very early stage of development , called gastrulation , when the embryonic cells first begin to organize themselves into three tissue layers that will go on to form all the different parts of the organism . Devine et al . found that shortly after gastrulation begins , heart precursor cells are present and are already assigned to particular regions of the heart . This means that if there is a single pool of heart precursor cells , it specializes into different populations very early in the development of an embryo . Devine et al . show that during gastrulation , heart precursor cells are already split into two distinct populations: one containing the cells that go on to form the atria and left ventricle of the heart; the other consisting of the cells that will make up the right ventricle and the ‘outflow tract’ that will eventually form the great vessels leading into and out of the heart . These two populations are separated by a boundary , which Devine et al . suggest is established very early on , and will go on to form the septum that separates the left and right ventricles in the developed heart . As defects in the septum are the source of many congenital heart defects , a better understanding of the heart cell precursor populations and how they interact could help develop treatments for these conditions .

2014

Early patterning and specification of cardiac progenitors in gastrulating mesoderm

Optical super-resolution microscopy techniques enable high molecular specificity with high spatial resolution and constitute a set of powerful tools in the investigation of the structure of supramolecular assemblies such as viruses . Here , we report on a new methodology which combines Structured Illumination Microscopy ( SIM ) with machine learning algorithms to image and classify the structure of large populations of biopharmaceutical viruses with high resolution . The method offers information on virus morphology that can ultimately be linked with functional performance . We demonstrate the approach on viruses produced for oncolytic viriotherapy ( Newcastle Disease Virus ) and vaccine development ( Influenza ) . This unique tool enables the rapid assessment of the quality of viral production with high throughput obviating the need for traditional batch testing methods which are complex and time consuming . We show that our method also works on non-purified samples from pooled harvest fluids directly from the production line . The potential of super-resolution microscopy ( SRM ) to unravel details of the structure and replication of viruses was recognised early on in the development of the methodology ( Betzig et al . , 2006; Müller and Heilemann , 2013 ) . Since then , SRM has been used to provide unprecedented insights into viral protein architecture ( Laine et al . , 2015; Zhang et al . , 2015; Gray et al . , 2016; Albecka et al . , 2016 ) . Previous work has focused on those SRM techniques that achieve the highest theoretical resolution , such as Stimulated Emission Depletion ( STED ) ( Hell and Wichmann , 1994 ) and Single Molecule Localisation Microscopy ( SMLM ) ( Rust et al . , 2006; Heilemann et al . , 2008 ) . Whilst offering high fidelity data , the downside is the associated long acquisition time required by these methods , limiting their application to the imaging of static samples at low throughput . A much faster technique , although inferior in spatial resolution , is Structured Illumination Microscopy ( SIM ) ( Gustafsson , 2000; Heintzmann and Cremer , 1999 ) and this has been applied to study large viruses such as the prototypic poxvirus ( Gray et al . , 2016; Horsington et al . , 2012 ) . In addition to understanding the structure of viruses , there is also a need to identify and analyse classes of structures within large viral populations , especially in the biotechnology industry where virus quality is often compromised by large scale production operations and the virus product is often characterised by significant morphological heterogeneities . In particular , campaigns of influenza immunization rely heavily on the timely and efficient production of specific virus strains . Similarly , a deeper understanding of the structural heterogeneity of oncolytic viruses such as Newcastle Disease Virus ( NDV ) ( Ganar et al . , 2014; Lichty et al . , 2014 ) would enable optimization of the production processes and in turn improve the development of viriotherapy . However , quantifying and understanding this structural heterogeneity and relating it to virus efficacy requires the imaging of large numbers of viruses at sufficient spatial resolution to reveal characteristic morphological details . Typically , this is achieved by extracting batches from the production process , with elaborate subsequent purification and preparation steps before characterisation by Transmission Electron Microscopy ( TEM ) ( Gad , 2007; Goldsmith and Miller , 2009; Brenner and Horne , 1959 ) . Although TEM can achieve relatively high imaging throughput if the highest resolution is not necessary , its contrast remains unspecific and therefore does not typically permit discerning the presence of particular proteins in the virus envelope . Also , the typical signal-to-noise ratios achieved by TEM are not sufficient to permit automated , robust and efficient downstream analysis of structural features at the single particle level . It is therefore challenging for TEM to be of practical use during production operations . Here , we demonstrate that rapid high resolution imaging with Total Internal Reflection SIM fluorescence microscopy ( TIRF-SIM ) ( Shao et al . , 2011; Kner et al . , 2009; Young et al . , 2016 ) , combined with a machine learning ( ML ) approach to analyse and classify structures in virus batches offer a great opportunity to circumvent these problems . We present MiLeSIM ( Machine Learning Structured Illumination Microscopy ) as an efficient combination of SRM , ML-based classification ( Van Valen et al . , 2016; Sommer et al . , 2011 ) and advanced image analysis for the quantification of morphological heterogeneities in large virus populations . We use ML algorithms to perform a classification of super-resolved images of a heterogeneous virus population into particle classes with distinct and characteristic structural features ( e . g . spherical , filamentous ) . The classified subpopulations are then further analysed through image analysis pipelines that are specifically adapted for each structural class . We and others have shown that appropriate model fitting can lead to precision in structural parameters beyond the resolution of the images used ( Laine et al . , 2015; Manetsberger et al . , 2015 ) . The method combines speed and specificity and allows an in-depth exploration of large virus populations that is unachievable by TEM . The method has potentials in the industrial production of viruses , for example for oncolytic viriotherapy and vaccine development . First , we compare TIRF-SIM with alternative imaging modalities and show that it is the method of choice to investigate virus structure at high-throughput ( ~220 virus particles/s , see Supplementary Note 1 ) with a spatial resolution reaching ~90 nm . The large datasets obtained with TIRF-SIM are then fed into an ML algorithm for the automated classification of Newcastle Disease Virus ( NDV ) and live attenuated influenza virus ( LAIV ) vaccines , enabling further shape-specific quantitative analyses for a structural description of viral subpopulations . The purpose of our study is to validate the MiLeSIM approach as a powerful analysis tool for biotechnological processes involving virus production both in industry and in the research laboratory . First , we explored and compared three common SRM modalities for the structural investigation of purified NDV virus , namely direct stochastic optical reconstruction microscopy , dSTORM , stimulated emission depletion microscopy , STED and TIRF-SIM . NDV viruses were labelled for the envelope glycoprotein Hemagglutinin-Neuraminidase ( HN ) and imaged with all three SRM imaging techniques ( see Figure 1 ) . Labelling for HN allows us to directly and specifically observe the shape of the virus particles . For comparison , a conventional ( non-super-resolved ) TIRF wide-field image is also shown . Typical shapes observed with TIRF-SIM are shown in Figure 1 ( b ) . TIRF-SIM provides clear structural details to discern filamentous , spherical and rod-like structures in large NDV populations . A comparison of performance parameters ( resolution and imaging speed ) for the different methods is presented in Figure 1—figure supplement 1 ( a ) . It is clear that improving resolution beyond the ~90 nm offered by TIRF-SIM ( see Figure 1—figure supplement 1 ( b ) ) comes at a significant cost in acquisition times and throughput . Furthermore , although dSTORM and STED offer theoretically higher resolution than SIM , the images obtained with these methods do not reveal additional structural details that are not also resolved by TIRF-SIM images . This indicates that the ~2 fold resolution improvement provided by TIRF-SIM is sufficient for the structural study presented here . Traditionally , EM has been the method of choice for observing sub-diffraction structures of virus particles ( see Figure 1—figure supplement 2 for examples of particles ) . Here we show that TIRF-SIM can offer significant advantages compared to EM ( summarized in Table 1 ) . The improvement in molecular specificity allows an unambiguous identification of viral components; the high signal-to-noise ratio ( SNR ) furthermore enables a robust and straightforward application of further image analysis steps ( identification and classification of virus particles ) . Also , the capability of investigating unpurified and aqueous samples makes TIRF-SIM ideally suited to the present application . The images obtained with TIRF-SIM show a number of stereotypical virus structures in NDV samples labelled for HN , indicating a large morphological diversity in the virus populations that may stem from variability occurring during viral replication or at the purification stage . Understanding the origins and consequences of such heterogeneity informs not only on the life cycle of the virus but can also provide essential insights into the virus production process to manufacturers of virus-based therapeutics . An automated classification of virus shapes would enable the quantification of virus heterogeneity and permit further analysis of each individual class independently . The workflow to achieve these goals is shown in Figure 2 . Individual virus particles are first identified by automated segmentation and then fed to the ML routine for classification . We used a supervised ML algorithm ( here a random forest algorithm ( Breiman , 2001 ) ) to ensure the robustness of the method and for ease of implementation . We identified six major structural classes in the NDV samples which we divide into long and short filamentous , small and large spherical , rod-like and unknown structures . The unknown class is made of clumps of viral material with no consistent and identifiable shapes . A control sample that was prepared identically to the other samples except without virus particles present allowed us to identify that non-specific bindings of antibodies appear as rare , dim and small point-like structures that could easily be discriminated and excluded from further analysis . The filamentous ( long and short ) class is further analysed by automatic extraction of the linear backbone of structures and measurement of their length . The width of these filamentous structures appeared to be limited by the resolution of the imaging technique ( ~90 nm for TIRF-SIM , see Figure 1—figure supplement 1 , but also observed in higher resolution approaches such as dSTORM ) and therefore , we considered the filamentous class as 1D structures . The spherical structures were analysed by estimating their equivalent radius from the area of the particle . We note that other methods for estimation of the radius , for example the ellipsoid localization microscopy ( ELM ) analysis ( Manetsberger et al . , 2015 ) , could also be used here . The latter fits a shape model to imaging data to permit the extraction of structural parameters with precision higher than the inherent resolution of the imaging method ( Laine et al . , 2015; Manetsberger et al . , 2015 ) . A similar model-based fitting approach was used to fit rod-like viral particles and to obtain length and width parameters for this structural class ( see Materials and methods section and Figure 2—figure supplement 1 for details ) . The structural classification was performed using a supervised ML algorithm which allows for rapid and automated classification of large datasets . The choice of algorithm and the set of features ( often called predictors ) extracted for each identified particle were optimised to maximise the overall accuracy of the model based on the training dataset ( comprising of 370 manually annotated particles ) . Here , the model accuracy is defined as the fraction of correctly classified particles across all classes . Figure 3 ( a ) describes the list of chosen individual features ( selected from basic shapes features , Hu’s image moments ( Hu , 1962 ) , features obtained from the pre-trained convolutional neural network ( CNN ) AlexNet ( Krizhevsky et al . , 2012 ) and from Speeded Up Robust Features , SURF ( Bay et al . , 2008 ) ) . The predictors were selected based on the following criteria: basic structural features of the particles ( e . g . area , eccentricity ) and Hu’s moments were chosen because they are rotationally and translationally invariant . For the features from AlexNet and SURF a feature selection approach was designed based on maximising the standard deviation across the different structural classes . This approach constitutes a more rational choice compared to simple principal component analysis ( PCA ) , which does not typically take the information regarding the classes into account , therefore our method selects for predictors that have high potential for class discrimination . This data reduction narrowed down the number of predictors to six for AlexNet and six for SURF . A total of 24 predictors was finally chosen: seven based on basic shapes ( area , ratio of axis lengths , eccentricity , solidity , perimeter-to-area , mean intensity , standard deviation of pixel intensities ) , 5 of Hu’s image moments ( Hu1 , Hu4 , Hu5 , Hu6 and Phi4 ) , six features obtained from the pre-trained convolutional neural network ( CNN ) AlexNet and six from a SURF bag of features . The classification workflow is described in Figure 3—figure supplement 1 . Panels described in Figure 3 ( b ) show examples of scatter plot from arbitrarily chosen pairs of predictors highlighting that some predictors support classification across specific classes better than other combinations ( identifiable clusters of certain classes ) . Here , the training dataset was used to build a scatter plot of pairs of predictors with the knowledge of their true classification ( see colour scheme ) . For instance , the pair of predictors L1/L2 and Area shows a good separation between long filamentous ( dark blue labels ) and unknown structures ( black labels ) . The confusion matrix ( Figure 3 ( c ) ) highlights the effective true positive rate ( TPR ) and positive predictive values ( PPV ) across the different classes with a model accuracy of 88 . 4% . We note that some long filamentous viruses are misclassified as small filamentous , some small filamentous are misclassified as small spherical and that , on occasion , some unknown structures populate the predicted long and large spherical structure classes . Considering the simple shapes of these viruses , it is expected that a small fraction of particles are misclassified as structures with close resemblance . The scoring of the predictors presented in Figure 3 ( d ) indicates the average accuracy of each individual predictor . A high score indicates a high capacity to discriminate between different classes . The scoring was performed by measuring the accuracy of the classification for many combinations of predictors and distributing the accuracy score across the predictors tested ( see Materials and methods for details ) . In other words , if a combination of 2 predictors alone give an accuracy of 60% , a score of 30% is awarded to both individual predictors . This method was repeated and scores represent averages across >13 , 000 different combinations of predictors . We analysed a total of ~6500 particles using MiLeSIM and established that 49 . 7% of NDV particles presented a filamentous shape whereas the large spherical , small spherical and rods represent 18 . 6% , 7 . 8% and 7 . 3% of the total population , respectively ( Figure 4 ( a ) ) . In addition to structural classification , the high-resolution images also permitted a dimensional analysis to be performed at the single particle level . We estimated the particle radius from both small and large spherical particles by calculating the equivalent radius from the particle area; backbone extraction to the short and long filamentous particles , to estimate the particle length; and designed a model fitting for the rod structures . Figure 4 shows the distribution of structural parameters for each class . We observe that both long filamentous and large spherical are well described by a Gamma distribution whereas the small filamentous and small spherical are well described by a Gaussian distribution . The model-fitting applied to the rod-shaped particles ( see Materials and methods and Figure 2—figure supplement 1 for details ) allows the extraction of both the width and length of each particle . Therefore , it is possible to plot the distribution of structural parameters as a contour plot Figure 4 ( c ) ) . We estimated the mean and standard deviation of the structural parameters from the distributions and obtained: LLF = 650 ± 430 nm , LSF = 200 ± 100 nm , DLS = 338 ± 94 nm , DSS = 190 ± 10 nm . For the rod-shaped particles , we observed that the width WRD = 135 ± 30 nm and the length LRD = 610 ± 350 nm ( all rounded to two significant figures , ± represents the standard deviation of the distribution ) . These values are distributed around two populations as shown on the contour plot in Figure 4 ( d ) . However , we note that the radius analysis based on the area of the particle used here constitutes an overestimate of the physical radius of the particle due to the broadening caused by the point-spread function . It is possible to estimate a more accurate diameter of the underlying spherical structures by using the ELM analysis . The results obtained from the ELM analysis of the large spherical structures are shown in Figure 4—figure supplement 1 ( a ) . The ELM diameter obtained for the large spherical particles ( 220 ± 69 nm ) is in good agreement with an area-based diameter of 338 nm and a resolution of 90 nm . It should be noted that the small spherical distribution is centred on the value of optical resolution of our SIM microscope , which indicates that the small spherical structures the small spherical structures are smaller than the point-spread-function . We applied our approach to four different strains of Live Attenuated Influenza Virus ( LAIV ) immuno-labelled for the glycoprotein Hemagglutinin ( HA ) present on the exterior of the viral envelope . The shape of the virus particles obtained here were classified using the same classifier as for NDV . The LAIV virus population was dominated by spherical structures ( >60% ) . Figure 5 shows the distribution of particle sizes for four virus strains: a B-Victoria subtype ( B/Brisbane/60/2008 ) , a B-Yamagata subtype ( B/Phuket/3073/2013 ) and two subtype A H1N1 strains ( A/South Dakota/06/07 and A/Bolivia/559/2013 ) . The fractions of small and large spherical particles are shown , as well as the equivalent radii and representative images of the viruses . It is clear that B-Victoria particles consist of mostly large hollow particles with an equivalent radius of ~130 nm , a value that is in good agreement with the ELM analysis and a resolution of 90 nm ( Figure 4—figure supplement 1 ( b ) ) . In contrast , the B-Yamagata strain shows small and large particles of equal amount , indicating that the particles sizes are distributed around the region of overlap between small and large particles . This is confirmed by the nearly identical equivalent radius distributions . Both A strains appeared clearly dominated by small spherical particles with sizes close to the resolution limit of our imaging . However , our high-throughput approach reveals subtle differences in the distribution of small spherical structures where the A/South Dakota viruses appear more heterogeneous ( standard deviation ~10 nm ) , whereas the A/Bolivia viruses are sharply distributed ( standard deviation ~4 nm ) . We also investigated the potential of directly imaging pool harvested fluid ( PHF ) . LAIV are commonly propagated in embryonated hens’ eggs where progeny viruses are released into the allantoic fluid of the egg . This fluid is harvested from numerous eggs and pooled . This constitutes a very basic and commonly used virus material . It is easy to produce and does not undergo any downstream purification . Consequently , PHF is impure , containing a variety of egg-derived impurities . The high molecular specificity of fluorescence microscopy allowed us to visualize the structure of the viruses with the same image quality directly in PHF despite the presence of a large amount of impurities ( Figure 5—figure supplement 1 ( a ) ) . The structural analysis of the B-Victoria strains from MVB ( Figure 5 ) and PHF ( Figure 5—figure supplement 1 ( b ) ) allows us to decipher the effect of purifications steps on the structural properties of the population . The fraction of unknown structures dropped from 23% to 8% between MVB and PHF respectively . This change in fraction of unknown structure may be a result of the different densities of viruses on the cover slip . We observed a lower density of virus particles in the PHF preparations , which may lead to fewer aggregated classes and therefore fewer unidentifiable structures . In addition , whereas the PHF shows a nearly equal amount of small and large spherical structures , the MVB preparation is missing a large population of small spherical compared to the PHF . This is also reflected by the larger average diameters observed in the MVB compared to the PHF ( DSS = 191 ± 12 nm and 198 ± 14 nm and DLS = 241 ± 56 nm and 275 ± 49 nm for PHF and MVB respectively ) . MiLeSIM therefore enables the study of unpurified samples and allows probing the virus production at any intermediate levels of production and purification . This constitutes a strong advantage over EMtechniques which require the use of highly purified samples and elaborate preparation protocols . We have demonstrated the potential of high-throughput imaging of virus structures , taking advantage of the optimal combination of speed and resolution afforded by the TIRF-SIM imaging method . TIRF-SIM provided sufficient resolution to identify , discriminate and analyse individual viral structural classes with high specificity , even in non-purified samples . Our approach combines machine learning to classify NDV viruses , followed by a model-based or direct quantification of virus structural parameters . The method yielded similar results both in purified samples and in samples from unfiltered PHF offering promise for use as an assay during virus production . We were able to image up to ~220 particles/second at 90 nm resolution , vastly increasing imaging throughput compared to alternative super-resolution methods , improving sensitivity and specificity in comparison to EM . Furthermore EM does not feature the specificity to analyse virus samples in their aqueous , unaltered unpurified forms . We were able to observe large structural variabilities in the NDV population and also between different strains of LAIV . Our particular classification uses random forest with a selection of predictors from simple shape parameters , rotational and translational invariant image moments and features from AlexNet and common feature for image recognition such as SURF . The model accuracy is ~88 . 4% and the mis-classifications occur between classes that are similar ( between small spherical and small filamentous for instance ) . The structural parameters that we extract from the model fitting are precise beyond the image resolution as they take into account the finite optical resolution . This therefore reveals subtle differences in populations such as the two sub-classes observed in the rod-shaped class . This approach will be beneficial especially when heterogeneous populations are present and need to be quantified . In future , such information can be correlated with functional characteristics of produced virus classes and production parameters can accordingly be optimised . The approach thus holds great promise for the production of virus-based therapeutics . We note , however , that the methods presented are generally applicable to other systems and they are not restricted to a particular type of fluorescence microscopy , SRM or not . The purified NDV samples were prepared on cover slips as previously described ( Laine et al . , 2015 ) . Briefly , viruses were adhered on poly-L-lysine-coated Ibidi 8-well dishes , fixed , permeabilised and immuno-labelled for the envelope glycoprotein Hemagglutinin-Neuraminidase ( HN ) with primary antibodies ( mouse anti-hemagglutinin-neuraminidase HN , Abcam , UK ) followed by secondary labelling ( goat anti-mouse labelled with Alexa Fluor 647 for dSTORM , with Alexa Fluor 488 for TIRF-SIM and with ATTO647-N for STED , Abcam , UK ) . The LAIV samples were prepared identically but using primary antibodies originating from MedImmune in-house , non-commercially available monoclonals that target the viral glycoprotein Hemagglutinin ( HA ) present on the exterior of the viral envelope: F16 mouse antibody for B-Victoria , Infa0121 mouse antibody for B-Yamagata and FY1 human antibody ( Kallewaard et al . , 2016 ) for A South Dakota and A Bolivia . The corresponding secondary antibodies were used ( donkey anti-mouse DyLight 488 labelled or rabbit anti-human DyLight 488 labelled antibodies , ThermoFisher ) . All virus samples originated from the monovalent bulk ( MVB ) and are therefore highly purified , unless indicated in the text , where the direct pool harvest fluid ( PHF ) was used . Our custom-built TIRF-SIM system was described previously ( Young et al . , 2016 ) . We used an Olympus UAPON 100x TIRF NA = 1 . 49 and an Orca Flash 4 . 0 camera , with a sample pixel size of 64 nm . A total of 9 SIM images were acquired ( three phases , three orientations ) with a camera exposure time of 200 ms and ~250 µW of 488 nm laser , measured at the back aperture of the objective . The SIM images were obtained using the reconstruction code provided by Dr Lin Shao ( Shao et al . , 2011 ) , providing images with doubled resolution and 32 nm final pixel size using a Wiener filter of 0 . 01 . The STED imaging was performed on our custom-built STED microscope as described previously ( Mahou et al . , 2015 ) . The dSTORM imaging was performed on a custom-built single-molecule microscope previously described ( Ströhl et al . , 2017; Wong et al . , 2017 ) and with mercaptoethylamine ( MEA ) buffer as previously described ( Laine et al . , 2015 ) . The dSTORM image reconstruction was carried out using rapidSTORM 3 ( Wolter et al . , 2012 ) . The resolution achieved by the TIRF-SIM microscope was assessed by identifying the edge of the spatial frequency support using the SIMcheck plugin ( Ball et al . , 2015 ) , as shown in Figure 1—figure supplement 1 . For STED microscopy , the resolution was estimated from cross-sections of 20 nm beads and reporting the full width at half maximum ( FWHM ) . The dSTORM resolution reported here was obtained from the FWHM of the localization precision , estimated by ( Thompson et al . , 2002 ) . All segmentations , predictors extractions and classifications were performed using MATLAB ( Mathworks ) . The code is freely available ( Laine , 2018 ) . A general diagram of the method is shown in Figure 1—figure supplement 1 . The segmentation was obtained by an initial Otsu binarization and refined by active contour . This allowed a better outline of the particles and efficient separation of particles in close proximity . The particles that were judged too small or too dim to be real particles ( based on criteria obtained from the control sample ) were excluded from further analysis . The basic shape features were extracted using the MATLAB function regionprops . Hu’s image moments were computed from the 71 × 71 pixels particle image centred on the centre of mass of the particle . The absolute values of the logarithm of the moments were used in the classification . For the features obtained from AlexNet ( Krizhevsky et al . , 2012 ) , the individual 71 × 71 pixels images were resized to 227 × 227 pixels and used as all three color layers of the RGB images taken by AlexNet . Then , feature extraction was performed using AlexNet as a pre-trained network . 4096 features were obtained and data reduction was performed to limit the number of predictors used . For this , the features were averaged within each individual class and the standard deviation of every feature across the classes was computed . The six features with the highest standard deviation was selected . For the SURF features , first a bag of visual words was created from the training dataset , this bag was then used to check the presence of visual words in the 71 × 71 pixels images of individual particles . Similarly to AlexNet features , we selected only the six visual word features with the highest standard deviation across the different classes for classification . This allowed the computation of a total of 24 features for ML . The classification was performed using a random forest algorithm . The training dataset was made of 370 manually labelled individual particles and was used to train the random forest across 60 epochs . The classification was validated by 10-fold cross validation on the same dataset . The confusion matrix obtained from this cross-validation is shown in Figure 3 . At the training stage , the training dataset was augmented 5-fold by transforming the images with image translation and rotation randomly picked between 0 and 1 pixel and between 0 and 360 degrees respectively . The accuracy of the model was estimated by calculating the fraction of correctly classified particles across all classes . accuracy=Number of particles correctly classifiedTotal number of particles The predictors were scored by computing the accuracy of the random forest trained on the training dataset but with only subsets of features . Out of the 24 predictors all combinations of 2 , 3 , 4 , 24 , 23 and 22 predictors were tested corresponding to a total of 13 , 227 combinations of predictors . For each combination of predictors , the accuracy obtained was split equally across the different predictors used , producing a ‘local’ accuracy for each feature . This local score was average across all combinations using a specific feature to obtain the global score . Si=1Nci∑j=1NcaijPjnj Where Si is the global score of the feature i , Nci is the total number of combinations tested involving feature i , aij is a factor reflecting the presence of the feature i in the combination j . aij is equal to one if i is present in j , 0 otherwise . Pj is the accuracy of the combination j , Nc is the total number of combination tested and nj is the number of features present in the combination j . All quantitative analyses were performed using MATLAB ( Mathworks ) . The code is freely available ( Laine , 2018 ) . The length of the filamentous structures were extracted by measuring the geodesic distance along the skeletonized image of the filament . The ELM analysis is freely available ( Manetsberger et al . , 2015 ) and the code was adapted to insert within the workflow of our approach . For ELM analysis , we observed no significant ellipticity in the spherical virus particles and fitted spherical shapes to extract the radius of the particles ( Figure 4—figure supplement 1 ) . The equivalent radius r of the spherical particles were simply calculated from the area A of the segmented particle . r=Aπ The image model for the rod-shaped particles is presented in Figure 1—figure supplement 1 . Briefly , the backbone of the particle was extracted by image thinning and then dilated by a disk-shaped kernel of radius equal to half of the width of the rod . The length of the rod could be adjusted by shortening the ends of the backbone or by extrapolating it outwards to lengthen it . The interior pixels of the image obtained were removed to leave the outline of the particle shape . This outline was then convolved with a Gaussian kernel in order to take into account the effect of the image resolution ( here 90 nm ) . The intensity , the width and length of the model image were adjusted to minimize the sum of the square difference of intensity χ2 . χ2=∑ijImi , j-Id ( i , j ) 2 Where i and j refer to the indices in the image , Im ( i , j ) is the image model , and Id ( i , j ) is the data image . The imaging throughput of the method can be assessed in terms of number of particles imaged per second . The field-of-view achievable in our TIRF-SIM system is ~32 µm x 32 µm and a high quality sample preparation can yield a virus particle density of ~1 particle/µm2 . Therefore , with an acquisition time of 200 ms/SIM raw frame ( with a total of 9 frames ) , we assess that our single frame particle throughput can reach ~500 imaged particles/s . However , the acquisition of two consecutive fields-of-view are affected by imaging dead time as a consequence of stage movement and refocussing . In the study presented here , this step was done manually and took approximately 2–3 s . Therefore , a practical throughput achievable for the imaging is of the order of ~220 particles/s . We note however that both acquisition times and the stage movement time can be easily reduced by increasing illumination power and automation respectively . This makes the 500 particles/s not an unreasonable estimation for the achievable throughput of a further optimised acquisition . The throughput of the method can also be regarded as the time necessary to perform the complete study from sample preparation to analysis . Table 2 indicates typical times necessary to perform the individual steps of the workflow . This table indicates that a full structural analysis of a particular sample can be obtained within a day .

Viruses are like the Trojan horses of the biological world; they sneak their genetic code into a living cell and then hijack it , forcing that cell to produce their own viral proteins . Yet , if scientists replace the harmful genes in a virus with other genes , the virus can be transformed into a powerful tool for biology and medical science . For example , viruses can be turned into vaccines that prime the immune system to ward off future infections . Viruses could also be made to deliver the genetic code needed to repair faulty cells , and thus treat the cause of an illness from inside the body . Nevertheless , it is complicated to produce viruses like these on a large scale . The individual viruses in one batch can be very different shapes and sizes; they can also end up displaying different proteins on their outer surface – which is the part of the virus that our immune system will see first . To optimise the production of standardised viruses , scientists need a way to test the viruses throughout the manufacture process . At the moment , the best way to do this would be with electron microscopes . Yet these microscopes cannot tell exactly which proteins are in the outer surface of the virus . Also , these methods often need purified samples of virus , so cannot be used to look at the viruses until the final stage of production . Laine et al . now report a method that can test virus production at every step of the process . This new method uses a different type of microscopy called super-resolution imaging , which is quicker than electron microscopy and more able to deal with impurities , but can still see objects that are 500 times smaller than the width of a human hair . First , Laine et al . took pictures of many viruses with this new imaging technique , sorted the images into groups based on their appearance , and then trained computer algorithms with the pre-sorted groups ( a technique called “supervised learning” ) . Next , the trained algorithms were shown new images of viruses and asked to classify them . The algorithms could separate images of a mixed population of viruses into six groups according to their shape and size , and then analyse each group in a specific way . For example , they would measure and report the length of filament-shaped viruses , the radius of spherical viruses and the length and width of rod-shaped viruses . The first set of test images were of Newcastle Disease Virus , which is currently under development as a treatment for cancer . But further testing revealed that the algorithm also works for the influenza virus , which is used to make flu vaccines . The algorithm could classify the viruses in pure and impure samples , and the imaging technique could handle over 200 viruses each second . This approach of combining super-resolution imaging with artificial intelligence could help scientists to understand what makes good vaccines and how best to optimise the production of viruses for medical purposes . It could also allow researchers to respond more rapidly to outbreaks of viral infections . The next step is to build this work into a system that can be used by the pharmaceutical industry .

2018

Structured illumination microscopy combined with machine learning enables the high throughput analysis and classification of virus structure

Coordinated gastrointestinal smooth muscle contraction is critical for proper nutrient absorption and is altered in a number of medical disorders . In this work , we demonstrate a critical role for the RGD-binding integrin α8β1 in promoting nutrient absorption through regulation of gastrointestinal motility . Smooth muscle-specific deletion and antibody blockade of α8 in mice result in enhanced gastric antral smooth muscle contraction , more rapid gastric emptying , and more rapid transit of food through the small intestine leading to malabsorption of dietary fats and carbohydrates as well as protection from weight gain in a diet-induced model of obesity . Mechanistically , ligation of α8β1 by the milk protein Mfge8 reduces antral smooth muscle contractile force by preventing RhoA activation through a PTEN-dependent mechanism . Collectively , our results identify a role for α8β1 in regulating gastrointestinal motility and identify α8 as a potential target for disorders characterized by hypo- or hyper-motility . Coordinated gastrointestinal motility orchestrates nutrient absorption by mixing ingested food with digestive enzymes ( Armand et al . , 1994; Kong and Singh , 2008 ) and by propelling the food bolus from the stomach to the small intestine , the primary site of nutrient absorption . Dysfunctional gastrointestinal motility occurs in a number of common disease processes ( Ambartsumyan and Rodriguez , 2014; Fan and Sellin , 2009 ) , is difficult to treat , and is characterized by either accelerated motility leading to rapid intestinal transit times , diarrhea , and malabsorption ( Spiller , 2006 ) or delayed motility leading to nausea , vomiting , and aspiration of stomach contents ( Enweluzo and Aziz , 2013; Janssen et al . , 2013 ) . A better understanding of the molecular mechanisms that regulate gastrointestinal motility has significant clinical implications for disorders characterized by hypo- or hyper-motility . Milk fat Globule Epidermal Growth Factor like 8 ( Mfge8 ) is an integrin ligand ( Atabai et al . , 2009 ) that is highly expressed in breast milk ( Atabai et al . , 2005 ) . Mfge8 coordinates absorption of dietary fats by promoting enterocyte fatty acid uptake after ligation of the αvβ3 and αvβ5 integrins ( Khalifeh-Soltani et al . , 2014 ) . Mfge8 also modulates smooth muscle contractile force ( Kudo et al . , 2013 ) . In mice deficient in Mfge8 ( Mfge8-/- ) , airway and jejunal smooth muscle contraction are enhanced in response to contractile agonists after these muscle beds have been exposed to inflammatory cytokines but not under basal conditions ( Kudo et al . , 2013 ) . Contraction of antral smooth muscle is a key determinant of the rate at which a solid food bolus exits the stomach and transits through the primary site of nutrient absorption , the small intestine ( Haba and Sarna , 1993; Kelly , 1980; Burks et al . , 1985 ) . Since Mfge8 promotes enterocyte fatty acid uptake and can regulate smooth muscle contraction , we were interested in examining whether Mfge8 reduces the force of basal antral smooth muscle contraction , thereby slowing gastrointestinal motility and allowing a greater time for nutrient absorption . α8β1 is a member of the RGD-binding integrin family that is prominently expressed in smooth muscle ( Zargham et al . , 2007; Zargham and Thilbault , 2006; Schnapp et al . , 1995 ) . The most definitive in vivo role described for α8β1 is in kidney morphogenesis where deletion of this integrin subunit leads to impaired recruitment of mesenchymal cells into epithelial structures ( Müller et al . , 1997; Humbert et al . , 2014 ) . Osteopontin , fibronectin , vitronectin , nephronectin , and tenascin-C have all previously been identified as ligands for α8β1 ( Schnapp et al . , 1995; Denda et al . , 1998; Kiyozumi et al . , 2012 ) . In this work we show that Mfge8 is a novel ligand for α8β1 and that Mfge8 ligation of α8β1 reduces the force of gastric antral smooth muscle contraction , the extent of gastric emptying , and the rate at which a food bolus transits through the small intestine . We further show that mice with smooth muscle-specific deletion of α8 integrin subunit ( Itga8flox/flox—Tg ( Acta2-rtTA , TetO-Cre ) fail to properly absorb ingested fats and carbohydrates and are partially protected from weight gain in a model of diet-induced obesity . α8β1 slows gastrointestinal motility by increasing the activity of Phosphatase and tensin homolog ( PTEN ) leading to reduced activation of the Ras homolog gene family member A ( RhoA ) . To determine whether Mfge8 regulates the force of antral smooth muscle contraction , we isolated gastric antral rings and measured the force of antral contraction in a muscle bath . Antral rings isolated from Mfge8-/- mice had increased force of contraction in response to both methacholine ( MCh ) and KCl as compared with wild type ( WT ) controls ( Figure 1A and B ) . The thickness of antral smooth muscle was not different when comparing Mfge8-/- and WT mice indicating that the enhanced contraction was not due to smooth muscle hypertrophy ( Figure 1—figure supplement 1A and B ) . Incubation with recombinant Mfge8 ( rMfge8 ) , but not a recombinant construct where the integrin-binding RGD sequence was mutated to RGE , rescued enhanced contraction indicating that the effect of Mfge8 on gastric smooth muscle was integrin-dependent ( Figure 1A , B ) . Induction of Mfge8 expression in the smooth muscle of Mfge8-/-—Tg ( Acta2-rtTA , TetO-Mfge8 ) transgenic mice , abbreviated Mfge8-/-sm+ , where Mfge8 expression was driven by a tetracycline-inducible Mfge8 transgene coupled with an α-smooth muscle-rtTA transgene ( Figure 1—figure supplement 1C ) also rescued enhanced contraction ( Figure 1C; Figure 1—figure supplement 1D ) . Of note , unlike antral rings , duodenal rings from Mfge8-/- mice did not have enhanced contraction ( Figure 1—figure supplement 1E ) consistent with our previously reported findings in jejunal smooth muscle rings ( Kudo et al . , 2013 ) . We next determined whether enhanced antrum contractility was associated with altered gastric emptying and small intestinal transit times ( SIT ) , two in vivo measures of gastrointestinal motility . Mfge8-/- mice had significantly more rapid gastric emptying and more rapid SIT ( Figure 1D–G ) . Administration of rMfge8 by gavage and transgenic smooth muscle expression of Mfge8 significantly reduced the rate of gastric emptying and prolonged SIT in Mfge8-/- mice ( Figure 1D–G ) . Administration of rMfge8 by gavage also significantly reduced gastric emptying and prolonged SIT in WT mice ( Figure 1D and F ) . 10 . 7554/eLife . 13063 . 003Figure 1 . Mfge8 regulates gastrointestinal motility . ( A–C ) Force of antral smooth muscle ring contraction with and without the addition of rMfge8 or RGE construct in Mfge8-/- and WT in response to MCh ( A , N = 4–5 ) or KCl ( B , N = 4–5 ) or after in vivo induction of smooth muscle Mfge8 expression in Mfge8-/-sm+ mice in response to MCh ( C , N = 5 ) . ( D , E ) The rate of gastric emptying in Mfge8-/- and WT with and without the addition of rMfge8 or RGE construct ( D , N = 10 ) or after smooth muscle transgenic ( Mfge8-/-sm+ ) expression of Mfge8 ( E , N = 7 ) . ( F–G ) Small intestinal transit time in Mfge8-/- and WT with and without the addition of rMfge8 or RGE construct ( F , N = 5–10 ) or after smooth muscle transgenic expression of Mfge8 ( G , N = 4–5 ) . Female mice were used for all experiments . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . Data are expressed as mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 13063 . 00310 . 7554/eLife . 13063 . 004Figure 1—figure supplement 1 . Mfge8-/- smooth muscle morphology and smooth muscle expression of Mfge8 in Mfge8-/-sm+ mice . ( A , B ) Morphometric analysis of antrum from WT and Mfge8-/- mice . ( A ) Representative image of antral smooth muscle in WT and Mfge8-/- mice . ( B ) Thickness quantitation of muscularis mucosae , circular layer , and longitudinal layer of antral smooth muscle . ( C ) Western blot of antrum lysates probing for Mfge8 after transgenic induction of smooth muscle Mfge8 in Mfge8-/-sm+ after doxycycline administration . ( D ) Force of antral smooth muscle ring contraction after in vivo induction of smooth muscle Mfge8 expression in Mfge8-/-sm+ mice in response to KCl ( N = 4–5 ) . ( E ) Force of duodenal smooth muscle ring contraction in response to MCh ( N = 4 ) . Both male and female mice were used for these experiments . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . Data are expressed as mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 13063 . 004 The αvβ3 and αvβ5 integrins are the known cell surface receptors for Mfge8 ( Atabai et al . , 2005; Hanayama et al . , 2004; Hanayama et al . , 2002 ) and mediate the effect of Mfge8 on fatty acid uptake ( Khalifeh-Soltani et al . , 2014 ) . We therefore investigated whether these integrins mediated the effect of Mfge8 on gastrointestinal motility . Antrum contraction was similar in WT , Itgb3-/- , Itgb5-/-and Itgb3-/-::Itgb5-/- mice ( Figure 2—figure supplement 1A ) . Gastric emptying and SIT were also similar in Itgb3-/- , Itgb5-/-and Itgb3-/-::Itgb-/- mice and rMfge8 significantly reduced the rate of gastric emptying and prolonged SIT in each mouse line ( Figure 2—figure supplement 1B and C ) . These data indicate that the effect of Mfge8 on smooth muscle contraction occurs via a novel RGD-binding , integrin partner . We have previously shown that Mfge8 does not bind the RGD-binding integrins αvβ6 , αvβ8 , and α5β1 ( Atabai et al . , 2009 ) , leaving α8β1 and αvβ1 as the potential RGD-binding receptors for the effect of Mfge8 on smooth muscle contraction . We initially focused on α8β1 because of its high expression in smooth muscle ( Schnapp et al . , 1995; Kitchen et al . , 2013 ) . To determine if α8β1 is a receptor for Mfge8 , we used a solid-phase assay to analyze the direct binding of Mfge8 to purified α8 . We included purified αvβ3 and α5β1 as positive and negative controls , respectively . Mfge8 bound to α8β1 and αvβ3 , but not to α5β1 ( Figure 2A ) . To further confirm this interaction , we evaluated cell adhesion of SW480 cells , a human colon cancer cell line , transfected with α8 or β3 to Mfge8 ( Figure 2B ) . Control SW480 cells express the Mfge8 ligand αvβ5 as well as α5β1 and bind Mfge8 in an αvβ5-dependent manner . Mock-transfected control SW480 cells adhered to Mfge8 and adherence was blocked by anti-β5 antibody ( ALULA ) . In the presence of ALULA , β3-transfected cells adhered to Mfge8 ( αvβ3 is a known receptor for Mfge8 , as above ) , and adherence was blocked by an anti-β3 antibody ( LM609 ) . α8-transfected SW480 cells also adhered to Mfge8 in the presence of ALULA , and adherence was blocked by the addition of an α8 blocking antibody ( YZ3 ) ( Nishimichi et al . , 2015 ) . The YZ3 antibody blocks both human and mouse α8 . These results indicate that α8β1 specifically mediates cell adhesion to Mfge8 . Next we analyzed adhesion of α8-transfected SW480 cells to Mfge8 at various concentrations in the presence of ALULA ( Figure 2C ) . The α8-transfected cells adhered to Mfge8 in a dose-dependent fashion . Of note , expression of β5 , β3 , and α8 , as evaluated by flow cytometry ( Figure 2—figure supplement 2 ) and the extent of binding to Mfge8 ( Figure 2B and C ) were similar in these cell lines . 10 . 7554/eLife . 13063 . 005Figure 2 . Mfge8 binds to α8 integrin to regulate gastrointestinal motility . ( A ) Purified α8 , αvβ3 , or α5β1 were used for solid-phase binding assays with purified Mfge8 at indicated concentrations in the presence or absence of 10 mM EDTA . ( B ) Adhesion of SW480 ( mock ) , α8 transfected SW480 cells ( α8 ) or β3 transfected SW480 cells ( β3 ) adhesion to wells coated with rMfge8 ( 5 µg/ml ) in the presence or absence of integrin blocking antibodies ( 5 µg/ml ) against β5 ( ALULA ) , β3 ( LM609 ) or α8 ( YZ83 ) . ( C ) Dose-dependent binding of SW480 cells to wells coated with a dose range of rMfge8 in the presence of a β5 blocking antibody . ( D ) Western blot of integrin expression in human gastric smooth muscle cells ( HGSMC ) , SW480 cells and α8 transfected SW480 ( SW480_α8 ) cells . ( E ) Human gastric smooth muscle cell adhesion to rMfge8-coated wells in the presence of blocking antibodies against the αv , β1 , β5 , α8 , or α5 integrin subunits . ( F ) Immunofluorescence staining of antral sections from Mfge8-/- and α8sm-/- mice with or without rMfge8 gavage proved for α8 ( green ) , human-FC-Mfge8 recombinant construct ( red ) and DAPI ( blue ) . Arrows point co-localization of Mfge8 and α8 . ( G ) Force of antral contraction in WT and α8sm-/- mice in response to MCh ( N = 3–4 ) . ( H ) The rate of gastric emptying in α8sm-/- and WT mice with and without the addition of rMfge8 ( N = 4–5 ) . ( I ) Small intestinal transit time in α8sm-/- and WT mice with and without the addition of rMfge8 ( N = 4–5 ) . ( J ) Force of antral contraction in WT mice after IP injection of α8 blocking or control antibody in response to MCh ( N = 4–5 ) . ( K ) The rate of gastric emptying in WT mice after IP injection of α8 blocking or IgG1 isotype control antibody ( N = 7 ) . ( L ) Small intestinal transit time in WT mice after IP injection of α8 blocking or IgG1 isotype control antibody ( N = 7 ) . Female mice were used in G , H and I and male mice were used for all remaining panels . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . Data are expressed as mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 13063 . 00510 . 7554/eLife . 13063 . 006Figure 2—figure supplement 1 . Normal gastrointestinal motility in Itgb3-/- , Itgb5-/- and Itgb3-/-::Itgb5-/- mice . ( A ) Force of antral smooth muscle ring contraction in Itgb3-/- , Itgb5-/- , and Itgb3-/-::Itgb5-/-mice in response to MCh . ( B ) The rate of gastric emptying in Itgb3-/- , Itgb5-/- and Itgb3-/-::Itgb5-/- mice with and without the addition of rMfge8 ( N = 5–6 ) . ( C ) Small intestinal transit time in Itgb3-/- , Itgb5-/- and Itgb3-/-::Itgb5-/- mice with and without the addition of rMfge8 ( N = 5–6 ) . Both male and female mice were used for these experiments . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . Data are expressed as mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 13063 . 00610 . 7554/eLife . 13063 . 007Figure 2—figure supplement 2 . Integrin expression levels in SW480 cells . Flow cytometric analysis of mock transfected cells and cells transfected with either Itgb3 or Itga8 . Gray histogram represents isotype control . DOI: http://dx . doi . org/10 . 7554/eLife . 13063 . 00710 . 7554/eLife . 13063 . 008Figure 2—figure supplement 3 . Enhanced antral contraction in α8sm-/- mice . ( A ) Floxed band , ( B ) recombination band , and ( C ) western blot of gastric smooth muscle of α8sm-/- after two weeks of doxycycline administration . ( D ) Force of antral contraction in WT and α8sm-/- mice in response to KCl ( N = 3–4 ) . Female mice were used for these experiments . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . Data are expressed as mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 13063 . 008 To confirm that these findings were relevant in smooth muscle cells , we evaluated adhesion of primary human gastric smooth muscle cells to Mfge8 . Primary human gastric smooth muscle cells expressed the β5 , β1 , αv , and α8 integrin subunits ( Figure 2D ) and adhered to Mfge8 ( Figure 2E ) . Adherence was significantly inhibited by blocking antibodies to the β5 , β1 , αv , and α8 subunits but not the α5 integrin ( Figure 2E ) . Simultaneous blockade of the αv and α8 integrins had a significantly greater effect on adhesion than blocking each integrin individually ( Figure 2E ) . To determine whether Mfge8 and the α8integrin colocalize in gastric smooth muscle , we injected Mfge8-/- mice with our recombinant Mfge8 protein containing a human FC domain , prepared tissue sections , and probed with an anti-Fc and anti-α8 antibody . As shown in Figure 2F , Mfge8 and α8 colocalized in gastric smooth muscle ( Figure 2F ) . To evaluate whether α8β1 mediates the effect of Mfge8 on gastric smooth muscle , we created a Itga8flox/flox—Tg ( Acta2-rtTA , TetO-Cre ) transgenic mouse line , abbreviated as α8sm-/- , containing α8 floxed/floxed alleles , a tetracycline-inducible Cre transgene and an α-smooth muscle-rtTA transgene . The addition of doxycycline chow resulted in smooth muscle-specific recombination of α8 ( Figure 2—figure supplement 3A–C ) . Gastric antral smooth muscle from α8sm-/- mice had enhanced contraction in response to MCh and KCl ( Figure 2G and Figure 2—figure supplement 3D ) . Unlike WT samples , rMfge8 did not significantly reduce the force of contraction in α8sm-/- antral smooth muscle ( Figure 2G and Figure 2—figure supplement 3D ) . α8sm-/- mice had enhanced gastric emptying and more rapid SIT ( Figure 2H and I ) . Oral gavage with rMfge8 did not significantly slow gastric emptying or prolong SIT in α8sm-/- mice ( Figure 2H and I ) . Administration of α8 blocking antibody to WT mice significantly increased the force of antral contraction , accelerated gastric emptying , and reduced SIT ( Figure 2J–L ) . In sum , these data indicate that disruption of α8β1 integrin signaling accelerates gastrointestinal motility . Enhanced antral smooth muscle contraction could be the result of an increase in the frequency of intracellular calcium oscillations after release of calcium from intracellular sources or the result of an increase in calcium sensitivity due to inactivation of the enzyme myosin light chain phosphatase ( Kudo et al . , 2013; Bhattacharya et al . , 2014; Kudo et al . , 2012; Somlyo and Somlyo , 2000; Somlyo and Somlyo , 2003 ) . Antral rings from Mfge8-/- and α8sm-/- mice had exaggerated contraction to both MCh and KCl suggesting altered calcium sensitivity as the mechanism by which Mfge8 reduced contraction since these agonists increase intracellular calcium through different mechanisms . KCl works primarily by inducing opening of voltage-gated calcium channels leading to influx of extracellular calcium while MCh induces release of intracellular calcium stores after receptor binding . To determine whether enhanced antral contraction was due to an increase in smooth muscle calcium sensitivity , we assessed the phosphorylation status of the regulatory subunit of myosin light chain phosphatase , MYPT , and myosin light chain ( MLC ) ( Kudo et al . , 2013; Bhattacharya et al . , 2014; Kudo et al . , 2012; Somlyo and Somlyo , 2000; Somlyo and Somlyo , 2003 ) . Antral smooth muscle from Mfge8-/- and α8sm-/- mice had increased phosphorylation of both MYPT and MLC in response to MCh as compared with WT smooth muscle ( Figure 3A and B ) . Antral smooth muscle from Itgb3-/-::Itgb5-/- mice did not have increased phosphorylation of MYPT or MLC as compared with WT samples and MYPT and MLC phosphorylation that was present in response to MCh was reduced to a similar extent by rMfge8 in WT and Itgb3-/-::Itgb5-/- mice ( Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 13063 . 009Figure 3 . α8 integrin regulates antrum smooth muscle calcium sensitivity by preventing RhoA activation . ( A , B ) Western blot of antrum muscle strips obtained from ( A ) Mfge8-/- and ( B ) α8sm-/-mice and incubated with MCh . ( C , D ) Western blot of antrum smooth muscle strips obtained from ( C ) Mfge8-/- and ( D ) α8sm-/- treated with MCh demonstrating active and total RhoA using a GST pull-down assay . ( E ) Force of antral smooth muscle ring contraction with and without the addition of ROCK inhibitor Y-27632 ( N = 3–4 ) . ( F ) The rate of gastric emptying in Mfge8-/- and WT with and without the IP injection of ROCK inhibitor ( Y-27632 ) or control inhibitor ( N = 5–11 ) . ( G ) Small intestinal transit times Mfge8-/- and WT with and without IP injection of ROCK inhibitor ( Y-27632 ) or control inhibitor ( N = 6–11 ) . Female mice were used for all experiments . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . Data are expressed as mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 13063 . 00910 . 7554/eLife . 13063 . 010Figure 3—figure supplement 1 . Normal calcium sensitivity in Itgb3-/-::Itgb5-/- antrum smooth muscle . Western blot of antrum muscle strips obtained from WT and Itgb3-/-::Itgb5-/- mice incubated with MCh . N = 3 . Both male and female mice were used for these experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 13063 . 010 The small GTPase RhoA is a prominent regulator of MYPT phosphorylation and inhibition of RhoA has been shown to reduce the force of gastric smooth muscle contraction ( Ratz et al . , 2002; Tomomasa et al . , 2000; Büyükafşar and Levent , 2003 ) . RhoA activation , assessed by a GST pull-down assay , was significantly increased in Mfge8-/- and α8sm-/- antral smooth muscle as compared with WT controls while total RhoA protein expression was unchanged ( Figure 3C and D ) . rMfge8 reduced RhoA activation in WT and Mfge8-/- antrum but not α8sm-/- antral smooth muscle ( Figure 3C and D ) . Pharmacological inhibition of ROCK , the kinase downstream of RhoA responsible for phosphorylation and inactivation of MYPT , with Y-27632 , inhibited antral contraction in both WT and Mfge8-/- smooth muscle reducing Mfge8-/- antral contraction to WT levels ( Figure 3E ) . Intraperitoneal ( IP ) administration of Y-27632 also reduced gastric emptying and prolonged SIT in WT and Mfge8-/- mice with a relatively greater effect in Mfge8-/- mice ( Figure 3F and G ) . Taken together , these data indicate that in gastric antral smooth muscle , the absence of α8β1 integrin results in enhanced RhoA activation leading to increased smooth muscle calcium sensitivity , antral contraction , gastric emptying , and more rapid small intestinal transit times . PI3 kinase ( PI3K ) is a positive regulator of smooth muscle contraction ( Wang et al . , 2006; Kawabata et al . , 2000 ) . To determine whether the Mfge8-α8β1 axis modulates smooth muscle contraction through PI3K , we incubated antral smooth muscle rings with the PI3K inhibitor wortmannin . Wortmannin significantly reduced contraction in Mfge8-/- , α8sm-/- , and WT antral smooth with a proportionally greater effect on antrum from Mfge8-/- and α8sm-/- as compared with antrum from WT mice ( Figure 4A and B ) . PI3K activation leads to phosphorylation of AKT . Antral rings from Mfge8-/- and α8sm-/- mice had enhanced phosphorylation of AKT at serine 473 ( Figure 4C and D ) . rMfge8 reduced AKT phosphorylation in Mfge8-/- but not α8sm-/- samples ( Figure 4C and D ) . Wortmannin also prevented the enhanced RhoA activation in Mfge8-/- and α8sm-/- antral smooth muscle ( Figure 4E ) . 10 . 7554/eLife . 13063 . 011Figure 4 . Mfge8 ligation of α8β1 integrin inhibits PI3 kinase activity . ( A–B ) Force of antral smooth muscle ring contraction with and without the addition of PI3K inhibitor wortmannin ( wort 100 ng/ml ) in response to MCh in WT and Mfge8-/- ( A , N = 4–5 ) or WT and α8sm-/- ( B , N = 4–5 ) . ( C–D ) Western blot of antrum muscle strips obtained from WT and Mfge8-/- ( C ) and WT and α8sm-/- mice ( D ) incubated with MCh . ( E ) Western blot of antrum from WT , Mfge8-/- and α8sm-/- treated with wortmannin ( 100 ng/ml ) and MCh demonstrating active and total RhoA using a GST pull-down assay . Male mice were used in panel A and B . The remaining panels include both male and female mice . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . Data are expressed as mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 13063 . 011 Phosphatase and tensin homolog ( PTEN ) is the major negative regulator of PI3K ( Leevers et al . , 1999 ) . To determine whether Mfge8 ligation of α8β1 opposed PI3K activation through modulation of PTEN , we measured PTEN activity using an ELISA that quantifies the ability of lysates to convert PI ( 3 , 4 , 5 ) P3 to PI ( 4 , 5 ) P2 ( Maehama and Dixon , 1998 ) . PTEN activity was reduced in both Mfge8-/- and α8sm-/- antral rings ( Figure 5A and B ) . rMfge8 significantly increased PTEN activity in antrum from WT and Mfge8-/- mice with no effect on antrum from α8sm-/- mice ( Figure 5A and B ) . In both WT and Mfge8-/- mice there was a significant inverse correlation between the extent of PTEN activity and the rate of gastric emptying and SIT ( Figure 5C and D ) . rMfge8 also increased PTEN activity in primary human gastric smooth muscle cells , an effect that was blocked using a blocking antibody to α8 but not to α5 or β5 integrin subunits ( Figure 5E ) . Of note , treatment with fibronectin or vitronectin , both ligands of α8β1 , did not increase PTEN activity suggesting a specific effect for Mfge8 ( Figure 5—figure supplement 1 ) . IP administration of α8 blocking antibody also decreased antral PTEN activity in WT mice ( Figure 5F ) . We next used siRNA to knockdown PTEN expression in primary human gastric smooth muscle cells ( Figure 5—figure supplement 2 ) to evaluate the effect on smooth muscle calcium sensitivity . PTEN knockdown lead to increased MLC and MYPT phosphorylation in response to 5-HT ( Figure 5G ) as well as to increased RhoA activation ( Figure 5H ) . Unlike control samples , rMfge8 did not reduce the degree of MYPT or MLC phosphorylation or RhoA activation in human gastric smooth muscle after PTEN knockdown ( Figure 5G and H ) . These data indicate that α8β1 prevents RhoA activation in gastric smooth muscle by increasing the activity of PTEN . 10 . 7554/eLife . 13063 . 012Figure 5 . Mfge8 modulates PTEN activity . ( A–B ) PTEN activity in antral smooth muscle of WT and Mfge8-/- ( A , N = 5 ) and WT and α8sm-/- ( B , N = 7 ) with and without the addition of rMfge8 and RGE construct . ( C–D ) Correlation between the PTEN activity and the rate of gastric emptying ( C , N = 11 ) and the small intestinal transit time in WT mice ( D , N = 13 ) . ( E ) PTEN activity in antral smooth muscle strips with addition of rMfge8 in presence of blocking antibody against α8 , α5 and β5 . ( F ) PTEN activity in antral smooth muscle strips of WT mice after IP injection of α8 blocking or IgG1 isotype control antibody . ( N = 5 ) . ( G ) Western blot in human gastric smooth muscle cells ( HGSMC ) treated with siRNA targeting PTEN with or without rMfge8 and then treated with 5-HT ( 100 µM ) . ( H ) Western blot of human gastric smooth muscle cells ( HGSMC ) treated with PTEN siRNA and with 5-HT demonstrating active and total RhoA using a GST pull-down assay . Both male and female mice were used for these experiments . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . Data are expressed as mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 13063 . 01210 . 7554/eLife . 13063 . 013Figure 5—figure supplement 1 . Mfge8 increases PTEN activity . PTEN activity assay in Human Gastric Smooth Muscle Cells after treatment with rMfge8 , RGE construct , fibronectin or vitronectin ( N = 5 ) . **p<0 . 01 , ***p<0 . 001 . Data are expressed as mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 13063 . 01310 . 7554/eLife . 13063 . 014Figure 5—figure supplement 2 . siRNA knockdown of PTEN . Western blot showing efficiency of PTEN-targeting siRNA and control siRNA in Human Gastric Smooth Muscle Cells . DOI: http://dx . doi . org/10 . 7554/eLife . 13063 . 014 We next wanted to evaluate the functional consequences of altered motility on nutrient absorption in α8sm-/- mice . Since we have previously reported impaired fat absorption in Mfge8-/- mice ( Khalifeh-Soltani et al . , 2014 ) , we first assessed the ability of α8sm-/- mice to absorb dietary fats . After an olive oil gavage , α8sm-/- mice had significantly higher fecal triglyceride ( TG ) concentrations ( Figure 6A ) as well as lower serum TG levels ( Figure 6B ) as compared with WT mice . Fecal TG levels were also significantly higher in α8sm-/- mice on a normal chow diet control ( CD ) as compared with WT mice ( Figure 6C ) . Of note , primary enterocytes isolated from α8sm-/- mice did not have a defect in fatty acid uptake indicating that the increase in stool fat was unlikely to be due to a defect in enterocyte fatty acid uptake ( Figure 6D ) . Furthermore , IP injection of olive oil resulted in similar serum TG levels in α8sm-/- mice as compared with WT mice ( Figure 6E ) indicating that clearance of lipids by tissue outside of the intestinal tract was preserved in α8sm-/- mice . Taken together , these data indicate that α8sm-/- mice develop steatorrhea . 10 . 7554/eLife . 13063 . 015Figure 6 . α8sm-/-mice are protected from diet-induced obesity . ( A ) Fecal triglycerides in WT and α8sm-/- mice after an olive oil gavage ( N = 8 ) . ( B ) Serum triglycerides levels in WT and α8sm-/- mice after an olive oil gavage ( N = 5 ) . ( C ) Fecal triglycerides in WT and α8sm-/- mice on a normal chow control diet ( N = 6 ) . ( D ) Primary enterocyte fatty acid uptake in the isolated enterocytes from WT and α8sm-/- mice ( N = 5 ) . ( E ) Serum triglycerides levels after IP administration of olive oil in WT and α8sm-/- mice ( N = 5 ) . ( F , G ) Fecal ( F , N = 8 ) and enterocytes ( G , N = 8 ) 2NBDG content in WT and α8sm-/- mice after gavage with a 2NBDG -methylcellulose mixture . ( H ) Glucose uptake assay in isolated primary enterocytes from WT and α8sm-/- mice ( N = 8 ) . ( I , J ) Fecal ( I , N = 8 ) and enterocytes ( J , N = 8 ) 2NBDG content in WT and Mfge8-/- mice after gavage with a 2NBDG-methylcellulose mixture . ( K , L ) Fecal 2NBDG content in WT and α8sm-/- ( K , N = 5 ) and Mfge8-/- ( L , N = 6 ) after gavage with a 2NBDG in PBS . ( M ) Weight gain in female WT and α8sm-/- mice on a normal chow diet ( CD ) ( N = 6–8 ) or HFD ( N = 8–12 ) . ( N ) Fecal energy content in WT and α8sm-/- mice on a normal chow diet ( CD ) ( N = 5–6 ) or HFD ( N = 4–5 ) . Each sample represents stool combined from 3 mice . Female mice were used for all experiments . ( O ) Fecal triglycerides in WT and Itgb3-/-::Itgb 5-/- integrin-deficient mice with normal chow control diet ( N = 5–6 ) . For all in vivo experiments , each group of 5 mice represents 1 independent experiment . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . Data are expressed as mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 13063 . 01510 . 7554/eLife . 13063 . 016Figure 6—figure supplement 1 . Protection from weight gain in α8sm-/- mice on a HFD . ( A ) Weight gain in WT and α8sm-/- male mice on a CD ( N = 6–8 ) or HFD ( N = 8–10 ) . ( B–C ) Body composition of WT and α8sm-/- mice aged 14 weeks on a HFD ( B , N = 8–12 ) or on a CD ( C , N = 6–8 ) . *p<0 . 05 , **p<0 . 01 . Data are expressed as mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 13063 . 016 To evaluate whether malabsorption was specific for fat or represented a more generalized impairment of nutrient uptake , we measured stool glucose levels after gavage with a 2- ( N- ( 7-Nitrobenz-2-oxa-1 , 3-diazol-4-yl ) Amino ) -2-Deoxyglucose ( 2NBDG ) fluorescent glucose analog . 2NBDG was mixed with methylcellulose to form a semisolid bolus sensitive to antral contraction . α8sm-/- mice had increased stool glucose levels ( Figure 6F ) coupled with reduced enterocyte glucose levels ( Figure 6G ) . Enterocytes isolated from α8sm-/- mice and cultured in vitro did not have a defect in 2NBDG uptake ( Figure 6H ) . Mfge8-/- mice also had increased stool 2NBDG and reduced enterocyte 2NBDG levels ( Figure 6I and J ) when 2NBDG was gavaged as a semisolid mixed with methylcellulose but not when administered as a liquid preparation in PBS ( Figure 6K and L ) . Mfge8-/- mice gain approximately 50% less weight on a high-fat diet ( HFD ) as compared with WT controls ( Khalifeh-Soltani et al . , 2014 ) . To evaluate the relative contribution of altered motility to this phenotype , we placed α8sm-/- mice on a HFD . Both female and male α8sm-/- mice were significantly protected from weight gain on a HFD ( Figure 6M , Figure 6—figure supplement 1A ) . Reduced weight gain on a HFD in α8sm-/- mice was associated with reduced body fat as measured by Dexa scanning ( Figure 6—figure supplement 1B ) . A modest reduction in body weight was also apparent in α8sm-/- mice on a CD as compared with WT controls , became statistically significant at 22 weeks of age ( Figure 6M and S8A ) , and was associated with decreased body fat on DEXA scan ( Figure 6—figure supplement 1C ) . α8sm-/- mice also had increased stool energy content as measured by bomb calorimetry on both a HFD and CD ( Figure 6N ) . To further explore the contribution of impaired enterocyte fatty acid uptake to steatorrhea in Mfge8-/- mice , we measured stool TG content of Itgb3-/-::Itgb5-/- mice . Itgb3-/-::Itgb5-/- mice on a CD had significantly greater stool TG than WT controls ( Figure 6O ) suggesting that both impaired fatty acid uptake mediated through these integrins and altered motility mediated by the α8β1 integrin contribute to the development of steatorrhea in in Mfge8-/- mice . In this work we identify a key role for the α8β1integrin in promoting nutrient absorption through regulation of gastrointestinal motility . Smooth muscle-specific deletion of α8 results in an increase in gastric antral contractile force , more rapid gastric emptying , and faster small intestinal transit times coupled with impaired absorption of both fats and carbohydrates and increased stool energy content . α8sm-/- mice are partially protected from weight gain on a HFD and have reduced body weight on a CD . We further show that the milk protein Mfge8 is a novel ligand for α8β1and that binding of Mfge8 to α8β1 is responsible for the effects of this integrin on gastrointestinal motility . The α8 integrin forms heterodimers with the β1 integrin and was initially identified in axon tracts where the integrin promotes axonal growth during embryogenesis ( Bossy et al . , 1991 ) . Previous work has shown expression of α8 in both vascular and visceral smooth muscle including the muscularis mucosa of the GI tract ( Schnapp et al . , 1995 ) . In vitro studies suggest that α8 promotes smooth muscle proliferation ( Zargham et al . , 2007 ) and maintains vascular smooth muscle in a differentiated , contractile , non-migratory phenotype ( Zargham et al . , 2007; Zargham and Thilbault , 2006; Zargham et al . , 2007 ) . In vascular smooth muscle , α8 has been shown to co-immunoprecipitate with RhoA and siRNA knockdown of α8 leads to less membrane localization and thereby less active RhoA ( Zargham et al . , 2007 ) . shRNA-mediated knockdown of α8 in intestinal epithelial cells has also been reported to lead to reduction in active RhoA ( Benoit et al . , 2009 ) . In contrast to these findings ( Zargham et al . , 2007; Benoit et al . , 2009 ) , we found that in gastric smooth muscle , Mfge8 ligation of α8β1 prevents RhoA activation and that this effect is mediated by opposing the activity of PI3K . It is unclear to us why α8β1 has opposing effects on RhoA activation in different cell types . RhoA becomes activated by GTP loading and reverts to an inactive state when GDP-bound . The activation status of RhoA is regulated by a large family of RhoA GTPase Activating Factors ( RhoGAPs ) and RhoA GTPase guanine nucleotide exchange factors ( RhoGEFs ) which convert GTP to GDP or load or RhoA would GTP respectively ( Puetz et al . , 2009 ) . We speculate that the most likely explanation for α8β1 preventing RhoA activation in gastric smooth muscle while presumably favoring RhoA activation in vascular smooth muscle and intestinal epithelial cells is differential effects of the α8β1 on and/or differential expression of RhoGAPs and RhoGEFs in specific cell types . These data suggest that the role for the α8β1 integrin in regulation of RhoA and contractility can vary depending on the context within which the integrin is activated . Support for an α8β1-PI3K-RhoA axis regulating motility is evidenced by the ability of the PI3K inhibitor wortmannin to abrogate the exaggerated antral contraction in Mfge8-/- and α8sm-/- antral rings coupled with the increase in AKT phosphorylation and RhoA activation in Mfge8-/- and α8sm-/- antral rings . We also found that PI3K inhibition after Mfge8 ligation of α8β1 occurs through an increase in the activity of the phosphatase PTEN which directly opposes PI3K by converting PIP3 to PIP2 . Furthermore , PTEN activity in antral samples had an inverse correlation with the extent of gastric emptying and rate of SIT suggesting an essential role for PTEN in regulating gastrointestinal motility . Identification of α8β1 as a new binding partner for Mfge8 that slows gastrointestinal motility coupled with malabsorption in α8sm-/- mice indicate that the effect of Mfge8 on promoting absorption of dietary fats occurs through dual cooperative mechanisms . Mfge8 slows gastrointestinal motility thereby increasing the time for absorption of dietary fats and directly induces enterocyte fatty acid uptake ( Khalifeh-Soltani et al . , 2014 ) . Since Mfge8 is a secreted molecule , we cannot rule out an effect on antral contraction secondary to Mfge8 binding of integrin receptors on local neurons in the gastrointestinal tract . However , the fact that α8sm-/- mice phenocopy the motility phenotype of Mfge8-/- mice strongly suggests that the dominant effect of Mfge8 on antral contractility is mediated through ligation of integrin receptors on smooth muscle . The relative contribution of altered motility versus impaired fatty acid uptake to the malabsorption phenotype in Mfge8-/- mice is difficult to parse out from our data . After 12 weeks on a HFD , Mfge8-/- gained approximately 50% less weight than WT controls ( Khalifeh-Soltani et al . , 2014 ) while α8sm-/- mice gained approximately 40% less weight than WT controls . One interpretation of this data is that the dominant mechanism underlying protection from weight gain in Mfge8-/- mice on a HFD is accelerated motility . However the elevated stool TG content in Itgb3-/-::Itgb5-/- mice suggests a critical contribution of lipid absorption to steatorrhea since these mice have impaired enterocyte fatty acid uptake but normal motility . Another possibility is that ligands in addition to Mfge8 can bind α8β1 leading to inhibition of gastrointestinal motility and resulting in a relatively greater malabsorption phenotype secondary to altered motility in α8sm-/- mice than in Mfge8-/- mice . Of note , a recent publication looking at global α8 null mice and global α8 null mice in the ApoE background assessed body weights in each of these mouse lines ( Menendez-Castro et al . , 2015 ) . Consistent with our findings in α8sm-/- mice on a control diet , global α8 null mice and global α8 null mice in the ApoE background had reduced body weight as compared with control mice . In global α8 null mice , there was no statistically significant difference in body weights ( by our calculation a P value of 0 . 059 ) , but it is important to note that these mice do not appear to have reached the age at which we saw significant differences in body weights in α8sm-/- mice ( Menendez-Castro et al . , 2015 ) . Furthermore , though no statistical significance was reported in the comparison of body weights of the global α8 null mice in the ApoE background at 1 year of age , a 2-tailed t-test of the presented data comparing mice homozygous for the α8 null mutation with wild type mice ( both in the ApoE no background ) produced a P value of 0 . 016 ( Menendez-Castro et al . , 2015 ) . Consistent with accelerated motility as the cause of malabsorption , α8sm-/- mice had increased stool triglyceride and 2NBDG levels after gavage with olive oil/2NBDG respectively . Unlike primary enterocytes from Mfge8-/- mice , primary enterocytes from α8sm-/- mice did not have impaired fatty acid , further supporting altered motility as the cause of steatorrhea inα8sm-/- mice . We previously reported that serum glucose levels after gavage of a liquid glucose/PBS mixture were similar in Mfge8-/- and WT mice ( Khalifeh-Soltani et al . , 2014 ) . To assess whether exaggerated antral contraction in Mfge8-/- mice led to impaired absorption of glucose , we administered 2NBDG in a mixture with methylcellulose . Unlike the previously used liquid glucose mixture ( Khalifeh-Soltani et al . , 2014 ) , the 2NBDG/methylcellulose provides a semisolid substrate that depends on antral contraction for propulsion along the small intestinal tract . As with α8sm-/- mice , Mfge8-/- mice had increased stool 2NBDG levels . However , when we gavaged mice with 2NBDG in a liquid form in PBS , stool 2NBDG levels were similar in Mfge8-/- mice as compared with WT control mice suggesting that the effect of Mfge8 on enterocyte glucose uptake is through altered motility rather than direct uptake . One interesting observation from our work is the opposing effects Mfge8 has on PI3K activation in smooth muscle cells as compared with adipocytes . Mfge8 slows gastrointestinal motility by increasing PTEN activity leading to inhibition of PI3K activity while Mfge8 increases fatty acid uptake by activating PI3K ( Khalifeh-Soltani et al . , 2014 ) . The effect on fatty acid uptake and PI3K activation is mediated through the αvβ3 and αvβ5 integrins ( Khalifeh-Soltani et al . , 2014 ) while the effect on gastrointestinal motility and PI3K inhibition is mediated through the α8β1integrin . Of note , both the αvβ3 and αvβ5 integrins are expressed on gastric smooth muscle . However , even though Mfge8 is a ligand for both of these integrins , single and double knockouts of αvβ3 and αvβ5 did not have the gastric smooth muscle phenotype of Mfge8-/- mice , suggesting that even if Mfge8 binds these integrins on smooth muscle cells , binding does not lead to alterations in calcium sensitivity and smooth muscle contractility . Furthermore , in relation to modulation of PI3K activity , our data suggest that in smooth muscle cells the effect of Mfge8 ligation of α8β1is dominant over any effect that Mfge8 ligation of αvβ3 and αvβ5 may havesince baseline phosphorylated AKT levels ( reflecting PI3K activation ) are increased in antral tissue from Mfge8-/- mice as compared with WT controls . Furthermore , rMfge8 treatment reduces AKT phosphorylation in antral rings while it increases AKT phosphorylation in 3T3-L1 adipocytes ( Khalifeh-Soltani et al . , 2014 ) . Finally , we have ruled out the possibility that Mfge8 is a ligand for any other RGD-binding integrin ( Atabai et al . , 2009 ) other than αvβ1 and non-integrin receptors that have not been identified for Mfge8 . Another interesting observation that comes from our data is that oral administration of Mfge8 by gavage successfully prolongs small intestinal transit time and reduces the extent of gastric emptying . We have shown that recombinant Mfge8 reaches the smooth muscle by immunofluorescence . However , we are not entirely sure how Mfge8 gets to the smooth muscle layer . One possibility is that it is absorbed into the bloodstream and takes a hematogenous route to the smooth muscle . In our previously published work , oral Mfge8 gavage did not lead to discernible serum Mfge8 levels 30 min after gavage ( Khalifeh-Soltani et al . , 2014 ) . While this would argue against a hematogenous route to the smooth muscle , it is possible that blood levels are below the sensitivity of the Mfge8 ELISA . Alternatively , measurement of serum levels at earlier time points may show recombinant protein in the bloodstream . Another possibility is that the epithelium takes up the recombinant Mfge8 and secretes it from the basal side and that Mfge8 reaches the smooth muscle either directly or through entering the circulation on the basal side . From a therapeutic viewpoint , our findings provide new targets for treatment of diseases associated with altered gastrointestinal motility . Activation of the pathway we describe with recombinant Mfge8 could be used to treat disorders characterized by increased gastrointestinal transit time and malabsorption such as short bowel syndrome . Blockade of the α8β1 integrin could be used as a treatment for gastroparesis secondary to conditions such as diabetic neuropathy , medication side effects , and a number of neurological diseases . Dissociation of the dual effects of Mfge8 at the receptor level provides flexibility for designing therapeutic strategies that can target only the effect on fatty acid uptake or gastrointestinal motility or both simultaneously . In addition , our data provides preliminary proof of principle evidence to support evaluation of α8 blockade as a therapy for gastroparesis . All animal experiments were approved by the UCSF Institutional Animal Care and Use Committee in adherence to NIH guidelines and policies . All mice were maintained on a C57BL/6J background . Mfge8-/- mice were obtained from RIKEN ( Hanayama et al . , 2002 ) . Tg ( TetO-cre ) 1Jaw/J and Tg ( Acta2-rtTA ) #Des mouse lines have been described previously ( Perl et al . , 2002; Chen et al . , 2012 ) . Mfge8-/-sm+ transgenic mice were created by cloning the Mfge8 long isoform into the PTRE2 vector with subsequent microinjection of DNA by the Gladstone Institute Gene-Targeting Core . Tg ( TetO-Mfge8 ) transgenic mice containing the tetracycline-inducible Mfge8 construct were crossed with a Mfge8-/- mice line created using a gene disruption vector ( Atabai et al . , 2005; Silvestre et al . , 2005 ) and mice carrying the Tg ( Acta2-rtTA ) #Des transgene . Itga8flox/flox mice have been previously described ( Chan et al . , 2010 ) . α8sm-/- mice were created by crossing Itga8flox/flox mice with mice carrying Tg ( TetO-cre ) 1Jaw/J and Tg ( Acta2-rtTA ) #Des transgenes . Itgb3-/- , Itgb5-/- , and Itgb3-/-::Itgb5-/- mice in the 129 SVEV strain have been previously described ( Huang et al . , 2000; Sugimoto et al . , 2012 ) . For smooth muscle induction of Mfge8 or Cre- mediated recombination of Itga8flox/flox—Tg ( Acta2-rtTA , TetO-Cre ) mice were placed on doxycycline chow for 2 weeks prior to experiments . Human gastric smooth muscle cells were obtained from ScienCell Research Laboratories and have been characterized by immunofluorescent method with antibodies to α-smooth muscle actin and desmin . Cells were received at passage 1 and are negative for HIV-1 , HBV , HCV , mycoplasma , bacteria , yeast and fungi . Experiments were conducted with cells between passages 3 and 4 . SW480 cells were generously provided by Yasuyuki Yokosaki and Dean Sheppard . SW480 cells were characterized by STR profiling and isoenzyme analysis by ATCC , and were negative for mycoplasma . We suspended freshly isolated antral ring slices ( 2–3 mm in length ) on plexiglass rods in a double-jacketed organ bath ( Radnoti 8 unit tissue organ bath system ) in Krebs-Henseleit solution maintained at 5% CO2-95% O2 , 37°C , and a pH of 7 . 4–7 . 4533 . We attached rings by a silk thread to a FT03 isometric transducer . Concentration-response curves of multiple chambers were continuously displayed and recorded . We set initial tension at 0 . 5 g for antral rings before adding contractile agonists . We then added a range of concentrations of MCh ( 10–4 to 10-9 M ) and KCl ( 3 . 75–60 mM ) to induce contraction . For selected studies , wortmannin ( 100 ng/ML ) , Y-27632 ( 100 nm ) or recombinant Mfge8 constructs ( 10 µg/ml ) were added 15 min prior to addition of contractile agonists . In the α8 blocking antibody experiments ( in Figure 2J ) we injected 10 mg/kg IP α8 blocking antibody or I IgG1 isotype control antibody ( Cell signaling , 5415 ) 30 min before the antral ring contraction assay . Mice were fasted for 12 hr prior to experiments but had free access to water . Mice were gavaged with 250 μl of methylcellulose mixed with phenol red ( 0 . 5 g/L phenol red in 0 . 9% NaCl with 1 . 5% methylcellulose ) . We euthanized mice 15 min after administration of the test meal , dissected out the stomach and removed the stomach after ligation of the cardiac and pyloric ends to ensure that any retained meal did not leak out of the stomach during removal . We then cut the stomach into pieces and homogenized with 25 ml of 0 . 1 N NaOH and added 0 . 5 ml of trichloroacetic acid ( 20% w/v ) and centrifuged at 3000 rpm for 20 min . We then added 4 ml of 0 . 5 N NaOH to 1 ml of the supernatant and measured absorbance at 560 nm to assess phenol red content in the stomach . The percentage gastric emptying was derived as ( 1-X/Y ) *100 where X represents absorbance of phenol red recovered from the stomach of animals sacrificed 15 min after a test meal . Y represents mean ( n = 5 ) absorbance of phenol red recovered from the stomachs of control animals which were euthanized immediately following gavage with the test meal . In experiments using rMfge8 and RGE constructs , we administered each construct by gavage ( 50 μg/kg body weight in a total volume of 200 μl in PBS ) before administration of phenol to mice . Y-27632 was administered IP ( 100 nm ) 15 min prior to gavage . In the α8 blocking antibody experiments ( in Figure 2K ) we injected 10 mg/kg IP α8 blocking antibody 30 min before administration of phenol red to mice . Mice were fasted for 12 hr prior to experiments but had free access to water . We then gavaged mice with 250 μl Carmine meal ( 6% Carmine red and 0 . 5% methylcellulose in water ) . 15 min after administration of gavage , we euthanized mice and dissected out the small intestine from the pylorus to the ileocecal junction , identifying the location to which the meal had traversed , and securing that position with thread to avoid changes in the length of the transit due to handling . The small intestinal transit ( SIT ) was calculated from the distance traveled by Carmine meal divided by total length of the small intestine multiplied by 100 . In experiments using rMfge8 and RGE constructs , we administered each construct by gavage ( 50 μg/kg body weight in a total volume of 200 μl in PBS ) before administration of the Carmine meal to mice . Y-27632 was administered IP ( 100 nm ) 15 min prior to gavage . In the α8 blocking antibody experiment ( in Figure 2L ) we injected 10 mg/kg IP α8 blocking antibody 30 min before administration of Carmine meal to mice . We collected primary enterocytes by harvesting the proximal small intestine from anesthetized mice , emptying the luminal contents , washing with 115 mM NaCl , 5 . 4 mM KCl , 0 . 96 mM NaH2PO4 , 26 . 19 mM NaHCO3 and 5 . 5 mM glucose buffer at pH 7 . 4 and gassing for 30 min with 95% O2 and 5% CO2 . We then filled the proximal small intestine with buffer containing 67 . 5 mM NaCl , 1 . 5 mM KCl , 0 . 96 mM NaH2PO4 , 26 . 19 mM NaHCO3 , 27 mM sodium citrate and 5 . 5 mM glucose at pH 7 . 4 , saturated with 95% O2 and 5% CO2 , and incubated in a bath containing oxygenated saline at 37°C with constant shaking . After 15 min , we discarded the luminal solutions and filled the intestines with buffer containing 115 mM NaCl , 5 . 4 mM KCl , 0 . 96 mM NaH2PO4 , 26 . 19 mM NaHCO3 , 1 . 5 mM EDTA , 0 . 5 mM dithiothreitol and 5 . 5 mM glucose at pH 7 . 4 , saturated with 95% O2 and 5% CO2 , and we placed them in saline as described above . After 15 min , we collected and centrifuged the luminal contents ( 1 , 500 r . p . m . , 5 min , room temperature ) and resuspended the pellets in DMEM saturated with 95% O2 and 5% CO2 ) . We fasted 6- to 8-week-old mice for 4 hr and then each mouse received an oral gavage of 200 μl olive oil or 2 μg per g body weight 2NBDG and 2 μg per g body weight rhodamine-PEG ( Methoxyl PEG Rhodamine B , MW 5 , 000 g mol−1 ) with 0 . 2% fatty acid–free BSA by gavage . We collected feces from 20 min to 4 hr after 2NBDG was administered . We homogenized 50 mg of feces in PBS containing 30 mM HEPES , 57 . 51 mM MgCl2 and 0 . 57 mg ml−1 BSA with 0 . 5% SDS and sonicated for 30 s; we then centrifuged at 1000 g for 10 min . We transferred supernatants to 96-well plates and measured fluorescence values immediately using a fluorescence microplate reader for endpoint reading ( Molecular Devices ) . We then subtracted baseline fluorescence from untreated mice from measured fluorescence . We also measured enterocytes’ 2NBDG content after isolation of primary cells as described above , using excitation and emission wavelengths of 488 nm and 515 nm , respectively . For rhodamine-PEG , the excitation and emission wavelengths were 575 nm and 595 nm , respectively . We fasted 6–8 week old mice for 4 hr and administered 200 μl olive oil by oral gavage or IP injection and collected tail vein blood at indicated times . Serum TG concentration was determined by Wako L-Type TG determination kit ( Wako Chemicals USA ) . We collected the feces from 20 min to 4 hr after Olive oil administration . 50 mg of feces were homogenized with chloroform/methanol ( 2:1 ) in a 20:1 v/w ratio , the whole mixture was incubated overnight at 4°C with gentle shaking . Then , 0 . 2 volume of 0 . 9% NaCl was added and centrifuged at 500 g for 30 min After extracting the organic phase , samples were evaporated under nitrogen until dry and reconstituted in PBS containing 1% Triton X-100 for TG measurement by Wako L-Type TG determination kit ( Wako Chemicals USA ) . Direct binding of Mfge8 with α8 was assessed by solid-phase binding in non-tissue coated microplates . Either recombinant α8 , αvβ3 , or α5β1 were attached to the plates and purified Mfge8 was added for 2 hr at room temperature in the presence or absence of 10mM EDTA . For α5β1 , 1 mM MgCl2+ and 1 mg/mL CaCl2+ was added to activate β1 . Following 5 washes with PBS + 1% BSA and 0 . 05% Tween , the extent of Mfge8 binding was detected using a biotinylated antibody against Mfge8 ( 1:1000 , 1 hr at 37°C ) . Then streptavidin-HRP was added for 20 min at room temperature followed by 3 , 3’ , 5 , 5’ tetramethylbenzidine substrate solution . Absorbance was then measured at 450 nm in a microplate reader . Cell adhesion assays were performed as described ( Yokosaki et al . , 1994 ) with slight modifications . Briefly , 1 × 105 cells were seeded into each well of 96 well MaxiSorp enzyme-linked immunosorbent assay plates ( Nunc ) coated with substrate proteins at 37°C for 1 hr and then incubated for 1 hr at 37°C . Attached cells were stained with 0 . 5% crystal violet and solubilized in 2% Triton X-100 for taking optical density at 595 nm . For blocking experiments , cells were incubated with antibodies ( 5 μg/ml ) before plating for 15 min on ice . HGSMCs were obtained from commercial sources ( Science Cell Research Laboratories ) and maintained in minimum essential medium supplemented with 10% FBS at 37°C with 5% CO2 . We plated the cells in six-well plates 1 day prior to infection . We transfected cells with 100 nM PTEN siRNA ( ON-TARGETplus Human PTEN , Thermo Fisher Scientific ) or controls ( ON-TARGETplus Scramble Control siRNA , Human , Thermo Fisher Scientific ) in antibiotic- and norepinephrine-free culture medium using Lipofectamine-2000 ( Invitrogen ) . 6 hr later , we change the medium to fully supplemented medium and conducted assays 48 hr after transfection . Mice were starved for 4 hr before gavage with recombinant Mfge8 ( 50 µg/kg ) . Antrum samples were removed 30 min post-gavage and fixed with paraformaldehyde and paraffin-embedded or immediately frozen . Sections ( 5–10 µm ) were cut and stained for α-smooth muscle actin ( Sigma-Aldrich ) , integrin α8 ( YZ3 ) and human FC ( Rockland ) . The recombinant Mfge8 consists of a fusion protein containing full length Mfge8 and a human Fc domain ( Atabai et al . , 2009 ) . Five images were taken per antrum section and each image was divided into fifths . Thickness of individual muscle layers was quantified at each point using a scale of 314 pixels to 100 µm . Averaged thickness is reported in Figure 1—figure supplement 1 . The RhoA activation assay was performed according to the manufacturer’s instructions ( Cytoskeleton ) . Briefly , we dissected out the gastric antrum , gently removed the mucosal layer and incubate them with methacholine ( 100 μm ) for 15 min and then homogenized the muscle layer in lysis buffer ( 50 mM Tris-HCl , pH 7 . 5 , 10 mM MgCl2 , 0 . 5 M NaCl , 1% Triton X-100 , and protease and phosphatase inhibitor cocktail ( Thermo ) . Human Gastric Smooth Muscle Cells ( HGSMC ) were treated with 5-hydroxytryptamine ( 100 μM ) for 15 min and then lysed . We collected the supernatants after centrifugation and incubated with GST-Rhotekin bound to glutathione-agarose beads at 4°C for 1 hr . We washed the beads with a wash buffer containing 25 mM Tris , pH 7 . 5 , 30 mM MgCl2 , and 40 mM NaCl . GTP-bound RhoA was detected by immunoblotting . We isolated antral lysates or human gastric smooth muscle cell lysates and measured conversion of PI ( 3 , 4 , 5 ) P3 to PI ( 4 , 5 ) P2 ( PTEN activity ELISA , Echelon ) after incubation with recombinant proteins ( rMfge8 , RGE , Fibronectin , or Vitronectin , R&D Systems , Inc . 10 μg/ml ) or blocking antibodies against α8 , β3 , and β5 ( 10 g/ml ) . In this competitive ELISA , we incubated lysates on a PI ( 4 , 5 ) P2 coated microplate and then added a PI ( 4 , 5 ) P2 detector protein . PI ( 4 , 5 ) P2 produced by PTEN in lysate binds to the detector protein and thus prevents it from binding immobilized PI ( 4 , 5 ) P2 on the plate . We then used a peroxidase-linked secondary to measure PI ( 4 , 5 ) P2 detector protein binding to the plate in a colorimetric assay where the colorimetric signal is inversely proportional to the amount of PI ( 4 , 5 ) P2 produced by PTEN . We lysed tissues in cold RIPA buffer ( 50 mM Tris HCl pH 7 . 5 , 150 mM NaCl , 1% NP-40 , 1% sodium deoxycholate , 0 . 1% SDS ) supplemented with complete miniprotease and phosphatase inhibitor cocktail ( Pierce , Rockford , IL ) . We incubated lysates at 4°C with gentle rocking for 30 min , sonicated on ice for 30 s ( in 5 s bursts ) and then centrifuged at 12 , 800 rpm for 15 min at 4°C . We determined protein concentration by Bradford assay ( Bio-Rad , Hercules , CA ) . We separated 20 μg of protein by SDS-PAGE on 7 . 5% resolving gels ( Bio-Rad ) and transblotted onto polyvinylidene fluoride membranes ( Millipore ) . We incubated the membranes with a 1:1 , 000 dilution of antibodies against Akt ( catalog 9272 , Cell Signaling ) , phospho-Akt Ser473 ( clone 193H12 , Cell Signaling ) , MLC ( catalog 3672S Cell Signaling ) phospho-MLC ( clone 519 , Cell Signaling ) , MYPT ( catalog 2634S , Cell Signaling ) phospho-MYPT ( catalog 5163 , Cell Signaling ) , RhoA ( clone 67139 , Cell Signaling ) , PTEN ( clone 138G6 , Cell Signaling ) , or GAPDH ( clone 14C10 , Cell Signaling ) followed by a secondary HRP-conjugated antibody . For evaluation of total Akt , MLC or MYPT we stripped and reprobed membranes that had been blotted for phospho-versions of these proteins . Blots were developed using the enhance chemical luminescence system ( Amersham ) . We created and expressed recombinant Mfge8 and RGE protein constructs in High Five cells as previously described ( Atabai et al . , 2009 ) . All constructs were expressed with a human Fc domain for purification across a protein G sepharose column . For experiments in Figure 3A and B , Mfge8 was expressed in Freestyle 293 cells with His-tag and purified by Ni-NTA column . We placed 8-week-old α8sm-/- mice on a high-fat diet formula containing 60% fat calories ( Research Diets ) for 12 weeks . Mouse were placed on doxycycline chow ( 2 g/kg , Bioserve ) for two weeks prior to beginning the HFD and subsequently had doxycycline in their water ( 0 . 2 mg/ml ) for the duration of the experiments . We performed bone , lean and fat mass analysis with a GE Lunar PIXImus II Dual Energy X-ray Absorptiometer . We freeze-dried feces from mice on a HFD and pulverized them with a ceramic mortar and pestle . We measured caloric content of feces with an 1108 Oxygen Combustion Bomb calorimeter . We assessed data for normal distribution and similar variance between groups using GraphPad Prism 6 . 0 . We used a one-way ANOVA to make comparisons between multiple groups . When the ANOVA comparison was statistically significant ( P < 0 . 05 ) , we performed further pairwise analysis using a Bonferroni t-test . We used a two-sided Student's t-test for comparisons between 2 groups . For analysis of weight gain over time in mice , we used a two-way ANOVA for repeated measures . We used GraphPad Prism 6 . 0 for all statistical analyses . We presented all data as mean ± s . e . m . We selected sample size for animal experiments based on numbers typically used in the literature . We did not perform randomization of animals .

Animals absorb nutrients from the food they eat in a complicated process that involves multiple steps . In the mouth , teeth break down the food into smaller chunks . Then the food passes through the stomach and small intestine , where enzymes break it down into individual molecules that are small enough to be absorbed by cells that line the small intestine . These cells then package the molecules and release them into the bloodstream so that they can be distributed to the rest of the body . Muscles in the wall of the small intestine control how quickly food travels through this part of the gut . If food moves too quickly , the cells that line the intestine have less time to absorb the food molecules and may fail to absorb enough nutrients . If the food moves too slowly , an individual may experience nausea or vomiting , or the contents of their stomach may spill into their lungs . In 2014 , researchers reported that a protein in breast milk called Mfge8 helps to boost the number of fat molecules absorbed from food . Now , Khalifeh-Soltani et al . – including some of the same researchers involved in the earlier work – show that Mfge8 also slows the rate at which food travels through the small intestine in mice . Mfge8 binds to another protein called integrin α8β1 to control how often the intestine muscles contract . Genetically engineered mice that lacked integrin α8β1 developed diarrhea and food passed through their intestines more quickly than in normal mice . Furthermore , these mice did not gain as much weight as normal mice when they were fed a high-fat diet . Khalifeh-Soltani et al . ’s findings show that Mfge8 has a dual role in controlling the absorption of food molecules in the small intestine . The next challenge is to find out whether drugs that alter the activity of integrin α8β1 could be used to help treat patients with diseases in which food moves too quickly , or too slowly , through the gut .

2016

α8β1 integrin regulates nutrient absorption through an Mfge8-PTEN dependent mechanism

Sequencing studies have implicated haploinsufficiency of ARID1B , a SWI/SNF chromatin-remodeling subunit , in short stature ( Yu et al . , 2015 ) , autism spectrum disorder ( O'Roak et al . , 2012 ) , intellectual disability ( Deciphering Developmental Disorders Study , 2015 ) , and corpus callosum agenesis ( Halgren et al . , 2012 ) . In addition , ARID1B is the most common cause of Coffin-Siris syndrome , a developmental delay syndrome characterized by some of the above abnormalities ( Santen et al . , 2012; Tsurusaki et al . , 2012; Wieczorek et al . , 2013 ) . We generated Arid1b heterozygous mice , which showed social behavior impairment , altered vocalization , anxiety-like behavior , neuroanatomical abnormalities , and growth impairment . In the brain , Arid1b haploinsufficiency resulted in changes in the expression of SWI/SNF-regulated genes implicated in neuropsychiatric disorders . A focus on reversible mechanisms identified Insulin-like growth factor ( IGF1 ) deficiency with inadequate compensation by Growth hormone-releasing hormone ( GHRH ) and Growth hormone ( GH ) , underappreciated findings in ARID1B patients . Therapeutically , GH supplementation was able to correct growth retardation and muscle weakness . This model functionally validates the involvement of ARID1B in human disorders , and allows mechanistic dissection of neurodevelopmental diseases linked to chromatin-remodeling . It is becoming clear that SWI/SNF chromatin-remodeling complexes have a major impact on human diseases , from cancer to neuropsychiatric disorders to body size regulation ( Kadoch and Crabtree , 2015; Ronan et al . , 2013 ) . SWI/SNF chromatin-remodeling complexes use the energy of ATP to remodel nucleosome density and position to control epigenetic states , lineage differentiation , and cellular growth during development and cancer ( Kadoch and Crabtree , 2015; Ho and Crabtree , 2010 ) . ARID1B is a 236 kDa protein that contains an AT-rich DNA interactive domain ( ‘ARID’ domain ) and facilitates proper genomic targeting of ARID1B containing SWI/SNF complexes . ARID1B is the most commonly mutated gene in Coffin-Siris syndrome ( CSS ) , a monogenic syndrome characterized by growth retardation , facial dysmorphism , and intellectual disability ( ID ) ( Santen et al . , 2012; Tsurusaki et al . , 2012; Wieczorek et al . , 2013 ) . In addition , ARID1B is among the most frequently mutated genes in autism spectrum disorders ( ASD ) and non-syndromic ID ( O'Roak et al . , 2012; Deciphering Developmental Disorders Study , 2015 ) . In these diseases , ARID1B mutations are scattered across the gene without clear accumulation in particular domains ( Ronan et al . , 2013; Santen et al . , 2013 ) . Since these are often predicted to be nonsense or frameshift mutations , heterozygous ARID1B loss-of-function is hypothesized to be the causative genetic mechanism . Up to this point , if and how ARID1B mutations translate into various human phenotypes is unknown , and animal models have not yet been used to model or devise novel treatments for these ‘ARID1B-opathies’ . Here , we employ genetically engineered mouse models to elucidate the phenotypic impact of Arid1b mutations . We developed an Arid1b haploinsufficient mouse that exhibits neuropsychiatric abnormalities reminiscent of ASD , as well as the developmental and growth retardation phenotypes seen in CSS . Although not previously considered a cardinal feature of this syndrome , a meta-analysis of 60 patients by Santen et al . shows that on average , stature is considerably shortened in CSS patients by about two standard deviations ( Santen et al . , 2013 ) . After showing the clinical relevance of our mouse model , we focused on potentially reversible etiologies of behavioral and growth phenotypes . We observed GHRH-GH-IGF1 axis deficiencies in Arid1b heterozygous mice and also found evidence for this in humans . GH supplementation in mice rescued growth retardation and muscle weakness , which are also salient features of human ARID1B-opathies . Though successful in Mecp2 mutant mice that model Rett syndrome ( Tropea et al . , 2009; Castro et al . , 2014 ) , intervening on the GH-IGF1 axis was not able to reverse neuropsychiatric defects associated with Arid1b . Our findings not only functionally validate ARID1B’s involvement in human disease , they suggest underappreciated clinical manifestations of human ARID1B mutations that can be approached from a treatment-perspective . Using Cas9 germline gene-editing , we generated whole-body knockout and conditional floxed mice ( Figure 1A ) . Arid1b is prominently expressed in the cortex , cerebellum , and hippocampus ( Lein et al . , 2007; Ka et al . , 2016 ) ( Figure 1—figure supplement 1A ) . In heterozygous mice , Arid1b mRNA transcripts were reduced in liver , whole brain , pituitary gland , dentate gyrus , and hypothalamus ( Figure 1B ) . Protein levels showed a similar pattern in whole brain extracts , and homozygous P0 pups showed an absence of ARID1B protein ( Figure 1C ) . Homozygous mice were born but died perinatally ( Figure 1D ) . To model the genetics of haploinsufficient human ARID1B-opathies , we generated whole-body heterozygous ( Arid1b+/- ) mice [birth ratios from Arid1b+/- x Arid1b+/+ crosses: 389/661 ( 58 . 9% ) WT , 272/661 ( 41 . 1% ) Arid1b+/-] , which survived into adulthood and appeared healthy but were small for age ( Figure 1E ) . There were no abnormalities in electrolytes , liver function tests , or blood counts ( Figure 1—figure supplement 1B–D ) . 16/272 ( 6 . 6% ) of Arid1b+/- mice had hydrocephalus , the displacement of brain parenchyma by accumulated cerebrospinal fluid , a condition that frequently accompanies Dandy-Walker malformations seen in CSS patients ( Schrier Vergano et al . , 2013 ) ( Figure 1—figure supplement 1E ) . 10 . 7554/eLife . 25730 . 003Figure 1 . Arid1b+/- mice exhibit physical manifestations of developmental delay , autistic-like features , and abnormal behavioral phenotypes . ( A ) Schematic of Arid1b whole body heterozygous mice in which exon five is deleted ( hereafter referred as Arid1b+/- ) and Arid1b floxed mice . ( B ) Relative Arid1b mRNA levels in selected organs and brain regions as assessed by qPCR . ( C ) Relative Arid1b levels in p0 mouse limb ( top panel ) and whole brain extracts at p45 as assessed by western blot analysis . ( D ) Appearance of WT and Arid1b+/- mice at postnatal day 0 . ( E ) Appearance of WT and Arid1b+/- littermates at 1 month of age . ( F ) Juvenile social interaction testing for 10 WT and 9 Arid1b+/- male mice . ( G ) Grooming test for 10 WT and 9 Arid1b+/- female mice . ( H , I ) The ultrasonic vocalization ( USV ) test measuring the duration and frequency of vocal communication in 63 WT and 33 Arid1b+/- male and female mice during separation of pups from dams at postnatal day 4 . ( J ) Representative traces of WT and Arid1b+/- mice in the open field and time spent in the indicated areas for 20 WT and 20 Arid1b+/- 8 week old male mice . ( K ) Representative traces of WT and Arid1b+/- mice in the elevated plus maze and time spent in the indicated areas for 20 WT and 20 Arid1b+/- 8 week old male mice . ( L ) Dark-light box testing for 20 WT and 20 Arid1b+/- 8 week old male mice . Values represent mean ± SEM . Asterisks indicate significant differences between indicated littermate genotypes , *p-value ≤ 0 . 05; **p-value ≤ 0 . 01; ***p-value ≤ 0 . 001; ****p-value ≤ 0 . 0001; ns , not significant . Student’s t-test ( two-tailed distribution , two-sample unequal variance ) was used to calculate p-values unless otherwise indicated in the figure legend . DOI: http://dx . doi . org/10 . 7554/eLife . 25730 . 00310 . 7554/eLife . 25730 . 004Figure 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25730 . 00410 . 7554/eLife . 25730 . 005Figure 1—figure supplement 1 . Additional characterization of Arid1b+/- mice . ( A ) Arid1b in situ hybridization showing a sagittal brain section from an 8 week old male mouse . Hippocampus and cerebellum showed high levels of signal . Image taken from the Allen Brain Atlas . ( B ) Serum electrolytes including potassium ( K+ ) , creatinine ( CREA ) , and blood urea nitrogen ( BUN ) measured in 100 day-old females ( n = 9 WT and 9 Arid1b+/- ) . ( C ) Liver function tests including Aspartate transaminase ( AST ) , Alanine amino transferase ( ALT ) , and Albumin ( ALB ) in 100 day old females ( n = 9 WT and 10 Arid1b+/- ) . ( D ) Blood cell counts including white blood cells ( WBC ) , red blood cells ( RBC ) , and platelets ( PLT ) measured in 100 day old females ( n = 9 WT and 10 Arid1b+/- ) . ( E ) 0/389 WT and 16/272 ( 6 . 6% ) Arid1b+/- mice had grossly appreciable hydrocephalus . Values represent mean ± SEM . Asterisks indicate significant differences between indicated littermate genotypes: *p-value ≤ 0 . 05; **p-value ≤ 0 . 01; ***p-value ≤ 0 . 001; ****p-value ≤ 0 . 0001; ns , not significant . Student’s t-test ( two-tailed distribution , two-sample unequal variance ) was used to calculate p-values unless otherwise indicated in the figure legend . DOI: http://dx . doi . org/10 . 7554/eLife . 25730 . 00510 . 7554/eLife . 25730 . 006Figure 1—figure supplement 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25730 . 00610 . 7554/eLife . 25730 . 007Figure 1—figure supplement 2 . Additional neurobehavioral testing on Arid1b+/- mice . ( A ) Marble burying test for 10 WT and 9 Arid1b+/- female mice . ( B ) Representative spectrograms of WT and Arid1b+/- vocalizations . ( C ) Number of vocalization calls of 63 WT and 33 Arid1b+/- male and female mice . ( D ) Locomotor activity quantified by number of beam breaks in a familiar home-cage environment within periodic time intervals over the course of two hours . Two-way repeated measures ANOVA was used to calculate the p-value ( left ) . Total number of beam breaks during two hours , analyzed with Student’s t-test ( right ) . 8 week old male mice ( n = 20 WT and 20 Arid1b+/- ) were used . ( E ) Time spent in the target quadrant vs . other quadrants in the probe testing of the water maze test at day 14 . The target quadrant contains a hidden platform . 20 WT and 20 Arid1b+/- mice were examined ( left ) . Number of target quadrant crossings in the probe testing of the water maze test at day 14 ( right ) . 20 WT and 20 Arid1b+/- mice were examined . ( F ) Percent freezing time was quantified in the contextual fear conditioning test ( n = 10 WT and 11 Arid1b+/- 8 week old male mice ) . ( G ) Percent freezing time was quantified in the cued fear conditioning test ( n = 10 WT and 11 Arid1b+/- 8 week old male mice ) . ( H ) Foot shock sensitivity testing ( n = 10 WT and 11 Arid1b+/- 8 week old male mice ) . Values represent mean ± SEM . Asterisks indicate significant differences between indicated littermate genotypes: *p-value ≤ 0 . 05; **p-value ≤ 0 . 01; ***p-value ≤ 0 . 001; ****p-value ≤ 0 . 0001; ns , not significant . Student’s t-test ( two-tailed distribution , two-sample unequal variance ) was used to calculate p-values unless otherwise indicated in the corresponding figure legend . DOI: http://dx . doi . org/10 . 7554/eLife . 25730 . 00710 . 7554/eLife . 25730 . 008Figure 1—figure supplement 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25730 . 008 Given the associations between Arid1b mutations and ASD , we examined behaviors related to this disorder . To examine social interactions , we quantified the time spent interacting with a juvenile target mouse . Compared to WT littermate controls , Arid1b+/- mice spent significantly less time interacting with unfamiliar juvenile mice ( Figure 1F ) , suggesting impaired social behavior . To enrich the connections between Arid1b+/- mice and ASD-like phenotypes , we also performed grooming and marble burying tests that examined repetitive behaviors ( Silverman et al . , 2010 ) . Consistent with other ASD mouse models , Arid1b+/- mice exhibited increased self-grooming ( Figure 1G ) and potentially as a consequence , buried less marbles ( Figure 1—figure supplement 2A ) . A similar pattern of repetitive behaviors was seen with Synapsin knockout mice , another mouse model of ASD ( Greco et al . , 2013 ) . Another feature of ASD is abnormal communication and language . Several mouse models of ASD and language disorders show alterations in one or more vocalization parameters , including the number , duration , frequency , amplitude , and other characteristics of ultrasonic vocalizations ( USVs ) ( Konopka and Roberts , 2016; Araujo et al . , 2015 ) . Furthermore , ASD patients who have retained speech tend to exhibit abnormalities in voice quality and pitch ( Kanner , 1968; Bonneh et al . , 2011 ) . USVs emitted by Arid1b+/- mice are longer in duration , and have abnormal pitch ( Figure 1H , I and Figure 1—figure supplement 2B ) . Interestingly , Arid1b+/- mice emitted the same total number of USVs compared to WT mice ( Figure 1—figure supplement 2C ) , suggesting altered modulation rather than absent vocalizations . Anxiety-like behavior , a comorbidity of ASD , was examined using three separate tasks in male mice . In the open field test , Arid1b+/- mice spent significantly more time in the periphery while avoiding the anxiety-provoking center ( Figure 1J ) . In the elevated plus maze , Arid1b+/- mice spent more time in the anxiety-relieving , walled arms of the maze ( Figure 1K ) . In the dark-light box test , Arid1b+/- mice avoided exploring the brightly lit chamber ( Figure 1L ) . WT and mutant mice traveled equal distances both initially and over a 2 hr time period , making locomotor differences less likely a confounder in simple environments ( Figure 1—figure supplement 2D ) . These tests consistently demonstrated higher levels of anxiety-like behavior in Arid1b+/- mice compared to their WT littermates . Given the associations between Arid1b haploinsufficiency and intellectual disability , we assessed cognitive functions in Arid1b+/- mice . The Morris water maze test , a contextual fear-conditioning test , and a cued fear-conditioning test each did not reveal defects in memory and learning ( Figure 1—figure supplement 2E–G ) . The genotypes were equally able to sense the electric shock applied during fear conditioning ( Figure 1—figure supplement 2H ) . Overall , these tests showed that Arid1b+/- mice displayed abnormal social , vocal , and behavioral phenotypes , but did not clearly have cognitive or memory deficiencies . In an effort to understand how behavioral abnormalities arose , we examined Arid1b+/- brains for other neurodevelopmental abnormalities . Because some patients with ARID1B mutations exhibit corpus callosum hypoplasia or agenesis ( Schrier Vergano et al . , 2013 ) , we examined brains of Arid1b+/- mice and identified a significant reduction in corpus callosum volume ( Figure 2A ) . Consistent with studies showing that small hippocampus , dentate gyrus , and cortex size are associated with anxiety and depressive disorders in mice and humans ( Persson et al . , 2014; Travis et al . , 2015; Boldrini et al . , 2013; Schmaal et al . , 2017 ) , Arid1b+/- mice have smaller dentate gyri ( Figure 2B ) and both Arid1b+/- and Arid1b-/- pups had reduced cortical thickness with reduced TBR1 marked neuronal cellularity ( Figure 2—figure supplement 1A–D ) . Less proliferating cells were also seen in the subgranular zone of the dentate gyrus ( Figure 2C , D , F , G ) , especially in posterior regions ( Figure 2E , H ) . Thus , reduced corpus callosum size , dentate gyrus size , cortex thickness , and proliferation are neuroanatomical and cellular correlates of the behavioral phenotypes seen in Arid1b mutants . 10 . 7554/eLife . 25730 . 009Figure 2 . Arid1b haploinsufficiency results in neuroanatomical abnormalities implicated in neuropsychiatric diseases . ( A ) Relative corpus callosum volume quantified through Cavalieri analysis ( n = 8 WT and 7 Arid1b+/- brains from 50 day old females ) . ( B ) Dentate gyrus volume quantified through Cavalieri analysis ( n = 7 WT and 7 Arid1b+/- brains from 50 day old females ) . ( C ) Representative Ki67 immunostaining . ( D ) Quantitation of Ki67+ total cell number ( 8 WT and 7 Arid1b+/- brains from 50 day old females ) . ( E ) Bregma analysis was used to determine cell proliferation ( Ki67 ) as a function of location in the subgranular zone of the dentate gyrus . Two-way ANOVA with uncorrected Fischer’s Least Significant Difference ( LSD ) was used to calculate the statistics . ( F ) Representative BrdU immunostaining . WT and Arid1b+/- mice received one injection per day of the thymidine analog , bromodeoxyuridine ( BrdU ) , for five days and brains were harvested three days following the last injection ( 6 WT and 4 Arid1b+/- brains from 50 day old females ) . ( G ) Quantification of BrdU+ total cell number . ( H ) Bregma analysis was used to determine cell proliferation ( BrdU ) as a function of location in the subgranular zone of the dentate gyrus ( n = 6 WT and 4 Arid1b+/- ) . Values represent mean ± SEM . Asterisks indicate significant differences between indicated littermate genotypes , *p-value ≤ 0 . 05; **p-value ≤ 0 . 01; ***p-value ≤ 0 . 001; ****p-value ≤ 0 . 0001; ns , not significant . Student’s t-test ( two-tailed distribution , two-sample unequal variance ) was used to calculate p-values unless otherwise indicated in the figure legend . DOI: http://dx . doi . org/10 . 7554/eLife . 25730 . 00910 . 7554/eLife . 25730 . 010Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25730 . 01010 . 7554/eLife . 25730 . 011Figure 2—figure supplement 1 . Arid1b+/- and Arid1b-/- brains have defects in cortical development . ( A ) Hoechst staining of WT , Arid1b+/- , and Arid1b-/- brain sections showing cortical thickness differences outlined by the region in the white box . ( B ) Quantification of cortical thickness in WT , Arid1b+/- , and Arid1b-/- brain sections ( the n is at the bottom of each bar ) . ( C ) Close up view of cortical regions in the white box from ( A ) above , showing reduced cortical thickness and cellularity . ( D ) TBR1 staining in WT and Arid1b-/- cortex showing neuron numbers . DOI: http://dx . doi . org/10 . 7554/eLife . 25730 . 01110 . 7554/eLife . 25730 . 012Figure 2—figure supplement 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25730 . 012 RNA-seq was performed to examine the impact of Arid1b haploinsufficiency on transcriptional output in the hippocampus . Differential gene expression analysis showed 56 significantly down- and 79 upregulated mRNAs ( edgeR FDR < 0 . 05; Figure 3A ) . As expected , Arid1b was one of the most downregulated genes . Globally , differentially regulated genes were associated with nervous system development as well as psychological , behavioral , and developmental disorders ( Figure 3B ) . Arid1b+/- tissues also showed specific alterations in Ephrin , nNOS , axonal guidance and glutamate receptor signaling pathways ( Figure 3C ) . 14 of 140 ( 10% ) differentially regulated genes were among the highest ranking candidate autism risk genes identified in the SFARI gene database ( Basu et al . , 2009 ) ( Figure 3D ) . To determine if some of these genes could be directly regulated by SWI/SNF , we analyzed the ChIP-Seq targets of Smarca4 ( Brg1 ) , a core SWI/SNF complex subunit ( Attanasio et al . , 2014 ) . 91 of 140 ( 65% ) differentially regulated genes showed direct binding by Brg1 ( Figure 3E ) , with positional enrichment at transcriptional start sites ( TSSs ) ( Figure 3F , G ) . Arid1b-mediated SWI/SNF transcriptional activities appeared to directly regulate numerous neuropsychiatric related genes , including ones implicated in ASD ( Figure 3H , I ) . Our data also show that haploinsufficiency is sufficient to cause broad gene expression disruption , but future studies will be required to determine the exact downstream genes that account for the neuropsychiatric phenotypes . 10 . 7554/eLife . 25730 . 013Figure 3 . Arid1b haploinsufficiency results in changes in the expression of SWI/SNF regulated genes implicated in neuropsychiatric diseases . ( A ) All significantly up- and downregulated genes in the Arid1b+/- hippocampus are ranked according to p-value ( least to most significant from left to right ) . ( B ) Most enriched diseases and biological functions in hippocampus . ( C ) Most differentially regulated pathways in the hippocampus . ( D ) 14 of 140 ( 10% ) differentially regulated genes were among the highest ranking autism risk genes identified in the SFARI database . Category S: syndromic , Category 1: high confidence , Category 2: strong candidate , Category 3: suggestive evidence , Category 4: minimal evidence , Category 5: Hypothesized ( Basu et al . , 2009 ) . ( E ) Pie chart showing that 91 of 140 ( 65% ) differentially regulated genes in hippocampus are direct targets of Brg1 , a core SWI/SNF complex subunit . Brg1 target genes were identified using ChIP-Seq in mouse e11 . 5 forebrain ( Attanasio et al . , 2014 ) . ( F ) Metaplot showing enrichment of Brg1 at the TSSs of genes regulated by Arid1b . ( G ) Heatmap showing Brg1 promoter binding in these genes . ( H ) Differential mRNA expression of representative genes involved in neurodevelopment and ASD ( Data from: SFARI database , updated September , 2016 ) ( Basu et al . , 2009 ) . ( I ) Brg1 peaks suggesting direct binding of SWI/SNF at the promoters of ASD-related genes . Values represent mean ± SEM . Asterisks indicate significant differences between indicated littermate genotypes , *p-value ≤ 0 . 05; **p-value ≤ 0 . 01; ***p-value ≤ 0 . 001; ****p-value ≤ 0 . 0001; ns , not significant . Student’s t-test ( two-tailed distribution , two-sample unequal variance ) was used to calculate p-values unless otherwise indicated in the corresponding figure legend . DOI: http://dx . doi . org/10 . 7554/eLife . 25730 . 013 Having establishing that Arid1b haploinsufficient mice recapitulate salient aspects of human ARID1B-opathies , we were particularly interested in identifying reversible pathological mechanisms and therapeutic opportunities . Since we identified neuroanatomical and neural expression aberrations in Arid1b+/- mice , we also asked if any non-neuropsychiatric syndromic features were potentially related to neurodevelopmental abnormalities . As mentioned previously , Arid1b+/- mice developed reduced nose-to-rump length and weight ( Figure 4A , B ) . Arid1b+/- mice had disproportionally small kidneys and hearts , but no other gross organ defects ( Figure 4—figure supplement 1A ) . Profiling using metabolic cages showed that Arid1b+/- mice had equivalent food intake and water consumption ( Figure 4—figure supplement 1B , C ) , suggesting that size differences were unlikely due to food intake or energetic differences . 10 . 7554/eLife . 25730 . 014Figure 4 . Growth retardation in Arid1b+/- mice is due to GH-IGF1 axis deficiency with a neuronal source . ( A ) Body length ( nose-to-rump ) curve of females ( n = 9 WT and 9 Arid1b+/- ) . ( B ) Body weight growth curve for males ( n = 14 WT and 14 Arid1b+/- ) and females ( n = 20 WT and 20 Arid1b+/- ) . For ( A ) and ( B ) , repeated ANOVA with Bonferroni’s post-hoc analysis was used . ( C ) Plasma IGF1 as measured by ELISA ( n = 16 WT and 19 Arid1b+/- 28–41 day old male and female mice ) . ( D ) Igf1 mRNA in WT and Arid1b+/- livers as measured by qPCR ( n = 5 WT and 5 Arid1b+/- livers from 45 day old female mice ) . ( E ) Plasma GH as measured by ELISA ( n = 15 WT and 14 Arid1b+/- 28–41 day old male and female mice ) . ( F ) Gh mRNA in WT and Arid1b+/- pituitary as measured by qPCR ( n = 6 WT and 5 Arid1b+/- pituitary from 33-44 day old female mice ) . ( G ) Plasma GH ( n = 9 WT and 9 Arid1b+/- 2 week old male mice ) . ( H ) Plasma GH before and after stimulation by human GHRH ( n = 19 WT and n = 20 Arid1b+/- mice at baseline , n = 11 WT and n = 10 Arid1b+/- mice 5 and 15 min after GHRH administration ) . ( I ) Ghrh mRNA in WT and Arid1b+/- mediobasal hypothalamus as measured by qPCR . ( n = 8 WT and 7 Arid1b+/- samples from 33-44 day old female mice ) . ( J ) Body weight curve for female Arid1bFl/+ ( n = 9 ) and Albumin-Cre; Arid1bFl/+ ( n = 15 ) mice . ( K ) Body weight curve for female Arid1bFl/+ ( n = 10 ) and Nestin-Cre; Arid1bFl/+ ( n = 6 ) mice . ( L ) Plasma IGF1 levels for 40–45 day old female Arid1bFl/+ ( n = 7 ) and Albumin-Cre; Arid1bFl/+ ( n = 7 ) mice . ( M ) Plasma IGF1 levels for 30–45 days old female Arid1bFl/+ ( n = 6 ) and Nestin-Cre; Arid1bFl/+ ( n = 6 ) mice . ( N ) Plasma GH levels for female Arid1bFl/+ ( n = 5 ) and Nestin-Cre; Arid1bFl/+ ( n = 5 ) mice . Values represent mean ± SEM . Asterisks indicate significant differences between indicated littermate genotypes , *p-value ≤ 0 . 05; **p-value ≤ 0 . 01; ***p-value ≤ 0 . 001; ****p-value ≤ 0 . 0001; ns , not significant . Student’s t-test ( two-tailed distribution , two-sample unequal variance ) was used to calculate p-values unless otherwise indicated in the figure legend . DOI: http://dx . doi . org/10 . 7554/eLife . 25730 . 01410 . 7554/eLife . 25730 . 015Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25730 . 01510 . 7554/eLife . 25730 . 016Figure 4—figure supplement 1 . Growth and metabolic analysis of Arid1b+/- mice . ( A ) Arid1b+/- organ/body weight ratios presented as % of WT organ/body weight ratio . ( B ) Food intake and ( C ) water consumption was quantified in a metabolic cage over 4 days . The absolute data was normalized to body weight ( n = 12 WT and 12 Arid1b+/-8 week old male mice ) . Values represent mean ± SEM . Asterisks indicate significant differences between indicated littermate genotypes: *p-value ≤ 0 . 05; **p-value ≤ 0 . 01; ***p-value ≤ 0 . 001; ****p-value ≤ 0 . 0001; ns , not significant . Student’s t-test ( two-tailed distribution , two-sample unequal variance ) was used to calculate p-values unless otherwise indicated in the figure legend . DOI: http://dx . doi . org/10 . 7554/eLife . 25730 . 01610 . 7554/eLife . 25730 . 017Figure 4—figure supplement 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25730 . 017 A critical factor that regulates both body size and brain development is Insulin-like growth factor ( IGF1 ) . We found a significant reduction in plasma IGF1 levels in Arid1b+/- mice ( Figure 4C ) , and confirmed a reduction in Igf1 mRNA in the liver , which is a major source of IGF1 ( Figure 4D ) . To discern if there was a hypothalamic , pituitary , peripheral , or combinatorial problem that led to IGF1 deficiency , we performed a series of endocrinologic tests . In the same cohorts of mice with IGF1 deficiency , fasting GH was not significantly different in Arid1b+/- mice ( Figure 4E ) . In addition , there were no significant Gh mRNA expression differences between WT and Arid1b+/- pituitary glands despite differences in Arid1b mRNA levels ( Figure 4F ) . We also confirmed that plasma GH was not altered in younger 2 week old Arid1b+/- mice , an age where GH levels are more critical for growth ( Figure 4G ) . The combination of low IGF1 and normal GH levels pointed to a peripheral defect without appropriate GH compensation from the pituitary gland . Because GH was not elevated as would be expected if there was only a peripheral IGF1 producing defect , we asked if the Arid1b+/- pituitary was capable of making and secreting sufficient amounts of GH in the context of GH stimulation testing . In multiple cohorts of Arid1b+/- mice , GH levels were never significantly different at baseline and also increased normally at multiple time points after stimulation with Growth hormone-releasing hormone ( GHRH ) ( Figure 4H ) . This indicated a normal ability for the pituitary to respond to exogenous GHRH . Next , we attempted to determine if the hypothalamus was not producing enough Ghrh . We dissected the mediobasal hypothalamus containing Ghrh expressing neurons to examine Ghrh expression . We found that Ghrh mRNA levels were not significantly different ( Figure 4I ) , indicating a lack of appropriate GHRH response to IGF1 deficiency , suggesting a central defect that contributed to growth impairment . In addition , we sought genetic evidence for a partial central ( hypothalamic or pituitary ) root cause of IGF1 deficiency by generating organ-specific Arid1b mutant models . We used Nestin-Cre in the brain , Albumin-Cre in the liver , and Ckmm-Cre in the skeletal muscles to spatially control Arid1b haploinsufficiency . Neither Albumin-Cre; Arid1bFl/+ ( Figure 4J ) nor Ckmm-Cre; Arid1bFl/+ mice ( data not shown ) showed growth or morphological defects , while Nestin-Cre; Arid1bFl/+ mice recapitulated the growth impairments seen in whole body Arid1b+/- mice ( Figure 4K ) , suggesting at least a neuronal contribution to growth impairment . While Albumin-Cre; Arid1bFl/+ mice showed no IGF1 differences , Nestin-Cre; Arid1bFl/+ mice had reduced plasma IGF1 levels ( Figure 4L , M ) . Moreover , Nestin-Cre; Arid1bFl/+ mice showed an inappropriate lack of GH increase in the face of this IGF1 deficiency ( Figure 4N ) . Since liver specific Arid1b+/- mice did not replicate the whole body Arid1b+/- mice , it is possible that a combination of central and multi-organ peripheral defects in the GHRH-GH-IGF1 axis were required to fully recapitulate the growth impairment of whole body Arid1b+/- mice . Given plasma IGF1 deficiency in Arid1b+/- cohorts , we first tested if IGF1 replacement could rescue physical aspects of developmental delay and abnormal behavioral phenotypes . Neither body size ( Figure 5A ) nor elevated plus maze abnormalities ( Figure 5B ) were rescued after treating WT and Arid1b+/- cohorts with recombinant human IGF1 ( rhIGF1 ) . This was not surprising because it is known that exogenous IGF1 is unstable and often does not efficiently reach target tissues responsible for growth ( Kletzl et al . , 2017 ) . 10 . 7554/eLife . 25730 . 018Figure 5 . GH therapy reverses growth retardation and muscle weakness . ( A ) Body weights at p50 ( WT + vehicle ( n = 12 ) , WT + rhIGF1 ( n = 12 ) , Arid1b+/- + vehicle ( n = 13 ) and Arid1b+/- + rhIGF1 ( n = 13 ) ) . ( B ) Time spent in the closed arms of elevated plus maze at p50 ( WT + vehicle ( n = 27 ) , WT + rhIGF1 ( n = 29 ) , Arid1b+/- + vehicle ( n = 21 ) and Arid1b+/- + rhIGF1 ( n = 22 ) ) . For ( A ) and ( B ) , 0 . 5 mg/kg rhIGF1 was administrated daily starting from postnatal day 11 . ( C ) Schema showing the duration and dose of daily recombinant GH treatment . ( D ) Body weights at p50 ( WT + vehicle ( n = 20 ) , WT + rmGH ( n = 16 ) , Arid1b+/- + vehicle ( n = 16 ) and Arid1b+/- + rmGH ( n = 16 ) ) . ( E ) Nose-to-rump lengths at p50 ( WT + vehicle ( n = 7 ) , WT + rmGH ( n = 7 ) , Arid1b+/- + vehicle ( n = 6 ) and Arid1b+/- + rmGH ( n = 6 ) ) . ( F ) Forelimb grip strength at p50 ( WT + vehicle ( n = 9 ) , WT + rmGH ( n = 9 ) , Arid1b+/- + vehicle ( n = 9 ) and Arid1b+/- + rmGH ( n = 9 ) ) . ( G ) Hindlimb grip strength at p50 ( WT + vehicle ( n = 9 ) , WT + rmGH ( n = 9 ) , Arid1b+/- + vehicle ( n = 9 ) and Arid1b+/- + rmGH ( n = 9 ) ) . ( H ) Time spent in the closed arms of elevated plus maze at p50 ( WT + vehicle ( n = 19 ) , WT + rmGH ( n = 16 ) , Arid1b+/- + vehicle ( n = 19 ) and Arid1b+/- + rmGH ( n = 16 ) ) . Values represent mean ± SEM . Asterisks indicate significant differences between indicated littermate genotypes: *p-value ≤ 0 . 05; **p-value ≤ 0 . 01; ***p-value ≤ 0 . 001; ****p-value ≤ 0 . 0001; ns , not significant . Two-way ANOVA was used to calculate the p-value . DOI: http://dx . doi . org/10 . 7554/eLife . 25730 . 01810 . 7554/eLife . 25730 . 019Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25730 . 01910 . 7554/eLife . 25730 . 020Figure 5—figure supplement 1 . Two ARID1B mutant CSS patients were GH deficient and responsive to GH replacement . ( A ) Growth curve of a CSS patient with an ARID1B mutation . GH treatments started at 2 years of age . Green shaded area represents the normal range . ( B ) Growth curve of a CSS patient with a 6q21 . 5 deletion , which includes the ARID1B locus . GH treatments started at 5 . 5 years of age . DOI: http://dx . doi . org/10 . 7554/eLife . 25730 . 020 The fact that GH was not elevated in the context of low IGF1 suggested to us that there was not adequate GH production or compensation . Thus , we asked if GH supplementation could rescue some of the physical aspects of developmental delay . WT and Arid1b+/- cohorts were treated with recombinant mouse GH ( rmGH ) ( Figure 5C ) . After 40 days of treatment , Arid1b heterozygous mice gained significantly more body weight and nose-to-rump length than did WT mice ( Figure 5D , E ) , demonstrating that exogenous GH supplementation was sufficient to rescue growth retardation in Arid1b+/- mice . Given this selective efficacy for mutant mice , we asked if GH could potentially improve muscle weakness often associated with CSS . We found that at baseline , Arid1b+/- mice also had muscle weakness identified through grip strength testing . Replacement with GH was able to selectively increase muscle strength in mutant mice ( Figure 5F , G ) . Despite improvements in physical manifestations , GH replacement was not able to reverse behavioral phenotypes such as anxiety , as measured in the elevated plus maze ( Figure 5H ) . This suggested that correcting the GHRH-GH-IGF1 axis was not sufficient to rescue neuropsychiatric manifestations , but was able to reverse growth retardation mediated by Arid1b deficiency . In an analysis of 60 ARID1B CSS patients , height was shown to be significantly reduced ( Santen et al . , 2014 ) . In addition , some non-syndromic patients with missense mutations in ARID1B exhibited growth deficiency due to partial GH deficiency ( Yu et al . , 2015 ) . We also obtained clinical information from additional CSS patients , two with ARID1B mutations ( from the www . arid1bgene . com database ) and one with a mutation in SMARCA4 , which encodes another SWI/SNF component . All three of these cases had deficiencies in the GH-IGF1 axis and clear beneficial responses to GH replacement therapy ( growth curves for the ARID1B patients are shown in Figure 5—figure supplement 1A , B ) . These data from humans and mice suggest deficiencies at various parts of the GHRH-GH-IGF axis , leading to growth impairment responsive to GH supplementation . In summary , we have developed the first mouse model of Arid1b haploinsufficiency , one of the most common genetic lesions found in ASD , ID , and CSS . Several aspects of our model recapitulated the features of the overlapping disorders associated with ARID1B ( Table 1 ) . Our model is faithful to the heterozygosity seen in most ARID1B-opathies . It is interesting that another model of ASD involving a chromatin remodeling gene called CHD8 also requires haploinsufficiency ( Katayama et al . , 2016 ) and suggests that dose could play a critical mechanistic role in phenotypes resulting from mutations in epigenetic regulators . Our study also raises interesting questions about how genotype relates to phenotype in diseases involving ARID1B . Some children with ARID1B mutations have a subset but not all features of CSS , and others have different disorders such as ASD or isolated corpus callosum agenesis . For example , Yu et al . reported cases of idiopathic short stature without cognitive defects that were attributed to de novo ARID1B missense mutations , suggesting that there are either differences between mutations or important genetic interactions that play a key role in defining the phenotypic expression of these mutations ( Yu et al . , 2015 ) . It is also possible that differences between the function of Arid1b in mouse and human brain development account for intellectual and cognitive discrepancies . Our mouse model affords the ability to interrogate these types of questions in different strain backgrounds and with genetic interactors . We are hopeful that this will advance the understanding of ARID1B and SWI/SNF in human diseases . 10 . 7554/eLife . 25730 . 021Table 1 . Major clinical features associated with ARID1B mutations and phenotypes seen in Arid1b+/- mice . Abbreviations: CSS: Coffin-siris syndrome , ID: Intellectual disability , ASD: Autism spectrum disorder . DOI: http://dx . doi . org/10 . 7554/eLife . 25730 . 021Human ARID1B featuresDiagnosisReferencesMouse Intellectual/cognitive disabilityCSS , ID ( Hoyer et al . , 2012; Santen et al . , 2012; Halgren et al . , 2012 ) No Growth retardationCSS ( Tsurusaki et al . , 2012 ) Yes Coarse facial featuresCSS ( Sim et al . , 2015 ) Unknown Muscle hypotoniaCSS ( Hoyer et al . , 2012 ) Yes HydrocephalusCSS ( Imai et al . , 2001 ) Yes Corpus callosum agenesis or hypoplasiaCSS ( Halgren et al . , 2012; Santen et al . , 2012 ) Yes Brachydactyly , hypoplastic nail/fingerCSS ( Hoyer et al . , 2012; Santen et al . , 2012; Brautbar et al . , 2009 ) Unknown Abnormal vocalization , speech impairmentASD , CSS ( Santen et al . , 2012 ) Yes AnxietyASD , CSS ( O'Roak et al . , 2012; Deciphering Developmental Disorders Study , 2015 ) Yes Social interaction deficitsASD ( O'Roak et al . , 2012; Deciphering Developmental Disorders Study , 2015 ) Yes Repetitive behaviorsASD ( O'Roak et al . , 2012; Deciphering Developmental Disorders Study , 2015 ) Yes We also uncovered a role for the GHRH-GH-IGF1 axis in ARID1B-opathies . Previously , height was shown to be reduced in ARID1B patients ( Santen et al . , 2014 ) , but there are only anecdotal findings of GH deficiencies ( Figure 5—figure supplement 1A , B ) ( Yu et al . , 2015 ) . After the clinical identification of ARID1B or SWI/SNF mutations , interventions for short stature are usually not investigated . Thus , it is likely that GHRH-GH-IGF1 deficiency is under-diagnosed and rarely treated in this patient population . Conversely , ARID1B mutations are not suspected in patients with non-syndromic short stature , and could represent a more common causative mechanism than previously suspected . The findings here should motivate deeper interrogation of the GHRH-GH-IGF1 axis and potentially GH supplementation in syndromic patients with CSS or non-syndromic patients with ARID1B mutations and short stature . Our study does not pinpoint the exact source of the peripheral IGF1 deficiency since liver-specific Cre lines did not recapitulate the IGF1 deficiency seen in the whole body Arid1b+/- mice . Given the Nestin-Cre results , it is possible that reduced IGF1 production in the brain led to reduced plasma IGF1 and the inability to compensate with GHRH and GH exacerbated growth retardation . Another possibility that subtle peripheral defects in the liver and muscle will only manifest when combined with defects in other organs such as the brain . Future studies with tissue-specific conditional experiments could help to resolve these questions . Overall , our study provides a preclinical model for mechanistic and therapeutic dissection of ARID1B related diseases , and offers a translatable avenue to alleviate growth related aspects of developmental delay . All animal procedures were based on animal care guidelines approved by the Institutional Animal Care and Use Committee at University of Texas Southwestern Medical Center ( UTSW ) . Constitutive and conditional Arid1b knockout mice were generated by the UTSW Transgenic Core using CRISPR/Cas9 genome editing . Guide RNAs were designed to target sequences before and after exon 5 of Arid1b , creating a frame-shift mutation to induce nonsense-mediated decay . Guide RNAs , S . pyogenes Cas9 mRNA , and oligo donors containing LoxP sequences were injected into single celled zygotes . C57BL/6J mice were used to generate these mice . To generate WT and Arid1b+/- study mice , C57BL/6J WT females were crossed to Arid1b+/- males . Arid1b+/- mice were tail genotyped using the primers CTTGGTCTTACCCATTTGCACAGT ( forward ) and GATGGAGGATCCTTACTACAGGGGGATT ( reverse ) . Amplicon size for WT allele is 710 bp and deletion band is 310 bp . Arid1b floxed mice were tail genotyped using the primers 5’-CTT GGT CTT ACC CAT TTG CAC AGT-3’ ( forward ) and 5’-AGT GCC TAG GAA GGC AGA GTT TGA GAG-3’ ( reverse ) . Amplicon size for WT allele is 475 bp , and for floxed allele is 554 bp . The whole pituitary and the mediobasal hypothalamus ( MBH ) , which includes both the arcuate nucleus ( ARC ) and the ventromedial ( VMH ) hypothalamus , were dissected and subjected to RT-qPCR analysis . Total RNA was isolated using TRIzol reagent ( Invitrogen ( Carlsbad , California ) , catalog no . 15596018 ) . cDNA synthesis was performed with 1 μg of total RNA using the iScript RT Supermix ( BioRad , catalog no . 1708840 ) . SYBR Green based quantitative real-time PCR was performed . Gene expression levels were measured using the ΔΔCt method . Mouse Arid1b primers were: forward , 5’-GTTGGCTCTCCTGTGGGAAGCAA-3’; reverse , 5’- GTGACTGGCTCAAGGCAGGAT-3’ . Tissues were lysed in TPER tissue protein extraction reagent and homogenized in FastPrep tissue homogenizer . Western blots were performed in standard fashion . Primary antibodies were prepared in 5% BSA in PBS-T . The following primary antibody was used: Arid1b Antibody ( KMN1 ) X ( cat #: sc-32762 X , RRID:AB_2060367 ) . Following secondary antibody was used: anti-mouse IgG , HRP-linked antibody ( Cell Signaling , #7076 , RRID:AB_330924 ) . Perinatal mice were decapitated after anesthesia . Heads were fixed in 4% PFA overnight before brains were extracted . Adult mice were anesthetized and underwent intracardial perfusion with ice-cold 0 . 1 M phosphate-buffered saline ( PBS ) followed by 4% PFA for fixation . Extracted brains were immersed for 24 hr in 4% PFA in 0 . 1M PBS at 4°C for post-fixation , followed by least three days of immersion in 30% sucrose in 0 . 1 M PBS with 0 . 01% sodium azide for cryoprotection . Brains were cut into 40 μm- ( adult ) or 20 μm-thick ( perinatal ) sections with a cryostat ( model CM3050S; Leica ) . Brain sections were permeabilized with 0 . 25% Triton X-100 in 1×PBS and then blocked for 2 hr with 5% BSA/3% normal goat serum ( NGS ) in 0 . 25% Triton X-100 in 1×PBS . Primary antibodies for TBR1 ( 1:150 dilution , rabbit , polyclonal , Abcam Cat# ab31940 RRID:AB_2200219 ) were applied to sections overnight at 4°C . To count cell number , brain sections were incubated with Hoechst 33342 ( 1 μg/ml; Cell Signaling Technology Cat# 4082S RRID:AB_10626776 ) alone or together with secondary antibodies ( Alexa 488 ) for 2 hr at room temperature . For the analysis of proliferation , adult mice received one BrdU injection for five consecutive days , and three days following the last injection , brains were fixed and harvested . Slide-mounted IHC for BrdU- , Ki67- , and doublecortin- immunoreactive ( + ) cells in dentate gyrus was performed as described previously ( DeCarolis et al . , 2014; Walker et al . , 2015 ) . Briefly , every ninth section of the hippocampus was slide-mounted onto charged slides and left for two hours to dry . Antigen retrieval was performed using 0 . 01 M citric acid ( pH 6 . 0 ) at 100°C for 15 min , followed by washing in PBS at room temperature ( RT ) . Hydrogen peroxide ( 0 . 3% H2O2 ) incubation was performed for 30 min to inhibit endogenous peroxidase activity . For BrdU IHC , permeabilization with 0 . 1% Trypsin in 0 . 1 M TRIS and 0 . 1% CaCl2 and denaturation with 2N HCl in 1x PBS were performed in order to allow antibody access to nuclear DNA . One hour of blocking non-specific binding was performed by incubation in 3% donkey serum , 0 . 3% Triton-X in PBS . Following these steps , slides were incubated with rat-α-BrdU ( 1:400; Bio-Rad / AbD Serotec Cat# OBT0030 RRID:AB_609568 ) , rabbit-α-Ki67 antibody ( 1:500; Thermo Fisher Scientific Cat# RM-9106-S0 RRID:AB_2341197 ) in 3% serum and 0 . 3% Tween-20 overnight . Primary antibody incubation was followed by 1 x PBS rinses and a 1 hr incubation with a biotin-tagged secondary antibody targeting the respective primary antibody . Following rinses with 1x PBS , incubation with an avidin-biotin complex occurred for 90 min . Incubation with diaminobenzidine for 5–10 min was used to visualize immunoreactive cells . The counterstain Fast Red was used for nuclear visualization ( ~3 min incubation ) . Lastly , slides were placed through an ethanol dehydration series and coverslipped with DPX . BrdU+ and Ki67+ cells were quantified using an Olympus BX-51 microscope at 40X by an observer blind to experimental groups as previously described ( Walker et al . , 2015 ) . Immunopositive cells were quantified in every 9th coronal section in the subgranular zone of the granular cell layer in the dentate gyrus , spanning the entire anterior-posterior axis of the hippocampus ( −0 . 82 mm to −4 . 78 mm from Bregma ) . Manual stereological counting was performed under bright field , and total cell counts were multiplied by nine to account for the whole hippocampus . Doublecortin+ cells were quantified using the Optical Fractionator method , and dentate gyrus and corpus callosum volume were assessed using Cavalieri analysis . Stereoinvestigator was used to perform both of these techniques . Investigators were blinded to genotypes during sectioning , counting , and analysis . RNA from four WT and four Arid1b+/- hippocampi from 78 to 82 day old females were purified with a QIAGEN miRNeasy Mini Kit . NuGEN libraries were made with these RNAs . These indexed libraries were multiplexed in a single flow cell lane and received 75 base single-end sequencing on a NextSeq 500 using the High Output Kit v2 ( 75 cycles ) at the CRI Sequencing Facility . Raw sequencing reads were trimmed to remove adaptor and low quality sequences ( Phred score <20 ) using trim_galore package ( http://www . bioinformatics . babraham . ac . uk/projects/trim_galore/ ) . Trimmed reads were aligned to mouse reference genome GRCm38/mm10 with HiSAT2 Kim et al . , 2015 . After duplicates removal by SAMtools ( Li et al . , 2009 ) and Picard ( http://broadinstitute . github . io/picard . ) , read counts were generated for the annotated genes based GENCODE V20 ( Harrow et al . , 2012 ) using featureCounts ( Liao et al . , 2014 ) . Differential gene analysis was performed use edgeR ( Robinson et al . , 2010 ) , using FDR < 0 . 05 as cutoff . Enriched pathways were analyzed through the use of QIAGEN’s Ingenuity Pathway Analysis ( IPA , QIAGEN RedwoodCity , www . qiagen . com/ingenuity ) . Heatmaps to visualize the data were generated by using GENE-E ( www . broadinstitute . org/cancer/software/GENE-E ) . RNA-Seq data is deposited to GEO database and can be accessed through GEO accession number ( GSE92238 ) . Brg1 ChIP-seq data from the forebrain was downloaded from GEO with the accession number GSM912547 in GSE37151 . Files were remapped to the mm10 genome build by CrossMap . Brg1 target genes were predicted by using BETA-minus program on the Cistrome Analysis Pipeline , an integrative platform for transcriptional regulation studies . Heatmap and Metaplot were generated using deeptools , a flexible platform for exploring deep-sequencing data . Metabolic cage studies were performed in a temperature-controlled room containing 36 TSE metabolic cages ( The TSE Labmaster System of Germany ) maintained by UTSW Metabolic Core . Three days prior to study , mice were introduced to metabolic cages and after three day acclimation , cages were connected to TSE system and parameters were recorded for a total of five days . Investigators were blinded to mouse genotypes . Muscle strength was measured by a grip strength test performed by the Neuro-Models Core Facility at UT Southwestern Medical Center in a blinded fashion . Test was conducted using a wire mesh grid connected to a horizontally-aligned force meter ( San Diego Instruments , San Diego , CA ) . The grid was secured at a 45 degree angle , and the top rung of the grid was used for all testing . Mice were held at the base of the tail and supported ventrally while being moved into position to grasp the wire grid . Once the rug was successfully grasped , mice were gently pulled in a horizontal plane until the animal’s grip was released from the grid . Peak force ( in gram-force units , gf ) was captured by the force meter and recorded for later analysis . Forelimb and hindlimb tests were conducted separately , with each being measured five times over a 2–3 min period . Investigators were blinded to mouse genotypes and the identity of treatment groups . Arid1b+/- mice with apparent hydrocephaly were excluded . Experiment was performed once . Ultrasonic vocalizations ( USVs ) were recorded from both male and female pups isolated from their mothers at P4 , during the daylight period of the light/dark cycle . Dams and their litters were acclimated for 30 min in the test room . Each pup was removed from the cage containing its mother and littermates and placed in a clean plastic container in a wooden sound-attenuating recording chamber . Each pup was first acclimated in the recording chamber for 30 s then recorded for 10 min . Recordings were acquired using an UltraSoundGate CM16/CMPA condenser microphone ( Avisoft Bioacoustics ) positioned at a fixed height of 8 cm above the pups , and were amplified and digitized ( sampled at 16 bits , 250 kHz ) using UltraSoundGate 416 hr 1 . 1 hardware and Avisoft-RECORDER software ( Avisoft Bioacoustics ) . The data were transferred to Avisoft-SASLab Pro ( version 5 . 2 ) to analyze spectrograms of vocalizations with settings of 0% overlapping FlatTop windows , 100% frame size , and 256 points fast Fourier transform ( FFT ) length . The following measures were recorded for each group: number of USV calls , mean duration of USV calls , and mean peak frequency . Recordings were performed with the experimenter blinded to mouse genotypes . The adult male test mouse was placed into a fresh home cage and habituated in the test room with red light for 15 min before testing . A three week old male juvenile mouse was placed into the opposite end of the cage that the test mouse was already in . Active interactions between the mice were scored manually with 2 min of total test time . Only interactions when the test mouse is interacting with the juvenile , but not other way around were scored . Non-strict male littermates were used . Arid1b+/- mice with obvious hydrocephaly were excluded . The experiment was performed twice . The test was performed between 10:00 am and 2:00 pm . Adult female mouse was singly placed into a new standard cage , without nestlets , food , or water , acclimated for 10 min , then videotaped for another 10 min . The amount of time spent grooming was recorded continuously to calculate the total time spent grooming . Grooming is considered self-grooming of any part of the body ( including the face , head , ears , full-body ) . Data is plotted as percent of total time spent grooming . The test mouse was acclimated for 30 min in the testing room . One standard housing cage for each test mouse was filled with clean bedding material . 15 clean marbles were arranged on top of the bedding in each cage , forming five even rows and three columns . Mice were placed individually into the prepared cages and kept undisturbed for 30 min . After the testing period , they were returned to their original cages . A still image of the test cage was taken to record the number and pattern of buried marbles . A marble was considered buried if more than 2/3 of its depth is covered . Results were calculated and plotted as the percentage of marbles buried per genotype . This test was performed by UTSW Rodent Core Facilty . Experimenters were blinded to mouse genotypes . The experiment was repeated two times and combined data is included . Mice were placed individually into a clean , plastic mouse cage ( 18 cm x 28 cm ) with minimal bedding . Each cage was placed into a dark Plexiglas box . Movement was monitored by 5 photobeams in one dimension ( Photobeam Activity System , San Diego Instruments , San Diego , CA ) for 2 hr , with the number of beam breaks recorded every 5 min . The movement is characterized in three ways: repetitive beam breaks of a single beam is classified as stereotypy , consecutive beam breaks of two or more beams is classified as ambulatory movements , and total beam breaks during each 5 min interval . Number of total beam breaks during 5 min interval was reported . Arid1b+/- mice with apparent hydrocephaly were excluded . The test was performed by UTSW Rodent Core Facilty . Experimenters were blinded to mouse genotypes . Experiment was repeated two times and combined data is included . Mice were placed in the periphery of a novel open field environment ( 44 cm x 44 cm , walls 30 cm high ) in a dimly lit room and allowed to explore for 5 min . The animals were monitored from above by a video camera connected to a computer running video tracking software ( Ethovision 3 . 0 , Noldus , Leesburg , Virginia ) to determine the time , distance moved and number of entries into three areas: the periphery ( 5 cm from the walls ) , the center ( 14 cm x 14 cm ) and non-periphery ( area excluding periphery ) . The open field arenas were wiped and allowed to dry between mice . Locomotor activity test was performed prior to open field activity test in these cohorts . Arid1b+/- mice with apparent hydrocephaly were excluded . Test was performed by UTSW Rodent Core Facilty . Experimenters were blinded to mouse genotypes . Experiment was repeated two times and combined data is included . Mice were placed in the center of a black Plexiglas elevated plus maze ( each arm 30 cm long and 5 cm wide with two opposite arms closed by 25 cm high walls ) elevated 31 cm in a dimly lit room and allowed to explore for 5 min . The animals were monitored from above by a video camera connected to a computer running video tracking software ( Ethovision 3 . 0 , Noldus , Leesburg , Virginia ) to determine time spent in the open and closed arms , time spent in the middle , and the number of entries into the open and closed arm . The apparatus was wiped and allowed to dry between mice . Locomotor activity and open field activity test were performed prior to elevated plus maze in these cohorts . Arid1b+/- mice with apparent hydrocephaly were excluded . Test was performed by UTSW Rodent Core Facilty . Experimenters were blinded to mouse genotypes . Experiment was repeated two times and combined data is included . Mice were placed into a black Plexiglas chamber ( 25 cm x 26 cm ) and allowed to explore for 2 min . After the habituation period , a small door was opened allowing them to access the light side of the apparatus ( 25 cm x 26 cm lit to approximately 1700 lux ) for 10 min . The animals were monitored by 7 photobeams in the dark compartment and 8 photobeams on the light side connected to a computer which recorded the time spent in each compartment , latency to enter the light side and the number of entrances to each compartment ( Med-PC IV , Med Associates , St . Albans , VT ) . The dark-light apparatus was wiped and allowed to dry between mice . Locomotor activity , open field activity , and elevated plus maze tests were performed prior to dark-light activity test in these cohorts . Arid1b+/- mice with apparent hydrocephaly were excluded . Test was performed by UTSW Rodent Core Facilty . Experimenters were blinded to mouse genotypes . A circular pool was filled with room temperature water to a depth of approximately 12 inches . A platform ( 10 cm diameter ) was placed in one quadrant of the pool with the top of the platform about 2 cm below the water level . White non-toxic paint was added to enhance the contrast with the animal and to hide the location of the platform . Each day the mice were placed in the pool and allowed to swim for 1 min to find the platform . The swim path and time until locating the platform wasrecorded via a videocamera and computer running videotracking software ( Ethovision , Noldus ) . If the mouse does not find the platform within a minute , they were gently guided or placed on the platform for 10 s , then removed from the pool and return to their home cage . Each animal was placed in the pool for a total of four times each day for 13 days . Twenty-four hours after the last training day , a probe test was conducted in which the platform was removed from the pool and each mouse was allowed to swim for 1 min to determine whether the animal has learned the location of the platform . The time each animal spends in the quadrant which had contained the platform on training days and the number of times that the animal crosses the location which had contained the platform are indicators of how well the animal has learned the spatial location of the platform . To control for visual problems , the mice weregiven 4–6 trials after the probe test using the same pool and platform , however a large black block was placed on top of the platform to clearly mark the location . The location of the platform was moved on each trial . Test was performed by UTSW Rodent Core Facilty . Experimenters were blinded to mouse genotypes . Fear conditioning was measured in boxes equipped with a metal grid floor connected to a scrambled shock generator ( Med Associates Inc . , St . Albans ) . For training , mice were individually placed in the chamber . After 2 min , the mice received three tone-shock pairings ( 30 s white noise , 80 dB tone co-terminated with a 2 s , 0 . 5 mA footshock , 1 min intertrial interval ) . The following day , memory of the context was measured by placing the mice into the same chambers and freezing was measured automatically by the Med Associates software . Forty-eight hours after training , memory for the white noise cue was measured by placing the mice in a box with altered floors and walls , different lighting , and a vanilla smell . Freezing was measured for 3 min , then the noise cue was turned on for an additional 3 min and freezing was measured . Test was performed by the UTSW Rodent Core Facility . Experimenters were blinded to mouse genotypes . The mice were placed individually into boxes equipped with a metal grid floor connected to a scrambled shock generator ( Med Associates Inc . , St . Albans ) . After approximately 1 min , the mice received a series of footshocks ( 2 s each ) with increasing intensity . The initial shock intensity was 0 . 05 mA and the amplitude was increased by 0 . 05 mA for each consecutive footshock with 15 s intershock interval . The first shock intensity that each animal displayed each behaviour ( flinch , jump , and vocalization ) was recorded . Once the animal displayed all three behaviours , it was removed from the chamber . Plasma IGF-1 concentration was determined using mouse/rat IGF-1 Quantikine ELISA Kit ( R and D Biosystems , Cat #: MG100 ) without fasting . Mice were fasted for 36 hr before GH stimulation testing . Fifteen minutes after anesthesia with pentobarbital ( 50 mg/kg given once i . p . ) , 0 . 14 g/kg GHRH ( Phoenix Pharmaceuticals , Cat #:031–02 ) was injected i . p . Blood was sampled retro-orbitally using a capillary tube before , 5 , and 15 min after injection . Plasma GH concentration was measured using a Rat/Mouse GH ELISA KIT ( Millipore , Cat #: EZRMGH-45K ) . See Figure 5C . Mice at the age of P10 were ranked from highest to lowest body weight and even numbered mice were placed into the treatment group and odd numbered mice were placed into the vehicle group . Arid1b+/- mice with apparent hydrocephaly were excluded . Recombinant human IGF1 was purchased from Peprotech ( Catalog:#100–11 ) . It was prepared according to datasheet and injected intraperitoneally . Vehicle or 0 . 5 mg/kg rhIGF1 dissolved in vehicle were injected starting from P11 daily . Mice at the age of P11 were ranked from highest to lowest body weight and even numbered mice were placed into the treatment group and odd numbered mice were placed into the vehicle group . Arid1b+/- mice with apparent hydrocephaly were excluded . Recombinant mouse growth hormone was obtained from National Hormone and Peptide Program ( NHPP ) . It was prepared according to the NHPP datasheet . Injection was performed subcutaneously . 30 ug GH/mouse/day was injected between P11 to P14 , 50 ug GH/mouse/day was injected between P14 to P21 , 70 ug GH/mouse/day was injected between P21 to P50 . Grip strength test was performed at P60 after 10 days without treatment . Unless specified otherwise in the figure legends , statistical analyses were performed using unpaired , two-tailed , Student’s t-test or ANOVA . The data bars and error bars indicate mean ± standard error mean ( SEM ) . *p-value ≤ 0 . 05; **p-value ≤ 0 . 01; ***p-value ≤ 0 . 001; ****p-value ≤ 0 . 0001; ns , not significant . No statistical methods were used to predetermine sample sizes; however , sample sizes were estimated based on similar experiments reported in the relevant literature in the field ( Araujo et al . , 2015 and Katayama et al . , 2016 ) .

DNA does not just float freely inside our cells . Instead , it is wound around proteins called histones and packaged tidily into a form called chromatin . This packaging allows genes to be switched on or off by making it easier or harder to access different stretches of the genetic code . A group of proteins called the SWI/SNF chromatin-remodeling complex are responsible for the packing and unpacking of DNA during development , dictating the fate of thousands of genes . Mutations that affect one component of this complex , a protein known ARID1B , are associated with a rare genetic condition called Coffin-Siris syndrome , and may also have a role to play in autism spectrum disorders and intellectual disability . However , there were previously no animal models that can be used to study this mutation in the laboratory . Celen , Chuang et al . have now genetically modified mice to remove one of their two copies of the gene that encodes the mouse equivalent of ARID1B . This change replicates the mutation that is most commonly seen in people with Coffin-Siris syndrome . Celen , Chuang et al . report that the mutant mice with just one working copy of the gene showed many features also seen in Coffin-Siris syndrome , including a smaller size and weaker muscles . The mutant mice also repeated certain behaviors , like grooming themselves , and showed unusual interactions with other mice . Further tests showed that the mutant mice had lower than expected levels of growth hormone in their blood . The mice were then treated with growth hormone supplements to find out if this could reverse any of their symptoms . Indeed , this treatment made the mice larger and stronger , but did not change their behavior . Some doctors are already treating people with Coffin-Siris syndrome with growth hormone , and these new findings suggest that this treatment counteracts defects caused directly by the mutation affecting ARID1B . Moreover , this mouse model will allow the role of ARID1B to be investigated further in the laboratory , and could be used as a tool to discover , develop and test new treatments for Coffin-Siris syndrome .

2017

Arid1b haploinsufficient mice reveal neuropsychiatric phenotypes and reversible causes of growth impairment

Glucagon secretion dysregulation in diabetes fosters hyperglycemia . Recent studies report that mice lacking glucagon receptor ( Gcgr-/- ) do not develop diabetes following streptozotocin ( STZ ) -mediated ablation of insulin-producing β-cells . Here , we show that diabetes prevention in STZ-treated Gcgr-/- animals requires remnant insulin action originating from spared residual β-cells: these mice indeed became hyperglycemic after insulin receptor blockade . Accordingly , Gcgr-/- mice developed hyperglycemia after induction of a more complete , diphtheria toxin ( DT ) -induced β-cell loss , a situation of near-absolute insulin deficiency similar to type 1 diabetes . In addition , glucagon deficiency did not impair the natural capacity of α-cells to reprogram into insulin production after extreme β-cell loss . α-to-β-cell conversion was improved in Gcgr-/- mice as a consequence of α-cell hyperplasia . Collectively , these results indicate that glucagon antagonism could i ) be a useful adjuvant therapy in diabetes only when residual insulin action persists , and ii ) help devising future β-cell regeneration therapies relying upon α-cell reprogramming . Glucagon , a 29-amino acid-long hormone synthetized in pancreatic α-cells through cleavage of its precursor , proglucagon , by prohormone convertase 2 ( PC2 ) , counterbalances the effects of insulin on blood glucose homeostasis by stimulating hepatic glycogenolysis and gluconeogenesis ( Gromada et al . , 2007 ) . In addition , the two hormones act in a paracrine fashion to reciprocally regulate α- and β-cell function ( Unger and Orci , 2010 ) . Hypersecretion of glucagon in diabetes exacerbates hepatic glucose output , thereby fostering hyperglycemia and ketogenesis ( Unger et al . , 1970; Unger , 1971; Sherwin et al . , 1976; D'Alessio , 2011 ) . In consequence , antagonists of glucagon signaling are currently being tested in clinical trials for diabetes ( Campbell and Drucker , 2015 ) . The importance of glucagon signaling in diabetes was recently highlighted in studies performed with glucagon receptor knockout ( Gcgr-/- ) mice and in animals lacking α-cells due to pancreatic aristaless-related homeobox ( Arx ) deficiency . Surprisingly , these animals did not exhibit the usual signs of diabetes , such as hyperglycemia or glucose intolerance , after streptozotocin ( STZ ) -mediated β-cell destruction ( Conarello et al . , 2006; Lee et al . , 2011; 2012; Hancock et al . , 2010 ) . These findings lead to hypothesize that glucagon is responsible for the features of diabetes ( Unger and Cherrington , 2012 ) . Although suppression of glucagon action is likely to attenuate the consequences of insulin deficiency , its primary role in the hyperglycemia is uncertain . Indeed , because STZ causes an incomplete β-cell ablation due to variations in administration protocols and in genetic background-dependent sensitivity ( Deeds et al . , 2011; Cardinal et al . , 1998; Gurley , 2006 ) , it is possible that the “diabetes resistance” phenotype of Gcgr-/- mice relies on the action of insulin from residual β-cells . Thus , to determine whether lack of glucagon signaling would also prevent hyperglycemia and diabetes in the context of a more severe insulin deficiency , we used a transgenic model of diphtheria toxin ( DT ) -mediated β-cell ablation , termed RIP-DTR , which leads to an almost complete β-cell elimination ( Thorel et al . , 2010; Chera et al . , 2014 ) . Also , because adult RIP-DTR mice spontaneously reconstitute new insulin-producing cells by α-cell transdifferentiation in this condition of severe insulin insufficiency , we explored whether the compensatory α-cell hyperplasia due to glucagon signaling blockade ( Furuta et al . , 1997; Gelling et al . , 2003; Longuet et al . , 2013 ) influences the reprogramming of α-cells toward insulin production . Here we show that near-total β-cell loss triggers severe hyperglycemia and all the metabolic features of type 1 diabetes ( cachexia , glucose intolerance , and death ) in mice with constitutive or induced glucagon signaling deficiency . We report that the absence of hyperglycemia observed in glucagon-deficient mice after STZ treatment can be explained through the persistence of a residual β-cell mass , which ensures a low level of insulin action . Recent reports indicate that Gcgr-/- mice do not develop hyperglycemia after STZ-mediated β-cell loss . Here we aimed at determining the effect of the absence of glucagon action in the context of a more extreme insulin deficiency . For this purpose , we crossed Gcgr-/- mutant animals ( Gelling et al . , 2003 ) with RIP-DTR mice , in which diphtheria toxin ( DT ) injection triggers the near-total ( >99% ) β-cell loss ( Thorel et al . , 2010 ) . RIP-DTR;Gcgr-/- mice , like Gcgr-/- mice , displayed lower basal glucose levels than controls ( RIP-DTR;Gcgr+/+ and RIP-DTR;Gcgr+/-; not shown ) ( Gelling et al . , 2003 ) . Upon DT-induced β-cell ablation , both control and knockout animals developed severe hyperglycemia , with a slower kinetics in RIP-DTR;Gcgr-/- mice ( Figure 1A ) . Animals of both groups lost weight at similar rates ( Figure 1B ) , and died in absence of exogenous insulin treatment ( Figure 1C ) . By contrast , administration of long-acting insulin , although insufficient to normalize blood glucose levels , permitted survival and body weight maintenance ( Figure 1—figure supplement 1 ) . As soon as insulin treatment was discontinued , blood glucose levels and body weight quickly deteriorated in all groups . Altogether , these findings indicate that Gcgr-/- mice are not protected against hyperglycemia after near-total β-cell loss , but develop classical signs of type 1 diabetes and require insulin therapy . 10 . 7554/eLife . 13828 . 003Figure 1 . Gcgr-/- mice become diabetic after massive β-cell ablation . ( A ) Random-fed glycemia ( left ) and area under the glycemia curve ( AUC ) between days 0 and 7 after DT ( right ) in untreated ( Untr . ) and DT-treated RIP-DTR;Gcgr+/- and RIP-DTR;Gcgr-/- females . ( B ) Body weight ( left ) and AUC body weight ( days 0–7 after DT; right ) . † , all mice of the group were dead at this time point ( see Figure 1C ) . *p<0 . 05; **p<0 . 01; Mann-Whitney U test . C: Survival curve of RIP-DTR;Gcgr+/- and RIP-DTR;Gcgr-/- mice after DT treatment ( N=5–6 ) . Survival analysis of DT-treated animals ( Gcgr+/- versus Gcgr-/- ) : p=0 . 044; Log-rank test . DOI: http://dx . doi . org/10 . 7554/eLife . 13828 . 00310 . 7554/eLife . 13828 . 004Figure 1—figure supplement 1 . Insulin administration stabilizes body weight and allows survival of DT-treated Gcgr-/-mice . Glycemia ( left ) and body weight ( right ) of RIP-DTR;Gcgr+/+ ( blue triangles , N=7 ) , RIP-DTR;Gcgr+/- ( black squares , N=9 ) , and RIP-DTR;Gcgr-/- ( red circles , N=9 ) males following DT-mediated β-cell ablation and exogenous insulin treatment . Grey areas indicate the period during which mice were treated with insulin detemir ( 5 U/kg/day between days 6 and 25 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13828 . 004 Constitutive Gcgr deletion leads to increased embryonic lethality , and defects in pancreatic development and islet-cell maturation ( Vuguin et al . , 2006; Vuguin and Charron , 2011; Ouhilal et al . , 2012 ) . Since these abnormalities may encompass long-lasting compensatory metabolic adaptations , we conditionally inhibited glucagon action in adult mice that had developed normally using a glucagon receptor antagonizing monoclonal antibody ( anti-GCGR mAb ) . We first assessed its activity in C57BL/6 wild type mice ( Figure 2—figure supplement 1A ) . In agreement with a previously described antibody ( Gu et al . , 2009; Yan et al . , 2009 ) , anti-GCGR treatment led to a reduction in basal glycemia ( Figure 2—figure supplement 1B ) , and triggered α-cell hyperplasia and hypertrophy , as observed in Gcgr-/- animals ( Figure 2—figure supplement 1C–D ) ( Gelling et al . , 2003 ) . In addition , antibody-treated Gcgr+/+ mice showed altered responses , like Gcgr-/- animals , to intraperitoneal glucose and insulin tolerance tests ( Figure 2—figure supplement 1E–F ) . Anti-GCGR administration in Gcgr+/+ mice therefore phenocopies the main metabolic and cellular alterations of Gcgr-/- mice and thus represents a valuable tool for inducing glucagon signaling antagonism in vivo . To assess whether induced glucagon receptor blockade prevents diabetes upon near-total β-cell ablation , we pre-treated adult RIP-DTR mice with the anti-GCGR mAb for 3 weeks , and then injected them with DT ( Figure 2A ) . In agreement with the above results using RIP-DTR;Gcgr-/- animals , all mice became severely hyperglycemic and lost weight after DT , regardless of antibody treatment ( Figure 2B–C ) . Moreover , only insulin administration allowed for survival following β-cell ablation , not glucagon receptor inhibition ( Figure 2—figure supplement 2 ) . Collectively , these observations indicate that the lack of glucagon signaling is not sufficient per se to prevent severe hyperglycemia and diabetes following extreme β-cell loss , and contrast with previous studies in which Gcgr-/- , or anti-GCGR-treated mice did not develop the metabolic manifestations of the disease when β-cell ablation was mediated by STZ ( Conarello et al . , 2006; Lee et al . , 2011; 2012; Wang et al . , 2015 ) . 10 . 7554/eLife . 13828 . 005Figure 2 . Anti-GCGR mAb-treated mice become diabetic after massive β-cell ablation . ( A ) Experimental design . ( B-C ) Random-fed glycemia ( B ) and body weight ( C ) after DT in C57BL/6 males pre-treated with vehicle or mAb ( N=3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13828 . 00510 . 7554/eLife . 13828 . 006Figure 2—figure supplement 1 . Anti-GCGR mAb administration recapitulates the metabolic and cellular phenotypes of Gcgr-/- mice . ( A ) Experimental design . 9 mg/kg anti-GCGR mAb was injected i . p . 3 times per week for 3 weeks in C57BL/6 animals . ( B ) Left: Random fed glycemia of vehicle- ( black squares ) or mAb-treated males ( red triangles ) . The grey area indicates the period of antibody treatment . Right: Area under the glycemia curves . *p<0 . 05; Mann-Whitney U test . C and D . Confocal images of pancreatic islet sections from vehicle- ( C ) and mAb-treated ( D ) males . α-cell hyperplasia and hypertrophy ( compare C’ and D’ , the dashed lines represent the cell perimeters ) are observed in islets from mAb-treated mice . Scale bars: 20 μm . E and F . Intraperitoneal glucose tolerance test ( E ) and insulin tolerance test ( F ) performed in Gcgr+/+ ( black squares , N=9 ) , Gcgr-/- ( grey circles , N=10 ) , and mAb pre-treated Gcgr+/+ ( red triangles , N=10 ) males . *p<0 . 05; **p<0 . 01; ***p<0 . 001; Gcgr+/+ versus mAb-treated Gcgr+/+ mice . † , p<0 . 05; †† , p<0 . 01; ††† , p<0 . 001; Gcgr+/+ versus Gcgr-/- mice; two-way ANOVA . The difference between Gcgr-/- and mAb-treated Gcgr+/+ mice is not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 13828 . 00610 . 7554/eLife . 13828 . 007Figure 2—figure supplement 2 . Insulin administration is required to stabilize body weight and allow survival of anti-GCGR-treated mice after DT . ( A-C ) Exogenous insulin , but not anti-GCGR mAb treatment , stabilizes body weight and improves survival after extreme β-cell loss . ( A ) Experimental design . ( B ) Evolution of body weight following DT administration in RIP-DTR males treated with anti-GCGR mAb and/or exogenous insulin ( N=5–12 ) . ( C ) Survival curves . Survival analyses are indicated next to the legend: n . s . , not significant; **p<0 . 01; ***p<0 . 001; Log-rank test . Insulin was administered as subcutaneous implants ( antibody-untreated mice ) , or as injections of long-acting insulin ( antibody-treated mice ) because insulin implants lead to hypoglycemia and death in mice with deficient glucagon signaling . DOI: http://dx . doi . org/10 . 7554/eLife . 13828 . 007 The different impact of STZ and DT treatments on glycemia in Gcgr-/- mice may result from a difference in completeness of β-cell destruction . To test this hypothesis , we compared the relative ablation efficiencies of these two methods . To maximize β-cell destruction , we treated Gcgr+/- and Gcgr-/- mice with two high doses of STZ ( 200 and 150 mg/kg , one week apart ) . Following the first injection , control mice became severely hyperglycemic . By contrast , Gcgr-/- animals remained normoglycemic even after the second STZ injection , as previously reported ( not shown ) ( Lee et al . , 2011; 2012 ) . RIP-DTR;Gcgr-/- animals remained markedly hyperglucagonemic after STZ- or DT-mediated β-cell loss and α-cell mass was not affected ( Figure 3—figure supplement 1A–B ) . Histologically , we observed that nearly 90% of islet sections were totally devoid of β-cells after DT , versus only 45% after STZ ( Figure 3A ) . Accordingly , the β-cell mass and pancreatic insulin content were reduced by 98–99% after DT , but only by 70–80% after STZ ( Figure 3B–C ) . In addition , plasma insulin levels were just above detection threshold after DT , but readily detectable after STZ ( Figure 3D ) . We made similar observations in mice with normal glucagon signaling ( Figure 3—figure supplement 2 ) . Together , these results indicate that β-cell destruction is more complete after DT- than after STZ-treatment in Gcgr-/- mice . 10 . 7554/eLife . 13828 . 008Figure 3 . DT administration leads to a more complete β-cell ablation than STZ . ( A ) Islet sections stained for insulin ( red ) and glucagon ( green ) from untreated , STZ- , or DT-treated RIP-DTR;Gcgr-/- females , 6 days after the last STZ or DT injection . Scale bars: 20 μm . ( B-D ) β-cell mass ( B ) , pancreatic insulin content ( C ) and fed plasma insulin levels ( D ) in untreated ( Untr . ) , STZ- , or DT-treated RIP-DTR;Gcgr-/- males and females , 6 days after the last injection . STZ administration: two injections ( 200 and 150 mg/kg ) . *p<0 . 05; **p<0 . 01; Mann-Whitney U test . DOI: http://dx . doi . org/10 . 7554/eLife . 13828 . 00810 . 7554/eLife . 13828 . 009Figure 3—figure supplement 1 . RIP-DTR;Gcgr-/- mice remain hyperglucagonemic and α-cell mass is not affected after STZ- or DT-treatment . ( A-B ) fed plasma glucagon levels ( A ) and α-cell mass ( B ) in untreated ( Untr . ) , STZ- , or DT-treated RIP-DTR;Gcgr+/- and RIP-DTR;Gcgr-/- males and females , measured 6 days after the last injection . STZ administration: two injections ( 200 and 150 mg/kg ) . **p<0 . 01; Mann-Whitney U test . DOI: http://dx . doi . org/10 . 7554/eLife . 13828 . 00910 . 7554/eLife . 13828 . 010Figure 3—figure supplement 2 . Higher efficiency of β-cell ablation after DT- than after STZ-treatment in mice with normal glucagon signaling . ( A-B ) β-cell mass ( A ) and pancreatic insulin content ( B ) in untreated ( Untr . ) , STZ- , or DT-treated RIP-DTR;Gcgr+/- females , measured 6 days after the last injection . STZ administration: two injections ( 200 and 150 mg/kg ) . **p<0 . 01; Mann-Whitney U test . DOI: http://dx . doi . org/10 . 7554/eLife . 13828 . 010 Because β-cell ablation was incomplete after STZ , we aimed at determining whether the action of residual circulating insulin might , in combination with glucagon signaling deficiency , protect Gcgr-/- mice from diabetes . To test this hypothesis , we inhibited insulin action using the insulin receptor antagonist drug S961 ( Schäffer et al . , 2008 ) . In vivo , S961 administration induces hyperglycemia in wild type animals and closely recapitulates the phenotype of mice with liver-specific insulin receptor deletion ( Yi et al . , 2013; Michael et al . , 2000 ) . In agreement with its previously reported action , S961 administration in Gcgr+/- mice triggered a strong increase in glycemia ( Figure 4A; blue dashed vs black continuous line ) . Interestingly , Gcgr-/- animals exhibited a smaller but significant increase in glycemia , indicating that glucagon deficiency has a beneficial effect in this situation of relative insulin deficit ( purple dashed vs red continuous line ) . Although STZ-treated Gcgr-/- mice remained normoglycemic , as previously reported ( Conarello et al . , 2006; Lee et al . , 2011; 2012 ) , they developed severe hyperglycemia after insulin receptor inhibition ( continuous vs dotted purple line ) . This suggests that residual insulin action , likely originating from STZ-escaping β-cells , is still present after STZ administration in Gcgr-/- animals , and is necessary to prevent hyperglycemia and diabetes . 10 . 7554/eLife . 13828 . 011Figure 4 . Inhibition of insulin action triggers hyperglycemia in STZ-treated Gcgr-/-mice . ( A ) Random-fed glycemia after STZ and/or S961 administration in Gcgr+/- and Gcgr-/- females ( left ) , and area under the glycemia curve ( AUC ) during S961 treatment ( right ) . ( B-D ) Hepatic Pepck ( top ) and Glucokinase ( bottom ) mRNA levels relative to those of untreated Gcgr+/- ( control ) mice ( N=4–6 ) . ( B ) Glucagon deficiency: Gcgr-/- background . ( C ) Insulin deficiency: β-cell ablation or insulin signaling inhibition . ( D ) Combined deficiency: β-cell ablation and/or insulin signaling inhibition in a Gcgr-/- background . ( E-G ) FoxO1 mRNA levels in skeletal muscle , relative to those of untreated Gcgr+/- mice ( N=4–6 ) . STZ administration: 200 mg/kg at day 0 and 150 mg/kg at day 7 . S961 treatment: osmotic pump ( days 15 to 21 ) . *p<0 . 05; **p<0 . 01; Mann-Whitney U test . Only groups that exhibited a > twofold regulation as compared to controls ( dashed lines ) were tested . DOI: http://dx . doi . org/10 . 7554/eLife . 13828 . 01110 . 7554/eLife . 13828 . 012Figure 4—figure supplement 1 . Higher hepatic PEPCK protein expression after DT in both Gcgr+/- and Gcgr-/- mice . Western blot analysis showing PEPCK and Tubulin expression in the liver of untreated ( untr . ) and DT-treated RIP-DTR-Gcgr+/- and RIP-DTR-Gcgr-/- females ( left ) . Quantification of PEPCK band intensities relative to Tubulin is shown on the right . DOI: http://dx . doi . org/10 . 7554/eLife . 13828 . 01210 . 7554/eLife . 13828 . 013Figure 4—figure supplement 2 . Liver glycogen concentration is reduced after DT-treatment in both RIP-DTR-Gcgr+/- and RIP-DTR-Gcgr-/- mice . Liver glycogen concentration in different conditions of insulin and/or glucagon deficiency ( N=4 ) . *p<0 . 05; Mann-Whitney U test . DOI: http://dx . doi . org/10 . 7554/eLife . 13828 . 01310 . 7554/eLife . 13828 . 014Figure 4—figure supplement 3 . Expression of genes negatively regulated by insulin signaling in skeletal muscle . mRNA levels of genes inhibited by insulin in skeletal muscle ( gastrocnemius ) , relative to those of untreated Gcgr+/- females ( normalized to Actb , Gapdh , and Gusb ) ( N=4–6 ) . Irs2 , Insulin receptor substrate 2; Fbxo32 , F-box only protein 32 ( Atrogin-1 ) ; Trim63 , Tripartite motif-containing 63 ( MuRF1 ) ; 4e-bp1 , Eukaryotic translation initiation factor 4E binding protein 1 ( Eif4ebp1 ) ; Gadd45a , Growth arrest and DNA-damage-inducible 45 alpha; p21 , Cyclin-dependent kinase inhibitor 1A ( Cdkn1a ) . p27 , Cyclin-dependent kinase inhibitor 1B ( Cdkn1b ) . *p<0 . 05; **p<0 . 01; Mann-Whitney U test . Only groups that exhibited a > twofold regulation as compared to controls ( dashed lines ) were tested . DOI: http://dx . doi . org/10 . 7554/eLife . 13828 . 014 To better characterize the effect of insulin insufficiency in a glucagon-deficient context , we evaluated hepatic transcript levels of Phosphoenolpyruvate carboxykinase ( Pepck ) and Glucokinase ( Gck ) , two hormone-sensitive enzymes whose transcription is regulated by the relative levels of glucagon and insulin signaling ( Rucktäschel et al . , 2000; Chakravarty et al . , 2005; Iynedjian et al . , 1995 ) . Liver is a relevant organ to assess the impact of insulin and glucagon deficiency because re-expression of the glucagon receptor in the liver of STZ-treated Gcgr-/- mice , and conditional inactivation of the insulin receptor in hepatocytes are both sufficient to trigger hyperglycemia ( Lee et al . , 2012; Michael et al . , 2000 ) . In conditions of glucagon deficiency ( increased insulin/glucagon ratio; Gcgr-/- mice ) , we observed a decreased expression of the gluconeogenic enzyme Pepck and an increased expression of the glycolytic enzyme Gck as compared to Gcgr+/- controls ( Figure 4B ) , which is consistent with a previous study ( Yang et al . , 2011 ) . By contrast , upon induced insulin deficiency ( decreased insulin/glucagon ratio ) , as in STZ- , S961- , or DT-treated Gcgr+/- animals , Pepck and Gck exhibited the opposite regulation ( Figure 4C ) . We observed the strongest effect after DT , which caused a 1000-fold decrease in Gck expression , suggesting that it led to a more complete suppression of insulin action than STZ or S961 . When inducing insulin insufficiency in a Gcgr-/- background , a situation of combined insulin and glucagon deficiency , we observed Pepck and Gck mRNA levels similar to those measured in untreated Gcgr+/- control mice , except after DT , which induced a strong downregulation of Gck expression in Gcgr-/- livers ( Figure 4D ) . We also confirmed the increase in hepatic PEPCK expression after DT at the protein level ( Figure 4—figure supplement 1 ) . Similarly , DT- , but not STZ-treatment depleted liver glycogen stores in RIP-DTR;Gcgr-/- animals ( Figure 4—figure supplement 2 ) . These results suggest that lack of glucagon action can compensate for the effect of partial insulin insufficiency on the expression of rate-limiting enzymes and hepatic glycogen metabolism , but not after near-total β-cell loss , a situation where the effect of insulin deficiency outweighs that of glucagon deficiency . We then assessed insulin signaling activity in skeletal muscle by measuring the expression of the transcription factor Forkhead box protein O1 ( FoxO1 ) and of several of its target genes , such as Insulin receptor substrate 2 ( Irs2 ) , which are induced upon insulin insufficiency ( Long et al . , 2011 ) . FoxO1 mRNA levels were similar in untreated Gcgr+/- and Gcgr-/- mice ( Figure 4E ) . In Gcgr-/- animals , STZ or S961 administration did not significantly affect FoxO1 expression . By contrast , FoxO1 and its targets were strongly upregulated upon combined STZ and S961- , or DT-treatment , reflecting a more severe insulin insufficiency ( Figure 4G and Figure 4—figure supplement 3 ) . Together , these results indicate that lack of glucagon signaling efficiently compensates for the consequences of insulin insufficiency only if residual insulin action persists after β-cell loss . As Gcgr-/- mice exhibit resistance to STZ-induced hyperglycemia , we assessed the impact of glucagon signaling blockade on C57BL/6 mice made hyperglycemic with a single injection of either 175 or 225 mg/kg STZ . Once the animals were hyperglycemic , we implanted them with an osmotic pump containing the anti-GCGR mAb . In mice injected with 175 mg/kg STZ , antibody treatment strongly reduced , but did not completely normalize , blood glucose levels ( Figure 5A and B ) . By contrast , animals that had received 225 mg/kg STZ remained severely hyperglycemic ( >30 mM ) after anti-GCGR mAb administration . As expected , residual pancreatic insulin content negatively correlated with the dose of STZ ( Figure 5C ) . We thus observed beneficial effects of glucagon signaling inhibition only in diabetic mice that had retained a relatively higher pancreatic insulin after STZ-mediated β-cell loss . Strikingly , the impact of glucagon signaling inhibition on the glycemia of diabetic mice was dependent on very small measurable differences in residual pancreatic insulin , as seen after 175 and 225 mg/kg STZ ( respectively 1 . 79% and 0 . 45% of the pancreatic insulin content of non-ablated controls ) . As seen in Gcgr-/- animals , anti-GCGR mAb administration resulted in a lower expression of hepatic Pepck ( Figure 5—figure supplement 1 ) . In addition , the highest STZ dose triggered a stronger glucokinase downregulation than the 175 mg/kg dose in mAb-treated mice . 10 . 7554/eLife . 13828 . 015Figure 5 . Anti-GCGR mAb treatment does not normalize hyperglycemia after efficient STZ-mediated β-cell ablation . ( A ) Random-fed glycemia in C57BL/6 males treated with STZ ( single injection at day 0: 175 or 225 mg/kg ) and/or anti-GCGR mAb ( osmotic pump , days 6 to 14; N=3–6 ) . ( B ) Area under the glycemia curves during mAb treatment . ( C ) Pancreatic insulin content . *p<0 . 05; **p<0 . 01; Mann Whitney U test . DOI: http://dx . doi . org/10 . 7554/eLife . 13828 . 01510 . 7554/eLife . 13828 . 016Figure 5—figure supplement 1 . Hepatic Pepck and Glucokinase expression after STZ and/or anti-GCGR mAb treatment . Liver Pepck ( left ) and Glucokinase ( right ) mRNA levels in mice treated with STZ ( single injection at day 0: 175 or 225 mg/kg ) and/or anti-GCGR mAb ( osmotic pump , days 6 to 14 ) relative to those of untreated Gcgr+/- mice ( N=4–7 ) . *p<0 . 05; **p<0 . 01; Mann-Whitney U test . Only groups that exhibited a > twofold regulation as compared to controls ( dashed lines ) were tested . DOI: http://dx . doi . org/10 . 7554/eLife . 13828 . 016 Collectively , our findings support the notion that , regardless the method of β-cell ablation ( STZ or DT ) , the beneficial effects of inhibiting glucagon action , either genetically or pharmacologically , rely upon residual insulin action . We have previously shown that massive β-cell ablation triggers insulin expression in a small fraction of the α-cell population , with the appearance of glucagon/insulin bihormonal cells ( Thorel et al . , 2010 ) . We report above that in such a situation of near-total β-cell loss , lack of glucagon action fails to normalize glycemia . We then assessed whether the α-cell expansion triggered by glucagon signaling inhibition could have a beneficial effect on α-cell reprogramming . One month after DT-mediated β-cell ablation , we observed bihormonal cells in RIP-DTR;Gcgr+/+ and RIP-DTR;Gcgr-/- mice ( Figure 6A ) . Because RIP-DTR;Gcgr-/- animals have α-cell hyperplasia ( Gelling et al . , 2003; Longuet et al . , 2013 ) and the number of bihormonal cells was proportional to the number of α-cells in both groups ( Figure 6B ) , we observed a significant increase in the absolute number of bihormonal cells in RIP-DTR;Gcgr-/- mice ( Figure 6C ) . Consistent with these observations , they had a higher pancreatic insulin content ( Figure 6D ) . These results indicate that there is an increased number of α-cells engaged into reprogramming in mice lacking glucagon signaling . We also observed the appearance of bihormonal cells in DT-treated adult RIP-DTR mice undergoing anti-GCGR mAb treatment ( Figure 6E–F ) . We confirmed the α-cell origin of these newly formed bihormonal cells using a previously described tetracycline-activated system , which allows the specific and efficient doxycycline ( DOX ) -dependent irreversible tracing of α-cells with YFP ( Figure 6—figure supplement 1A–B ) ( Thorel et al . , 2010 ) . One month after DT injection in Gcgr-/- mice , we observed that a significant fraction of insulin-producing cells were also YFP-positive and therefore derived from cells that had previously expressed glucagon ( Figure 6—figure supplement 1C ) . We confirmed these observations in animals in which conditional GCGR inhibition was applied after DT-mediated β-cell ablation ( Figure 6—figure supplement 1D ) . 10 . 7554/eLife . 13828 . 017Figure 6 . Absence of glucagon signaling does not block the appearance of new glucagon-insulin bihormonal cells after β-cell ablation . ( A ) Islet sections exhibiting glucagon-insulin co-expressing cells ( arrowheads ) from RIP-DTR;Gcgr+/+ and RIP-DTR;Gcgr-/- females ( 1 m after DT ) . Scale bars: 20 μm . ( B-D ) Percentage of glucagon+ cells that co-express insulin ( B ) , bihormonal cells per islet section ( C ) , and pancreatic insulin content ( D ) in RIP-DTR;Gcgr+/+ and RIP-DTR;Gcgr-/- females ( 1 m after DT , N=5–6 ) . ( E-F ) Percentage of glucagon+ cells that co-express insulin ( E ) , and bihormonal cells per islet section ( F ) in vehicle- or anti-GCGR mAb- treated RIP-DTR males ( 2 weeks after DT , N=3 ) . *p<0 . 05; Mann-Whitney U test . DOI: http://dx . doi . org/10 . 7554/eLife . 13828 . 01710 . 7554/eLife . 13828 . 018Figure 6—figure supplement 1 . Newly formed bihormonal cells in Gcgr-/- mice are reprogrammed α-cells . ( A ) Transgenes required to irreversibly lineage-trace pancreatic α-cells with YFP before β-cell ablation . Inverted triangles represent loxP sites . ( B ) Experimental design . Upon DOX administration , the transgenic rtTA protein expressed in α-cells binds to the TetO promoter and activates Cre expression , which in turn recombines the STOP sequence in the R26-YFP transgene , leading to irreversible YFP expression . C and D: Example of YFP-traced cells that co-express insulin , as observed after β-cell ablation in a RIP-DTR;Gcg-rtTA;TeTO-Cre;R26-YFP;Gcgr-/- female ( C ) or in an anti-GCGR mAb-treated RIP-DTR;Gcg-rtTA;TetO-Cre;R26-YFP male ( D ) . Higher magnification of the dotted areas is shown on the right side of panels C and D . YFP was detected using an anti-GFP antibody . Scale bars: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 13828 . 018 Together , these findings indicate that although glucagon signaling blockade does not prevent hyperglycemia in diabetic mice that exhibit extreme insulin deficiency , it results in enhanced formation of new insulin-producing cells by increasing the absolute number of converting α-cells . Glucagon receptor inhibition decreases hyperglycemia in various animal models of diabetes ( Gu et al . , 2009; Johnson et al . , 1982; Brand et al . , 1994; Sloop et al . , 2004; Mu et al . , 2011; Sorensen et al . , 2006 ) , as well as in patients with type 2 diabetes ( Kelly et al . , 2015 ) . The extent of these benefits remains however disputed in situations where the β-cell population is nearly completely depleted , as in long-standing type 1 diabetes ( Wang et al . , 2012; Meier et al . , 2005 ) . Previous studies have shown that STZ-mediated β-cell ablation does not induce diabetes in the Gcgr-/- mouse model ( Conarello et al . , 2006; Lee et al . , 2011; 2012 ) , giving rise to the hypothesis that mice cannot develop hyperglycemia in absence of glucagon action ( Unger and Cherrington , 2012 ) . Here , we show that Gcgr-/- and anti-GCGR mAb-treated animals develop severe hyperglycemia after massive DT-mediated β-cell ablation ( Figures 1 and 2 ) . Our results suggest that the disparity in blood glucose levels observed between STZ- and DT-treated Gcgr-/- animals originate from a difference in β-cell destruction efficiency ( Figure 3 ) . Recent studies reached conflicting conclusions regarding the beneficial effect of glucagon signaling blockade in severely diabetic mice: Wang et al reported that anti-GCGR mAb treatment was sufficient to normalize glycemia of STZ-treated BALB/c animals ( Wang et al . , 2015 ) , whereas Steenberg et al did not observe improvements in glucose tolerance after GCGR antagonism or glucagon immunoneutralisation in C57BL/6 mice ( Steenberg et al . , 2016 ) . These discrepancies may be explained by differences in completeness of β-cell ablation linked to the protocol of injection ( single high dose versus multiple low doses ) and/or to strain-dependent sensitivity; it was indeed reported that BALB/c mice are less sensitive to STZ than C57BL/6 animals ( Cardinal et al . , 1998; Gurley , 2006 ) . Here , we injected C57BL/6 mice with two different high doses of STZ that triggered a severe hyperglycemia; after anti-GCGR mAb treatment , however , we observed a decrease in glycemia only in animals treated with the lowest STZ dose . These results indicate that a small difference in pancreatic insulin , such as that observed after 175 and 225 mg/kg STZ , can cause a major difference in glycemia in animals lacking glucagon signaling , thereby highlighting the importance of residual insulin action and providing a potential explanation for discrepancies between previous studies ( Figure 5 ) . Remarkably , STZ-treated Gcgr-/- mice became hyperglycemic upon S961-mediated insulin receptor antagonism , illustrating the requirement of residual insulin action for maintenance of normoglycemia in these animals ( Figure 4 ) . Collectively , these findings demonstrate that a total absence of glucagon action is not sufficient to prevent hyperglycemia in case of severe insulin deficiency . Although Gcgr-/- mice developed diabetes upon massive β-cell ablation , lack of glucagon action reduced or normalized glycemia in conditions of less severe insulin deficiency . In particular we observed that i ) anti-GCGR mAb administration reduced hyperglycemia in C57BL/6 mice treated with the lowest STZ dose ( Figure 5 ) and ii ) S961 treatment caused a less severe increase in glycemia in Gcgr-/- than in Gcgr+/- animals ( Figure 4A ) . Our data on hepatic expression of Pepck and Gck , two rate-limiting enzymes of gluconeogenesis and glycolysis , respectively , suggest that lack of glucagon signaling counterbalances the effects of insulin insufficiency after STZ or S961 . This would prevent , or limit , the rise in net hepatic glucose output by decreasing gluconeogenesis and glycogenolysis , and by increasing glycolysis and glycogenesis . Absence of glucagon action is however not sufficient to compensate severe insulin deficiency after DT , as reflected by Gck downregulation and reduced hepatic glycogen content , thereby contributing to the elevation of blood glucose . Interestingly , the mRNA levels of FoxO1 target genes in skeletal muscle were strongly upregulated , reflecting insulin signaling insufficiency , after DT and STZ+S961 , the two conditions that led to hyperglycemia ( Figure 4 and Figure 4—figure supplement 3 ) . In addition , gonadal adipose tissue was markedly depleted in experimental conditions leading to hyperglycemia ( in Gcgr-/- mice after STZ+S691 and DT; in Gcgr+/- mice after STZ and DT; not shown ) . Together , these findings provide new insights into the mechanisms by which lack of glucagon signaling protects against elevated blood glucose levels in situations of insulin insufficiency . Recent studies have shown that protection against STZ-mediated hyperglycemia also rely on the high levels of circulating glucagon-like peptide-1 ( GLP-1 ) in Gcgr-/- animals ( Gu et al . , 2010; Ali et al . , 2011; Jun , 2014; Omar et al . , 2014 ) . Yet , these high levels of GLP-1 combined with a lack of glucagon action were insufficient to maintain normoglycemia after near-total β-cell loss . Finally , we report here that lack of glucagon signaling does not compromise the ability of α-cells to convert to insulin production after DT-mediated near-total β-cell loss . Indeed , YFP-traced α-cells become glucagon/insulin bihormonal cells after DT in RIP-DTR;Gcgr-/- mice and in animals treated with the anti-GCGR antibody ( Figure 6 and Figure 6—figure supplement 1 ) . The proportion of α-cells co-expressing insulin after DT is comparable between mice with either intact , reduced or absent glucagon signaling , indicating that glucagon does not play an essential role in the α-to-β transdifferentiation process . Interestingly , because glucagon signaling inhibition leads to a compensatory α-cell hyperplasia ( Furuta et al . , 1997; Gelling et al . , 2003; Longuet et al . , 2013 ) , the absolute number of newly formed insulin-producing cells through α-cell conversion was augmented in RIP-DTR;Gcgr-/- mice . As previously described in adult mice ( Chera et al . , 2014 ) , we also observed the δ-to-β conversion in β-cell-ablated RIP-DTR;Gcgr-/- mice ( not shown ) . In conclusion , although inhibition of glucagon action alone is insufficient to prevent diabetes in conditions of near-total insulin deficiency , it is beneficial when residual insulin action persists , as in STZ-treated Gcgr-/- animals . Combination of glucagon inhibition with insulin therapy may however increase the risk of hypoglycemia . We encountered this problem when using subcutaneous insulin pellets in DT-treated RIP-DTR;Gcgr-/- mice: they became hypoglycemic and died likely as a consequence of the constitutive insulin release from the pellets , which could not be compensated by glucagon action . Our findings suggest that diabetes therapy through glucagon suppression would be unsafe if exogenous insulin has to be supplemented , but may be beneficial in patients with sufficient residual insulin action . In case of near-total insulin deficiency , transient glucagon receptor blockade could also serve as a means to increase the α-cell mass before triggering insulin production in these cells , a strategy that might be envisioned as a novel therapy to treat diabetes . Gcgr-/- ( Gelling et al . , 2003 ) , RIP-DTR ( Rat insulin promoter - diphtheria toxin receptor ) ( Thorel et al . , 2010 ) , Gcg-rtTA ( Glucagon promoter - reverse tetracycline transactivator ) ( Thorel et al . , 2010 ) , TetO-Cre ( Tetracycline operator - Cre recombinase ) ( Perl et al . , 2002 ) , and R26-YFP ( Rosa26 promoter - yellow fluorescent protein ) ( Srinivas et al . , 2001 ) mice were described previously and bred on a C57BL/6-enriched mixed genetic background . As pups born from Gcgr-/- mothers die perinatally ( Vuguin et al . , 2006 ) , Gcgr+/- females were used for breeding . C57BL/6 mice were purchased from Janvier Labs ( France ) . All mice used in this study were adult ( 10–20 week old ) males or females . They were housed and treated in accordance with the guidelines and regulations of the Direction Générale de la Santé , state of Geneva . Blood glucose was measured from tail blood using a handheld glucometer ( detection range: 0 . 6 to 33 . 3 mM , values exceeding 33 . 3 mM were artificially set to 34 mM ) . For β-cell ablation in RIP-DTR mice , DT ( D0564 , Sigma , St . Louis , MO ) was injected i . p . in 3 injections of 125 ng each , at days 0 , 3 , and 4 . STZ ( S0130 , Sigma ) was used as an alternative method of β-cell ablation . It was freshly diluted in citrate buffer and administered in 5-h fasted mice . Two different protocols were used depending on the genetic background: i ) Gcgr+/- and Gcgr-/- mice: two i . p . injections of 200 and 150 mg/kg , one week apart; ii ) C57BL/6 mice: single i . p . injection ( 175 or 225 mg/kg ) . For inducible α-cell labeling in Gcg-rtTA;TetO-Cre;R26-YFP mice , DOX ( D9891 , Sigma ) was added to drinking water ( 1 mg/ml ) for 2 weeks followed by at least 2 weeks of clearance before DT injection . Anti-GCGR monoclonal antibody A-9 was generated at Eli Lilly and Company ( Yan H , Hu S-FS , Boone TC , Lindberg RA , inventors; Amgen Inc . , assignee . Compositions and methods relating to glucagon receptor antibodies . United States patent US 8158759 B2 , 2012 Apr 17 ) . It was delivered either via i . p . injections , thrice weekly ( 9 mg/kg per injection ) , or using a s . c . implanted osmotic pump ( model 2002 , Alzet , Cupertino , CA ) containing 11 mg/ml of anti-GCGR mAb in PBS ( estimated delivery rate: 5 . 5 μg/h for 2 weeks ) . The insulin receptor inhibitor S961 was a kind gift of Lauge Schäffer ( Novo Nordisk , Denmark ) ( Schäffer et al . , 2008 ) . Mice were implanted s . c . with an osmotic pump ( model 1007D , Alzet ) loaded with 40 nmol S961 ( estimated delivery rate: 0 . 25 nmol/h for 1 week ) . Long-acting insulin detemir ( Levemir , Novo Nordisk ) was freshly diluted in NaCl 0 . 9% and injected s . c . twice per day ( 1 . 7 U/kg in the morning , 3 . 3 U/kg in the evening ) . Insulin pellets ( LinShin Canada Inc . , Canada ) were implanted s . c . For the ipGTT , mice were fasted overnight ( 15 hr ) and then injected i . p . with 2 mg/kg D-glucose . For the ITT , mice were fasted for 5 hr and injected i . p . with 0 . 7 U/kg insulin ( Humalog , Eli Lilly ) . Following euthanasia , collected pancreata were processed as described ( Desgraz and Herrera , 2009 ) . Paraffin and cryostat sections were 5 and 10 μm-thick , respectively . Primary antibodies: guinea pig anti-insulin ( 1:400 , Dako , Denmark ) , mouse anti-glucagon ( 1:250 to 1:1000 , Sigma ) , and rabbit anti-GFP ( 1:200 , Molecular Probes Inc . , Eugene , OR ) . Secondary antibodies were coupled to Alexa Fluor dyes 488 , 568 , or 647 ( 1:500 , Molecular Probes Inc . ) ; or to FITC , Cy3 , or Cy5 ( 1:500 , Jackson ImmunoResearch , West Grove , PA ) . Images were acquired on a confocal microscope ( TCS SPE , Leica Microsystems , Germany ) . For cell mass measurement , 8 to 12 equally spaced sections per pancreas were imaged on a Leica M205 FA stereo microscope . Islets were manually selected using ImageJ ( NIH ) and thresholding was applied to measure the insulin- and glucagon-positive areas . After dissection , liver and skeletal muscle ( gastrocnemius ) were immediately stored in RNAlater ( Sigma ) . Tissues were homogenized with a Polytron and total RNA was extracted with the Qiagen ( Germany ) RNeasy mini kit ( standard kit for liver , fibrous tissue kit for muscle ) . Reverse transcription was performed using the Qiagen QuantiTect RT kit . qPCR reactions and analyses were performed as described ( Thorel et al . , 2010 ) ; each sample was run in triplicate . For normalization , eight housekeeping genes were tested and the three more stable across our experimental conditions were defined using geNorm ( Vandesompele et al . , 2002 ) : β-Glucuronidase ( Gusb ) , Glyceraldehyde-3-phosphate dehydrogenase ( Gapdh ) , and Non-POU-domain-containing , octamer binding protein ( Nono ) for liver; β-actin ( Actb ) , Gapdh , and Gusb for skeletal muscle . Primer sequences are indicated in Supplementary file 1 . Protein extracts from total pancreas were prepared as described ( Strom et al . , 2007 ) . Blood samples were collected in EDTA-coated tubes and plasma was separated by centrifugation . Insulin and glucagon concentrations were measured using Ultrasensitive Mouse Insulin and Glucagon ELISA kits ( Mercodia , Sweden ) , respectively . Glycogen concentration was measured from the supernatatant of homogenized liver tissue using a glycogen asssay kit ( Sigma ) . Liver samples were lyzed in radioimmumoprecipitation ( RIPA ) buffer with protease inhibitors ( Thermo Fisher Scientific , Waltham , MA ) . Protein concentration was measured using a BCA assay ( Thermo Fisher Scientific ) . Proteins were resolved on a TruPAGE gel ( Sigma ) and transferred to a PVDF membrane . The membrane was blocked in Tris-buffered saline with 0 . 1% Tween containing 5% bovine serum albumin . Primary antibodies were rabbit anti-PEPCK ( 1:1500 , Abcam , UK ) and mouse anti-tubulin ( 1:2500 ) , both incubated overnight at 4°C; secondary antibodies were horseradish peroxidase-conjugated anti-rabbit ( 1:5000 ) and anti-mouse ( 1:5000 ) . Proteins were detected using ECL plus substrate ( Thermo Fisher Scientific ) and images were acquired on a LAS-4000 imager ( Fujifilm , Japan ) . Data are presented as mean ± SEM . P values were calculated with GraphPad Prism 6 ( GraphPad Software , La Jolla , CA ) . The following statistical tests were applied: unpaired , two-tailed , Mann-Whitney U test for two sample comparisons; one- or two-way ANOVA with post hoc Bonferroni correction for multiple comparisons; Log-rank ( Mantel-Cox ) test for survival analyses .

After meals , digested food causes sugar to accumulate in the blood . This triggers the release of the hormone insulin from beta cells in the pancreas , which allows liver cells , muscle cells and fat cells to use and store the sugar for energy . Other cells in the pancreas , called alpha cells , release a hormone called glucagon that counteracts the effects of insulin by telling the liver to release sugar into the bloodstream . The balance between the activity of insulin and glucagon keeps blood sugar levels steady . Diabetes results from the body being unable to produce enough insulin or respond to the insulin that is produced , which results in sugar accumulating in the blood . Diabetes also increases the production of glucagon , which further increases blood sugar levels . Recently , some researchers have reported that mice that lack the receptor proteins through which glucagon works do not develop diabetes , even when they are treated with a drug called streptozotocin that wipes out most of their beta cells . This suggests that the high blood sugar levels seen in diabetes result from an excess of glucagon , and not a lack of insulin . Drugs that block the action of glucagon have been found to reduce the symptoms of mild diabetes in mice and are now being tested in humans . However , it is less clear whether this treatment has any benefits in animals with more severe diabetes . Streptozotocin destroys most of a mouse’s beta cells but a significant fraction of them persist , while a different system relying on diphtheria toxin destroys more than 99% of these cells . Damond et al . have now found that treating mice that lack glucagon receptors with diphtheria toxin causes the mice to develop severe diabetes . Mice that lacked glucagon receptors that had been treated with streptozotocin also developed diabetes after they had been treated with an insulin-blocking drug . Further experiments showed that blocking glucagon receptors in typical mice with diabetes reduces blood sugar , but only if there is some insulin left in their bodies . Damond et al . also found that the glucagon receptor-lacking mice have more alpha cells , which have the ability to convert into insulin-producing cells after the widespread destruction of beta cells . Together , the experiments suggest that blocking glucagon could be a useful treatment for diabetes , but only in individuals who still have some insulin-producing cells . Such treatment would help reduce the release of sugar from the liver and increase the production of insulin in converted alpha cells in the pancreas . Damond et al . are now investigating how alpha cells convert into beta cells , with the aim of learning how to make beta cells regenerate more efficiently .

2016

Blockade of glucagon signaling prevents or reverses diabetes onset only if residual β-cells persist

The glucose-sensing Mondo pathway regulates expression of metabolic genes in mammals . Here , we characterized its function in the zebrafish and revealed an unexpected role of this pathway in vertebrate embryonic development . We showed that knockdown of mondoa impaired the early morphogenetic movement of epiboly in zebrafish embryos and caused microtubule defects . Expression of genes in the terpenoid backbone and sterol biosynthesis pathways upstream of pregnenolone synthesis was coordinately downregulated in these embryos , including the most downregulated gene nsdhl . Loss of Nsdhl function likewise impaired epiboly , similar to MondoA loss of function . Both epiboly and microtubule defects were partially restored by pregnenolone treatment . Maternal-zygotic mutants of mondoa showed perturbed epiboly with low penetrance and compensatory changes in the expression of terpenoid/sterol/steroid metabolism genes . Collectively , our results show a novel role for MondoA in the regulation of early vertebrate development , connecting glucose , cholesterol and steroid hormone metabolism with early embryonic cell movements . The glucose-sensing Mondo pathway is a key regulator of energy metabolism ( Abdul-Wahed et al . , 2017; Richards et al . , 2017 ) . It consists of three basic helix-loop-helix/leucine zipper ( bHLH/Zip ) family transcription factors ( TFs ) : the two Mondo family factors ‘Mlx interacting protein’ ( Mlxip; MondoA ) and ‘Mlx interacting protein-like’ ( Mlxipl; or ChREBP for ‘carbohydrate response element binding protein’ ) and their binding partner ‘Max-like protein X’ ( Mlx ) ( Havula and Hietakangas , 2018; Richards et al . , 2017 ) . After their activation by glucose , likely via glucose-6-phosphate ( G6P ) ( Dentin et al . , 2012; Li et al . , 2010b; Stoltzman et al . , 2008 ) , the Mondo family factors regulate gene expression as MondoA:Mlx or ChREBP:Mlx heterodimers by binding E-box type enhancer elements ( ‘carbohydrate response elements’ , ‘ChoREs’ ) in the regulatory regions of their target genes ( Billin et al . , 2000; Jeong et al . , 2011; Koo and Towle , 2000; Ma et al . , 2006; Poungvarin et al . , 2015; Yamashita et al . , 2001 ) . While recent studies have highlighted a crucial role for metabolism and metabolic signaling in embryonic development ( Miyazawa and Aulehla , 2018 ) , research on Mondo pathway function has concentrated on adult animals ( Hunt et al . , 2015; Iizuka et al . , 2004; Imamura et al . , 2014; Song et al . , 2019 ) . As it is currently unknown if and how MondoA , ChREBP or Mlx contribute to metabolic regulation of embryogenesis , we investigated a potential developmental role of the Mondo pathway in zebrafish , a model for early vertebrate development ( Nüsslein-Volhard , 2002 ) . After fertilization , a cap of cells ( the blastoderm ) forms on top of the yolk cell , which it subsequently engulfs in a morphogenetic movement called epiboly ( Bruce , 2016 ) . Mechanistically , epiboly involves processes both in the enveloping layer ( EVL ) as well as the deep cell layer ( DEL ) of the blastoderm and in a teleost specific structure of the yolk cell , the yolk syncytial layer ( YSL ) . By the end of epiboly , germ layers and body axes have been formed . Most developmental pathways as well as many facets of metabolism and its regulation are highly conserved between mammals and zebrafish ( Gut et al . , 2017; Schier and Talbot , 2005; Solnica-Krezel and Sepich , 2012 ) . However , studies addressing Mondo pathway function in the zebrafish are still lacking . Herein , we uncover a role of the Mondo pathway in vertebrate development . Mondo pathway gene expression and function was characterized in zebrafish embryos and cultured cells . Morpholino oligonucleotide ( MO ) mediated loss-of-function of MondoA severely impaired epiboly movements . Subsequent transcriptome analysis revealed genes deregulated upon loss of MondoA function and highlighted an enzyme involved in cholesterol biosynthesis , Nsdhl ( NAD ( P ) dependent steroid dehydrogenase-like ) , as a potential key mediator of MondoA function in epiboly . Functional analysis of Nsdhl suggests that Nsdhl-mediated cholesterol synthesis downstream of MondoA is required for the synthesis of sufficient levels of the steroid hormone pregnenolone , which stabilizes YSL microtubules necessary for epiboly . Maternal-zygotic ( MZ ) mutant embryos homozygous for a small deletion allele of mondoa leading to a premature stop codon showed a severe aberrant epiboly phenotype , albeit with low penetrance that our transcriptome analysis indicates to result from compensatory changes in the expression of cholesterol and steroid biosynthesis genes . For each of the Mondo pathway members MondoA , ChREBP and Mlx one single orthologue is present in the zebrafish genome ( GRCz11/danRer11 ) . We cloned the full cDNAs for zebrafish MondoA and ChREBP ( GenBank Accession KF713493 [mondoa] , KF713494 [chrebp] ) based on the ( partially ) predicted sequences . Phylogenetic analysis showed that zebrafish mondoa , chrebp and mlx cluster with their mammalian and chicken homologs ( Figure 1A ) and revealed a high level of protein sequence conservation between zebrafish and human orthologs ( Figure 1B ) , especially of the glucose-sensing module ( GSM ) specific to the Mondo family and of the DNA binding bHLH/Zip domains . This finding suggests that the functions of these proteins are conserved . To examine whether the zebrafish Mondo pathway factors function similarly in the regulation of gene transcription as their mammalian orthologs , we studied the pathway in zebrafish PAC2 cells ( Lin et al . , 1994 ) . All three Mondo pathway factors are expressed in these cells ( Figure 1—figure supplement 1A ) . To monitor Mondo signaling , we generated a luciferase reporter gene construct driven by two ChoREs fitting the mammalian consensus ( Ma et al . , 2006 ) and a TATA box minimal promoter ( Figure 1C ) . This construct was active in HepG2 cells , a mammalian cell culture model commonly used for Mondo pathway studies ( Kim et al . , 1996; Yu and Luo , 2009; Figure 1—figure supplement 1B ) . In PAC2 cells , bioluminescence equally increased in a dose-dependent manner upon glucose treatment ( Figure 1C ) . No significant changes in bioluminescence levels were detected upon glucose treatment in cells expressing a constitutively active luciferase reporter ( pGL3-Control; Figure 1D ) , excluding a general unspecific increase in transcriptional activity by glucose treatment . We next tested whether overexpression of Mondo pathway members enhances glucose induced pathway activity . Transient overexpression of MondoA and Mlx activated transcription from the reporter also under low glucose conditions ( Figure 1E ) , as shown by the increased bioluminescence compared with transfection of the 2xChoRE reporter alone ( p≤0 . 001 ) . Overexpression of the MondoA and Mlx factors together led to an even more pronounced effect on bioluminescence ( p≤0 . 01 ) , revealing synergistic effects . High glucose levels caused significant reporter gene induction in control cells ( p≤0 . 01; Figure 1E ) . Importantly , strong glucose induction of the reporter was also shown by cells overexpressing either Mlx ( p≤0 . 01 ) or MondoA ( p≤0 . 001 ) alone as well as both factors together ( p≤0 . 001; Figure 1E ) . An unrelated control reporter construct ( pGRE-Luc , Weger et al . , 2012 ) was not responsive to any of the treatments ( Figure 1—figure supplement 1C ) . Together , the data indicate that limited availability of endogenous pathway components restrains pathway activity , and that both basal activity and the response to higher glucose levels are potentiated when more sensor proteins are available . In addition , we explored the effect of morpholino oligonucleotide ( MO ) mediated loss-of-function of mondoa to confirm that Mondo pathway function is required for glucose induction of reporter expression . Cells transfected with a control 5 bp mismatch MO ( mondoa-mis ) showed a 2 . 0-fold induction of bioluminescence from the 2xChoRE reporter plasmid by high glucose levels , while in cells transfected with a MO directed against the translation start site ( mondoa-mo ) this glucose response was abolished ( Figure 1F ) . Taken together , our data demonstrate that ChoRE mediated glucose induction of gene expression is regulated by the Mondo pathway also in zebrafish . After confirming the similarity of zebrafish Mondo pathway function to mammals in cultured cells , we studied its role during development . We reanalyzed a previously published zebrafish developmental transcriptome dataset ( White et al . , 2017 , ENA accession number ERP014517 ) and detected maternal transcripts of mondoa , chrebp and mlx as well as expression of the three genes throughout development , following distinct temporal patterns ( Figure 1G–I ) . Whole mount in situ hybridization analysis ( WISH ) for mondoa revealed a ubiquitous expression pattern ( Figure 1J ) . Similar results were observed for chrebp and mlx ( Figure 1—figure supplement 1D , E ) . At the 50% epiboly stage , all three genes are expressed throughout the embryo ( Figure 1K; Figure 1—figure supplement 1F ) . Next , we tested whether glucose treatment increases the expression of mammalian Mondo pathway target gene homologs in early zebrafish embryos , namely hexokinase 2 ( hk2 ) ( Sans et al . , 2006 ) , fatty acid synthase ( fasn ) ( Ma et al . , 2006 ) and thioredoxin-interacting protein a ( txnipa ) ( Stoltzman et al . , 2008 ) . We injected the glucose analogue 2-deoxy-glucose ( 2-DG ) , which is metabolized to the G6P analogue 2-DG6P but not further ( Chi et al . , 1987; Stoltzman et al . , 2008 ) , thereby avoiding activation of other pathways relying on downstream metabolization of glucose . This treatment significantly increased expression of hk2 ( p<0 . 001 ) , fasn ( p<0 . 01 ) and txnipa ( p<0 . 001; Figure 1L–N ) , but not of the control gene eef1a1l1/ef1a ( eukaryotic translation elongation factor 1 alpha 1 ( Figure 1O ) , ruling out a nonspecific general increase in gene expression by 2-DG treatment . Furthermore , txnipa expression was not induced by 2-DG in embryos injected with mondoa-mo , showing that MondoA function is required for its induction ( Figure 1—figure supplement 1G ) . To directly examine if 2-DG treatment regulates transcription by ChoRE enhancer elements in the early embryo , we injected the 2xChoRE luciferase reporter construct together with 2-DG . Embryos treated with 2-DG ( n = 72 ) showed a 2 . 3-fold increase ( p≤0 . 01 ) in bioluminescence over the control ( n = 78; Figure 1P ) . Together , these results suggest that Mondo pathway mediated glucose signaling is present in embryos as early as at the sphere stage . To examine Mondo pathway function during development , we injected embryos with MOs targeting mondoa and chrebp function . While injections of a MO directed against chrebp did not result in an aberrant phenotype ( Figure 2—figure supplement 1A , B ) , mondoa-mo injected embryos showed a striking delay in epiboly movements when compared with uninjected or mondoa-mis injected embryos ( Figure 2A–D; Video 1 ) . This phenotype was dose-dependent , with the strongest delay occurring at the highest concentration ( Figure 2A ) . To test for efficiency and specificity of the MO-mediated knockdown , a number of controls were employed . GFP expression from a reporter construct carrying the target sequence was efficiently abolished by co-injection of mondoa-mo , but not mondoa-mis ( Figure 2—figure supplement 1C ) . An involvement of P53-mediated apoptosis in the observed phenotypes was ruled out by co-injection of mondoa-mo and a MO directed against p53 ( p53-mo ) , leading to the same phenotype as observed upon injection with mondoa-mo alone ( Figure 2—figure supplement 1D ) . Knockdown of the mondoa cofactor mlx with a MO directed against the translation start site of mlx ( mlx-mo ) showed a dose-dependent impairment of epiboly similar to the one observed with mondoa-mo , albeit somewhat less severe ( Figure 2C , D ) . These results are consistent with the model that MondoA forms heterodimers with Mlx to exert its function also in epiboly . Finally , to obtain additional evidence that the observed effects are indeed dependent on MondoA function , we attempted to rescue epiboly movements in mondoa-mo injected embryos by co-injecting a mondoa mRNA in which the sequence targeted by mondoa-mo was mutated . Epiboly progression was monitored with WISH against the blastoderm margin marker no-tail ( ntl ) . Embryos injected with only mondoa-mo reached 38% epiboly , whereas embryos co-injected with mondoa mRNA arrived at 47% epiboly ( Figure 2E ) . These values are consistent with those reported in rescue experiments addressing other genes implicated in epiboly ( Hsu et al . , 2006; Schwend et al . , 2011 ) and might be explained by decreased mRNA stability and/or decreased translation efficiency of the rescue construct in vivo . In conclusion , these observations indicate that mondoa-mo specifically targets mondoa gene function , and reveal that loss of MondoA function severely impacts on epiboly during zebrafish development . Next , we aimed at clarifying which epiboly-related processes are perturbed by loss of mondoa function . Since the Mondo pathway has been implicated in the regulation of energy metabolism in adult mammalian tissues , we tested if the aberrant epiboly phenotype of mondoa morphants is a consequence of generally impaired energy metabolism . Our results showed that energy charge was unaltered in the morphants ( Figure 3—figure supplement 1A ) . Alternatively , abnormal cell proliferation or defects in migratory behavior might limit the spreading capacity of the blastoderm . To count cell numbers in the embryos by automated quantification , we imaged transgenic embryos carrying a H2A transgene which labels cell nuclei ( Tg ( h2afva:h2afva-GFP ) ) ( Pauls et al . , 2001 ) . We employed digital scanned laser light sheet microscopy ( DSLM ) ( Kobitski et al . , 2015 ) , allowing the optical sectioning of the embryo with a high spatial and temporal resolution . As shown by this analysis , mondoa morphants had similar numbers of cells as control embryos ( Figure 3A ) . Accordingly , the cell density was higher in mondoa morphants compared to mismatch controls , as movement to the vegetal pole was impaired while cells continued to divide ( Figure 3B , C; Videos 2–4 ) . These observations strongly argue against deficient cell proliferation . In addition , we observed germ ring-like thickenings ( Figure 2A , C; arrowheads ) as well as internalization of cells at the blastoderm margin in mondoa morphants ( Figure 3D; Figure 3—figure supplement 1B–D; Videos 5 and 6 ) . Furthermore , the average migration speed of deep cells was identical for morphants and controls , demonstrating normal individual cell migration capacity in morphants ( Figure 3—figure supplement 1E ) . EVL , DEL and YSL have all been implicated in mechanisms of epiboly ( Bruce , 2016 ) . Thus , we turned to examining how the three different embryonic cell layers were affected in mondoa morphants . Imaging of Tg ( h2afva:h2afva-GFP ) embryos allowed us to assign nuclei to the EVL or to the DEL and to distinguish the YSL nuclei . Our observations revealed that EVL cells continued to progress and could reach a position vegetally to the YSL nuclei in the morphants ( Figure 4—figure supplement 1A–F , Video 7 ) . Consistently , EVL integrity as revealed by rhodamine-phalloidin staining ( Lepage and Bruce , 2010; Lepage et al . , 2014 ) was intact in the morphants ( Figure 4—figure supplement 1G–H ) . These findings indicate that EVL epiboly still occurs in the morphants . However , deep cells of mondoa morphants lagged behind both EVL and YSL layers , consistent with models suggesting rather passive movement of these cells into spaces left open by the other two layers ( Bensch et al . , 2013; Song et al . , 2013; Figure 4A , B; Figure 4—figure supplement 1A–F ) . Nevertheless , in mondoa morphants a few deep cells were detected more vegetally than the YSL nuclei . Also , YSL nuclei showed a much broader and more disorganized distribution than in control injected or wild-type embryos ( Figure 4—figure supplement 1A–F ) . The abnormal shape of the YSL was already evident at a slightly earlier stage ( compare Figure 4A with Figure 4B ) . YSL nuclei appeared to cluster and to break away from the blastoderm margin ( Figure 4B , white arrow ) , in contrast to the more even distribution in a relatively narrow band around the margin in wild-type embryos . These findings pointed to a defect in YSL structure or function as a potential reason for the impaired epiboly in mondoa morphants . To test whether MondoA function in the YSL itself is important for epiboly , we injected mondoa-mo at the 1 k-cell stage into the yolk cell only ( Figure 4C ) . At this late time of YSL injection , additional mondoa transcript and protein have been produced after mid-blastula transition ( MBT ) , which may limit knockdown efficiency compared with injection into the zygote . Strikingly , mondoa knockdown limited to the YSL still led to epiboly delay and developmental arrest compared with uninjected or control injected embryos ( Figure 4D ) . Thus , MondoA function in the YSL indeed is important for correct epiboly movements . Impaired YSL structure or function may lead to impaired vesicle trafficking at the YSL-EVL border ( Lepage and Bruce , 2010; Solnica-Krezel et al . , 1995; but see also Lepage and Bruce , 2014 ) or to defective patterning of the overlying blastoderm ( Carvalho and Heisenberg , 2010 ) , both of which can perturb epiboly . However , we did not find any evidence for major perturbations of marginal endocytosis or blastoderm patterning in the morphants ( Figure 4—figure supplement 1J–Q ) . To obtain cues how MondoA function might impact on epiboly we carried out an unbiased examination of differential gene expression upon mondoa knockdown by applying RNA-seq . Embryos were injected either with mondoa-mo or mondoa-mis and left to develop until the control embryos reached sphere stage , just before the morphological phenotype becomes apparent ( Figure 5—figure supplement 1A–C; Supplementary file 1 ) . We validated a random subset of the genes expressed differentially between the two conditions by RT-qPCR , confirming the RNA-seq data ( Figure 5—figure supplement 1D , E ) . Interestingly , the gene most down-regulated by mondoa-knockdown was nsdhl ( NAD ( P ) dependent steroid dehydrogenase-like; Figure 5A ) . Nsdhl is part of the cholesterol biogenesis pathway , catalyzing steps in the conversion of the cholesterol precursor lanosterol to zymosterol ( Figure 5B ) . Consistently , a global metabolic pathway gene set enrichment analysis revealed that both the ‘sterol biosynthesis’ pathway , of which Nsdhl is part , and the upstream ‘terpenoid backbone biosynthesis’ pathway are coordinately downregulated in the morphants ( Supplementary file 2 , Figure 5B , Figure 7—figure supplement 1A , B ) . Cholesterol has an important role as substrate for the synthesis of pregnenolone , the first steroid hormone within the steroid hormone biosynthesis pathway . Pregnenolone and other steroids have been implicated in the regulation of zebrafish epiboly ( Eckerle et al . , 2018; Hsu et al . , 2006; Schwend et al . , 2011; Weng et al . , 2013 ) . Therefore , we hypothesized that mondoa knockdown might affect epiboly by interfering with Nsdhl-dependent cholesterol and steroid hormone biogenesis . To test this hypothesis , we first validated regulation of nsdhl expression by MondoA mediated glucose sensing . We observed a significant increase of basal nsdhl transcript levels upon 2-DG injection ( p≤0 . 05 ) and upon overexpression of both MondoA and Mlx ( p≤0 . 001 ) in early embryos , which was augmented further when both treatments were combined ( p≤0 . 001; Figure 5C ) . No changes in expression were observed for ef1a under these conditions ( Figure 5D ) , confirming the target gene specific effects of the treatments . Together , these results strongly suggest that the Mondo pathway mediates glucose regulation of nsdhl expression , thereby linking glucose sensing with cholesterol synthesis . We next examined whether nsdhl knockdown can affect epiboly using a MO targeted against a splice site of the nsdhl gene ( nsdhl-mo ) . We observed a highly similar phenotype to the mondoa knockdown phenotype , with nsdhl-mo injected embryos arrested at 50% epiboly when control embryos had completed epiboly ( Figure 5E , F ) . As mondoa , nsdhl is ubiquitously expressed in the embryo , including the YSL ( Figure 5—figure supplement 1F , G ) , and injection of nsdhl-mo into the yolk cell at the 1 k-cell stage resulted in an epiboly delay and developmental arrest as observed for YSL specific mondoa knockdown ( Figure 5G ) . Thus , the function of Nsdhl in the biosynthesis pathway to cholesterol in the YSL appears to be essential for epiboly movements . Altered cholesterol biosynthesis in the YSL by loss of MondoA/Nsdhl function might affect steroid hormone-dependent epiboly mechanisms . Therefore , we next tested whether pregnenolone treatment can rescue mondoa morphants . Treatment with pregnenolone indeed partially rescued epiboly in mondoa-mo injected embryos ( Figure 5H ) at a level comparable to the rescue with mondoa mRNA . Accordingly , the aberrant epiboly phenotype of nsdhl morphants was also partially rescued by pregnenolone treatment ( Figure 5I ) , confirming its role downstream of the MondoA/Nsdhl pathway . Pregnenolone was shown to regulate epiboly by mediating microtubule stability ( Hsu et al . , 2006; Schwend et al . , 2011; Weng et al . , 2013 ) . To examine whether yolk cell microtubule integrity is affected in mondoa morphants , we stained embryos against α-tubulin . In uninjected embryos ( n = 9/9 ) , the YSL forms a regular band below the blastoderm , with yolk syncytial layer nuclei ( YSN ) surrounded by microtubule organizing centers ( MTOCs ) ( Solnica-Krezel and Driever , 1994; Strähle and Jesuthasan , 1993; Figure 6A ) . Below this area , dense arrays of microtubules spread from the MTOCs along the animal-vegetal axis into the yolk cytoplasmic layer ( YCL ) , forming root-like bundles . Compared with the control embryos ( Figure 6C , D ) , mondoa-mo injected embryos showed a rather unstructured YSL with a less well defined border ( Figure 6E , F ) . YSN were less regularly arranged , and their associated MTOCs formed stellar structures ( n = 14/16; Figure 6B ) , apparently reflecting shortened microtubules that did not form the arrays of root-like bundles along the animal-vegetal axis seen in uninjected controls . Importantly , the YSL of morphants treated with pregnenolone appeared to be better organized compared to the untreated morphants because the YSN and their corresponding MTOCs were arranged as in the uninjected control ( n = 6/9; Figure 6B , G , H ) . Furthermore , in pregnenolone treated embryos , the microtubule arrays were longer and formed array-like structures along the animal-vegetal axis of the embryos similar to the control . Taken together , these results strongly suggest that MondoA functions in epiboly via the stabilization of yolk microtubules by pregnenolone formed downstream of cholesterol biosynthesis . Finally , we wanted to validate the knockdown results with a genetic loss-of-function model of mondoa using CRISPR-Cas9 . We identified a mutant allele transmitted through the germline that carried a 5 bp deletion in exon 3 ( Figure 7A ) , leading to a frameshift in the coding sequence and a predicted premature stop eliminating both the DNA binding domain and most of the glucose-sensing domain . Embryos homozygous for this allele did not show an aberrant embryonic phenotype and were raised to sexual maturity ( Figure 7B , C ) . By contrast , we found a few maternal-zygotic ( MZ ) mutants from incrosses of homozygous mutant parents that showed a severe epiboly delay and an arrest in mid-epiboly ( Figure 7B , C ) . The low penetrance of the epiboly phenotype in the MZ mutants is indicative of compensatory mechanisms that counteract the loss of function of mondoa . To test this assumption , we injected mondoa-mo into MZmondoa mutants . Indeed , we observed a resistance of the mutants to the perturbance of epiboly caused by this MO in wild-types ( Figure 7D ) . This observation strongly suggests the presence of strong compensatory or buffering mechanisms that enable embryonic development even when MondoA function is genetically perturbed . It also further confirms the specificity of the morphant phenotype . To begin to explore potential compensatory mechanisms allowing epiboly progression in MZmondoa mutants , we performed total RNA sequencing of MZmondoa embryos with severe and no epiboly anomalies and of wild-type embryos to determine differential gene expression ( Supplementary file 1 ) . To compare the gene expression signatures of MZmondoa mutants with and without aberrant epiboly phenotype to those of mondoa morphants and their respective controls , we employed the rank-rank hypergeometric overlap ( RRHO ) algorithm ( Plaisier et al . , 2010 ) . The analysis revealed that differential gene expression significantly overlapped between mutants and morphants , demonstrating that both conditions reflect a lack of MondoA function . Specifically , mutant embryos showed an overlap in differential gene expression with respect to the morphants mainly in the upregulated gene fraction . There were also some differences apparent between the mutants with and without an aberrant epiboly phenotype . We observed that the overlap in downregulated genes was weaker between morphants and mutants without aberrant epiboly phenotype ( compare lower left quadrant in Figure 7—figure supplement 2B and C ) . This observation indicated that mainly the downregulated genes in mondoa morphants were transcriptionally compensated in MZmondoa mutants with unaffected epiboly . In contrast , the affected mutants showed a weaker overlap in the upregulated gene signatures , suggesting that only a fraction of the upregulated genes in MZmondoa mutants was actually important for phenotype compensation . We next explored these differential gene expression patterns in more detail . As paralogues of genes have been implicated in compensation ( El-Brolosy et al . , 2019; Rossi et al . , 2015 ) , we looked at expression of other Mondo pathway components . Neither chrebp , the paralogue of mondoa , nor the MondoA partner mlx were significantly upregulated in MZmondoa mutants , even though trends ( p=0 . 06 ) for elevated mRNA levels were seen for mlx in both mutant phenotypes and for chrebp in the affected phenotype only ( Figure 7E ) . We next turned our attention to genes in the cholesterol biosynthesis pathway based on our previous observations in morphant embryos . Our analysis showed that nsdhl gene expression was comparable to wild-type embryos in both MZmondoa mutant phenotypes ( affected and unaffected epiboly ) and , thus , was compensated in both ( Supplementary file 1 and Figure 7—figure supplement 2E–F ) . Therefore , differences in nsdhl expression between MZmondoa mutants cannot explain the differences in the epiboly phenotype . This observation indicates that the restoration of expression to wild-type levels of this most downregulated gene upon mondoa knockdown by itself was not sufficient for compensation . Consistently , in both mutant phenotypes the entire sterol biosynthesis pathway of which nsdhl forms part ( Figure 5B ) lacked the coordinated downregulation that was seen in the morphants ( unaffected: p=0 . 87 , affected: p=0 . 79; Figure 7F and Figure 7—figure supplement 1A ) . But what is causing the epiboly arrest if MZmondoa mutants with affected epiboly share such compensatory transcriptional signatures with MZmondoa mutants without aberrant epiboly phenotype ? mondoa morphants showed a significant downregulation in terpenoid backbone biosynthesis ( p=0 . 016 ) as well as in sterol biosynthesis pathways ( p=0 . 022; Supplementary file 2 ) . By contrast , MZmondoa mutants with unaffected epiboly showed a coordinated upregulation of terpenoid pathway genes ( i . e . , they showed a statistically significant enrichment for upregulation in a directional gene set enrichment test; p=0 . 012; Figure 7F–G , Figure 7—figure supplement 1B and Supplementary file 2 ) . Intriguingly , mutants with affected epiboly lacked this significant coordinated upregulation in the terpenoid backbone biosynthesis pathway that was present in unaffected mutants ( p=0 . 178; Figure 7F–G and Figure 7—figure supplement 1B ) . This key difference highlights the upregulation of gene expression in the terpenoid backbone biosynthesis pathway as a compensation response specifically required to restore the normal epiboly phenotype . We additionally observed that gene expression of steroid hormone biosynthesis pathway genes showed a coordinated upregulation in MZmondoa mutants with unaffected epiboly ( p=0 . 033; Figure 7F and Figure 7—figure supplement 1C ) , probably to compensate for reduced flux through the biosynthesis pathways upstream . This upregulation was also shown by MZmondoa mutants with aberrant epiboly phenotype ( p=0 . 017; Figure 7F and Figure 7—figure supplement 1C ) . Strikingly , it was even increased when compared to the mutants with unaffected epiboly ( p=0 . 049 ) , and thereby led to strongly increased expression levels compared to wild-type . This is illustrated by the expression of cyp11a1 , the enzyme that is immediately upstream of pregnenolone synthesis ( Supplementary file 2 , Figure 7—figure supplement 2G ) . However , in the absence of coordinated regulation of genes in the terpenoid pathway , this regulation was apparently not sufficient to rescue epiboly progression . Taken together , despite some successful compensatory gene expression regulations , MZmondoa mutants with affected epiboly fail to properly regulate the entire set of genes necessary for epiboly progression . They are not lacking compensation per se , but their compensation is inadequate and fails . In summary , results from both MO mediated and genetic loss-of-function of MondoA revealed MondoA as a key developmental regulator of progression through epiboly by regulating expression of cholesterol/steroidogenesis pathway genes important for yolk cell microtubule function . Here , we have characterized the Mondo pathway in zebrafish and revealed a developmental function for Mondo signaling , implying a novel role for this glucose-sensing pathway in addition to its functions in adult animals ( Song et al . , 2019 ) . Previously , studies in non-vertebrates have implicated homologs of Mondo pathway factors in later stages of embryonic or larval development: Loss of function of the Mlx orthologue in Drosophila is lethal at the late pupal stage ( Havula et al . , 2013 ) , while knockdown of a distant Mlx homolog in C . elegans affected movements of ray one precursor cells ( Pickett et al . , 2007 ) , which will form a subset of male-specific sense organs in the tail . To the best of our knowledge , ours is the first study demonstrating conserved Mondo signaling mechanisms between zebrafish and mammals and linking its function to early embryonic development of vertebrates . Our findings demonstrate that MondoA acts upstream of the synthesis of pregnenolone , an important regulator of yolk cell microtubule stability and epiboly movements ( Hsu et al . , 2006; Schwend et al . , 2011; Weng et al . , 2013 ) , by regulating the expression of a whole set of genes in the terpenoid/sterol biosynthesis pathway , as exemplified by the cholesterol synthesis enzyme nsdhl . It is tempting to speculate that the role of MondoA in cholesterol/steroid metabolism we here uncover in the zebrafish is a feature conserved across vertebrates , albeit not much is known in mammals . One in vitro study described ‘sterol biosynthetic process’ as one category of genes upregulated in MondoA deficient HeLa cells under conditions of acidosis ( Wilde et al . , 2019 ) . For the MondoA paralogue Chrebp , a study in vivo reported that ‘cholesterol biosynthesis’ genes are upregulated in the liver of Chrebp knockout mice which were fed with a high fructose diet ( Zhang et al . , 2017 ) . While these studies were performed under specific disease conditions ( i . e . , a human cancer cell line under acidosis conditions or animals fed with a high fructose diet ) , they support our observations that the Mondo pathway plays an important role in the regulation of cholesterol/steroid metabolism also under non-disease conditions . Furthermore , these observations indicate gene expression changes due to compensatory mechanisms upon loss of Mondo pathway function , as observed in our study and discussed in the following section . The low penetrance of epiboly phenotypes in MZmondoa mutant zebrafish embryos indicates the presence of efficient compensatory pathways , triggered by the genetic lesion but not by antisense-mediated knockdown , that may be operating in mammals as well ( El-Brolosy and Stainier , 2017 ) . Nonsense mediated decay products of the mutated transcript are proposed to trigger transcriptional upregulation of homologous genes that compensate for the mutant gene’s function ( El-Brolosy et al . , 2019 ) . However , chrebp and mlx mRNA levels were not significantly changed compared to wild-type in mutants with unaffected epiboly , implicating the presence of alternative compensatory mechanisms . Interestingly , many terpenoid , sterol and steroid biogenesis genes showed lower than wild-type levels of expression in the morphants and equal or higher levels in the mutants with unaffected epiboly , suggesting that increased steroidogenesis activity rescues epiboly in a majority of MZ mutant embryos . In the remaining mutant embryos , the lack of coordinated upregulation of terpenoid backbone biosynthesis genes is indicative of an incomplete compensation attempt . Apparently , concerted changes across the entire cholesterol/steroid biosynthesis pathway are needed to rescue epiboly progression in the absence of MondoA function . The mere restoration of nsdhl expression levels , for example , appears not to be sufficient for such a rescue , perhaps because a few genes in the sterol biosynthesis pathway still show deregulation of expression in MZ mutant embryos of both phenotypes ( Figure 7—figure supplement 1A ) . Therefore , even though globally there is no coherent up- or downregulation in sterol biosynthesis gene expression of mutants compared to wild-type embryos based on gene set enrichment analysis , pathway function in mutants may still be suboptimal and require the compensatory upregulation in the upstream terpenoid pathway to achieve sufficient synthesis flux . Furthermore , random variation in the amount of material deposited by the mother or in protein expression of key genes may limit compensation in the mutants with affected epiboly . A more detailed understanding of these processes awaits the determination of metabolite fluxes in morphants , mutants and wild-type embryos , specifically in the YSL . Importantly , incomplete penetrance is a frequent phenomenon in human genetic diseases and has been attributed to a wealth of mechanisms , which have been suggested as targets for therapy ( Cooper et al . , 2013 ) . Future studies how disturbed MondoA function is buffered during development might open leads to manipulating dysregulated Mondo pathway function in adult metabolic regulation . We observed that mondoa knockdown in the YSL was sufficient to cause an epiboly phenotype , pointing towards a crucial role of MondoA in this tissue . Even though the YSL and epiboly movements are specific for teleosts , there are several indications that the mechanisms detected in the zebrafish are relevant in mammals as well . For example , steroidogenic enzymes are expressed in extraembryonic tissues in both fish and mice , the fish YSL ( Hsu et al . , 2006; Hsu et al . , 2009 ) and the giant trophoblast cells of the murine placenta ( Arensburg et al . , 1999 ) . Indeed , it has been proposed that zebrafish epiboly is analogous to the spreading of the trophectoderm during implantation ( Kane and Adams , 2002 ) . Consistent with this idea , epiboly movements are severely impaired upon knockdown of the zebrafish homolog of solute carrier family 3 member 2 ( Slc3a2 ) ( Takesono et al . , 2012 ) , a factor crucial for trophoblast cell adhesion , migration and fusion in mammals ( Kabir-Salmani et al . , 2008; Kudo et al . , 2003 ) . Thus , it is conceivable that developmental functions of MondoA are conserved among vertebrates . Testing this hypothesis awaits a close examination of early embryonic and placental phenotypes in MondoA mutant mice , the analysis of which has so far been limited to adult phenotypes ( Ahn et al . , 2019; Imamura et al . , 2014 ) , and of potential functions of maternal MondoA mRNA or protein in mammals ( Li et al . , 2010a ) . Given the small litter size of rodents , phenotypes with low penetrance may easily be overlooked and are difficult to study in these animals . Importantly , mutations in the NSDHL gene in humans have been linked to an X-linked dominant condition called CHILD syndrome ( ‘congenital hemidysplasia with ichthyosiform erythroderma and limb defects’ , reviewed in Herman , 2000 ) that is usually lethal in males ( Bornholdt et al . , 2005 ) . Murine male embryos carrying Nsdhl mutations die shortly after implantation , and those carrying moderate or mild alleles showed reduced placental thickness with a poorly vascularized fetal labyrinth ( Caldas et al . , 2005; Liu et al . , 1999 and references therein ) . Interestingly , also in female mouse embryos carrying a maternal null allele placental area was severely reduced ( Cunningham et al . , 2010 ) . In these embryos , paternal X chromosome inactivation in yolk sac endoderm and trophoblast-derived lineages causes complete loss of function only in these tissues . Given the potential analogous functions of placenta and YSL ( Carvalho and Heisenberg , 2010 ) , our results regarding Nsdhl function in epiboly make it tempting to speculate that Nsdhl contributes to the development of the mammalian placenta by regulating microtubule stability . Remarkably , stability of microtubules apparently plays a role in trophoblast differentiation and implantation based on in vitro studies ( Bates and Kidder , 1984; Douglas and King , 1993 ) . In summary , our results identify a novel role for the Mondo pathway in vertebrate development and link glucose sensing to cholesterol/steroid biogenesis . These results contribute to the growing recognition of metabolic regulatory functions in development ( Miyazawa and Aulehla , 2018 ) and broaden our understanding of Mondo signaling . All zebrafish husbandry was performed in accordance with the German animal protection standards and approved by the Government of Baden-Württemberg , Regierungspräsidium Karlsruhe , Germany ( Aktenzeichen35-9185 . 64/BH KIT ) , by the Home Office , United Kingdom ( Scientific Procedures , Act 1986 ) , and by the Service de la consommation et des affaires vétérinaires du canton de Vaud , Switzerland . Wild-type fish were descendants of the AB strain ( University of Oregon , Eugene ) or the golden strain and have been reared for several years in the laboratory . Fish were bred and raised in E3 medium ( Nüsslein-Volhard , 2002 ) ( 5 mM NaCl , 0 . 17 mM KCl , 0 . 33 mM CaCl2 , 0 . 33 mM MgSO4 , 0 . 1% methylene blue ) . Staging of embryos was performed on a dissecting microscope according to Kimmel et al . , 1995 . MZmondoa and wild-type RNA was extracted using RNA Clean and Concentrator-25 columns with an included DNase I treatment ( ZYMO RESEARCH ) for the corresponding RNA-seq experiment , as described in the manufacturer’s protocol . RNA that was used for cDNA synthesis was extracted from cells and whole embryos with the TRIzol reagent ( Life Technologies ) , as well as RNA that was prepared for the RNA-seq experiment of MondoA morphants . TRIzol extraction was performed as described in the manufacturer’s protocol , with minor modifications . If not otherwise stated , 30 embryos were harvested in 1 . 5 ml TRIzol . Embryos were disrupted in the TRIzol solution with micropestles ( Eppendorf ) , and 5 × 105 cells in 25 cm2 flasks were harvested with 1 . 5 ml TRIzol solution . Samples were then stored overnight at –80°C . After thawing the samples on ice , they were passed through a syringe four times ( Braun Sterican , 0 . 45 × 25 mm ) . The integrity of the total RNA was checked on an agarose gel . A NanoDrop spectrometer was used to determine concentrations and A260/A280 ratios of the samples . Only RNA with an A260/A280 ratio above 1 . 6 was used . cDNA was synthesized using random primers and Superscript III reverse transcriptase according to the manufacturer’s protocol ( Life Technologies ) . Transcript levels of genes of interest were determined with the StepOnePlus Real-Time PCR System ( Applied Biosystems ) using the QuantiTect SYBR Green PCR Kit ( Qiagen ) . Copy numbers were normalized using β-actin transcript levels . The melting curves of the amplicons were determined after amplification . Ct values were determined using StepOne Software 2 . 1 ( Applied Biosystems ) and data were analyzed by the 2-ΔΔCt method as previously reported ( Livak and Schmittgen , 2001 ) . Primer sequences: β-actin: fw: 5'-gcctgacggacaggtcat-3' , rv: 5'-accgcaagattccataccc-3'; eef1a: fw: 5'-ccttcgtccaatttcagg-3' , rv: 5'-ccttgaaccaccccatgt-3'; mlxip: fw: 5'-ctgccagacgtatcacttcg-3' , rv: 5'-ggcgaatcttgtctctccac-3'; mlxipl: fw: 5'-ccacagccagtgttgatgat-3' , rv: 5'-tagctgtgcgatgaatgtcc-3'; mlx: fw: 5'-agacttctccatggccacac-3' , rv: 5'-ctgatgggccttcacaatg-3'; pck1: fw: 5'-tgacgtcctggaagaacca-3' , rv: 5'-gcgtacagaagcgggagtt-3'; hk2: fw: 5'-gaaattggcctcattgtcg-3' , rv: 5'-catccaccagctccaagtg-3'; fasn: fw: 5'-aggccatcgttgatggag-3' , rv: 5'-tgtagacgccagttttgctg-3'; txnipa: fw: 5'-ccaactagacgaacatccaaca-3' , rv: 5'-agacaccagctgcccttg-3'; sox3: fw: 5'-cttagcgcacaactttgcag-3' , rv: 5'-caccagtcccgtgtgtctc-3'; tmprss4b: fw: 5'-tcaaagtttcctctcagccagt-3' , rv: 5'-ccaccacctcaccacagtc-3'; adh5: fw: 5'-cacgctcctctggataaagtgt-3' , rv: 5'-gtagagcccgcttcaacct-3'; cfl1: fw: 5'-atgatctacgccagctccaa-3' , rv: 5'-tcacttgccactcgtgctta-3'; vent: fw: 5'-aggagaaatgcagcacagc-3' , rv: 5'-tcactctccacatcggtgtatc-3'; lmnl3: fw: 5'-cctggccaactacatcgag-3' , rv: 5'-gatctgcatggaggatttgtc-3'; dynlrb1: fw: 5'-aacagcggtttgtcttcaca-3' , rv: 5'-atggggattccttctgcatt-3'; gabarapl2: fw: 5'-cgacaaaactgtccctcagtc-3' , rv: 5'-ccgtcctcgtcttgttcttt-3'; aurkaip1: fw: 5'-cgagagtggtttcacgactg-3' , rv: 5'-ctggctctgtgtctgcaatg-3'; nsdhl: fw: 5'-ggggacctctgtgacaaaca-3' , rv: 5'-aggtgaggcacaatgaaaca-3' . If not otherwise stated , all injections were performed as follows: Injection mixes were prepared with the desired concentrations of the compound of interest ( see below ) and 0 . 1% ( w/v ) phenol red as an injection control . Each injection mix was injected with borosilicate glass capillary tubes ( ~0 . 6 mm diameter ) into the yolk of 1 cell stage zebrafish eggs with a gas-driven microinjector ( Eppendorf Femtojet express ) as described ( Müller et al . , 1999 ) . After the injection , unfertilized or damaged eggs were removed around the high to sphere stages and were not considered for the subsequent analysis . WISH was performed on embryos ( n ≥ 20 ) as previously described ( Oxtoby and Jowett , 1993 ) , with minor modifications: The Proteinase K digest step was omitted for early embryonic stages and replaced with four PTw ( 1xPBS , 0 . 1% Tween-20 ) washes , while it was reduced to 6 min for 24 hr old embryos . Epon embedding of embryos was carried out as described ( Westerfield , 2007 ) . 5 µm thick sections were imaged with a compound microscope ( Leica DM5000 ) . Percentage of epiboly progression of the embryos was determined by performing a ntl in situ hybridization to label the blastoderm margin and using the following formula as previously described ( Hsu et al . , 2006 ) : Epiboly ( % ) = ( The length between animal pole and blastoderm margin ) / ( The length between animal pole and vegetal pole ) . Microtubule staining of early embryos by WIHC was carried out as recommended by McMenamin et al . , 2003 . Uninjected embryos and mondoa-mo injected embryos were fixed at late blastula stage just before the onset of epiboly and stained for α-tubulin to visualize the microtubules . The primary antibody used was an anti-α-tubulin antibody ( Sigma-Aldrich , 1:500 ) , the secondary antibody was anti-mouse Alexa Fluor 488 ( Life technologies , 1:1000 ) . Phalloidin staining of actin filaments was performed according to Nakajima and Burke , 1996 . Briefly , embryos were fixed in 4% PFA over night at 4°C . The next day , fixed embryos were washed three times in PTw and then incubated with Rhodamine-Phalloidin ( 1:500 , R415 , Molecular Probes ( Life Technologies ) ) for 15 min . Embryos were mounted in 0 . 5% ( w/v ) low melting agarose in E3 and imaged with a confocal microscope ( Leica TCS2 SP5 ) . As observed for epiboly movement disruption , the severity of the YSL disorganization appears to be dose-dependent , with lower doses of MO leading to a less severe phenotype . Embryos were treated with the glucose analog 2-deoxy-glucose ( 2-DG , Sigma ) to activate the Mondo pathway . As Glucose-6-phosphate ( G6P ) is thought to be the signal activating the Mondo pathway , and because 2-DG can only be metabolized to 2-DG-6-phosphate , but not further , this treatment should avoid activation of other pathways relying on further metabolization of glucose ( Li et al . , 2010b; Stoltzman et al . , 2008 ) . For recording of in vivo bioluminescence from a ChoRE reporter , a mixture composed of the 2xChoRE luciferase reporter construct ( 100 ng/µl; see above , ‘cell culture’ ) , phenol red ( 0 . 1% ) and either 150 mM 2-DG or water was injected into zygotes . The injected embryos were transferred into 96-well plates containing E3 medium with 0 . 5 mM luciferin ( E3L ) at high/oblong stage for bioluminescence measurement as described ( Weger et al . , 2013a; Weger et al . , 2012; Weger et al . , 2013b ) . Bioluminescence was measured at sphere stage and subtracted against background luminescence from wells without embryos . To determine whether nsdhl is a downstream target of MondoA and is regulated by glucose , 150 ng/µl each of mondoa and mlx transcript were injected either with water as a control or with 150 mM 2-DG . RNA was extracted at sphere stage and processed for RT-qPCR against nsdhl . To determine if glucose induction of txnipa expression was affected by mondoa knockdown , embryos were injected with 150 mM 2-DG or with water along with either mondoa-mis or mondoa-mo . CRISPR/Cas9 mediated gene editing was performed essentially as described ( Gagnon et al . , 2014 ) , with the following modifications . Guide RNA target regions in exon 3 of mlxip/mondoa ( 5’-gagtccatattgccatccaa-3’ ) were determined using the CHOPCHOP web tool ( http://chopchop . cbu . uib . no/ ) ( Labun et al . , 2019 ) . The guide RNA templates that include an SP6 promotor and Cas9 scaffold were prepared by GeneArt Strings DNA fragment gene synthesis ( ThermoFisher ) ( 5’-tataagcttccatggatttaggtgacactatagagtccatattgccatccaagttttagagctagaaatagcaagttaaaataaggctagtccgttatcaacttgaaaaagtggcaccgagtcggtgcttttctagacgatgcccttgagagccttcaacccagtcctatagtgagtcgtattaggatcctac-3’ ) . The templates were amplified by PCR and the amplification product was transcribed using the MEGAscript SP6 Transcription Kit ( ThermoFisher ) . 60 ng/µl guide RNA were injected into 1 cell stage wild-type zebrafish embryos together with 0 . 9 µg/µl Cas9 protein ( GeneArt Platinum Cas9 Nuclease; ThermoFisher ) , 0 . 2 M KCl and 0 . 1% phenol red ( SIGMA-Aldrich ) . Guide RNA efficiency was determined using High Resolution Melting ( HRM ) analysis ( fw: 5’-tgcctttctttcctgaagatgt-3’; rv: 5’-gctttttccattcaaaaccagt-3’ ) . Injected ( F0 ) zebrafish were outcrossed with wild-type zebrafish . F1 mutant carriers were identified by HRM and PCR targeting the mutated region ( PCR: fw: 5’-cgcatactcctttatcttgc-3’; rv: 5’-catgctttttccattcaaaacc-3’ ) on genomic DNA from fin biopsies followed by sequence analysis . Further matings of the F1 fish generated homozygous F2 fish , which were raised to adulthood . Maternal zygotic ( MZ ) mutants were derived from incrosses of homozygous F2 fish . Morpholino oligonucleotides ( MOs ) were obtained from GeneTools Inc ( http://www . gene-tools . com/ ) . MO sequences were: chrebp-mo ( translation blocking MO ) : 5’-ttctgaatactctgcttttgccatc-3’; mondoa-mo ( translation blocking MO ) : 5’-ggtattgtcgagtagccatgttaaa-3’; mondoa-mis ( 5 bp mismatch MO ) : 5’-ggtaatctccagtacccatcttaaa-3’; mlx-mo ( translation blocking MO ) : 5’-cgcgctgttttccgtcattttggaa-3’; nsdhl-mo ( splice blocking MO ) : 5’- ctgaaacaattcacacctgtttgtc-3’; p53-mo ( Robu et al . , 2007 ) ( translation blocking MO ) : 5’-gcgccattgctttgcaagaattg-3’; tagged-mo ( 3’ lissamine tagged MO of unrelated sequence ) : 5’-gtagcttgtactcgcattcctatct-3’ . To target the transcript of interest specifically in the YSL , embryos were injected with MO into the yolk at the 1 k cell stage , as previously described ( Sakaguchi et al . , 2001 ) . In order to label the area where the MO is localized , either lis-mondoa-mo or lis-mondoa-mis was used , or 0 . 08 mM of tagged-mo was added to the injection mixture . Injection of tagged-mo alone did not cause an aberrant phenotype . Knockdown efficiency of the mondoa-mo was tested with 30 ng/µl of the mondoa target sequence-GFP fusion transcript encoded by the pCS2:GFP:MondoA vector . For mRNA transcription , this construct was linearized with NotI and transcribed with the mMESSAGE mMACHINE kit as recommended by the manufacturer ( Ambion ) . Rescue experiments with mondoa mRNA were performed using 250 ng/µl of transcript . mondoa mRNA was transcribed using the mMESSAGE mMACHINE Kit ( Ambion ) ( T7 ) on templates linearized with DraIII . Pregnenolone ( 20 µM ) rescue experiments were performed as previously reported ( Hsu et al . , 2006 ) . Knockdown efficiency of the chrebp-mo was tested with 100 ng/µl pCS2:GFP:ChREBP . Resistance of MZmondoa mutant embryos to mondoa knockdown was tested with 0 . 5 mM lis-mondoa-mo and 0 . 5 mM lis-mondoa-mis injections into MZmondoa mutants and wild-type controls . For live imaging of embryos by confocal microscopy , embryos were dechorionated and transferred with a glass pipette into a chamber containing 0 . 1% ( w/v ) low melting agarose in E3 . Images were taken on a Leica TCS SP5 upright microscope using either a HCX PL APO 20x/0 . 70 lambda blue IMM CORR or a HCX APO L 20x/0 . 50 W U-V-I objective at 26°C . Spatial and sensitivity resolution were strictly kept constant between samples of comparative analysis . Reflection images using 458/488/496 nm incident light were used to assess whether a nucleus is in the EVL or YSL . The reflected light images require no labeling and provide a confocal contrast at the boundary of different tissues ( Jester et al . , 1991 ) . We used the GFP fluorescence channel to identify the nucleus and the label-free reflection channel to visualize the extracellular matrix . Whether a nucleus belongs to EVL or YSL was individually determined in the XZ or YZ plane using the reflection channel . Images were analyzed and processed in Fiji/ImageJ ( Schindelin et al . , 2012 ) . To visualize endocytotic vesicles , embryos were dechorionated at five hpf and placed in 1 . 5% ( w/v ) Lucifer Yellow CH ( Sigma-Aldrich , L0259 ) dissolved in E3 medium for 15 min ( Solnica-Krezel and Driever , 1994 ) . After the incubation , embryos were washed briefly with excess E3 medium and mounted in 0 . 5% low-melting temperature agarose for confocal imaging . Time-lapse DSLM images of developing zebrafish embryos were acquired on a home-built microscope described in Kobitski et al . , 2015 . In the DSLM ( Keller et al . , 2008; Schmid et al . , 2013 ) , the sample is irradiated with two opposing beams of 488 nm laser light ( LuxX 488–60 laser , Omicron-Laserage Laserprodukte GmbH , Rodgau , Germany ) that were scanned up and down so as to excite fluorescence in the samples in a slab ( ‘light sheet’ ) with an average thickness of 6 . 6 µm across the field of view ( 1038 × 875 µm2 ) . Fluorescence emission from the illuminated region was collected perpendicular to the light sheet by using a water dipping objective ( 16x/0 . 8 w , Nikon ) and projected on a 5 . 5 megapixel sCMOS camera ( Neo , Andor , Belfast , UK ) , yielding 1 . 5 µm lateral image resolution . Image stacks of 500 frames at 2 µm displacement were collected at a rate of 20 frames per second by moving the sample through the light sheet . To achieve optimal image quality across the entire sample , images from two opposing views were acquired for each light sheet position by rotating the samples by 180° . Consequently , a complete 3D image stack was collected every 60 s . For DSLM imaging , dechorionated embryos were embedded in 0 . 1% ( w/v ) low melting agarose in 1x E3 medium , mounted in fluorinated ethylene propylene tubes as previously described ( Kaufmann et al . , 2012 ) , and placed inside a sample chamber . During image acquisition over 15 hr , the temperature in the sample chamber was kept constant at 26°C . Fluorescently labeled nuclei were automatically extracted from the 3D image stacks using the fast segmentation approach described previously ( Stegmaier et al . , 2014 ) . In a post-processing step , the extracted features from complementary rotation images were registered using a bead-based registration technique ( Preibisch et al . , 2010 ) . Moreover , redundant nuclei were fused on the basis of spatial overlap of segments; a feature based object rejection was performed to minimize the amount of false positive detections . Nucleus counts reflect the number of remaining objects after information fusion and object rejection . To track the detected nuclei over multiple time points , we made use of a nearest neighbor tracking algorithm implemented in the open-source MATLAB toolbox SciXMiner ( Mikut et al . , 2017; Stegmaier et al . , 2012 ) . To attain more comprehensive tracks , we used an additional heuristic to reconstruct nuclei that were missing in a single frame and we fused extracted sub-tracks in the spatio-temporal domain by applying an additional nearest neighbor matching on the start and end points of sub-tracks . It should be noted , though , that these tracks did not necessarily reflect perfect tracks on the single cell level over the whole duration , but merely are used to identify and trace the qualitative movements of nuclei in certain regions . To compute the cell density distribution , we counted the number of nuclei located within a bounding sphere with a radius of 40 µm centered on the nucleus of interest . For visualization , we aligned the virtual embryo by manually centering it on the origin and rotating the vegetal pole such that it was centered on the positive y-axis and finally converting the Cartesian coordinates into a spherical projection using azimuth , elevation and radius ( see schematic in Figure 3B ) . This representation allows us to unwrap the embryo by plotting the azimuth ( range: [-pi , pi] ) and the elevation ( range: [-pi/2 , pi/2] ) in a 2D graph . For speed determination of ingressing deep cells , we manually selected a minimum number of ten cells for each condition and tracked their movement over a time window of 100 frames , that is , for 200 min . We included only internalizing deep cells which could be tracked over the entire time frame in both conditions . The time point of ingression onset was visually determined using slice-based maximum intensity projection videos . The estimated speed reflects the arithmetic mean of approximately 1000 displacement steps of the selected cells during the time interval . ATP , ADP and AMP levels in mondoa-mo and mondoa-mis injected embryos were determined at sphere stage , just before the morphological phenotype becomes apparent , using a mass spectrometry approach . Metabolites were extracted from three pools of 30 zebrafish embryos of each condition with 0 . 3 ml of 0 . 1 M HCl in an ultrasonic ice bath for 10 min . The resulting homogenates were centrifuged twice for 10 min at 4°C and 16 . 400 g to remove cell debris . AMP , ADP and ATP were derivatized with chloroacetaldehyde as described ( Bürstenbinder et al . , 2007 ) and separated by reversed phase chromatography on an Acquity BEH C18 column ( 150 mm x 2 . 1 mm , 1 . 7 µm , Waters ) connected to an Acquity H-class UPLC system . Prior to separation , the column was heated to 42°C and equilibrated with five column volumes of buffer A ( 5 . 7 mM TBAS , 30 . 5 mM KH2PO4 pH 5 . 8 ) at a flow rate of 0 . 45 ml min−1 . Separation of adenosine derivates was achieved by increasing the concentration of buffer B ( 2/3 acetonitrile in 1/3 buffer A , by volume ) in buffer A as follows: 1 min 1% B , 1 . 6 min 2% B , 3 min 4 . 5% B , 3 . 7 min 11% B , 10 min 50% B , and return to 1% B in 2 min . The separated derivatives were detected by fluorescence ( Acquity FLR detector , Waters , excitation: 280 nm , emission: 410 nm , gain: 100 ) and quantified using ultrapure standards ( Sigma ) . Data were acquired and processed with the Empower3 software suite ( Waters ) . Energy charge values were calculated for each batch of embryos separately according to ( [ATP] + 0 . 5 [ADP] ) / ( [ATP] + [ADP] + [AMP] ) and were analyzed for differences between conditions by One-Way ANOVA . Bioluminescence measurements in embryos were analyzed using Mann–Whitney U test , as the data were not normally distributed . Data sets with a normal distribution , including RT-qPCR data , velocity comparison of ingressing deep cells , epiboly progression , cell based bioluminescence measurements and energy charge of embryos , were analyzed with a Student’s t-test ( for two groups ) or ANOVA ( for more than two groups ) . Multiple testing was corrected by the Benjamini-Hochberg procedure ( Benjamini and Hochberg , 1995 ) or an adequate ANOVA posttest . Bonferroni’s posttest was chosen if all groups had to be compared , while Dunnett’s posttest was chosen if the groups were compared only to the control . Data were analyzed and visualized by GraphPad software ( GraphPad Software , Inc ) , Microsoft Excel or R ( https://www . r-project . org/ ) . ‘n’ indicates number of biological repeats . Asterisks indicate p-values of *p≤0 . 05 , **p≤0 . 01 and ***p≤0 . 001 .

In most animals , a protein called MondoA closely monitors the amount of glucose in the body , as this type of sugar is the fuel required for many life processes . Glucose levels also act as a proxy for the availability of other important nutrients . Once MondoA has detected glucose molecules , it turns genetic programmes on and off depending on the needs of the cell . So far , these mechanisms have mainly been studied in adult cells . However , recent studies have shown that proteins that monitor nutrient availability , and their associated pathways , can control early development . MondoA had not been studied in this context before , so Weger et al . decided to investigate its role in embryonic development . The experiments used embryos from zebrafish , a small freshwater fish whose early development is easily monitored and manipulated in the laboratory . Inhibiting production of the MondoA protein in zebrafish embryos prevented them from maturing any further , stopping their development at an early key stage . This block was caused by defects in microtubules , the tubular molecules that act like a microscopic skeleton to provide structural support for cells and guide transport of cell components . In addition , the pathway involved in the production of cholesterol and cholesterol-based hormones was far less active in embryos lacking MondoA . Treating MondoA-deficient embryos with one of these hormones corrected the microtubule defects and let the embryos progress to more advanced stages of development . These results reveal that , during development , the glucose sensor MondoA also controls pathways involved in the creation of cholesterol and associated hormones . These new insights into the metabolic regulation of development could help to understand certain human conditions; for example , certain patients with defective cholesterol pathway genes also show developmental perturbations . In addition , the work highlights a biological link between cholesterol production and cellular responses to glucose , which Weger et al . hope could one day help to identify new cholesterol-lowering drugs .

2020

MondoA regulates gene expression in cholesterol biosynthesis-associated pathways required for zebrafish epiboly

Skeletal muscle regenerative potential declines with age , in part due to deficiencies in resident stem cells ( satellite cells , SCs ) and derived myogenic progenitors ( MPs ) ; however , the factors responsible for this decline remain obscure . TGFβ superfamily signaling is an inhibitor of myogenic differentiation , with elevated activity in aged skeletal muscle . Surprisingly , we find reduced expression of Smad4 , the downstream cofactor for canonical TGFβ superfamily signaling , and the target Id1 in aged SCs and MPs during regeneration . Specific deletion of Smad4 in adult mouse SCs led to increased propensity for terminal myogenic commitment connected to impaired proliferative potential . Furthermore , SC-specific Smad4 disruption compromised adult skeletal muscle regeneration . Finally , loss of Smad4 in aged SCs did not promote aged skeletal muscle regeneration . Therefore , SC-specific reduction of Smad4 is a feature of aged regenerating skeletal muscle and Smad4 is a critical regulator of SC and MP amplification during skeletal muscle regeneration . The regenerative capacity of adult skeletal muscle is endowed in a population of Pax7-expressing resident stem cells called satellite cells ( SCs ) ( Brack et al . , 2012 ) . Genetic studies utilizing lineage labeling , as well as cell ablation , have established that Pax7-expressing SCs are essential for various aspects of skeletal muscle regeneration ( Relaix and Zammit , 2012; Liu et al . , 2015 ) . At homeostasis , SCs reside in a quiescent state at the interface between skeletal muscle fibers ( myofibers ) and the surrounding basal lamina ( Brack et al . , 2012 ) . In response to degenerative stimuli , SCs activate and undergo proliferative expansion , providing myogenic progenitors ( MPs ) necessary for myofiber regeneration ( Brack et al . , 2012 ) . Skeletal muscle regeneration requires a balance between SC/MP amplification and terminal myogenic commitment in order to efficiently form multinucleated myofibers ( Brack et al . , 2012 ) . Moreover , there is evidence that this balance is compromised in aging skeletal muscle , where changes in SC function and their surrounding environment occur , yielding defective progenitors and stem cells ( Chakkalakal et al . , 2012; Sousa-Victor et al . , 2015; Almada and Wagers , 2016; Blau et al . , 2015 ) . Therefore , elucidating factors that regulate SC and derived MP fate is critical in order to develop interventions to combat aged skeletal muscle regenerative decline , in which these cells are lost . Transforming growth factor beta ( TGFβ ) superfamily signaling is crucial for the renewal and maintenance of various tissue-specific stem cell populations ( Oshimori and Fuchs , 2012 ) . Superfamily ligands include TGFβs , bone morphogenetic proteins ( BMPs ) , growth and differentiation factors ( GDFs ) , Activins , and Myostatin . These ligands bind to transmembrane type II receptors with differing specificities . This association leads to the eventual formation of transmembrane complexes composed of type I ( activin-like kinases , ALKs ) and type II receptor homodimers ( Massagué , 2008 ) . The formation of these complexes triggers the phosphorylation of receptor SMADs ( R-SMADs ) . The R-SMADs associated with TGFβ/Activin/Inhibin and BMP ligands are SMAD2/3 and SMAD1/5/8 respectively . Phosphorylated R-SMADs , associated with the cofactor SMAD4 , accumulate in the nucleus , where together with chromatin modifiers and other transcriptional co-factors they promote the expression of target genes . Depending on cell type and context , the TGFβ superfamily pathways regulate many cellular processes including differentiation , renewal , quiescence , and apoptosis ( Oshimori and Fuchs , 2012; Massagué , 2012 ) . In aged regenerating skeletal muscle , TGFβ superfamily signaling is widely considered to be abnormally elevated , and is thought to inhibit SC activation and terminal myogenic differentiation ( Kollias and McDermott , 2008; Trendelenburg et al . , 2012 , 2009 ) . However , it has also been proposed that during regeneration TGFβ superfamily ligands may enable SC and MP expansion ( Ono et al . , 2011; Sinha et al . , 2014 ) . Smad4 is recognized as the canonical cofactor for TGFβ superfamily signaling and was initially identified as a tumor suppressor in pancreatic cancer ( Malkoski and Wang , 2012; Miyaki and Kuroki , 2003; Hussein et al . , 2003; Nishita et al . , 2000 ) . Loss-of-function SMAD4 mutations lead to familial juvenile polyposis , which can be associated with hereditary hemorrhagic telangiectasia , and aggressive forms of various cancers ( Malkoski and Wang , 2012; Akhurst , 2004 ) . In myogenic culture systems derived from immortalized cell lines or using sequential pre-plate techniques , knockdown of Smad4 promotes myogenic differentiation ( Dey et al . , 2012; Ono et al . , 2011 ) . Furthermore , global reduction of expression through direct intramuscular injection of Smad4 siRNAs or viral vectors with Smad4 shRNAs into injured mouse skeletal muscle can promote the formation of larger regenerated muscle fibers relative to controls ( Dey et al . , 2012; Lee et al . , 2015 ) . However , given that multiple non-myogenic cell types , such as inflammatory cells and fibro/adipogenic progenitors , also contribute to SC and MP fate decisions during skeletal muscle regeneration; It is unclear which cellular mechanisms promote hypertrophy of regenerated myofibers with non-targeted Smad4 loss . In contrast , specific loss of Smad4 in MPs compromises myogenic differentiation during embryonic skeletal muscle development ( Han et al . , 2012 ) . Additionally , consistent with the critical role for Smad4 in stem and progenitor cell function , targeted deletion of Smad4 in hematopoietic , hair follicle , and neural stem and derived progenitor cell populations leads to their depletion during homeostasis and regeneration ( Karlsson et al . , 2007; Yang et al . , 2009; Mira et al . , 2010 ) . Moreover , targeted loss of Smad4 in myofibers leads to modest deterioration during growth and aggravation of denervation-induced atrophy in adults ( Sartori et al . , 2013 ) . Recently , gain-of-function SMAD4 mutations that prevent ubiquitination and subsequent degradation have been identified as the cause of the rare developmental disorder Myhre syndrome in humans ( Caputo et al . , 2014; Le Goff et al . , 2012 ) . Patients with Myhre syndrome are characterized by short stature , various musculoskeletal abnormalities , and hypertrophied musculature ( Caputo et al . , 2014; Le Goff et al . , 2012 ) . Although Smad4 clearly has crucial roles in skeletal muscle and tissue-specific stem and progenitor cell biology , to date no studies have explicitly examined whether or not there is a cell-autonomous requirement for Smad4 in SCs and derived MPs during skeletal muscle regeneration . In this study we show , in comparison to adult , evidence of failure to induce Smad4 expression in aged SCs and MPs during skeletal muscle regeneration . In order to examine the consequences of cell-specific Smad4 loss , we utilized transgenic mice expressing tamoxifen-inducible Cre recombinase under the control of Pax7 regulatory elements to perform targeted deletion of Smad4 in SCs . We found that specific disruption of Smad4 in adult SCs resulted in insufficient SC and derived MP amplification , which was accompanied by severe deficiencies in adult skeletal muscle regeneration . Unexpectedly , with specific loss of Smad4 in aged SCs in an environment of presumably high TGFβ activity , aged skeletal muscle regeneration was not improved . Deficiencies in aged skeletal muscle regeneration reflect in part a failure or delay of SC or SC-derived MP expansion due to multiple factors . These factors include impaired activation , premature terminal fate commitment , and the occurrence of senescence and apoptosis ( Sousa-Victor et al . , 2015 ) . Since SMAD-dependent signaling and target genes such as Id1 have been implicated in the regulation of the terminal fate and amplification of SC and MP populations ( Ono et al . , 2011; 2012; Clever et al . , 2010 ) , we examined the expression of Smad4 and Id1 in SCs and MPs from regenerating adult and aged skeletal muscles . Initially , we employed previously characterized flow cytometric analysis to examine age-related modification of SMAD4 protein levels in SCs and MPs ( Lin- , Sca1- , ITGA7+ ) isolated from adult and aged , uninjured and regenerating skeletal muscle . Regenerating muscle was examined at five days post injury ( 5dpi ) , a time point when new myofibers are rapidly forming through the expansion , differentiation , and fusion of SC-derived myogenic cells ( Murphy et al . , 2011; Cosgrove et al . , 2014; Bernet et al . , 2014; García-Prat et al . , 2016 ) . To induce skeletal muscle regeneration , a barium chloride ( BaCl2 ) solution was directly injected into tibialis anterior ( TA ) muscles , which is an established model of skeletal muscle degeneration and regeneration ( Murphy et al . , 2011 ) . Relative to SCs from adult uninjured TAs , an approximately 2 . 5-fold increase in SMAD4 protein was observed in SCs and MPs isolated from adult 5dpi TA muscles ( Figure 1A and C ) . In contrast , SMAD4 induction was not detected in SCs and MPs isolated from aged 5dpi relative to uninjured TA muscles ( Figure 1B and C ) . To further substantiate these findings , we conducted RTqPCR analysis of Smad4 expression as well as the SMAD-target Id1 , the loss of which is associated with deficiencies in skeletal muscle regeneration ( Clever et al . , 2010 ) . Both Smad4 and Id1 expression were higher in SCs and MPs from adult when compared to those from aged 5dpi TAs ( Figure 1D and E ) . We were unable to obtain consistent Ct values in the detectable range ( <36 ) for Smad4 or Id1 in SCs from uninjured skeletal muscle indicative of low to negligible expression ( data not shown ) . Therefore , a feature of age-related regenerative decline is the loss of Smad4 induction in SCs and MPs . 10 . 7554/eLife . 19484 . 003Figure 1 . Loss of Smad4 expression in aged satellite cells and myogenic progenitors during muscle regeneration . Representative profiles of SMAD4 protein levels by flow cytometric analysis in SCs and MPs isolated from C57BL6 ( A ) adult and ( B ) aged uninjured and 5-day post injured ( 5dpi ) ( TA ) muscles . ( C ) Quantification of SMAD4 protein levels in adult and aged SCs and MPs from uninjured and 5dpi TA muscles . ( D ) Quantification of Smad4 expression in FACs-sorted SCs and MPs isolated from 5dpi adult and aged TA muscles . ( E ) Quantification of SMAD target Id1 expression in FACs-sorted SCs and MPs isolated from 5dpi adult and aged TA muscles . N = 4 mice , for ( C ) *p<0 . 05 ANOVA and Fisher’s test , ( D ) and ( E ) *p<0 . 05 t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 19484 . 003 To begin investigating the consequences of Smad4 loss in adult SCs , Smad4flox/flox mice were bred with mice expressing a tamoxifen-inducible Cre under the control of Pax7 regulatory elements ( Pax7+/CreERT ) , enabling satellite cell-specific Smad4 disruption . To characterize Smad4 loss in Pax7+/CreERT; Smad4flox/flox ( P7:S4KO ) and Pax7+/+; Smad4flox/flox ( Ctl ) animals , adult mice were administered tamoxifen ( Tmx ) , SCs and MPs from 5dpi TAs were prospectively isolated , and the expression of Smad4 and the SMAD-target Id1 were interrogated ( Figure 2—figure supplement 1 ) . As a validation of successful deletion , flow cytometric analysis revealed that SMAD4 protein induction fails to occur in P7:S4KO SCs from 5dpi relative to uninjured skeletal muscles in comparison to the induction seen in Ctl muscles ( Figures 2A , B and C ) . As expected , we also observed a reduction in Smad4 and Id1 mRNA levels in P7:S4KO when compared to Ctl SCs and MPs from 5dpi TAs ( Figure 2D and E ) . We have also previously demonstrated a very high efficiency of Pax7CreERT recombination in adult SCs ( Liu et al . , 2015 ) . Therefore , both Smad4 expression and function are efficiently lost in P7:S4KO SCs and MPs . 10 . 7554/eLife . 19484 . 004Figure 2 . Disruption of Smad4 expression in P7:S4KO satellite cells and myogenic progenitors during muscle regeneration . Representative profiles of SMAD4 protein levels by flow cytometric analysis in SCs and MPs isolated from adult ( A ) Ctl and ( B ) P7:S4KO uninjured and 5dpi TA muscles . ( C ) Quantification of SMAD4 protein levels by flow cytometry of Ctl and P7:S4KO SCs and MPs from uninjured and 5dpi TA muscles . ( D ) Quantification of Smad4 expression in FACs-sorted SCs and MPs isolated from 5dpi Ctl and P7:S4KO TA muscles . ( E ) Quantification of SMAD target Id1 mRNA levels in FACs-sorted SCs and MPs isolated from 5dpi Ctl and P7:S4KO TA muscles . N = 4 mice , for ( C ) *p<0 . 05 to Ctl , **p<0 . 05 Adult Ctl ANOVA and Fisher’s test , for ( D ) and ( E ) *p<0 . 05 t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 19484 . 00410 . 7554/eLife . 19484 . 005Figure 2—figure supplement 1 . Gating of control and P7:S4KO SCs for intracellular protein analysis . Representative flow cytometry plots of Ctl and P7:S4KO SCs used to examine SMAD4 protein levels in Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 19484 . 005 Previous studies have shown that knockdown of Smad4 or TGFβ superfamily ligand supplementation in myogenic cells is able to promote or impair terminal myogenic commitment , respectively ( Ono et al . , 2011; Dey et al . , 2012; Lee et al . , 2015; Trendelenburg et al . , 2009 ) . To examine the terminal myogenic fate decisions of Smad4-deficient SCs , FACs-purified adult P7:S4KO and Ctl SCs and MPs were cultured for 96 hr in plating media ( 10% horse serum , FGF2 , DMEM ) , and supplemented with vehicle , TGFβ1 , or BMP4 ligands ( Chakkalakal et al . , 2012 , 2014; Zammit et al . , 2004 ) . Based on immunofluorescence analysis , all cells regardless of genotype or supplementation were labeled by either Pax7 ( SC renewal marker ) or Myogenin ( terminal myogenic commitment marker ) ( Figure 3A ) . Consistent with previous reports , both TGFβ1 and BMP4 supplementation led to a lower proportion of Ctl SC-derived cells that were Myogenin-positive ( Figure 3C ) ( Ono et al . , 2011 ) . Notably , cultures derived from P7:S4KO SCs contained a significantly lower proportion of Pax7+ cells; However , we observed a higher proportion of Myogenin+ cells compared to Ctl , regardless of supplementation ( Figure 3B and C ) . We subsequently tested whether the enhanced terminal myogenic commitment observed in P7:S4KO cultures was associated with greater myogenic fusion and formation of immature multinucleated muscle fibers ( myotubes ) . To achieve this , SCs were cultured at high density ( 10 , 000 cells/well ) to ensure adequate numbers of fusion-competent MPs . Cultures were then immunostained with pan anti-skeletal muscle myosin antibody to visualize myotubes ( Figure 3D ) . In accordance with an increased propensity to differentiate , quantification of nuclei within skeletal muscle myosin-expressing cells ( fusion index ) revealed this parameter to be ~two fold greater in P7:S4KO compared to Ctl myotubes ( Figure 3E ) . 10 . 7554/eLife . 19484 . 006Figure 3 . Enhanced terminal myogenic commitment in P7:S4KO satellite cells and myogenic progenitors . ( A ) Representative images of Pax7 ( red ) , Myogenin ( green ) , and DAPI ( blue ) immunofluorescence in FACs-sorted adult Ctl or P7:S4KO 96 hr SC cultures ( plated at 4000 cells/well ) treated with vehicle , TGFβ1 , or BMP4 . ( B ) Quantification of Pax7 immunofluorescence in Ctl and P7:S4KO FACs-sorted SC cultures . ( C ) Quantification of Myogenin immunofluorescence in Ctl and P7:S4KO FACs sorted SC cultures . ( D ) Representative images of Pax7 ( red ) , skeletal muscle myosin ( green ) , and DAPI ( blue ) immunofluorescence in adult FACs-sorted Ctl or P7:S4KO myotube cultures ( plated at 10000 cells/well ) . ( E ) Quantification of fusion index; myonuclei/myosin+ cell . N = 3 cultures , For ( B ) and ( C ) *p<0 . 05 significant to Ctl , **p<0 . 05 significant to vehicle Ctl , ANOVA Fishers test , for ( E ) *p<0 . 05 t-test , scale = 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 19484 . 00610 . 7554/eLife . 19484 . 007Figure 3—figure supplement 1 . Modulations in FGFR and FGF expression following SC-specific Smad4 loss in adult mice . Quantification of ( A ) Fgf1 , ( B ) Fgf2 , ( C ) Fgf6 , ( D ) Fgfr1 , and ( E ) Fgfr4 expression in FACs-purified adult SCs and MPs cultured for five days . N = 3 mice , *p<0 . 05 t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 19484 . 007 Specific Smad4 deletion in embryonic tongue MPs leads to impaired myogenic differentiation attributed to reduced Fgfr4 and Fgf6 expression ( Han et al . , 2012 ) . Given that we observed enhanced terminal myogenic commitment in culture , we then examined the expression levels of pertinent FGF ligands and receptors in five day cultures from Smad4-deleted adult SCs and derived MPs . Although we observed reductions in Fgfr4 and Fgf6 expression , we also identified a decrease in Fgfr1 and an induction of Fgf2 expression ( Figure 3—figure supplement 1 ) . Therefore , unlike what is seen in embryonic MPs from tongue muscle , loss of Fgf6 and Fgfr4 expression as a result of specific Smad4 disruption in adult SCs is associated with increased terminal commitment , reduced Fgfr1 and elevated Fgf2 expression . Next , we assessed the proliferative potential of Ctl and P7:S4KO SC cultures . FACs-purified adult P7:S4KO and Ctl SCs and MPs were cultured in plating media for 72 hr , and pulsed with the thymidine analog EdU for the last 4 hr ( Sousa-Victor et al . , 2014 ) . The vast majority of cells in 72 hr cultures , regardless of genotype , were labeled by the myogenic fate markers Pax7 or MyoD ( Figures 4A , C and D ) . In comparison to Ctl , the proportion of P7:S4KO cells that had incorporated EdU was significantly reduced ( Figure 4B ) . We performed further immunofluorescence analysis of MyoD and Pax7 labeling to assess the fate of Ctl and P7:S4KO SCs in 72 hr cultures ( Figure 4C ) . Although no difference was observed in the proportion of cells that were Pax7+/MyoD- , P7:S4KO cultures displayed higher percentages of Pax7-/MyoD+ cells and a lower proportion of Pax7+/MyoD+ cells ( Figure 4D ) . Collectively , these observations provide additional evidence that specific disruption of Smad4 in SCs promotes terminal myogenic commitment , which is associated with reduced proliferative potential . 10 . 7554/eLife . 19484 . 008Figure 4 . Reduced proliferation in P7:S4KO satellite cells and myogenic progenitors . ( A ) Representative images of MyoD ( green ) , EdU ( grey ) , and DAPI ( blue ) immunofluorescence of FACs-sorted adult Ctl or P7:S4KO 72 hr SC cultures pulsed with EdU for the last 4 hr . ( B ) Quantification of the proportion of cells that are EdU+ . N = 3 cultures , *p<0 . 05 t-test , scale = 50 μm . ( C ) Representative images of MyoD ( green ) , Pax7 ( red ) , and DAPI ( blue ) immunofluorescence of FACs-sorted adult Ctl or P7:S4KO 72 hr SC cultures , scale = 50 μm . ( D ) Quantification of the proportion of cells that are Pax7+ and/or MyoD+ . N = 3 cultures , *p<0 . 05 t-test . ( E ) Representative images of Crystal Violet-stained 7-day cultures of FACs-sorted adult Ctl or P7:S4KO myogenic cells plated at clonal density ( 10 cells/well ) , scale = 200 μm . ( F ) Quantification of cell growth . N = 96 cultures , *p<0 . 05 t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 19484 . 008 To further assess these cellular mechanisms , we sought to determine the consequences of Smad4 loss on SC clonal growth potential . FACs-purified adult Ctl and P7:S4KO SCs were sorted directly into 96 well plates at 10 cells per well , cultured for seven days , and stained with Crystal Violet to enable counting with conventional light microscopy ( Figure 4E ) . Consistent with our observation of reduced proliferative potential , we found the clonal growth potential of P7:S4KO SCs to be significantly reduced compared to Ctl ( Figure 4F ) . To initially examine SC fate in vivo , adult P7:S4KO and Ctl 5dpi TA muscles were processed to assess Pax7+ SC proliferation and number . To identify proliferating Pax7+ SCs at 5dpi , a pulse of the thymidine analog BrdU was i . p . administered at 2 hr prior to sacrifice . Quantification of the proportion of BrdU+/Pax7+ cells revealed a significant reduction in proliferating as well as total Pax7+ SCs in 5dpi P7:S4KO muscles ( Figure 5A–C ) . Additionally , we examined the expression of cell cycle inhibitors Cdkn1a ( p21 ) , Cdkn1b ( p27 ) , and Cdkn2a ( p16 ) . No induction in the expression of cell cycle inhibitors was observed in adult P7:S4KO SCs and MPs isolated by FACs from 5dpi TAs ( Figure 5—figure supplement 1 ) . Furthermore , no significant difference was observed between P7:S4KO and Ctl SC number in uninjured TAs 21 days after Tmx treatment ( Figure 5—figure supplement 2 ) . Therefore , the loss of SCs in P7:S4KO 5dpi TAs does not reflect initial declines in Pax7+ SCs upon recombination at homeostasis . 10 . 7554/eLife . 19484 . 009Figure 5 . Smad4 disruption induces loss of proliferating and total satellite cell number during muscle regeneration . ( A ) Representative images of Pax7 ( red ) , BrdU ( green ) , DAPI ( blue ) and Laminin ( grey ) immunofluorescence in adult 5dpi Ctl and P7:S4KO TA muscle sections . ( B ) Quantification of the proportion of BrdU+ Pax7+ cells in adult 5dpi Ctl and P7:S4KO TA muscle sections . ( C ) Quantification of Pax7+ cells in adult 5dpi Ctl and P7:S4KO TA muscle sections . N = 4 mice , *p<0 . 05 t-test , scale = 50 μm . Pax7+ cells ( white arrows ) and Pax7+BrdU+ ( yellow arrows ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19484 . 00910 . 7554/eLife . 19484 . 010Figure 5—figure supplement 1 . Smad4 disruption does not induce Cdkn1a , Cdkn1b , or Cdkn2a expression in SCs and MPs sorted from adult regenerating TA muscle . Quantification of ( A ) Cdkn1a ( p21 ) ( B ) Cdkn1b ( p27 ) , and ( C ) Cdkn2a ( p16 ) expression in SCs and MPs FACs-sorted from adult 5dpi Ctl and P7:S4KO TA muscles . N = 3 mice , *p<0 . 05 t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 19484 . 01010 . 7554/eLife . 19484 . 011Figure 5—figure supplement 2 . Smad4 disruption does not induce loss of Pax7+ SCs in uninjured TA muscles . ( A ) Representative images of Pax7 ( red ) , DAPI ( blue ) and Laminin ( grey ) immunofluorescence of adult uninjured Ctl and P7:S4KO TA muscle sections . ( B ) Quantification of Pax7+ cells in adult uninjured Ctl and P7:S4KO TA muscle sections . N = 4 mice , scale = 50 μm . Pax7+ cells ( white arrows ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19484 . 011 To examine MP number and fate , adult Ctl and P7:S4KO 5dpi muscle sections were processed for the detection of MyoD and active Caspase-3 ( aCasp , apoptosis marker ) by immunofluorescence ( Figure 6A ) . Consistent with the declines seen in Pax7+ SC proliferation and number , we found that MyoD+ cell number was reduced in P7:S4KO 5dpi muscles ( Figure 6C ) . Although we observed a significant reduction in MyoD+ cells , the proportion of MPs that were aCasp+ was similar in both Ctl and P7:S4KO 5dpi muscles ( Figure 6D ) . Therefore , SC-specific Smad4 disruption leads to impaired SC and MP amplification that does not coincide with heightened cell death . 10 . 7554/eLife . 19484 . 012Figure 6 . Smad4 disruption enhances terminal myogenic commitment during muscle regeneration . ( A ) Representative images of active Caspase 3 ( aCasp , green ) , MyoD ( red ) , DAPI ( blue ) , and Laminin ( grey ) immunofluorescence of adult 5dpi Ctl and P7:S4KO TA muscle sections . MyoD+ cells ( white arrows ) and MyoD+aCasp+ ( yellow arrows ) . ( B ) Representative images of Myogenin ( MyoG , green ) , Pax7 and MyoD ( red ) , DAPI ( blue ) and Laminin ( grey ) immunofluorescence of adult 5dpi Ctl and P7:S4KO TA muscle sections . ( C ) Quantification of MyoD+ cells in 5dpi Ctl and P7:S4KO TA muscle sections . ( D ) Quantification of the proportion of aCasp+ MyoD+ cells in 5dpi Ctl and P7:S4KO TA muscle sections . ( E ) Quantification of the proportion of MyoG+ Pax7 and MyoD+ cells in 5dpi Ctl and P7:S4KO TA muscles . Pax7 and MyoD+ cells ( white arrows ) and Pax7 and MyoD+ MyoG+ ( yellow arrows ) MyoG+ ( green arrows ) . N = 4 mice , *p<0 . 05 t-test , scale = 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 19484 . 012 To determine if reduced SC and MP amplification upon Smad4 loss coincides with an increased propensity for terminal commitment in vivo , adult Ctl and P7:S4KO 5dpi muscle sections were processed for the immunofluorescent detection of Pax7 and MyoD together ( Pax7+MyoD ) and Myogenin ( MyoG ) , recognized with separate fluorescent-conjugated secondary antibodies ( Figure 6B ) . The examination of Pax7/MyoD+ cells with Myogenin labeling enables the detection of total myogenic cells and based on the proportion that are Myogenin+ , the extent of terminal myogenic commitment can be determined . Consistent with our SC-derived culture data , we found that a significantly higher proportion of myogenic cells were terminally committed in P7:S4KO 5dpi muscles ( Figure 6E ) . Therefore , reduced SC and MP amplification resulting from Smad4 loss coincides with an increased drive toward terminal myogenic commitment during muscle regeneration . To assess the consequences of Smad4 deletion on proper muscle regeneration , we examined the size of Ctl and P7:S4KO 5dpi regenerated myofibers identified with embryonic Myosin Heavy Chain ( eMyHC , marker of recent regeneration ) and Laminin staining by immunofluorescence ( Figure 7A ) . In agreement with what would be expected given the impairment in SC-derived MP amplification , the size ( cross-sectional area ) of eMyHC-positive myofibers was reduced in adult 5dpi regenerated skeletal muscles following SC-specific Smad4 disruption ( Figure 7C ) . 10 . 7554/eLife . 19484 . 013Figure 7 . Smad4 disruption in satellite cells impairs skeletal muscle regeneration . Representative images of embryonic Myosin Heavy Chain ( eMyHC , green ) , DAPI ( blue ) and Laminin ( grey ) immunofluorescence of 5dpi Ctl and P7:S4KO ( A ) adult and ( B ) aged TA muscle sections . ( C ) Quantification of average eMyHC+ regenerated myofiber size in 5dpi adult and aged Ctl and P7:S4KO TA muscles . N = 4 mice , 250–300 myofibers . *p<0 . 05 ANOVA Fishers test , scale = 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 19484 . 01310 . 7554/eLife . 19484 . 014Figure 7—figure supplement 1 . Smad4 disruption reduces Smad4 and the SMAD target Id1 expression in adult and aged SCs and MPs sorted from regenerating TA muscle . Quantification of ( A ) Smad4 , ( B ) Id1 and ( C ) Cdkn2a ( p16 ) mRNA levels in SCs and MPs FACs-sorted from 5dpi adult and aged Ctl and P7:S4KO TA muscle sections . N = 3 mice , *p<0 . 05 ANOVA Fishers test . DOI: http://dx . doi . org/10 . 7554/eLife . 19484 . 014 In aged regenerating skeletal muscle , elevated TGFβ activity is thought to impair SC and myogenic progenitor amplification ( Carlson et al . , 2008; Pessina et al . , 2015 ) . Furthermore , non-specific knockdown of Smad4 through siRNAs or viral delivery of shRNAs to regenerating aged skeletal muscle promotes myofiber hypertrophy ( Dey et al . , 2012; Lee et al . , 2015 ) . Therefore , we sought to examine whether specific disruption of Smad4 in aged SCs could promote aged skeletal muscle regeneration , presumably in an environment of high TGFβ ligand activity , suggesting that the implications of signaling loss could differ from results observed in adult mice . To test this hypothesis , P7:S4KO and Ctl mice were aged to 22 months and treated with Tmx . Thereafter , a TA muscle was degenerated with intramuscular injection of BaCl2 , and subsequently allowed to regenerate for five days . In accordance with Smad4 disruption in adult mice , expression of Smad4 and the SMAD-target Id1 was reduced in FACs-isolated SCs and MPs from aged P7:S4KO 5dpi TAs , validating efficient and similar Cre-mediated Smad4 deletion at both ages ( Figure 7—figure supplement 1A and B ) . As described elsewhere , aged SCs displayed elevated expression of Cdkn2a ( p16 ) , however the extent of increased expression did not differ in response to Smad4 disruption ( Figure 7—figure supplement 1C ) ( Sousa-Victor et al . , 2014; Bernet et al . , 2014; Cosgrove et al . , 2014; Carlson et al . , 2008 ) . Based on quantification of regenerated eMyHC-positive myofiber size , specific loss of Smad4 in aged SCs did not promote aged regeneration ( Figure 7B and C ) . To determine if persisting regenerative deficits could be detected in aged and adult P7:S4KO mice , additional TA muscles were analyzed at 14 days following BaCl2 injury , a timepoint at which regeneration should essentially be complete . Regardless of genotype , uninjured TAs demonstrated similar age-related myofiber atrophy ( Figure 8A , B and C ) . Examination of 14dpi TAs revealed that , at all ages , insufficient muscle regeneration persisted following SC-specific Smad4 loss ( Figure 8B and D ) . Collectively , these data identify Smad4 as a critical muscle stem cell regulator , maintaining the appropriate balance between SC-derived MP amplification and terminal commitment during aged and adult skeletal muscle regeneration . 10 . 7554/eLife . 19484 . 015Figure 8 . Smad4 disruption in satellite cells leads to persistent deficits in skeletal muscle regeneration . Representative images of H and E-stained uninjured and 14dpi , and Sirius Red-stained 14dpi ( A ) adult and ( B ) aged TA muscle sections . Quantification and frequency distribution of ( C ) uninjured and ( D ) 14dpi myofiber size in adult and aged Ctl and P7:S4KO TA muscles . N = 4 mice , 700–900 myofibers . For ( C ) *p<0 . 05 to Adult , for ( D ) *p<0 . 05 to Ctl , **p<0 . 05 to Adult Ctl , ANOVA Fisher’s test , scale = 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 19484 . 015 Here we find Smad4 to be a factor that is normally induced in adult , but lost in aged , SCs and MPs in regenerating muscles . Furthermore , we demonstrate that in an inducible mouse model driving specific loss of Smad4 in SCs , a decline in the number of proliferating Pax7+ SCs and MPs , and consequently severe deficits in skeletal muscle regeneration are experienced regardless of age . Several studies have specifically manipulated components of TGFβ superfamily signaling in myogenic progenitors and SCs ( Guardiola et al . , 2012; Han et al . , 2012; Huang et al . , 2014 ) . However , until now , no reports have examined the cell-autonomous roles of Smad4 , the common cofactor for all branches of canonical TGFβ superfamily signaling , in adult or aged Pax7+ SCs and determined the subsequent consequences on skeletal muscle regeneration . Consistent with previous reports , Smad4 disruption did promote myogenic terminal commitment ( Dey et al . , 2012; Ono et al . , 2011; Lee et al . , 2015 ) . However , declines in proliferating SCs and MPs during regeneration were also observed . Myofibers are multinucleated cells formed by the fusion of many terminally committed myogenic cells . Therefore , the progression to the formation of mature multinucleated myofibers can be impeded at multiple levels , including impaired myogenic differentiation and/or reduced MP amplification through a variety of mechanisms: inhibition of cell cycle entry , cell death , and/or premature terminal commitment ( Sousa-Victor et al . , 2015; Brack et al . , 2012 ) . Given that we have detected differences in myogenic cell proliferation with no increased cell death or cell cycle inhibitor expression , we suspect that the reduced number of MPs in regenerating skeletal muscles following acute SC-specific Smad4 loss is likely due to a heightened propensity for terminal commitment . In contrast , specific disruption of Smad4 in embryonic mouse MPs , driven by Myf5-Cre , altered terminal myogenic commitment and myofiber formation in the tongue ( Han et al . , 2012 ) . Intuitively , these divergent results could reflect differences in the intrinsic mechanisms that regulate SCs and MPs in response to the environments of embryonic tongue growth versus adult TA regeneration . Another possibility is that the Myf5-Cre , in addition to targeting the myogenic lineage , drives recombination in further resident cell populations with the capacity to directly or indirectly influence skeletal muscle growth and regeneration ( Huang et al . , 2014 ) . It will be of interest to determine in what manner specific disruption of Smad4 in other resident non-myogenic cell populations may influence adult or aged skeletal muscle regeneration . Although in adult Smad4 null SC cultures we observed reductions in Fgf6 and Fgfr4 expression similar to embryonic Smad4 deleted muscle , we also detected a robust induction of Fgf2 . We have previously found that the presence of elevated levels of FGF2 in aged skeletal muscle is associated with a propensity for aged SCs to progress toward terminal fates ( Chakkalakal et al . , 2012 ) . Therefore , compensatory induction of FGF2 resulting from Smad4 loss may be a unique feature of adult SCs that drives terminal myogenic commitment , the mechanisms of which will require further investigation . Previous work has shown that loss of BMP signaling induces premature terminal myogenic commitment , preventing SC and MP amplification ( Ono et al . , 2011 ) . Furthermore , Id1 , which we found to be reduced in aged wildtype and adult P7:S4KO SCs and derived MPs , was identified as a SMAD target required for SC and MP amplification ( Ono et al . , 2011 ) . Indeed , through direct interactions Id1 can also promote Myogenin degradation ( Vinals and Ventura , 2004 ) . Therefore , loss of Id1 in aged and P7:S4KO SCs and MPs during regeneration could potentially stabilize Myogenin and thus promote excessive terminal commitment that impedes myogenic cell amplification . Although we observe loss of Smad4 and Id1 expression , some reports have described the presence of elevated TGFβ and phospho-Smad activity in aged SCs and MPs from regenerating skeletal muscle ( Carlson et al . , 2008; Pessina et al . , 2015 ) . Although less characterized , TGFβ can function through non-canonical Smad4-independent pathways ( Derynck and Zhang , 2003; Massagué , 2012 ) . For instance , pSmad3 , which is elevated in aged SCs and MPs , also associates with Drosha in a Smad4-independent complex required for microRNA processing ( Davis et al . , 2008 ) . Studies examining keratinocyte differentiation have shown Iκβ kinase to be a critical regulator of signaling through Smad2/3 that is Smad4-independent ( Descargues et al . , 2008 ) . Non-canonical TGFβ mediators such as TAK1 ( TGFβ activated kinase 1 ) and the downstream target p38 are both shown to be elevated in aged myogenic cells ( Cosgrove et al . , 2014; Bernet et al . , 2014; Trendelenburg et al . , 2012 ) . Notably , heightened p38 activity in aged SCs hinders myogenic cell amplification through mechanisms that include promotion of terminal myogenic commitment ( Cosgrove et al . , 2014; Bernet et al . , 2014 ) . Collectively , it will be critical to determine if loss of Smad4 in aged and P7:S4KO SCs and MPs could impair myogenic cell amplification through promotion of multiple Smad4-independent pathways . Although the TGFβ superfamily pathways are capable of influencing multiple cell types that contribute to the regeneration of skeletal muscle , few studies have examined specific inhibition of these pathways in adult or aged SCs and their derived progenitors . For example , it has been shown that SC-specific loss of Cripto , an inhibitor of Activin , Myostatin , and TGFβ signaling in myogenic cells , leads to smaller regenerated myofibers ( Guardiola et al . , 2012 ) . Additionally , genetic ablation of the BMP receptor Alk3 in embryonic myogenic cells , utilizing MyoD-Cre and Myf5-Cre mice , was found to result in ineffective adult skeletal muscle regeneration , however this occurs primarily through mechanisms other than major dysfunction within the SC pool ( Huang et al . , 2014 ) . This conclusion was based primarily on the lack of observable depletion or dysfunction of Alk3 null SCs in vivo or in culture , respectively , which is thought to occur due to stimulation of the expression of the related Alk6 as a compensatory mechanism to maintain SC function ( Huang et al . , 2014 ) . In this study , we determined that inducible Pax7-driven loss of Smad4 impedes SC-derived myogenic proliferation and proper skeletal muscle regeneration . Further studies will be needed to determine in what way targeted disruption of Alk3 , induced later in life , in adult or aged SCs may affect skeletal muscle regeneration at these ages , with potential compensation possibly being eliminated . Although not specifically targeting SCs or MPs , intramuscular delivery of Smad4 siRNA or shRNA at later time points during regeneration has been shown to stimulate myofiber hypertrophy ( Dey et al . , 2012; Lee et al . , 2015 ) . Furthermore , utilizing miR-26 disruption to increase , and miR-431 mimics to decrease , Smad4 expression at later timeponts during regeneration promotes myofiber atrophy and hypertrophy respectively ( Dey et al . , 2012; Lee et al . , 2015 ) . However , it is intuitive that since microRNAs often target many genes , the aforementioned manipulations likely do not alter Smad4 expression alone . Therefore , it remains unclear whether the balance between myogenic progenitor expansion and terminal fate commitment could be manipulated to be beneficial to adult and aged muscle if cell-specific Smad4 loss was induced at different timepoints during regeneration or by alternative , more transient , means . Here , we demonstrate that induction of Smad4 is compromised in aged SCs and myogenic cells during skeletal muscle regeneration . Disruption of Smad4 specifically in adult SCs leads to phenotypes observed in aged regenerating skeletal muscle including compromised proliferation and amplification of SC-derived MPs . Collectively , these data indicate that Smad4 function in SCs is essential for adult and aged skeletal muscle regeneration in the mouse . Moving forward , careful dissection of Smad4-mediated pathways and targets may yield novel factors that promote skeletal muscle genesis in the contexts of aging and disease . All procedures involving mice were carried out in accordance with guidelines set by the Animal Care and Use Committee at the University of Rochester . Adult ( 3–6 months ) and aged ( 22–24 months ) mice were housed in the animal facility with free access to standard rodent chow and water . All mouse strains were obtained from Jackson Laboratory ( Bar Harbor , ME ) : C57BL/6J , Pax7CreER ( 017763 ) , and Smad4flox/flox ( 017462 ) . Aged C57BL6 mice were obtained from the National Institute on Aging . PCR genotyping was performed using protocols described by Quanta , with primer sequences and annealing temps provided by JAX . To induce Cre recombination , mice were injected i . p . with 100 μL of 20 mg/mL ( ~60 μg/kg ) tamoxifen ( Tmx , Sigma-Aldrich , St . Louis , MO , T5648 in 90% Sigma corn oil/10% EtOH ) for five consecutive days , with clearance for five days prior to injury . To examine Pax7+ SC numbers at homeostasis , muscles were harvested 21 days after Tmx administration . Tmx-injected Pax7+/+;Smad4flox/flox littermates in each cohort were used as controls . Mice were anesthetized with an i . p . injection of ketamine ( 100 mg/kg ) and xylazine ( 10 mg/kg ) or by 1–3% isoflurane inhalation . Buprenorphine ( 0 . 1 mg/kg ) was administered prior to the procedure and approximately every 12 hr as needed . The skin overlaying the tibialis anterior ( TA ) muscle was shaved and the TA was directly injected with a 1 . 2% solution of BaCl2 in normal saline . At five days post-injury , injured and contralateral ( uninjured ) TAs were collected . TAs collected for immunofluorescence were incubated at 4°C overnight in 30% sucrose prior to embedding in OCT and flash freezing . To obtain highly purified SCs and MPs , primary cells were isolated from regenerating and uninjured muscles as described previously ( Chakkalakal et al . , 2012; Pessina et al . , 2015 ) . Hindlimb muscles were isolated and myofiber fragments were obtained by Type II Collagenase ( Gibco , Carlsbad , CA ) digestion , trituration , and multiple sedimentation . Mononucleated cells were liberated by further Type II Collagenase and Dispase ( Gibco ) digestion , trituration , sedimentation and filtration . Cells were stained with CD31 , Sca1 , CD45 ( BD Biosciences , San Jose , CA , #561410 , #562058 , Biolegend , San Diego , CA , #103132 ) , and Integrin α7 ( AbLab , Vancouver , Canada , clone R2F2 ) fluorescent-conjugated antibodies . Cells were collected using a FACSAria II Cell Sorter ( BD Biosciences ) . Live SCs/MPs were isolated using forward and side scatter profiles , negative selection for DAPI , CD31/45 and Sca1 , and positive selection for Integrin α7 . Injured and contralateral uninjured TAs were digested as described above . SCs/MPs were identified by an alternate cell surface staining panel selecting for Integrin α7 and negatively selecting for CD31 , CD45 , and Sca1 ( Biolegend #102420 , #103132 , #108127 ) . Cell fixation , permeabilization , and intracellular staining of Smad4 ( Santa Cruz Biotechnology , Santa Cruz , CA , sc-7966 PE ) using BD Fixation Buffer ( #554655 ) and BD Phosflow Perm Buffer III ( #558050 ) was carried out according to the BD Phosflow protocol . Analysis of cell surface and intraceullar staining was performed on an LSR II Flow Cytometer ( BD Biosciences ) . Mean Fluorescent Intensity ( MFI ) was quantified with FlowJo software . Negative controls stained with all cell surface markers but not intracellular antibodies were analyzed to assess background staining . Negative control MFI was subtracted to obtain final MFI values . FACs-purified SCs/MPs were plated at 4000 cells per well in eight-well Permanox chamber slides ( Nunc , Rochester , NY ) and cultured for five days in plating media ( 10% Horse Serum , 5 ng/mL FGF2 , DMEM ) . The ligands TGFβ1 ( 10 ng/mL ) and BMP4 ( 10 ng/mL ) , as well as DMSO vehicle ( 1 μL/mL ) , were added to cultures beginning on day 3 . Cultures were immunostained for Pax7 and Myogenin to characterize P7:S4KO and Ctl terminal myogenic commitment . To assess myotube formation , FACs-purified SCs/MPs were plated at 10 , 000 cells per well and cultured for five days in plating media , ensuring sufficient cell density to promote fusion . Myotube cultures were immunostained for skeletal muscle myosin and DAPI to calculate fusion index ( myonuclei per myosin-positive cell ) . To assess clonal cell growth , FACs-sorted SCs were plated at clonal density ( 10 cells per well in 96 well plates ) and the number of Crystal Violet-stained cells present in each individual well was determined after seven days in culture . In SC/MP cultures ( plated at 5000 cells per well ) , proliferative capacity was assayed by EdU ( 5-ethynyl-2'-deoxyuridine ) incorporation using the Click-iT EdU Alexa Fluor 647 Imaging Kit ( Molecular Probes , Carlsbad , CA , C10640 ) . Cells were grown in basal media for 72 hr and incubated with EdU for the last 4 hr . EdU detection was performed according to manufacturer protocol and was followed by immunostaining for Pax7 and MyoD . In vivo cell proliferation in Ctl and P7:S4KO mice was assayed by immunostaining for Pax7 , BrdU ( 5-bromo-2’-deoxyuridine ) , and Laminin after 250 μL of 2 mg/ml ( ~15 μg/kg ) BrdU ( Sigma ) was i . p . injected two hours prior to harvest of injured and contralateral uninjured TA muscles . Dissected TA muscles were incubated overnight at 4°C in 30% sucrose , flash frozen , cryosectioned at 10 μm , and stored at −80°C prior to staining . Muscle sections were fixed for 3 min in 4% paraformaldehyde ( PFA ) , and if needed , subjected to antigen retrieval: incubation in citrate buffer ( 10 mM sodium citrate , pH 6 . 0 ) in a steamer ( Oster #5712 ) for 15 min after 10 min preheating of buffer ( Liu et al . , 2015; Tang et al . , 2007 ) . Tissue sections were permeabilized with PBS-T ( 0 . 2% Triton X-100 ) for 10 min and blocked in 10% Normal Goat Serum ( NGS; Jackson Immuno Research , West Grove , PA ) in PBS-T for 30 min at room temperature . When mouse primary antibodies were used , sections were additionally blocked in 3% AffiniPure Fab fragment goat anti-mouse IgG ( H+L ) ( Jackson Immuno Research ) with 2% NGS in PBS at room temperature for 1 hr . Primary antibody incubation in 2% NGS/PBS was carried out at 4°C overnight or 2 hr at RT and sections were incubated with secondary antibodies in 2% NGS/PBS for 1 hr at RT . DAPI staining was used to label nuclei . All slides were mounted with Fluoromount-G ( SouthernBiotech , Birmingham , AL ) . At least four sections from three slides were analyzed per sample . Immunocytochemistry was performed following the same protocol with the exception of the Fab blocking step . Sections and cells were imaged on a Zeiss Axio Observer A . 1 microscope ( Germany ) . Sections of TA muscles were harvested at five or 14 days post-injury from adult or aged Ctl and P7:S4KO mice . Regenerated skeletal muscles harvested five days after injury were immunostained with embryonic Myosin Heavy Chain ( eMyHC ) antibodies to label actively regenerating fibers and average cross-sectional area of 250–300 myofibers was quantified with ImageJ . Regenerated skeletal muscles harvested 14 days after injury were processed for H and E or Sirius Red staining . For H and E , flash-frozen sections were fixed for 3 min in 4% PFA , stained with Mayers Hematoxylin and Alcoholic Eosin Y , dehydrated , equilibrated with xylene and mounted using Cytoseal 60 ( Richard-Allan Scientific , Kalamazoo , MI ) . For Sirius Red staining , a Picrosirius Red stain kit ( Polysciences , Warrington , PA ) was utilized . Frozen sections were fixed for 1 hr at 56°C in Bouin’s fixative , washed in water , stained for 1 hr in Picrosirius Red , washed in 1 M HCl , dehydrated , equilibrated and mounted . Bright-field images were collected with a Zeiss Axioskop 40 microscope . To obtain quantification of average cross-sectional area and frequency distribution of 14dpi regenerated fiber size , Myosin/Laminin immunostained TA sections were analyzed using ImageJ software . RNA was isolated from sorted SCs/MPs , or sorted SC/MP 5-day cultures for FGF analysis using phase separation in Trizol ( Invitrogen , Carlsbad , CA ) followed by cleanup with the RNeasy Plus Minikit ( Qiagen , Germany ) , according to manufacturer protocols . To prepare sorted cell RNA for RT-qPCR , first-strand complementary DNA was synthesized from ~50 ng of RNA using the SuperScript First-Strand cDNA Synthesis Kit ( Invitrogen ) . RT-qPCR was performed on a Step One Plus Real Time PCR machine ( Applied Biosystems , Carlsbad , CA ) using Platinum SYBR Green qPCR SuperMix-UDG with ROX master mix ( Invitrogen ) . Experiments were standardized to Gapdh . All reactions for RT-qPCR were performed using the following thermal cycler conditions: 50°C for 2 min , 95°C for 2 min , 40 cycles of a two-step reaction , denaturation at 95°C for 15 s , annealing at 60°C for 30 s . The following primers were used: Primer name Forward sequence Reverse sequence GapdhAGGTCGGTGTGAACGGATTTGTGTAGACCATGTAGTTGAGGTCSmad4ACACCAACAAGTAACGATGCCGCAAAGGTTTCACTTTCCCCAId1CCTAGCTGTTCGCTGAAGGCCTCCGACAGACCAAGTACCACCdkn1a ( p21 ) TCGCTGTCTTGCACTCTGGTGTCCAATCTGCGCTTGGAGTGATAGCdkn1b ( p27 ) TCAAACGTGAGAGTGTCTAACCCGGGCCGAAGAGATTTCTGCdkn2a ( p16 ) CGCAGGTTCTTGGTCACTGTTGTTCACGAAAGCCAGAGCGFgf1CCCTGACCGAGAGGTTCAACGTCCCTTGTCCCATCCACGFgf2GCGACCCACACGTCAAACTATCCCTTGATAGACACAACTCCTCFgf6CAGGCTCTCGTCTTCTTAGGCAATAGCCGCTTTCCCAATTCAFgfr1GCCTCACATTCAGTGGCTGAAGAGCACCTCCATTTCCTTGTCGGFgfr4TCCGACAAGGATTTGGCAGACCTGGCGGCACATTCCACAATCAC The following antibodies were used: Mouse anti-Pax7 ( 1:100 , Developmental Studies Hybridoma Bank ( DSHB ) , Iowa City , IA ) , rabbit or mouse anti-MyoD ( 1:250 , Santa Cruz , sc-304 or BD Biosciences #554130 ) , rabbit anti-Myogenin ( 1:250 , Santa Cruz , sc-576 ) , mouse anti-embryonic Myosin Heavy Chain BF-45/F1 . 652 ( 1:40 , Developmental Studies Hybridoma Bank ( DSHB ) , Iowa City , IA ) , rat anti-BrdU ( 1:250 , Abcam , Cambridge , UK , ab6326 ) , rabbit anti-Cleaved Caspase 3 ( 1:400 , Cell Signaling , Beverly , MA , #9664 ) , rat or rabbit anti-Laminin ( 1:1000 or 1:1500 , Sigma-Aldrich , L0663 or L9393 ) , rabbit anti-skeletal muscle myosin ( 1:250 , Sigma-Aldrich HPA1239 ) , AlexaFluor 594-conjugated goat anti-mouse IgG ( 1:1500 , Life Technologies , Carlsbad , CA , A-11032 ) , AlexaFluor 488-conjugated goat anti-mouse IgG ( 1:1500 , Life Technologies , A-11001 ) , AlexaFluor 488-conjugated goat anti-rabbit IgG ( 1:1500 , Life Technologies , A-11034 ) , AlexaFluor 488-conjugated goat anti-rat IgG ( 1:1500 , Life Technologies , A-11006 ) , AlexaFluor 647-conjugated goat anti-rat IgG ( 1:1500 , Life Technologies , A-21247 ) , AlexaFluor 647-conjugated goat anti-rabbit IgG ( 1:1500 , Life Technologies , A-21244 ) . Immunofluorescent images were analyzed using ImageJ software . Results are presented as mean + SEM . Statistical significance was determined by Student’s t-tests for simple comparison or by one-way ANOVA and Bonferroni multiple comparisons test for multiple comparisons with Graph Pad Prism software . p<0 . 05 was considered statistically significant .

Even in adulthood , injured muscles can repair themselves largely because they contain groups of stem cells known as satellite cells . These cells divide to produce progenitor cells that later develop , or differentiate , into new muscle fibers . However as muscles get older , this repair process becomes less effective , in part because the satellite cells do not respond as strongly to injury . It remains obscure precisely why the repair process declines with age . A protein called TGFβ is part of a signaling pathway that prevents the muscle progenitor cells from differentiating into muscle fibers , and TGFβ signaling is overactive in older muscles . Most TGFβ signaling operates via a protein called Smad4 , and Paris et al . now show that older satellite cells and progenitor cells from the muscles of old mice produce less Smad4 when they are regenerating . Next , the gene for Smad4 was deleted specifically from the satellite cells of mice . By examining the fate of these cells , Paris et al . found that Smad4 normally maintained the population of satellite cells by preventing them from differentiating into muscle fibers too soon . This was the case when both adult and aged muscle was regenerating . All in all , Smad4 is clearly important for directing satellite cells to regenerate properly; aged cells have less Smad4 and are less able to regenerate . Future studies are now needed to determine how disrupting Smad4 in other resident cell types may influence the regeneration of muscles in mice .

2016

Smad4 restricts differentiation to promote expansion of satellite cell derived progenitors during skeletal muscle regeneration

Long-term potentiation ( LTP ) is arguably the most compelling cellular model for learning and memory . While the mechanisms underlying the induction of LTP ( ‘learning’ ) are well understood , the maintenance of LTP ( ‘memory’ ) has remained contentious over the last 20 years . Here , we find that Ca2+-calmodulin-dependent kinase II ( CaMKII ) contributes to synaptic transmission and is required LTP maintenance . Acute inhibition of CaMKII erases LTP and transient inhibition of CaMKII enhances subsequent LTP . These findings strongly support the role of CaMKII as a molecular storage device . The mechanism by which the brain stores information has fascinated neuroscientists for well over a century . Two general ideas have emerged . The first proposes that information is held by ongoing activity in neuron ensembles . This , indeed , underlies short-term working memory ( D’Esposito and Postle , 2015; Inagaki et al . , 2019; Wang , 2001 ) . However , it is clear that such a mechanism cannot account for long lasting , enduring memories . For instance , it is well established that patients who recover from prolonged periods of isoelectric EEG and brain stem silence following barbiturate overdose have no detectable cognitive deficits , for example , Bird and Plum , 1968 . The second and generally accepted view , first enunciated by Cajal over a century ago ( Cajal , 1911 ) , is the modification of neuronal connections . How these changes might be stored has remained a mystery . Over the past few decades , Ca2+-calmodulin-dependent kinase II ( CaMKII ) has emerged as a very attractive molecular candidate for information storage ( Bayer and Schulman , 2019; Bhattacharyya et al . , 2020; Hell , 2014; Kennedy , 2013; Lisman et al . , 2002; Lisman et al . , 2012; Lisman and Goldring , 1988 ) . This dodecameric kinase is activated by Ca2+/CaM resulting in autophosphorylation and the activity of the kinase remains after the removal of Ca2+ ( Miller and Kennedy , 1986 ) , a state referred to as autonomy . Furthermore , evidence suggests that the enzyme can undergo activation-triggered subunit exchange such that subunits that did not experience the original Ca2+ transient are phosphorylated , a mechanism postulated to allow for continued phosphorylation in the face of protein turnover ( Bhattacharyya et al . , 2020; Bhattacharyya et al . , 2016; Stratton et al . , 2014 ) Thus , CaMKII has many attractive biochemical features expected for a memory molecule . What is the physiological evidence for such a role ? Long-term potentiation ( LTP ) , in which brief high-frequency synaptic stimulation results in a lasting increase in synaptic strength , is an attractive cellular model for learning and memory ( Choquet , 2018; Collingridge et al . , 2004; Huganir and Nicoll , 2013; Malinow and Malenka , 2002; Nicoll , 2017 ) . A role for CaMKII in LTP is compelling . Pharmacologically blocking CaMKII ( Malenka et al . , 1989; Malinow et al . , 1989; Otmakhov et al . , 1997 ) or genetically deleting CaMKII ( Giese et al . , 1998; Incontro et al . , 2018; Silva et al . , 1992 ) blocks LTP and constitutively active CaMKII both mimics and occludes LTP ( Lledo et al . , 1995; Pettit et al . , 1994; Pi et al . , 2010; Poncer et al . , 2002 ) . However , despite the attractive biochemical properties of CaMKII , whether CaMKII is involved in the maintenance of LTP and , by extension , memory storage remains problematic . There are two predictions if CaMKII is responsible for LTP maintenance and synaptic memory ( Lisman , 2017; Sanhueza and Lisman , 2013 ) . First , transiently blocking CaMKII after the induction of LTP should cause a long-lasting erasure . Attempts at reversing LTP have failed on numerous occasions ( Buard et al . , 2010; Chen et al . , 2001; Malinow et al . , 1989; Murakoshi et al . , 2017; Otmakhov et al . , 1997 ) , but see Feng , 1995 . Second , if CaMKII is involved in synaptic memory , one would expect it to leave a lasting trace at synapses ( Lisman , 2017; Sanhueza et al . , 2011; Sanhueza and Lisman , 2013 ) . Thus , silencing CaMKII should reduce synaptic transmission . Interestingly , the CaMKII/NMDAR complex does exist under basal conditions ( Leonard et al . , 1999 ) . However , the genetic deletion of CaMKII ( Achterberg et al . , 2014; Giese et al . , 1998; Silva et al . , 1992 ) , but see Hinds et al . , 1998 , or pharmacological inhibition of CaMKII does not affect baseline synaptic responses ( Buard et al . , 2010; Chen et al . , 2001; Feng , 1995; Malinow et al . , 1989; Murakoshi et al . , 2017; Otmakhov et al . , 1997; Wang and Kelly , 1996 ) . Finally , using a FRET-based CaMKII sensor , CaMKII activation in spines after LTP induction persists for only ~1 min ( Lee et al . , 2009 ) . Thus , physiological support for a role of CaMKII in maintaining LTP or synaptic transmission is lacking . Recent findings have prompted us to reevaluate CaMKII’s role in the maintenance of LTP . Deleting CaMKII with CRISPR ( Incontro et al . , 2018 ) or expressing a peptide inhibitor of CaMKII ( Goold and Nicoll , 2010; Sanhueza et al . , 2011 ) cause a substantial reduction in synaptic transmission . Furthermore , studies using a membrane permeable peptide inhibitor of CaMKII ( tatCN21 or antCN27 ) have reported a lasting depression in synaptic transmission ( Barcomb et al . , 2016; Gouet et al . , 2012; Sanhueza et al . , 2011; Sanhueza et al . , 2007 ) and evidence for a reduction in LTP maintenance ( Sanhueza et al . , 2011; Sanhueza et al . , 2007 ) . To reevaluate the role of CaMKII in synaptic memory we have used two different rapidly acting and reversible CaMKII inhibitors . As would be expected if LTP contributes to synaptic memory , inhibition of CaMKII depresses synaptic transmission and the properties of this persistent action of CaMKII are remarkably similar to those of LTP . Furthermore , we demonstrate that applying these inhibitors after inducing LTP fully reverses LTP , indicating that CaMKII is required for the persistence of LTP . Thus , our findings strongly support CaMKII as a molecular storage device . Reasons for why the present results differ from most previous studies are discussed . Previous studies established , both with the expression of inhibitory peptides ( Goold and Nicoll , 2010; Sanhueza et al . , 2011 ) or CRISPR deletion of CaMKII ( Incontro et al . , 2018 ) that CaMKII contributes ~50% to AMPAR responses . In order to study the basis for this activity as well as the role of CaMKII in LTP maintenance , rapid and reversible block of CaMKII is essential . Since peptide inhibitors have proved most effective , this requires making these peptides membrane permeable either by cell-penetrating peptides ( CPPs ) ( e . g . , tat ) or by protein lipidation ( e . g . , myristoylation ) . Delivery of peptides into cells has profound therapeutic potential and therefore has received a great deal of study ( Allen et al . , 2018; LeCher et al . , 2017; Nelson et al . , 2007; Patel et al . , 2019 ) . The effectiveness of CPPs is hotly debated , because , although these peptides clearly enter the cell by endocytosis , it is not certain to what degree these peptides actually have access to the cytoplasm . This appears not to be an issue with lipid modification . As an initial assay we bathed acute slices in a peptide inhibitor and measured the AMPAR/NMDAR ratio . We would expect the ratio to be reduced by ~50% , since expressing these peptides had no effect on NMDAR currents . We first tested the effectiveness of tatCN21 ( 5 μM ) ( Barcomb et al . , 2016; Buard et al . , 2010; Sanhueza et al . , 2011 ) in reducing the AMPAR/NMDAR ratio ( Figure 1—figure supplement 1 ) . In agreement with previous reports , we failed to see an effect on AMPAR responses . Since higher concentrations of tatCN21 have significant nonspecific effects ( Barcomb et al . , 2016 ) , we turned to other peptides . Previous work ( Goold and Nicoll , 2010 ) showed that prolonged ( 1–2 days ) incubation with myr-CN27 ( 10 µM ) caused a ~50% reduction in the AMPAR/NMDAR ratio . We , therefore , repeated this experiment with shorter incubations ( 1–2 hr ) and a lower concentration ( 1 µM ) and found that myr-CN27 reduced the ratio by ~50% ( Figure 1—figure supplement 1 ) . What might account for the lack of effect of tatCN21 ? To address this , we repeated the experiment using myr-CN21 ( 5 μM ) and found that it reduced the ratio by ~50% ( Figure 1—figure supplement 1 ) , suggesting that lipidation is a more effective mode of delivery than CPPs . We , therefore , selected myr-CN27 for our studies . Application of myr-CN27 to acute slices ( 1 μM ) caused a highly reproducible slowly developing depression of synaptic AMPAR mediated EPSCs , which stabilized after approximately 40 min ( Figure 1A , B ) . A number of experiments were carried out to ensure that the effect we observe with myr-CN27 is due to the specific inhibition of CaMKII . First , in many experiments we measured the size of the NMDAR EPSC , by depolarizing the cell to +40 mV before and after the application of the peptide . We found that myr-CN27 had no effect on the NMDAR EPSC ( Figure 1A , red circles , Figure 1B ) . Second , it is important to note that the magnitude of the decrease in transmission seen with myr-CN27 is similar to that reported with the genetic deletion of CaMKIIα ( Incontro et al . , 2018 ) suggesting that myr-CN27 fully silences CaMKII . Finally , we examined the effects of myr-CN27 in cells lacking CaMKIIα ( Figure 1C , D ) . CaMKIIα was deleted using CRISPR ( Incontro et al . , 2018 ) . We simultaneously recorded from control cells and cells lacking CaMKIIα in slice culture . In the absence of CaMKIIα ( green circle ) synaptic responses were ~50% of control ( black circles ) , as expected . Application of myr-CN27 depressed controls cells to the level recorded in cells lacking CaMKIIα ( Figure 1C , D , black circles ) , but it had no additional effect on cells lacking CaMKIIα ( green circles ) ( n = 6 ) . These results demonstrate that the effect of CN27 is fully explained by its selective and complete silencing of CaMKIIα . Previous studies have relied on field potential recording and thus we repeated our experiments with this approach ( Figure 1E ) . We still recorded a depression , but in this case the depression took tens of minutes to develop ( black symbols ) . We interleaved these experiments with whole cell recording and confirmed that the depression is considerably faster ( blue symbols ) . We interpret the difference in time course to the fact that whole cell recordings are made from superficial cells , while field potential responses are generated by populations of neurons throughout the depth of the slice . Penetration of peptides into the slice is expected to be slow . We next examined the effects of AIP , a peptide inhibitor designed from the autoinhibitory domain . We first expressed AIP in neurons in slice culture and compared AMPAR and NMDAR responses to those in simultaneously recorded neighboring control cells ( Figure 2A ) . Similar to the effect of expressing CN27 ( Goold and Nicoll , 2010 ) or applying myr-CN27 ( see above ) , there was a selective ~50% reduction in the AMPAR EPSC , but no change in the NMDAR response . Given these positive results , we bathed acute slices in myr-AIP ( 20 μM ) and measured the AMPAR/NMDAR ratio ( Figure 2B ) ( n = 12 ) . The ratio was significantly reduced . Acute application of myr-AIP in slice culture caused a slowly developing depression in synaptic activity ( Figure 2C , black circles ) ( n = 6 ) and this effect is due to the selective inhibition of CaMKII , because interleaved recording from cells in which CaMKIIα had been deleted ( Figure 2C , green circles ) , myr-AIP no longer had any effect ( n = 5 ) . These findings show that myr-CN27 and myr-AIP selectively and completely inhibit CaMKII , both in acute slices and in slice culture . It is interesting that , while CaMKII is expressed presynaptically and reported to effect transmitter release ( Benfenati et al . , 1992; Hinds et al . , 2003 ) , there was no presynaptic effect of the peptide inhibitors . The time course of inhibition for both myr-CN27 and myr-AIP is slow , requiring 10’s of minutes . Is this due to the slow access of these peptides to the cell interior , or is it due to the slow reversal of the action of CaMKII ? To address this , we turned to a novel light inducible inhibitor of CaMKII , in which the inhibitory peptide AIP2 ( Ishida et al . , 1998 ) is linked to the LOV-Jα helix domain of Phototropin ( referred to as paAIP2 ) ( Murakoshi et al . , 2017 ) . Blue light rapidly exposes the AIP2 peptide , which then returns to its closed inactive conformation in approximately 40 s following light exposure . We expressed paAIP2 in individual neurons with biolistics , which , unlike the myristoylated peptides , has the advantage of limiting the manipulation to individual postsynaptic cells . The synaptic properties of neurons expressing paAIP2 were not different from the properties of neighboring control cells ( Figure 3—figure supplement 1A , B ) ( n = 16 ) . Exposure of these cells to blue light resulted in a slowly developing inhibition of AMPAR responses ( Figure 3A ) ( n = 12 ) , similar to the kinetics of myr-CN27 and myr-AIP . The blue light had no effect on simultaneously recorded control cells ( Figure 3A , black circles ) or in cells coexpressing paAIP2 with CRISPR to delete CaMKIIα ( Figure 3—figure supplement 2 ) ( n = 5 ) , establishing the specificity of paAIP2 . These results with paAIP2 suggest that the slow time course is not due to slow access , but rather due to the slow reversal of the action of CaMKII . The persistent CaMKII activity has two possible origins . First , it could represent ongoing constitutive Ca2+ activation of CaMKII from various sources . It seems unlikely that resting Ca2+ levels ( Maravall et al . , 2000 ) stimulate CaMKII , since the sensitivity of CaMKII to Ca2+ requires considerably higher levels of Ca2+ ( Schulman , 1984 ) . To directly test whether constitutive CaMKII activity is Ca2+ dependent , we loaded cells with the Ca2+ chelator BAPTA ( 15 mM ) . However , myr-CN27 retained its depressive effect ( Figure 4—figure supplement 1A ) . This finding also rules out spontaneous background NMDAR activity as a source for the constitutive CaMKII activity . In addition , acute application of the NMDAR antagonist APV has little effect on AMPAR EPSCs ( Incontro et al . , 2018 ) . The constitutive activity might reflect an LTP process induced by slice preparation , which is known to depolarize cells and release glutamate . To address this possibility , we pretreated animals with a high dose of MK-801 , which is known to block in vivo NMDAR responses ( Davies et al . , 1988 ) . To verify that NMDARs were blocked by this procedure , we recorded NMDAR EPSCs from slices from these animals and found that the NMDAR EPSC was blocked ( Figure 4—figure supplement 1B ) . Having confirmed the absence of NMDAR responses the slices were perfused with MK-801 . Under such conditions myr-CN27 still exerted its inhibitory effect ( Figure 4—figure supplement 1C ) . Second , if CaMKII is involved in synaptic memory , one would expect that LTP acquired when the animal was alive , would contribute to synaptic transmission , as previously proposed ( Lisman , 2017; Sanhueza and Lisman , 2013 ) . If this is the case , there are a number of predictions . One would expect that the transient inhibition of CaMKII should result in long-lasting inhibition , akin to the resetting of a molecular switch . In a number of cells , we monitored the depression over time following the removal of myr-CN27 and found no recovery ( Figure 4A ) . One might argue that myr-CN27 fails to washout of the cells . To circumvent this possibility , we repeated these experiments with paAIP2 whose action rapidly reverses upon terminating the blue light ( ~40 s ) . Again , we failed to observe any recovery ( Figure 4B ) . We confirmed that this was not due to the continued action of the inhibitor , by demonstrating that LTP could be induced following light termination ( see below , Figure 5 ) . If the constitutive activity resulted from prior NMDAR-dependent LTP , one would expect that this inhibitory action should be absent in cells in which NMDARs have been deleted embryonically . We carried out in utero electroporation to express cre recombinase in neurons in GluN1 floxed mice . Transfected cells were recorded at P16–20 in acute slices . We first carried out paired recordings in the AMPAR inhibitor CNQX to ensure that cre expressing neurons lacked NMDAR EPSCs ( Figure 4C , left panel ) . We then compared the action of myr-AIP in transfected cells with that in interleaved control cells and normalized the baseline to 100% . Myr-CN27 had no effect in these cells . Application of myr-AIP had no effect in these cells ( Figure 4C , middle and right panels , green circles and traces ) , whereas it had its normal inhibitory effect in interleaved control cells ( Figure 4C , black circles and traces ) , further supporting the conclusion that the persistent CaMKII activity represents prior LTP . A caveat to these experiments is that the synaptic action of CaMKII requires its binding to the GluN2B C-terminus of the NMDAR . Thus , the loss of GluN2B in the GluN1 lacking neurons could equally explain the loss of effect of myr-CN27 . To address this concern we repeated these experiments , replacing GluN1 with a pore dead GluN1 mutant ( N812W ) ( Amin et al . , 2017 ) . While this mutant assembles normally with GluN2 subunits and traffics normally to the membrane , it generates little current ( Amin et al . , 2017 ) . We carried out in utero electroporation in GluN1 floxed mice to transfect Cre and GluN1 ( N812W ) . Thus , this genetic experiment is the equivalent of having APV present in utero and until acute slices were made at P14–18 . The NMDAR EPSC in transfected neurons was severely reduced , when compared to neighboring control cells recorded in CNQX ( Figure 4D , left panel ) . The depressant effect of myr-CN27 recorded in these transfected cells was absent ( Figure 4D , middle and right panels ) . These results suggest that in behaving animals , cells that lack NMDAR function , but with GluN2B intact , Ca2+ signaling independent of the NMDAR appears incapable of activating CaMKII , at least in regard to its action on synaptic function . Since LTP is saturable , one might expect that when constitutive activity is transiently silenced , the magnitude of LTP should be larger than that in control conditions . To test this prediction , we transfected paAIP2 in utero and prepared acute slices at days P15–25 . The slices were exposed to blue light for 40 min and then simultaneous recordings were made from a transfected cell and a control cell ( Figure 5A ) . In a typical experiment ( Figure 5B ) , as well as in the summary of all experiments ( Figure 5D ) , the size of baseline EPSCs in paAIP2 expressing cells following blue light exposure was , as expected , approximately half of that of the control neighboring cells . LTP was then induced in both cells . To compare the magnitude of LTP in the two cells we normalized the baseline EPSCs to 100% . As is clear , both in the single experiment ( Figure 5C ) and in the summary ( Figure 5E ) , the magnitude of LTP in cells expressing paAIP2 is approximately twice that of control cells . This interaction between the constitutive potentiation and LTP is consistent with their sharing the same underlying mechanism . Taken together these results strongly support the conclusion that while the animal is alive , LTP is acquired and leaves a synaptic memory trace . The experiments presented thus far are all consistent with the notion that synapses acquire LTP that contributes to synaptic transmission . If this is correct , then inhibiting CaMKII after LTP induction must reverse the potentiation . Yet , as discussed in the introduction , this experiment has failed on numerous occasions ( Buard et al . , 2010; Chen et al . , 2001; Malinow et al . , 1989; Murakoshi et al . , 2017; Otmakhov et al . , 1997 ) , but see Feng , 1995 . Thus , it was mandatory that we revisit this crucial experiment . We therefore carried out a ‘two pathway experiment’ , in which the responses from two independent pathways were recorded from a single cell ( Figure 6A ) . Following the recording of baseline responses , LTP was induced in one of the pathways ( Figure 6B , red circles ) by pairing depolarization of the cell to 0 mV with continued synaptic stimulation , while in the other control pathway , stimulation was stopped during the depolarization ( Figure 6B , black circles ) . Once LTP had been established , myr-CN27 was applied and the effect on both pathways recorded . Myr-CN27 caused a complete reversal of established LTP and as expected , a ~50% reduction in the control pathway ( Figure 6B ) . A summary of all the experiments ( Figure 6B , n = 11 ) demonstrates that in the presence of myr-CN27 the LTP pathway and control pathway converge to approximately 50% of the control . This experiment establishes the essential role of CaMKII in maintaining LTP . A previous field potential study using tatCN21 ( 5 μM ) failed to observe a reversal of LTP ( Buard et al . , 2010 ) . We , therefore , repeated our two pathway experiment using field potential recording . Again , we observed that myr-CN27 ( 1 μM ) caused a slow reduction the EPSPs in the control pathway and a reversal of LTP ( Figure 6—figure supplement 1 ) . Give that tatCN21 ( 5 μM ) had no effect on synaptic transmission ( Buard et al . , 2010; Figure 1—figure supplement 1 ) , it is not surprising that it failed to reverse LTP in the previous study ( Buard et al . , 2010 ) . The role of CaMKII in the maintenance of LTP and information storage has been one of the most vexing issues in the field of synaptic plasticity . On the one hand , the biochemical properties of CaMKII have for decades made this molecule an extremely attractive candidate for molecular storage ( Bhattacharyya et al . , 2020; Coultrap and Bayer , 2012; Hell , 2014; Lisman et al . , 2002; Lisman et al . , 2012 ) . On the other hand , numerous physiological experiments over the years have failed to support the role of CaMKII in the maintenance of LTP and by extension memory . To reevaluate the physiological role of CaMKII in synaptic memory , we have used two classes of CaMKII inhibitory peptides . With these inhibitors we find that CaMKII is required for the maintenance of LTP and provide evidence that LTP acquired while the animal was alive leaves a lasting synaptic memory trace . The first class of inhibitory peptide we used is derived from an endogenous CaMKII inhibitory protein referred to as CaMKIINtide ( CN27 ) ( Chang et al . , 1998; Goold and Nicoll , 2010; Pellicena and Schulman , 2014; Vest et al . , 2007 ) . The second class of inhibitory peptide is derived from the autoinhibitory domain of CaMKII ( Bayer et al . , 2001; Leonard et al . , 1999 ) . Although these inhibitory peptides were thought to bind to separate sites ( S and T sites ) , recent structural studies of the binding of GluN2B and other interacting peptides to CaMKII indicate that these peptides use similar interactions to bind across the substrate binding pocket of the CaMKII active site ( Özden et al . , 2020 ) . Thus , at present it is not possible to use these two classes of inhibitors to distinguish between the disruption of GluN2B binding and the blockade of kinase activity . These peptides included a myristoylated version of CN27 ( myr-CN27 ) , AIP ( myr-AIP ) and a recently developed photoactivatable peptide inhibitor ( paAIP2 ) ( Murakoshi et al . , 2017 ) . Both classes of peptides block kinase activity with high affinity ( Chang et al . , 2001; Ishida et al . , 1998 ) , but also interfere with the binding of CaMKII to GluN2B ( Sanhueza et al . , 2011; Vest et al . , 2007 ) . Our experiments clearly establish that inhibitory peptides fully reverse LTP . How long is CaMKII responsible for maintaining LTP ? Experiments addressing the role of constitutive CaMKII in the maintenance of LTP are constrained by the length of recording; approximately 1 hr for whole cell and a few hours for field potentials . However , if LTP is the substrate for memory , it should leave a lasting memory trace at synapses , acquired while the animal is alive ( Lisman , 2017; Sanhueza and Lisman , 2013 ) . A number of studies have reported a lasting depression following transient inhibition of CaMKII ( Barcomb et al . , 2016; Gouet et al . , 2012; Sanhueza et al . , 2011; Sanhueza et al . , 2007 ) . Furthermore , even under basal conditions , CaMKII is found in isolated PSDs ( Petersen et al . , 2003; Strack et al . , 1997 ) and synaptic puncta ( Bayer et al . , 2006 ) and this is in the autophosphorylated state ( Strack et al . , 2002; Trinidad et al . , 2005 ) . As discussed above all three inhibitory peptides ( myr-CN27 , myr-AIP , and paAIP2 ) depress synaptic transmission and this depression requires the presence of CaMKII . A number of trivial explanations , such as ongoing constitutive Ca2+ activation of the enzyme , were excluded . We then focused on the possibility that this constitutive action of CaMKII reflects prior LTP . First , the finding that the inhibition does not affect NMDAR responses is consistent with an LTP mechanism , since LTP preferentially enhances AMPAR responses ( reviewed in Nicoll , 2017 ) . Second , there should be no recovery from the transient inhibition of CaMKII , analogous to the resetting of a molecular switch . This is , indeed , the case . Third , cells in which the NMDAR has been deleted embryonically should be devoid of the constitutive activity . Again , this is the case . This is unlikely to be due to the loss of GluN2B subunit , because cells expressing a pore dead NMDAR mutant , in which assembles with GluN2B , had little CaMKII constitutive activity . This finding suggests that non-NMDAR sources of Ca2+ in the behaving animal are incapable of triggering CaMKII-dependent enhancement in synaptic transmission . This tight linking of NMDAR sources of Ca2+ to CaMKII is critical , because it prevents the degrading of the Hebbian nature of plasticity . Finally , it is well established that LTP is saturable . Thus , if constitutively action CaMKII represents LTP , removing this component should allow for larger LTP . We find that in cells in which CaMKII is transiently blocked LTP is approximately twice as large as in neighboring control cells . Taken together these results suggest that constitutive CaMKII represents LTP acquired at synapses while the animal was alive , thus supporting a role for CaMKII in synaptic memory . Based on the present results the term ‘basal synaptic transmission’ needs to be reevaluated . It has generally been assumed that the synapses studied in a hippocampal slice , in which much of the afferent drive from multiple inputs have been removed in the slicing , are at a ground ‘basal’ state . The present results indicate that the synaptic currents we measure are actually maintained by a persistent enhancement acquired prior to slicing . It should be pointed out that the contribution of CaMKII to synaptic transmission is unlikely to be a simple addition to the overall preexisting excitatory drive onto the cell . It has long been argued that such a scenario would be unstable and quickly saturate ( Bienenstock et al . , 1982; Cooper and Bear , 2012; Fusi et al . , 2005; Morrison et al . , 2008; Toyoizumi et al . , 2014; Turrigiano , 2008 ) . A number of well-established mechanisms exist to maintain stability , including heterosynaptic depression ( Scanziani et al . , 1996 ) , NMDAR-dependent long-term depression ( Malenka and Bear , 2004 ) and on a longer time scale cell wide homeostasis ( Turrigiano , 2008 ) . Thus , it is envisaged that the CaMKII memory trace is embedded in a network of synapses with no overall net change in the excitatory drive onto the cell or network . It is striking to compare the action of paAIP2 on LTP induction ( Murakoshi et al . , 2017 ) and on synaptic transmission ( present study ) . A brief light exposure ( ~1 min ) , delivered immediately prior to inducing LTP , is sufficient to prevent LTP ( Murakoshi et al . , 2017 ) . This finding indicates that the blocking of CaMKII activation by paAIP2 is very rapid . In contrast , the effect on synaptic transmission and the reversal of LTP required 10’s of minutes . What might explain the dramatic difference in the kinetics of inhibition of LTP induction compared to LTP maintenance and constitutive CaMKII ? There is a well-accepted sequence of events for the role of CaMKII in the induction of LTP ( Coultrap and Bayer , 2012; Hell , 2014; Lisman et al . , 2012; Nicoll , 2017 ) . Following NMDAR activation Ca2+ binds to CaM , which then binds to CaMKII activating the enzyme . This causes the autophosphorylation of T286 and the translocation of cytosolic CaMKII to the PSD where it binds to the C-tail of GluN2B . It seems reasonable to expect that under baseline conditions much of CaMKII and paAIP2 are freely diffusible in the spine cytoplasm , as are phosphatases . The activation of paAIP2 would quickly inhibit CaMKII autophosphorylation and its recruitment to the PSD and phosphatases would quickly reverse its effect . During baseline transmission and during the maintenance of LTP a fraction of CaMKII is bound to GluN2B . Evidence suggests that the CaMKII/GluN2B complex is protected from phosphatases ( Cheriyan et al . , 2011; Lisman and Raghavachari , 2015; Mullasseril et al . , 2007 ) . Thus , reversing action of this sequestered CaMKII would be more difficult than preventing phosphorylation of cytosolic CaMKII . The present results could shed light on the underlying mechanism of long-term depression ( LTD ) . Although this topic remains contentious ( Collingridge et al . , 2010; Dore et al . , 2016; Goodell et al . , 2017; Lisman and Zhabotinsky , 2001; Malenka and Bear , 2004; Stein et al . , 2021; Wong and Gray , 2018 ) , the long-held view is that LTD is mediated by the activation of phosphatases , in particular protein phosphatase 1 ( PP1 ) ( Mulkey et al . , 1994; Mulkey et al . , 1993 ) . One of the limitations to this model is the lack of evidence that synaptic transmission is maintained by kinase activity . Our demonstration that CaMKII maintains synaptic transmission provides a very simple model for LTD , that is , a depotentiation or reversal of this prior LTP . One of the features of LTD is that it takes many minutes for its induction . Our finding that it takes many minutes for CaMKII inhibitors to depress synaptic transmission provides an explanation for this property . A formal model for the interplay between CaMKII and PP1 was proposed some time ago ( Lisman and Zhabotinsky , 2001 ) . The absence of LTD in CaMKII knockout mice ( Coultrap et al . , 2014; Stevens et al . , 1994 ) supports such a model , but does not rule out other models . What might account for the failure of previous studies to detect a component of synaptic transmission driven by constitutive CaMKII ( Achterberg et al . , 2014; Buard et al . , 2010; Chen et al . , 2001; Feng , 1995; Giese et al . , 1998; Malinow et al . , 1989; Murakoshi et al . , 2017; Otmakhov et al . , 1997; Silva et al . , 1992; Wang and Kelly , 1996 ) , but see Hinds et al . , 1998 , or to detect a role of constitutive CaMKII in maintaining LTP ( Buard et al . , 2010; Chen et al . , 2001; Malinow et al . , 1989; Murakoshi et al . , 2017; Otmakhov et al . , 1997 ) , but see Feng , 1995 ? We suggest three factors . The first concerns the duration of peptide application , which in our hands takes 10’s of minutes to act , especially with field potential recording . The second concerns the concentration and time . As discussed above it is more difficult to reverse the constitutive action of CaMKII , than it is to block its activation . Finally , the membrane permeabilizing agent is important . In our hands myristoylation was more effective and specific than the CPP , tat . In summary our results overcome many of the obstacles that have prevented embracing CaMKII as a molecular storage device . Specifically , our results show that transient inhibition of CaMKII results in a lasting depression of synaptic transmission with properties consistent with the erasure of prior LTP acquired while the animal was alive . Furthermore , the finding that CaMKII inhibition after the induction of LTP reverses LTP establishes its role in LTP maintenance . Thus , these physiological results compliment the rich biochemical literature on CaMKII , making CaMKII a particularly attractive molecular storage device . It is important to note , that our results have focused on LTP and it remains open as to whether CaMKII actually stores memories . Recent experiments ( Rossetti et al . , 2017 ) using the viral expression of the catalytically dead CaMKII K42M mutant , presumed to act in a dominant negative manner , support its role in memory . All the experimental procedures on animals were approved by the UCSF Animal Care and Use Committee , BUA # BU002466-04C . For acute slice recordings , typically we use 5–10 animals to obtain complete dataset; for culture slice recordings , we usually used 5–7 animals to obtain complete dataset . The paAIP2 plasmid was obtained from Dr . Ryohei Yasuda and had been characterized ( Murakoshi et al . , 2017 ) . The CRISPR construct targeting at CaMKIIα was previously characterized ( Incontro et al . , 2018 ) . For biolistic experiments , all the plasmids were expressed in pCAGGS vector , which contains an internal ribosome entry site ( IRES ) followed by the fluorophore GFP . myr-CaMKIINtide ( myr-CN27 ) and myr-AIP were purchased from Calbiochem . Inc ( catalog# , 208921; catalog# , 189482 ) . MK-801 maleate was purchased from HelloBio . Inc ( catalog# , HB30004 ) . Myr-CN21 was custom ordered from Elim Biopharm , Inc by myristoylating the N-terminal of CN21: KRPPKLGQIGRSKRVVIEDDR amino acid sequence . Hippocampal cultured slices are obtained from 6- to 8-day-old rats ( Stoppini et al . , 1991 ) . Biolistic transfection was done 1 day after sectioning , by using a Helio Gene Gun with 1 μm DNA-coated gold particles ( BioRad ) . Slices were maintained at 34°C and the medium was changed every 2 days . Typically , slices are used for electrophysiological recording 6–8 days after transfection , except for CRISPR experiments , in which slices are maintained for another week before recording . E15 . 5 pregnant mice were anesthetized with 2 . 5% isoflurane in O2 and injected with buprenorphine for analgesia . The lateral ventricles of embryos were injected with 1 µl mix plasmid DNA ( 1 µg/µl ) with beveled micropipette . Each embryo was electoporated with 5 × 50 ms , 35–40 V pulse . Mice of 2–3 weeks of age were anesthetized with 4% isoflurane , decapitated , and the brain dissected free . The whole brain was sliced into 300 µm slices in cutting solution as described ( Granger et al . , 2013 ) ; recovery at 34°C for half an hour and then stored at room temperature . Solutions were continuously gassed with 95% O2/5% CO2 . Blue light pulses ( 0 . 1 Hz , 1 s , 20 mW/cm2 ) were generated by 473 nm blue DPSS laser ( Shanghai Laser & Optics Century , BL473T8-300FC ) . The blue light from a laser was delivered through a optical patch cable connected to the optical fibers . Light pulses were controlled by a Master-8 ( A . M . P . I ) . The blue light is applied for 1-s duration with interval of 10 s . MK-801 was dissolved in saline solution and administered as a single injection i . p . ( 10 mg/kg body weight ) 1 hr before brain slicing . Whole-cell voltage clamp recordings were obtained from either wild-type cells or fluorescent transfected pyramidal cell in CA1 region of hippocampus ( Schnell et al . , 2002 ) . Where indicated dual recordings were made from control and transfected cells . Pyramidal neurons were identified by location and morphology . All recordings were made at 20–25°C . Internal solution ( in mM ) : 135 CsMeSO4 , 8 NaCl , 10 HEPES , 5 QX314-Cl , 4 Mg-ATP , 0 . 3 Na-GTP , 0 . 3 EGTA , 0 . 1 spermine . Osmolarity was adjusted to 290–295 mOm and pH was buffered at 7 . 3–7 . 4 . External solution ( mM ) : 119 NaCl , 2 . 5 KCl , 4 CaCl2 , 4 MgCl2 , 1 NaH2PO4 , 26 . 2 NaHCO3 , 11 glucose , bubbled continuously with 95% O2/5% CO2 . For field recordings , the internal solution was 3 M NaCl with a large opening pipette tip . Synaptic currents were evoked every 10 s with bipolar stimulating electrodes placed in s . radiatum . To record EPSCs , picrotoxin ( 100 µM ) was added to the external solution; for recording of AMPAR EPSCs , the cell membrane was held at −70 mV , while for NMDAR EPSCs , it was hold at +40 mV; LTP is induced by stimulating at 2 Hz for 90 s while clamping the cell at 0 mV . For two independent synaptic pathway experiments , two bipolar stimulating electrodes were positioned to each side of the recording electrode with a distance of around 100 µm and alternately stimulated every 20 s . Current responses were collected with a Multiclamp 700B amplifier ( Axon Instruments ) , filtered at 2 kHz , and digitized at 10 kHz . Cells with series resistance larger than 20 MOhm were excluded from analysis . Data analysis was carried out in Igor Pro ( Wavemetrics ) , Excel ( Microsoft ) , and GraphPad Prism ( GraphPad Software ) . All paired recording data were analyzed statistically with a Wilcoxon Signed Rank Test for paired data . For unpaired data , a Mann–Whitney U test was used . Statistical parameters including the types of the statistical tests used , exact value of n , precision measures ( mean ± standard error of the mean ) and statistical significance are reported in the figure legends . All statistical tests performed were two sided , and with all tests a p value of <0 . 05 was considered statistically significant . All error bars represent standard error of the mean .

How the brain stores information is a question that has fascinated neuroscientists for well over a century . Two general ideas have emerged . The first is that groups of neurons hold information by staying active . The second is that they hold information by strengthening their connections to one another , making it easier for them to work together in the future . Scientists call this second idea 'long-term potentiation' . One of the molecules involved in long-term potentiation is a protein called calcium-calmodulin-dependent kinase II , or CaMKII for short . Blocking CaMKII , or deleting its gene , stops the connections between neurons from becoming stronger . This suggests neurons need CaMKII to learn , but it remains unclear whether neurons also use CaMKII to maintain neuronal memories after they have been created . If CaMKII does play a role in maintaining memories , blocking it after learning should reverse the learning process , but so far , experiments have not been able to show this . Tao et al . revisited these experiments to find out more . They examined slices of brain tissue from mice that had been treated with fast-acting CaMKII inhibitors . It took tens of minutes , but the inhibitors were able to reverse long-term potentiation , both for newly acquired neuronal memories and for older memories that had formed when the mice were alive . The choice of CaMKII inhibitor and the time lag could explain why scientists have not observed the effect before . Understanding long-term potentiation is a fundamental part of understanding learning and memory . It could also reveal more about the opposite phenomenon: long-term depression . This is a type of learning where the connections between neurons become weaker . Long-term depression also takes tens of minutes to occur , suggesting that future research into CaMKII might shed light on how it works .

2021

Synaptic memory requires CaMKII

Phytochrome proteins control the growth , reproduction , and photosynthesis of plants , fungi , and bacteria . Light is detected by a bilin cofactor , but it remains elusive how this leads to activation of the protein through structural changes . We present serial femtosecond X-ray crystallographic data of the chromophore-binding domains of a bacterial phytochrome at delay times of 1 ps and 10 ps after photoexcitation . The data reveal a twist of the D-ring , which leads to partial detachment of the chromophore from the protein . Unexpectedly , the conserved so-called pyrrole water is photodissociated from the chromophore , concomitant with movement of the A-ring and a key signaling aspartate . The changes are wired together by ultrafast backbone and water movements around the chromophore , channeling them into signal transduction towards the output domains . We suggest that the observed collective changes are important for the phytochrome photoresponse , explaining the earliest steps of how plants , fungi and bacteria sense red light . Phytochrome photosensor proteins are crucial for the optimal development of all vegetation on Earth ( Butler et al . , 1959; Gan et al . , 2014; Quail et al . , 1995 ) . Prototypical phytochromes can exist in two photochemical states with differential cellular signaling activity , called red light-absorbing ( Pr ) and far-red light-absorbing ( Pfr ) state ( Figure 1—figure supplement 1 ) . As a result , phytochromes can distinguish two colors of light , providing plants , fungi , and bacteria with primitive two-color vision . Light is detected by a bilin chromophore , which is covalently linked to the photosensory core of the protein ( Wagner et al . , 2005 ) , comprising of PAS ( Per/Arndt/Sim ) , GAF ( cGMP phosphodiesterase/adenyl cyclase/FhlA ) and PHY ( phytochrome-specific ) domains . Two propionate side chains additionally anchor the chromophore non-covalently to the protein ( Figure 1b ) . The signaling sites of the phytochrome are found in its C- and N-terminal output domains , which vary between species . Important for the signaling is a stretch of amino acids in the PHY domain , called the PHY-tongue , which changes from a β-sheet in Pr into an α-helix in Pfr state ( Essen et al . , 2008; Yang et al . , 2008; Takala et al . , 2014; Sanchez et al . , 2019 ) The chromophore connects to the PHY-tongue via a strictly conserved aspartatic acid , which is expected to play a crucial role in signal transduction . Key to phytochrome function is the primary photoresponse on picosecond time scales . Here , light signals are translated into conformational changes . The changes arise in the electronically excited bilin , but must then be transduced to the surrounding protein residues . This prepares the protein for a formation of the first intermediate ( Lumi-R for prototypical phytochromes ) , in which isomerization of the D-ring has likely occurred ( Rüdiger et al . , 1983; Dasgupta et al . , 2009; Yang et al . , 2012; Rockwell et al . , 2009; Ihalainen et al . , 2018 ) . The mechanism that leads to the first intermediate is currently not well understood , because crystallographic observations of phytochromes directly after photoexcitation have not been available . To address this gap of knowledge , we recorded time-resolved serial femtosecond X-ray crystallographic ( SFX ) data of the PAS-GAF domains of the phytochrome from Deinococcus radiodurans ( DrBphPCBD ) at 1 ps and 10 ps after femtosecond optical excitation . The experiments were performed in Japan , using the SPring-8 Angstrom Compact Free Electron Laser ( SACLA ) tuned to 7 KeV ( Tono et al . , 2015 ) . For homogeneous excitation of the crystals , we photoexcited micrometer-sized crystals in a grease jet with a photon density of 1 . 7 mJ/mm2 ( 1/e2 measure , see Materials and methods ) into the flank of the absorption peak at 640 nm ( Figure 1—figure supplement 1 ) . Taking into account the significant light scattering in the grease-buffer mixture ( Figure 1—figure supplement 1 ) , we estimate that the average number of photons per chromophore is 0 . 5–1 ( see Materials and methods ) . We recorded the SFX data at 1 ps for several excitation fluences ( Figure 2 ) . Lowering the excitation density tenfold from 1 . 7 mJ/mm2 photons to 0 . 2 mJ/mm2 resulted in a joint reduction of all difference signals . Critical signals , like the twist of the D-ring and the photodissociation of the pyrrole water from the chromophore sustained when lowering the excitation densities , indicating that the signal arises predominately from one-photon excitation . The refined structure in dark ( DrBphPdark ) , 2 . 07 Å resolution ( Table 1 ) , was very similar to our previous dark structure solved by SFX ( 5K5B , RMSD 0 . 646 Å and 0 . 610 Å for monomers A and B ) ( Edlund et al . , 2016 ) , but the present crystals contained two monomers in the asymmetric unit ( Figure 1a , Figure 1—figure supplement 2 , Figure 1—figure supplement 3 ) . The refined 1 ps structure was solved to 2 . 21 Å ( Table 1 ) . From the time-resolved data , we calculated Fourier difference electron density maps ( |Fo|l⁢i⁢g⁢h⁢t-|Fo|d⁢a⁢r⁢k ) , which report on the change of structure due to optical excitation ( see Materials and methods ) . Briefly , the diffraction data for light and dark were scaled to each other and subtracted , assuming preservation of the phases ( see Materials and methods for details ) . The map at 1 ps indicates many significant changes in difference electron density ( Figure 1a ) above the background level of 3 . 0 standard deviations ( σ ) ( Figure 1—figure supplement 4 ) . The changes cluster around the chromophore , with the strongest negative densities for the pyrrole water ( monomer A: −8 . 2σ , B: −9 . 4σ , Table 2 ) . The map at 10 ps contains similar significant features , but at weaker overall intensity ( pyrrole water A: −5 . 0σ , B:−6 . 5σ ) ( Figure 1—figure supplement 4 and Figure 1—figure supplement 5 ) . We ascribe this to a lower population of the activated state at 10 ps compared to 1 ps . Monomer A has a lower signal strength than monomer B , but provided a clearer difference map around the chromophore . We refined a structural model ( DrBphP1ps ) using extrapolated structure factors ( Figure 3—figure supplement 1; Pande et al . , 2016 ) . The refinement of the structure against the 1 ps data was successful using a photoexcitation density of 8% . However , we aborted our attempts to refine a structural model against the 10 ps data , as the model would have become unreliable due to an even lower photoactivation yield . We focus our discussion on monomer A and the 1 ps time point , although all conclusions are supported by monomer B and the features observed in the difference maps at 10 ps ( Table 2 , Figure 1—figure supplement 5 ) . First , we inspect the D-ring region at 1 ps ( Figure 3 ) . We observe strong negative difference density features on the atoms of the D-ring ( marked I , II , III ) , correlating with density gains at both faces of the ring ( IV , V , VI ) . These features strongly indicate that the D-ring twists . The positive feature IV homes the N-H and C = O groups , whereas V and VI indicate densities for the methyl and vinyl groups in the twisted ring ( Figure 3c ) . Excellent agreement was obtained between the observed difference map and the difference map calculated from DrBphP1ps ( Fc1⁢p⁢s-Fcd⁢a⁢r⁢k ) , when the D-ring twists ( C14-C15-C16-ND ) from around 20° in the dark to 60° monomer A ) and 90° monomer B ) at 1 ps ( Figure 3c and e ) . Although the twisting movement is clearly indicated by the difference map , we judge the precision of the angle to be low and approximately ±25° . Concomitant with the twist of the D-ring , the C-ring translates by approximately 0 . 69 Å as indicated by the correlated negative ( VII ) and positive ( VIII ) electron density ( Figure 3a ) . Furthermore , the C-ring propionate chain detaches from its conserved anchoring residues Ser272 and Ser274 ( IX and X , Figure 3a ) . The strictly conserved His260 retracts from its position ( XI and XII ) and Tyr263 moves upward at 1 ps ( XIII and XIV , Figure 3b ) . The water network connecting the C-ring propionate , the D-ring C = O , and His290 rearranges accordingly ( Figure 3a ) . The excellent agreement between calculated and observed difference maps confirms these observations ( Figure 3d and e , Figure 3—figure supplement 2 ) . We conclude that the twist of the D-ring causes detachment of the C-ring propionate from the protein scaffold by dislocation of the C-ring , facilitated by the associated hydrogen bonding network . Turning our attention to the B-ring , we find that the B-ring propionate breaks its salt bridge to Arg254 ( Figure 3—figure supplement 3 ) . However , this is not caused by movements of the chromophore backbone , as we observe little change on the B-ring itself . Instead , we find that a water bridge between the B- and C-ring propionates is broken as indicated by negative difference electron densities on the waters ( Figure 3—figure supplement 3 ) . Additionally , the highly conserved helix from Ser257 to Val269 , moves away from the chromophore by an average of 0 . 36 Å in monomer A and 0 . 62 Å in monomer B ( distances relative to the pyrrole water , Figure 3—figure supplement 4 ) . The changes of the D-ring are transduced to Ser257 via the side chains of His260 and Tyr263 , and as a result , the hydrogen bond of Ser257 to the B-ring propionate group breaks . The amino acids in the stretch are over 50% conserved ( Figure 3—figure supplement 4 ) , suggesting that it has evolved to transfer an ultrafast signal . We conclude that relaxation of the protein is necessary for the detachment of the B-ring propionate from the protein scaffold . Next to the changes around the D-ring , the maps reveal strong difference electron density on the A-ring ( XVIII and XIX ) , Asp207 ( XX to XXIII ) ( Figure 4a ) and the pyrrole water ( XV ) ( Figure 4b ) . When interpreted and modelled as downward movement of the A-ring and Asp207 and photodissociation of the pyrrole water from the chromophore , excellent agreement between calculated and observed difference electron density is obtained ( Figure 4—figure supplement 1 ) . The A-ring is covalently attached to the protein backbone in phytochromes ( Song et al . , 2014 ) , which renders complete isomerization impossible , but is sufficiently flexible to accommodate the proposed changes . The pyrrole water may either move to feature ( XVI ) , or occupy an anisotropic , worm-shaped feature which extends from the A-ring to the D-ring ( XVII ) ( Figure 4b ) . Furthermore , correlated negative and positive electron density features are observed on backbone atoms of the highly conserved stretch from Pro201 to His209 . These difference electron density features indicate that the residues move away from the centre of the chromophore by an average of 0 . 54 Å and 0 . 57 Å in monomers A and B , respectively ( Figure 3—figure supplement 4b ) . The stretch includes Asp207 and it is located between the A-ring and the PHY domain , which makes it plausible that the changes in the chromophore cause this protein rearrangement . The changes are complemented by significant rearrangements of a stretch of waters and a conserved Tyr176 ( Figure 4c ) . The structure of DrBphPCBD 1 ps after photoexcition reveals changes of the biliverdin chromophore and the surrounding residues . We find a twist of the D-ring , displacement of the C-ring , and associated changes of the water network which connects the D-ring , the C-ring propionate , and His290 . Further , we identify a disruption of the salt bridge between the B-ring propionate and the Arg254 , and significant changes around the A-ring , Asp207 and the pyrrole water . The changes are retained at 10 ps , even though at a lower population ( Figure 1—figure supplement 5 ) . The extensive and coordinated structural changes in the binding pocket ( Figure 3 ) manifest a liberation of the chromophore from the protein scaffold , which we propose to be necessary for the conformational rearrangements to occur in the downstream photoconversion to Pfr . Infrared spectral data indicate significant reorganization of the chromophore and several amino acids including the PHY-tongue region as early as in Lumi-R state , which is the first known ground state intermediate in the photoconversion from Pr to Pfr ( Ihalainen et al . , 2018; van Thor et al . , 2007 ) . However , the structure of the bilin and the binding pocket in Lumi-R is not known , because structural information is missing . Since the quantum yield of reaching the Lumi-R state is low ( on the order of 10% ) , spectroscopic investigation of the mechanism is difficult , and it is currently not fully established how the Lumi-R state is reached . Crystallographic data does not report on whether the chromophore is electronically excited or not , hence we cannot determine whether the structure that we observe is in a relaxed excited state or in a ground state . The time delay ( 1 ps ) supports that our structure presents an intermediate enroute to the Lumi-R state . The D-ring of the bilin chromophore isomerizes around the C15-C16 bond from ( Z ) in Pr to ( E ) in Pfr ( Rüdiger et al . , 1983; Burgie et al . , 2016; Takala et al . , 2014; Yang et al . , 2008; Essen et al . , 2008 ) . Circular dichroism spectroscopy and solid-state NMR spectroscopy have indicated that the position of the D-ring inverts from an ’α’-facial ( Pr ) to a ’β’-facial ( Pfr ) position in cyanobacterial and plant phytochromes , whereas it stays ’α’-facial in bacterial phytochromes ( Rockwell et al . , 2009; Song et al . , 2011; Song et al . , 2018 ) . Based on anticipated steric clashes with the C-ring methyl group , it has been proposed that the D-ring rotates counter-clockwise in plant and cyanobacterial phytochromes , but clockwise in bacterial phytochromes ( Rockwell et al . , 2009 ) . Moreover , spectroscopy has shown that the D-ring of the bilin chromophore is already isomerized in the Lumi-R state ( van Thor et al . , 2007; Heyne et al . , 2002; Yang et al . , 2012 ) . Seemingly contradictory , we now observe that the D-ring is rotated counter-clockwise by tens of degrees for the bacterial DrBphP at 1 ps time delay ( Figure 3c and e ) . The conformation is strongly supported by the difference map . It contains two positive peaks ( V and VI in Figure 3c ) , which indicate the new position of the vinyl and methyl group of the D-ring . We tested models in which the D-ring was rotated in a clockwise direction , but the agreement with the experimental difference map decreased . Thus , it may be that the D-ring indeed rotates counter-clockwise in bacterial phytochromes , similar to plant and cyanobacterial phytochromes . For complete isomerization , this would mean that the C-ring moves out of the way during the rotation . We observe significant movements of the C-ring , which may be an indication for that such a mechanism is possible . Raising a note of caution , we cannot fully exclude that the truncation of our phytochrome construct or the crystal packing influences the direction of rotation . NMR studies have reported conformational heterogeneity in the chromophore binding pocket of phytochromes in solution ( Song et al . , 2011; Lim et al . , 2018; Song et al . , 2018; Gustavsson et al . , 2020 ) . Crystallization could select one of the conformations , which may have a preferred rotation in the counter-clockwise direction . More experiments are needed to clarify this question . It is interesting to compare the structural changes at 1 ps time delay to the changes observed in the conversion between Pr and Pfr ( Burgie et al . , 2016; Takala et al . , 2014; Stojković et al . , 2014 ) . Major changes include a flipped D-ring , changes in conserved residues of the chromophore-binding pocket , for example Tyr176 , His201 and Phe203 , and refolding of the PHY tongue . The PHY tongue is not included in our construct , but Tyr176 and Phe203 are associated with difference electron density features in our maps ( Table 2 ) . However , the movements are much smaller at 1 ps compared to the Pr-to-Pfr transition . This is not unexpected , given the short time delays , but it shows that the residues are tightly coupled to the chromophore . Interestingly , the Pr and Pfr structures also reveal a sliding movement of the entire chromophore ( Yang et al . , 2011; Burgie et al . , 2016; Takala et al . , 2014 ) . This requires that the propionic groups have to break their bonds to the protein scaffold . Our data indicate that this is part of the primary photoresponse . The photodissociation of the pyrrole water from the chromophore is a surprising finding . The pyrrole water is ubiquitously found in phytochrome structures ( Essen et al . , 2008; Yang et al . , 2008; Otero et al . , 2016; Burgie et al . , 2016; Wagner et al . , 2005; Burgie et al . , 2014; Schmidt et al . , 2018; Yang et al . , 2011 ) . Our fluence dependent SFX data show that the negative density on the pyrrole water is the last signal to disappear when lowering the photon excitation densities 10-fold ( Figure 2 ) . This makes us confident that the photodissociation reaction is not caused by multi-photon effects . The removal of the water requires significant energy , because the hydrogen bonds to the A- , B- , and C-rings of the chromophore and the backbone C = O group of Asp207 have to be broken . We do not think that the twist of the D-ring causes this through direct steric interactions , because there is no contact between the pyrrole water and the D-ring . Rather , it may be triggered by an excited state charge redistribution between the pyrrole water and the chromophore , for example by ultrafast proton or electron transfer ( Toh et al . , 2010 ) . Such charge re-distributions are typically facilitated by changes in geometry ( Nosenko et al . , 2008 ) and may therefore be caused indirectly by structural changes of the A- , C- , or D-rings , but this requires further investigation . Conformational changes of the A-ring , Asp207 and the pyrrole water have not been considered to occur on picosecond time scales . The strictly conserved Asp207 is a key residue for signal transduction because it connects the chromophore to the PHY-tongue in Pr and Pfr ( Essen et al . , 2008; Yang et al . , 2008; Takala et al . , 2014 ) . Its displacement suggests , together with the relocation of the residue stretch surrounding it , that disruption of the GAF-PHY interface may occur as early as 1 ps after photoexcitation ( Figure 3—figure supplement 4b ) . With a hydrogen bond to the pyrrole water and in tight steric contact with the A-ring , Asp207 thereby acts as an extended arm of the chromophore . We propose that the photodissociation of the pyrrole water from the bilin and the change of the A-ring are integral parts of ultrafast phytochrome signaling toward the PHY domain . We demonstrate that within 1 ps , the D-ring twists , that the chromophore is liberated from the protein ( Figure 5a ) and that movements of the pyrrole water , the A-ring and Asp207 lead to signaling directed toward the PHY-tongue ( Figure 5b ) . When mapped on the structure of the complete photosensory core module ( Takala et al . , 2014 ) , both changes work together to destabilize the Arg466:Asp207 salt bridge . Tyr263 moves up , caused by the twist of the D-ring , and Asp207 moves down , caused by changes of the A-ring , retracting both residues from the salt bridge . Our data reveal a highly collective primary photoresponse for phytochromes . This is consistent with the fact that most point mutations of conserved residues alter , but do not inhibit , photoconversion ( Wagner et al . , 2008 ) . The ultrafast structural changes are more extensive than in bacteriorhodopsin , photoactive yellow proteins , and in a fluorescent protein ( Pande et al . , 2016; Nogly et al . , 2018; Coquelle et al . , 2018 ) . While previously observed ultrafast backbone movements have been interpreted as ’protein quakes’ for myoglobin and bacteriorhodopsin ( Barends et al . , 2015; Nogly et al . , 2018; Toh et al . , 2010 ) , the present backbone motion in the phytochrome binding pocket are much more directed ( Figure 3—figure supplement 4 ) . The changes occur in highly conserved regions of the protein and are part of the collective signaling response of the entire binding pocket . Phytochromes have to be able to stabilize the bilin and to direct its photoisomerization from two photochemical ground states , Pr and Pfr . These differ both structurally and electronically , which precludes a single reaction trajectory for isomerization in the two directions . With this in mind , the observed primary photoresponse is reasonable . The structural signal is highly delocalized already at 1 ps , causing near-simultaneous liberation of the chromophore and initial signal transduction . We propose that these reaction trajectories stabilize each other , navigating the protein into a productive reaction path . The multidimensional reaction trajectory is consistent with the low quantum yields for photoconversion ( Lamparter et al . , 1997 ) , which are characteristic for the phytochrome superfamily . Whereas the twisting motion of the D-ring has been the working model for phytochrome activation and is now confirmed , the photodissociation of the pyrrole water is highly surprising . We propose that both chemical events work together and enable phytochrome proteins to translate light information into structural signals , guiding the growth and development of plants , fungi , and bacteria on Earth . The H⁢i⁢s6-tagged PAS-GAF domain from D . radiodurans ( aa 1–321 ) in vector pET21b ( + ) ( Wagner et al . , 2005 ) was expressed and purified as previously described ( Lehtivuori et al . , 2013; Takala et al . , 2014 ) . The recombinant protein was expressed in Escherichia coli strain BL21 ( DE3 ) , either with or without Ho1 to yield holo- or apoprotein , respectively . Cells were lysed with Emulsiflex and cleared by centrifugation ( 20 , 000 rpm , 30 min , +4°C ) . Full biliverdin incorporation was ensured by adding 8 mg of biliverdin hydrochloride ( Frontier Scientific ) per litre of cell culture to the cell lysate , followed by overnight incubation on ice . The protein was then purified at room temperature with HisTrap HP column ( GE Healtcare ) in 30 mM Tris , 50 mM NaCl and 5 mM imidazole ( pH 8 ) and eluted with increasing imidazole concentration ( gradient elution over 5–500 mM ) . Size-exclusion chromatography was then conducted with a HiLoad 26/600 Superdex 200 pg column ( GE Healthcare ) in buffer ( 30 mM Tris pH 8 . 0 ) . Finally , the protein was concentrated to 30–50 mg/mL and flash-frozen in liquid nitrogen . Crystals were set up under green safe light and grown in dark . Batch crystallization was performed as described ( Edlund et al . , 2016 ) . 50 µL of purified protein ( 25–30 mg/mL ) was added to 450 µL of reservoir solution ( 60 mM Sodium acetate pH 4 . 95 , 3 . 3% PEG 400 , 1 mM DTT and 30% 2-methyl-2 , 4-pentanediol ) and immediately mixed . Initial microcrystals were grown on a tipping table at 4 °C for 48 hr . Once the microcrystals were formed , additional protein was added to increase crystal size . The microcrystals were first pelleted by brief centrifugation and 400 µL of supernatant was removed . 200 µL of diluted protein ( 14 mg/mL in 30 mM Tris pH 8 . 0 ) was then added to the microcrystals along with 200 µL of fresh reservoir solution . After 48 hr incubation on a tipping table at room temperature , crystals of diffraction quality ( 20–70 µm long needles ) were formed ( Figure 1—figure supplement 2 and Table 1 ) . Transient absorption experiments were performed on a home build setup based on a Ti:sapphire femtosecond laser system ( 1 kHz , 800 nm ) . The main beam was split into pump and probe beams . The pump beam was sent through the home build noncollinear optical parametric amplifier to produce excitation pulses at 640 nm central wavelength . The probe beam was focused on a 2 mm sapphire plate to generate broadband ( 400–760 nm ) white light which was split by 50/50 beamsplitter to reference and probe beams . The mutual polarization of the pump and probe beams was set to the magic angle ( 54 . 7° ) by Berek compensator . The probe beam was focused on a sample cuvette that was continuously translated in vertical axis to prevent sample degradation . The microcrystals were washed with crystallization buffer five time in order to remove the solubilized proteins . 2 . 5 µL of microcrystals including a small amount of crystallization buffer were placed between two CaF2 windows without a spacer . The OD of the sample was about 0 . 6 . Time-resolved absorption changes were measured by detecting probe and reference beams dispersed on the double-diode array; the time delay between pump and probe pulses was set by a computer controlled delay line placed in the probe beam path . All measurements were carried out in room temperature . In order to estimate the light intensity of the optical laser in the grease jet , the optical transmission of the grease , grease mixed with crystallization buffer , microcrystals in the grease matrix and pure microcrystals were measured with a transmittance diode-array UV-Vis spectrometer ( Cary 8454 , Agilent Technologies ) ( Figure 1—figure supplement 1c ) . 2 . 5 µl each of sample was placed between two CaF2 windows with a 50 µm Teflon spacer and squeezed together , resembling the characteristics of the jet during the XFEL experiments . The raw spectra of microcrystals ( Figure 1—figure supplement 1c ) in the buffer has been measured between two CaF2 windows without spacer to minimize the absorption loss , the pathlength was estimated to be ≤ 50µm . The optical laser parameters used for the experiment were as follows: wavelength was 640 nm , the laser-spot dimensions at the focus was 100 × 80 μm2 FWHM ( 170 × 136 μm2 at 1/e2 intensity ) , the pulse energy was 40 µJ , the nominal pulse duration was 70 fs ( not confirmed at the sample position ) , and the repetition rate was 30 Hz . The energy of a photon at 640 nm is 3 . 1×10-19⁢J . Using the photon density of the laser at 1/e2 convention of 1 . 7 mJ/mm2 , we obtain a photon fluence of 5 . 48×1015⁢photons⋅mm-2 . The extinction coefficient of biliverdin in the phytochrome at 640 nm ( ϵ640 ) is 27 . 7×103⁢M-1⁢cm-1 and the cross section is then σ640=l⁢n⁢ ( 10 ) ⋅ϵ640⋅1000/NA=1 . 06×10-14⁢mm2⋅molecule-1 , where NA is Avogadro’s number . Multiplying the photon fluence with the cross section yields 58 photons per molecule . Light scattering in the carrier matrix decreases the effective fluence of photons that interact with the crystals . Our absorption spectra of the grease ( Figure 1—figure supplement 1 ) indicate that the grease is transparent when untreated , but attenuates the light intensity by 2 orders of magnitude in the visible region when mixed with crystallization buffer or crystals ( pathlength 50 µm ) . This indicates that almost every photon is scattered . Therefore , even when neglecting the scattering of the jet surface , crystals will be exposed to a photon fluence that is significantly reduced . A reduction of the photon fluence by 2 orders of magnitude is a realistic assumption as we used grease jets with a diameter of 75 µm or 100 µm . Another factor that contributes to the reduction of the number of photons per chromophore and non-homogeneous illumination of the microcrystals is the orientation of the crystals and the high chromophore density in them . The first few chromophores in the light path will shade the remaining chromophores in the needle-shaped crystals . Since the X-rays probe every molecule in their path with approximately same likelihood , the average photon fluence per probed chromophore is reduced . Assuming that the effective fluence inside the grease jet is reduced by a factor of 100 , we estimate an average number of 0 . 5–1 photons per chromophore . This is consistent with the photoexcitation yield of 8% and with our experimental finding that the difference signal vanishes under the noise signal when reducing the photon fluence by a factor of 10 ( Figure 2 ) . Serial femtosecond crystallographic data were collected at SPring-8 Angstrom Compact Free Electron Laser ( SACLA ) in two beamtimes in October 2018 and May 2019 . The microcrystals were pelleted by brief centrifugation and the crystal pellet was mixed with 180 µL of grease . The grease/crystal mixture was loaded into a 4 mm sample reservoir for data acquisition . The sample was delivered to the X-ray beam at a flow rate of 2 . 5 µL/min or 4 . 2 µL/min for 75 µm and 100 µm diameter nozzles , respectively . The time resolution of the experiment was limited by the jitter of the XFEL of 100 fs r . m . s . The experimental settings were nominally the same for the 1 ps and 10 ps delay times and all data were recorded during 7 hr of beamtime . We also recorded data at 3 ps delay time , but these generated electron density maps of poor quality due to an unknown reason and were therefore not analyzed further . The background of the detector was estimated by averaging the first 150 dark images in each run and then subtracted from each diffraction pattern . Diffraction images with Bragg spots ( the ‘hits’ ) were found by a version of Cheetah adapted for SACLA ( Nakane et al . , 2016; Barty et al . , 2014 ) . These hits were indexed by the program CrystFEL ( version 0 . 6 . 3 ) ( White et al . , 2012 ) . Indexing was performed using Dirax and Mosflm ( Duisenberg , 1992; Battye et al . , 2011 ) . Spot finding in each diffraction image was done with the peakfinder8 algorithm using the parameters ( min SNR = 4 . 5 , threshold = 100 , minimum pixel counts = 3 ) . The indexed patterns were merged and scaled using partialator in CrystFEL and hkl files were produced . The figure of merits ( Table 1 ) were calculated by using compare_hkl and check_hkl in CrystFEL . The histograms of the unit cell parameters are presented in Figure 3 . All diffraction images have been deposited to CXIDB under ID 121 . The initial phases were solved by molecular replacement with Phaser ( McCoy et al . , 2007 ) and the PAS-GAF crystal structure ( PDB ID 5K5B ) ( Edlund et al . , 2016 ) as a search model . The structure was refined with REFMAC version 5 . 8 . 0135 ( Murshudov et al . , 2011 ) with a weight factor for the geometry restraints of 0 . 05 , accompanied by model building steps with Coot 0 . 8 . 2 ( Emsley et al . , 2010 ) . The final structure ( DrBphPdark ) had Rwork/Rfree of 0 . 161/0 . 192 and no Ramachandran outliers ( Table 1 ) . The coordinates and structure factors have been deposited in the Protein Data Bank under the accession code 6T3L . The difference structure factors ( Δ⁢F ) are computed from the measured structure factor amplitudes in dark and for preset delay times between laser and X-ray pulses as |Δ⁢Fo|=w⁢ ( |Fol⁢i⁢g⁢h⁢t|-|Fod⁢a⁢r⁢k| ) and with phases taken from the dark structural model ( DrBphPdark ) . |Fod⁢a⁢r⁢k| and |Fol⁢i⁢g⁢h⁢t| were brought to the absolute scale by first scaling |Fod⁢a⁢r⁢k| to |Fcd⁢a⁢r⁢k| and then scaling |Fol⁢i⁢g⁢h⁢t| to |Fod⁢a⁢r⁢k| using the CCP4 program Scaleit ( Winn et al . , 2011 ) . Difference Fourier density maps were calculated with a low resolution scaling cut-off at 18 Å . A weighting factor ( w ) was determined for each reflection to reduce the influence of outliers ( Ren et al . , 2001 ) . From the weighted Δ⁢F , a difference electron density map ( Δ⁢ρ ) is calculated using the program ‘fft’ from the ccp4 suite of programs ( Winn et al . , 2011 ) . Since Δ⁢F are on the absolute scale , Δ⁢ρ is on half the absolute scale as a result of the difference Fourier approximation ( Henderson and Moffat , 1971; Pandey et al . , 2020 ) . Extrapolated structural factors were assembled from amplitudes computed as |Fe|=|Fcd⁢a⁢r⁢k|+α*|Δ⁢Fo| . The Fcd⁢a⁢r⁢k denotes the calculated structure factors of the refined dark structure ( DrBphPdark ) . The phases were taken from DrBphPdark is inversely related to the population of the photoinduced state by ( 100/α ) *2 ( Pandey et al . , 2020; Henderson and Moffat , 1971 ) . We estimated α based on Fe map features in the chromophore-binding pocket . Too high values for α lead to physically unrealistic negative electron density . We converged to α=25 , which corresponds to 8% photoexcitation yield . Fe represents the pure structure factor of the photo-activated state ( Figure 3—figure supplement 1 ) . Refinement of a structural model was then performed in real and reciprocal space , using Coot ( Emsley et al . , 2010 ) and Phenix ( Adams et al . , 2010 ) . The equilibrium values for the restraints used in the refinement of the biliverdin chromophore were taken from a minimal energy biliverdin ground state geometry that was obtained at the B3LYP/6–31G* level of density functional theory . Torsional restraints for the excited state geometry with the twisted D-ring were obtained at the SA ( 5 ) - CASSCF ( 12 , 12 ) /cc-pVDZ level of ab initio theory . We removed the torsion restrains for the C/D-ring ( C14-C15-C16-C17; C14-C15-C16-ND; C13-C14-C15-C16; NC-C14-C15-C16 ) and for the A/B-ring ( C3-C4-C5-C6; NA-C4-C5-C6; C4-C5-C6-C7; C4-C5-C6-NB ) during refinement . The overall aim of the refinement was to maximize the agreement between the observed and calculated difference maps . To evaluate the agreement , we subtracted the calculated from the observed difference electron density ( Δ⁢ρo-Δ⁢ρc ) . The computation of this difference-difference maps require scaling of the maps to each other . To do so , the highest and lowest intensities of Δ⁢ρo were scaled to the corresponding maximum and minimum of Δ⁢ρc and the observed Δ⁢ρo were interpolated linearly according to this scaling . The resulting difference-difference electron density map was used to identify sites , which required further optimization in subsequent refinement steps . Calculation of Pearson Correlation Coefficient ( PCC ) values between the Δ⁢ρo and Δ⁢ρc were applied to guide refinement of specific regions , such as the D-ring and the whole chromophore region . To do so , the correlation was determined based on electron density within a sphere with a radius of 3 . 5 Å or 10 Å centred on the D-ring or pyrrole water , respectively . As a final step in the refinement procedure , we refined the models with REFMAC version 5 . 8 . 0135 ( Murshudov et al . , 2011 ) with high geometry restraints ( weight factor 0 . 005 ) . This was done against phased extrapolated structure factors , using the phases of the refined light and dark structure for computation of phased Δ⁢F as described ( Pande et al . , 2016 ) . The structures did not change much , although the R factors dropped in this last step of refinement to Rwork/Rfree of 0 . 230/0 . 256 ( Table 1 ) . The coordinates and structure factors have been deposited in the Protein Data Bank under the accession code 6T3U .

Plants adapt to the availability of light throughout their lives because it regulates so many aspects of their growth and reproduction . To detect the level of light , plant cells use proteins called phytochromes , which are also found in some bacteria and fungi . Phytochrome proteins change shape when they are exposed to red light , and this change alters the behaviour of the cell . The red light is absorbed by a molecule known as chromophore , which is connected to a region of the phytochrome called the PHY-tongue . This region undergoes one of the key structural changes that occur when the phytochrome protein absorbs light , turning from a flat sheet into a helix . Claesson , Wahlgren , Takala et al . studied the structure of a bacterial phytochrome protein almost immediately after shining a very brief flash of red light using a laser . The experiments revealed that the structure of the protein begins to change within a trillionth of a second: specifically , the chromophore twists , which disrupts its attachment to the protein , freeing the protein to change shape . Claesson , Wahlgren , Takala et al . note that this structure is likely a very short-lived intermediate state , which however triggers more changes in the overall shape change of the protein . One feature of the rearrangement is the disappearance of a particular water molecule . This molecule can be found at the core of many different phytochrome structures and interacts with several parts of the chromophore and the phytochrome protein . It is unclear why the water molecule is lost , but given how quickly this happens after the red light is applied it is likely that this disappearance is an integral part of the reshaping process . Together these events disrupt the interactions between the chromophore and the PHY-tongue , enabling the PHY-tongue to change shape and alter the structure of the phytochrome protein . Understanding and controlling this process could allow scientists to alter growth patterns in plants , such as crops or weeds .

2020

The primary structural photoresponse of phytochrome proteins captured by a femtosecond X-ray laser

Mechanical forces have emerged as coordinating signals for most cell functions . Yet , because forces are invisible , mapping tensile stress patterns in tissues remains a major challenge in all kingdoms . Here we take advantage of the adhesion defects in the Arabidopsis mutant quasimodo1 ( qua1 ) to deduce stress patterns in tissues . By reducing the water potential and epidermal tension in planta , we rescued the adhesion defects in qua1 , formally associating gaping and tensile stress patterns in the mutant . Using suboptimal water potential conditions , we revealed the relative contributions of shape- and growth-derived stress in prescribing maximal tension directions in aerial tissues . Consistently , the tension patterns deduced from the gaping patterns in qua1 matched the pattern of cortical microtubules , which are thought to align with maximal tension , in wild-type organs . Conversely , loss of epidermis continuity in the qua1 mutant hampered supracellular microtubule alignments , revealing that coordination through tensile stress requires cell-cell adhesion . As our understanding of the role of forces in development deepens , assessing accurate stress patterns in tissues has become increasingly important ( Roca-Cusachs et al . , 2017 ) . Stress patterns can be revealed through three approaches: 1- Computational models , for example with spring networks or finite elements , with relevant assumptions on tissue mechanics for animal ( e . g . Sherrard et al . , 2010 ) and plant ( e . g . Bozorg et al . , 2014 ) systems , 2- Strain measurements following local cuts at the subcellular ( e . g . Landsberg et al . , 2009 ) or organ ( e . g . Dumais and Steele , 2000 ) scale , 3- Strain measurement of deformable objects ( e . g . FRET-based molecular strain sensors [Freikamp et al . , 2017] , oil microdroplets [Campàs et al . , 2014] , elastomeric force sensors [Wolfenson et al . , 2016] ) . Previous work on animal single cells showed that hyperosmotic media can affect membrane tension and thus the molecular effectors of cell migration , like actin filaments , RAC activity or WAVE complex , suggesting that the corresponding mutants could be rescued by a modification of the osmotic conditions of the medium ( Batchelder et al . , 2011; Houk et al . , 2012; Asnacios and Hamant , 2012 ) . Consistently , adding sorbitol in growth media is sufficient to rescue defects in yeast endocytic mutants ( Basu et al . , 2014 ) . Here we take inspiration from these single cell studies and apply the same logic at the multicellular scale . Using an Arabidopsis mutant with severe cell adhesion defects , we partially rescue these defects by modifying the water potential of the growth medium and we deduce the maximal direction of tension in tissues from the gaping pattern following growth , without any external intervention . In plants , cell adhesion is achieved through the deposition of a pectin-rich middle lamella between contiguous cell walls ( Orfila et al . , 2001; Daher and Braybrook , 2015; Willats et al . , 2001; Chebli and Geitmann , 2017; Jarvis et al . , 2003; Knox , 1992 ) . QUASIMODO1 ( QUA1 ) encodes a glycosyltransferase that is required for pectin synthesis and cell adhesion ( Bouton et al . , 2002; Mouille et al . , 2007 ) . Here we reasoned that the resulting cell-cell gaps may in principle reveal the stress pattern in tissues . Yet , it has not been formally demonstrated that gap opening could be related to tissue tension . Furthermore , the severe defects in the mutant make it hard to deduce a stress pattern in such distorted tissues . We thus developed a protocol amenable to partially rescue the adhesion defect through water potential modulation , allowing us to relate adhesion to tissue tension on the one hand , and deduce a pattern of stress in various plant tissues on the other hand . This mutant also allowed us to investigate how the loss of adhesion affects the propagation of mechanical stress and thus tension-dependent cell-cell coordination . The quasimodo1 ( qua1 ) and qua2 mutants , respectively mutated in a galacturonosyltransferase and a pectin methyltransferase , are both required for the synthesis of a fraction of the cell wall pectins . They also display a comparable cell adhesion defect phenotype ( Bouton et al . , 2002; Mouille et al . , 2007 ) . For practical reasons , all the work reported in this study was performed with qua1-1 ( WS-4 background ) , although we observed similar phenotypes in the qua2-1 mutant ( Col-0 background ) . Because the qua1 mutant is very sensitive to sucrose in the medium , which leads to metabolic stress and growth arrest of the seedling ( Gao et al . , 2008 ) , we grew the seedlings on a medium containing no sucrose to focus on the cell adhesion phenotype . In these conditions , we could observe cell separation in the epidermis of hypocotyls , stems , cotyledons , and leaves ( Figure 1E; Figure 1—figure supplement 1 ) , consistent with the epidermal theory of growth where the epidermis is put under tension through the pressure exerted by inner tissues , and thus is load-bearing for aerial organs ( Kutschera and Niklas , 2007; Savaldi-Goldstein et al . , 2007; Maeda et al . , 2014 ) . Because endogenous tensile stress in plant epidermis originates both directly and indirectly from turgor pressure , we next modified the water potential of the medium , reasoning that epidermal integrity should be restored in the qua1 mutant if epidermal tension was decreased . Increasing concentrations of Polyethylene Glycol ( PEG ) in the growth medium reduced the growth rate of wild-type and qua1-1 seedlings , suggesting a reduction in turgor pressure in planta caused by a decrease in the water potential of the medium ( Figure 1—figure supplement 2A–C ) . Strikingly , in the lowest water potential condition , the overall qua1-1 phenotype was almost fully rescued ( Figure 1—figure supplement 2E–G ) . Using propidium iodide staining and confocal imaging of the cotyledon pavement cells , we further confirmed that such osmotic conditions reduced cell separations in qua1-1 ( Figure 1E–G ) . Note that qua1-1 pavement cells preferentially separated at the neck-lobe junction , consistent with previously calculated patterns of stress in this tissue ( Figure 1K and L , [Sampathkumar et al . , 2014] , Figure 1—figure supplement 3 , Video 1 and 2 ) . This suggests that low tensile stress in the epidermis is sufficient to restore cell adhesion in the mutant . Nonetheless , because PEG may have pleiotropic effects , we cannot exclude the possibility that the restoration of cell adhesion could be due to other factors . Therefore , we increased agar concentration in the medium , as an alternative way to affect water potential . Indeed , increasing the agar concentration reduces the water potential by decreasing its matrix potential ( a component of the water potential , [Owens and Wozniak , 1991] ) and in turn hinders the capacity of the plant to take up water , as does PEG-containing medium . As expected , we observed a similar restoration of the cell adhesion and seedling phenotype in qua1-1 , to that observed on PEG-containing medium ( Figure 1H , Figure 1—figure supplement 2H ) . To go beyond these qualitative observations , we developed a semi-automated pipeline of image analysis amenable to identify individual gaps between cells , quantify their areas and their main orientations ( see Material and methods , Figure 1I; Figure 1—figure supplement 4; Verger and Cerutti , 2018; copy archived at https://github . com/elifesciences-publications/Cell_separation_analysis ) . Based on images of five-day-old qua1-1 cotyledons ( Figure 1I ) , we found that , for a field of cells representing 138654 µm2 per image , seedlings grown on 1% agar medium exhibited 17906 ( ±8955 ) µm2 of cell separation per image ( i . e . ca . 13% of the surface area , n = 12 samples ) , while seedlings grown on 2 . 5% agar medium displayed 1457 ( ±1140 ) µm2 of cell separation per image ( i . e . ca . 1% of the surface area , n = 12 samples; Figure 1J , Welch’s t-test p-value=0 . 0004 ) , confirming the rescuing effect of a low water potential on cell adhesion . Last , to confirm that the propidium iodide staining truly reflected cell-cell adhesion defects , we analyzed the gaps in qua1-1 at different stages with confocal microscopy and , at high resolution , with atomic force microscopy . Our images matched previously published SEM images of qua1-1 mutants ( Bouton et al . , 2002 ) , with stretched and detached outer walls at the cell-cell junction ( Figure 1—figure supplement 5 ) . Altogether , these results strongly suggest that adhesion defects in qua1-1 indeed relate to the tensile status of the tissue . At this stage , we find a correlation between the medium water potential and adhesion defects in qua1-1 . To measure the impact on epidermal tension , we turned to atomic force microscopy ( AFM ) to obtain force-displacement curves on epidermal surfaces allowing us to extract a slope , which corresponds to the apparent stiffness of the material ( Figure 2 ) . We focused on cotyledons , as they are easier to manipulate under the AFM . To maintain tissue hydration , AFM live imaging was conducted in aqueous solutions: for 1% and 2 . 5% agar-grown cotyledons , cotyledons were submerged in water , while PEG-grown cotyledons were submerged in liquid Arabidopsis medium supplemented with mannitol to reach the same osmotic pressure as the PEG-infused medium ( see Materials and methods ) . To measure cell-level mechanical properties over the epidermal surface , we performed indentations with an AFM probe much smaller than cell size ( 0 . 8 μm probe diameter compared to >10 μm cell width; Figure 2A ) . Approximately 10 ~ 15 μN indentations were performed to achieve 1 ~ 2 μm indentation depth ( Figure 2A ) , deep enough to detect epidermal turgor pressure but relatively shallow compared to pavement cell thickness ( typically 6 ~ 10 μm , [Zhang et al . , 2011] ) . In these conditions , we are measuring the stiffness of single epidermal cells , and do not detect the stiffness of the rest of the tissue , notably the internal cell layers ( Beauzamy et al . , 2015 ) . As shown in Figure 2 , results obtained from wild-type seedlings grown on medium supplemented with 155 mOsm PEG or 2 . 5% agar were similar . Similar trends were obtained in qua1-1 , albeit with a globally reduced turgor pressure , apparent stiffness and cell wall tension ( Figure 2—figure supplement 1 ) . We also confirmed that the immersion medium had little impact on the measurement both in wild type and qua1-1 , at least in the short term ( Figure 2—figure supplement 1 ) . First , we found that turgor pressure levels in the epidermis were not affected by a change in osmotic or matrix potential . This suggests that water potential primarily affect internal tissues and/or that the epidermis can osmoregulate efficiently , as already shown before ( e . g . [Shabala and Lew , 2002] ) . Interestingly , when focusing on the outer wall , we found that apparent stiffness decreased by 15% in both PEG and high agar-grown seedlings , while cell wall tension in high agar and PEG-grown seedlings decreased by 18% and 16% respectively , thus demonstrating the impact of the corresponding treatments on epidermal tension . Although such differences in stiffness may appear small , comparable differences were obtained in other tissues , for example between central and peripheral zone of the shoot apical meristem , where differences in growth rates are of 200% to 300% ( Milani et al . , 2014 ) . Furthermore , the concentrations of PEG and agar were chosen so as to maintain growth; in other words , differences could be stronger for higher concentrations , but these would not be relevant for our study . Altogether , these data formally relate the decreased water potential in the medium to a decrease in outer wall tension in planta and rescue of gaping patterns in qua1-1 . Because increasing agar concentration is not toxic to the cell , as shown by propidium iodide staining ( see Figure 1G and H ) , we selected this protocol to alter water potential in the following experiments . Both shape-derived stress and growth-derived stress contribute to the final pattern of stress in any given field of cells . Shape-derived stress , or pressure stress , is calculated based on the assumption that an organ behaves like a pressure vessel , that is like a load-bearing envelope under tension . This is typically the case for individual plant cells ( cell walls resist internal turgor pressure and thus are under tension , see for example Sampathkumar et al . , 2014 ) for an example of stress prediction entirely based on cell shape ) and aerial organs ( in the epidermal theory of growth framework , ( Kutschera and Niklas , 2007 ) , see for example ( Hamant et al . , 2008 ) for an example of stress prediction entirely based on tissue shape ) ( Figure 3N ) . Growth-derived stress corresponds to mechanical conflicts arising from differential growth rates or directions ( [Rebocho et al . , 2017; Thimann and Schneider , 1938; Kutschera , 1992] , Figure 3N ) . Using the qua1-1 mutant , we dissected these two contributions to the global tensile stress patterns in various plant tissues . First , we analyzed the tensile stress patterns in the inflorescence stem apex , that is where the cells are still actively dividing below the shoot apical meristem ( Figure 3A ) . In normal in vitro growth condition , qua1-1 seedlings often generate aberrant inflorescence meristems and stems , morphologically reminiscent of calli ( see Figure 1P , and [Krupková et al . , 2007] ) . To test whether this phenotype also correlates with the medium water potential , we used in vitro grown seedlings to modulate the matrix potential of the growth medium . We also supplemented the medium with NPA , an inhibitor of polar auxin transport and floral organogenesis , to produce naked meristems that are more amenable to visualization and quantification ( see e . g . [Sassi et al . , 2014] ) . Strikingly , growth on low water potential medium almost completely restored the formation of normal meristems in qua1-1 with no or only minor loss of cell adhesion ( Figure 3C , see also Figure 1M–P ) , thereby allowing us to investigate the gaping pattern in qua1-1 inflorescence stems . In qua1-1 plants grown in normal condition ( 1% agar ) , we could not distinguish a clear pattern of cell separation , where in most cases the meristem shape was severely affected and some cells seemed to start proliferating randomly ( Figure 3B ) . However , when grown on 2 . 5% agar medium , longitudinal stripes of bright propidium iodide staining were observed in the qua1-1 mutant stems; such signal could not be detected in the wild type in the same growth conditions ( Figure 3C ) . Some of these stripes developed further into cracks between adjacent cells later on , confirming that the stripes correspond to slight separations between adjacent cells where propidium iodide can accumulate , due to the opening of the cuticle at cell junctions and likely unpacking of the cell wall polysaccharides ( Figure 1—figure supplement 5 ) . In order to quantify these orientations more precisely , we used our cell separation image analysis pipeline , this time focusing on the orientation of cell separation ( see Figure 1—figure supplement 4 ) . In the stem apex , we obtained a mean gap angle ( θG ) of 91 ± 7° ( n = 8 samples ) , relative to the transverse axis of the stem . Longitudinal cell separation reveals that the cells are being pulled apart transversely , indicating that maximal epidermal tension is transverse to the axis of the stem . This pattern is consistent with shape-derived stress , assuming that the epidermis acts as a load-bearing layer under tension in that tissue , and thus as in a cylindrical pressure vessel where maximal tension is also transverse ( Figure 3N ) . Note that the age of cell walls may bias our analysis . In particular , based on our results on stem apices , one could propose that older cell walls become less adhesive or more prone to separate . To explore that hypothesis further , we took advantage of our comparative analysis between different tissues to test whether that hypothesis could also hold true . While the stem apex grows relatively slowly , hypocotyls grow fast and primarily in one direction , through anisotropic cell expansion ( Gendreau et al . , 1997 ) . In hypocotyls , cell separations happened in several orientations , leading to epidermal cell naturally peeling out of the surface ( Figure 3; see also Figure 1—figure supplement 1 ) . This pattern is thus not consistent with the pressure vessel model in which stress depends only on shape . To explain this discrepancy , we explore the possible contribution of growth-derived stress: the anisotropic expansion of the inner tissues would pull the load-bearing epidermis longitudinally , thus exerting a longitudinal tensile stress on the epidermis ( for predictions of longitudinal stress patterns in growing cylindrical organs , see for example [Baskin and Jensen , 2013; Vandiver and Goriely , 2008] ) . To go beyond this qualitative assessment , we next focused on dark-grown hypocotyls , since they display a well characterized gradient of growth during their elongation , in which cells closest to the root ( rootward ) have already extensively elongated and are undergoing growth arrest , cells more toward the middle are rapidly elongating , and cells at the top ( shootward ) and in the apical hook are only starting to elongate ( Figure 3—figure supplement 1A , [Gendreau et al . , 1997; Bastien et al . , 2016] ) . To ensure phenotypic consistency , we observed the rootward part of dark-grown hypocotyls . Because cracks in qua1 emerge and develop through growth , we reasoned that the gaping pattern in this part of the hypocotyl would reflect the stress pattern in cells that previously experienced their maximal elongation phase , a posteriori ( Figure 3—figure supplement 1A ) . When grown on 1% agar , cell separation patterns were so extensive in dark-grown qua1-1 hypocotyls that the cell separation patterns could not be quantified properly ( Figure 3F ) . When grown on 2 . 5% agar , we could identify discrete cell separations happening almost exclusively transversely to the axis of dark-grown qua1-1 hypocotyl ( Figure 3G ) . Yet , the presence of cells peeling off suggests that longitudinal cell-cell separation also occurs along the longitudinal axis , likely after the initial transverse separations . Interestingly , this is not consistent with the hypothesis that older cell walls separate first , since all epidermal walls in the hypocotyl have the same age after embryogenesis . To check whether this pattern depends on mechano-chemical polarities in the anticlinal epidermal cell wall , we next overexpressed PECTIN METHYLESTERASE INHIBITOR 5 ( PMEI5 ) in qua1-1 , reasoning that PMEI overexpression should reduce heterogeneity of pectin esterification in the hypocotyl ( Wolf et al . , 2012; Müller et al . , 2013; Peaucelle et al . , 2015 ) . Using atomic force microscopy , it was previously shown that wild-type hypocotyls exhibit strong differences in apparent elastic moduli between transverse and longitudinal anticlinal walls in the epidermis , whereas overexpression of PMEI significantly reduced such mechanical polarities ( Peaucelle et al . , 2015 ) . As also shown before in PMEI overexpressor lines , we observed an increased twisting in the qua1-1 p35S::PMEI5 line . Yet , we could not detect significant differences in the gaping pattern between dark-grown qua1-1 and qua1-1 p35S::PMEI5 hypocotyls , further confirming that the gaping pattern primarily results from the tension pattern ( Figure 3—figure supplement 2 ) . The gaps in dark-grown hypocotyls were too large for our image analysis pipeline to precisely discern cell separation orientation ( Figure 3G ) . We thus manually counted the number of events of cell separations that happened either at the transverse or longitudinal junction between adjacent cells . In thirteen images from individual hypocotyls , we counted on average about 11 events of cell separation per image . In total , we found 135 events in which cells had separated along their shared transverse wall , and nine events in which they had separated along their shared longitudinal wall . However , among these nine events , seven were related to an adjacent event of transverse cell separation ( Figure 3—figure supplement 1D and E ) , while only two events were strictly longitudinal ( Figure 3—figure supplement 1B and C ) . Note that cells at the shootward portion of the hypocotyl , where rapid elongation has not started yet , very few cells were separated in qua1-1 ( Figure 3—figure supplement 1F ) , further supporting the role of anisotropic growth in generating the observed gaps . To further test whether these gaps can indeed be related to growth-derived stress , we took advantage of the ability of hypocotyls to modulate their growth rate according to light conditions: when grown in light , hypocotyls usually reach 1 . 5 to 2 millimeters in length , in contrast to dark-grown hypocotyls which can reach about two centimeters ( Gendreau et al . , 1997 ) . We reasoned that , in light conditions , the reduction of elongation should decrease the extent of gap opening in qua1-1 , while growth anisotropy would still prescribe longitudinal growth-derived stress . As in dark-grown hypocotyls , when grown on 1% agar , cell separation patterns were so extensive in light-grown qua1-1 hypocotyls that the cell separation patterns could not be quantified properly ( Figure 3J ) . When grown on 2 . 5% agar , transverse gaps were detected in light-grown qua1-1 hypocotyl ( Figure 3K ) . In contrast to dark-grown hypocotyls , we almost only observed slight cell separations in light-grown hypocotyls , as marked by bright propidium iodide staining , consistent with reduced elongation . Because the gaps were much smaller , we could use our pipeline and measured a mean θG of 5 ± 30° ( Figure 3K; Figure 1—figure supplement 4 , n = 8 samples ) . A similar response was observed in the qua1-1 PMEI-OE line , consistent with the pattern primarily resulting from growth-derived stress , and not from heterogeneities of pectin esterification ( Figure 3—figure supplement 2 ) . Altogether , our quantifications support the idea that transverse shape-derived tensile stress dominates in the stem apex epidermis , whereas longitudinal growth-derived stress dominates in the elongating hypocotyl epidermis . Next we investigated the gaping pattern in qua1-1 cotyledons where more complex shape and growth patterns occur . In comparison to stems and hypocotyls , cotyledon growth occurs mainly in 2D and is rather isotropic ( Zhang et al . , 2011 ) . Using our pipeline , we analyzed cell separation orientation , focusing on cotyledons at a very young stage ( 3-day old ) in order to observe very early cell separations before gaps become too large . No preferential gap orientations could be detected ( Figure 4C ) . Note that while an apparent bimodal distribution ( with more separations happening at a 0° and 90° angle ) , the population of angles could be considered uniformly distributed , as assessed using the Rao’s spacing test for uniformity ( non-parametric test due to the apparent bimodal distribution of the angle population , and adapted for directional data , p-value=0 . 5 , n = 12 samples ) . This contrasts with the stem apex and the light grown hypocotyl cell separations , for which the test revealed a non-uniform distribution ( p-value<0 . 001 for both ) . To obtain a different indicator of the spread of the population of angles , we also measured the resultant vector length ( R ) calculated on these populations of angles . R lies between 0 and 1; the higher it is the more clustered and unidirectional are the data , while a very low value reveals no preferential orientation . R reached 0 . 84 and 0 . 70 for the stem apex and light-grown hypocotyls respectively , whereas R was equal to 0 . 07 for cotyledons . Overall these tests reveal that cell separation and thus tensile stress in the cotyledon epidermis is globally isotropic . When growing seedlings on a 2 . 5% agar medium , while almost no more cell separation could be observed on the cotyledon blade ( see Figure 1H and J ) , the junction between the petiole and the isotropically expanding cotyledon blade exhibited large cell separations ( Figure 4D and I ) . The discrepancy in the extent of cell adhesion defects between blade and junction , as revealed by modulating the water potential , suggests that tensile stress is higher at the blade-petiole junction . When measuring the mechanical properties of the petiole-blade junction with AFM , we actually found that this region exhibits increased turgor pressure , apparent stiffness and surface tension ( Figure 4—figure supplement 1 ) . Interestingly , using the plasma-membrane-associated protein BREVIS RADIX LIKE 2 ( BRXL2 ) as a polar marker , the proximal region of leaves was recently shown to exhibit increased mechanical stress-driven planar cell polarity ( Bringmann and Bergmann , 2017 ) . This suggests that the predicted mechanical conflict in the petiole-blade junction of the cotyledon is also present in leaves . To check this , we analyzed the gaping pattern in the third leaf in qua1-1 and found a pattern roughly similar to that of cotyledons , with longitudinal cracks at the base of the leaf ( Figure 4—figure supplement 2 ) . Interestingly , in the blade of these leaves , we also observed radial cracks around trichomes , a pattern that is also consistent with the recently identified tensile stress pattern in trichome socket cells ( Hervieux et al . , 2017 ) . Furthermore , gaps at the junction appeared preferentially , but not exclusively , along the longitudinal axis of the petiole ( Figure 4D and I ) . Our pipeline revealed that cell adhesion defects displayed a mean θG of 85 ± 50° ( Figure 4D and I , n = 8 samples ) . Note however that the resultant vector length R was only equal to 0 . 21 and the Rao’s spacing test showed uniformity of the distribution of angles ( p-value of 0 . 5 ) , which suggest that the bias is weak . Altogether , the analysis of the gaping pattern suggests the presence of a mechanical conflict at the petiole-blade junction in cotyledons , both in intensity and direction ( Figure 4I ) . To test whether the inferred stress patterns in qua1-1 are consistent with predicted patterns of stress in the wild type , we next analyzed cortical microtubules ( CMTs ) in corresponding wild-type organs . CMTs have previously been shown to align along predicted maximal tensile stress direction ( Green and King , 1966; Williamson , 1990 ) . Such response was observed in sunflower hypocotyl ( Hejnowicz et al . , 2000 ) , Arabidopsis shoot apical meristems ( Hamant et al . , 2008 ) , leaves ( Jacques et al . , 2013 ) , cotyledons ( Sampathkumar et al . , 2014 ) and sepals ( Hervieux et al . , 2016 ) . To check whether tensile stress patterns in aerial organs , as inferred from qua1-1 cracks , match CMT orientations , we next analyzed the CMT behavior in plants expressing the p35S::GFP-MBD microtubule marker ( WS-4 ecotype ) , and quantified the average CMT array orientation ( θM ) and anisotropy using the FibrilTool ImageJ macro ( Figure 5—figure supplement 1 , [Boudaoud et al . , 2014] ) . First , we analyzed CMT orientations in stem apices and cotyledons , where CMT response to mechanical perturbation was already established ( Hamant et al . , 2008; Sampathkumar et al . , 2014 ) . CMTs at the stem apex were oriented transversely , matching the stress pattern revealed by the qua1-1 cell separation patterns ( Figure 3D; for plants grown on 2 . 5% agar medium: Mean θM = 3 ± 26° , Mean anisotropy = 0 . 20 ± 0 . 08 , R = 0 . 65 , n = 1669 cells in 13 samples ) . Note that similar trends were observed on stem apices from seedling grown on 1% and 2 . 5% agar medium , albeit with a slightly increased noise for seedlings grown on 1% agar medium ( Figure 3—figure supplement 3A , B , E , F; for plants grown on 1% agar medium: Mean θM = 2 ± 26° , Mean anisotropy = 0 . 21 ± 0 . 09 , R = 0 . 66 , n = 1604 cells in 12 samples ) . CMT exhibited different behaviors between the blade and the petiole-blade junction , consistent with the mechanical conflict revealed by the gaps in qua1-1 ( Figure 4E–H ) . When grown on 1% agar medium , CMTs in the blade exhibited a significant bias in their orientation ( Mean θM = 150 ± 43° , R = 0 . 32 , n = 898 cells in 12 cotyledons ) , whereas CMTs at the junction did not exhibit a significant bias in their orientation ( R = 0 . 06 , n = 550 cells in 12 cotyledons ) . Interestingly , the mean anisotropy of the CMT arrays was slightly increased at the junction , when compared to the blade , consistent with the measured increased tension and mechanical conflict in that region ( Mean anisotropy = 0 . 08 ± 0 . 04 ( blade ) vs . 0 . 11 ± 0 . 06 ( junction ) ; paired comparisons between blade and junction from each sample revealed a significant difference ( always higher in junction ) ( Wilcoxon-test , p-values<0 . 05 ) for 10 out of 12 samples in each condition ) . When grown on 2 . 5% agar medium , the biases in CMT orientation were weak , but significantly different , between the blade and in the junction ( blade: mean θM = 154 ± 51° , R = 0 . 19 , n = 639 cells in 13 cotyledons; junction: mean θM = 97 ± 45° , R = 0 . 28 , n = 425 cells in 13 cotyledons ) . As on 1% agar medium , the anisotropy of the CMT arrays was higher at the junction ( 0 . 18 ± 0 . 09 ) than in the blade ( 0 . 08 ± 0 . 05 ) . Paired comparisons between blade and junction from each sample revealed a significant difference ( always higher in junction ) for 12 out of 13 samples in each condition ( Wilcoxon-test , p-values<0 . 001 ) . In dark-grown hypocotyls , previous studies had reported that CMTs behave differentially at the inner and outer face of the epidermal cells of the hypocotyl and also display different responses at the hypocotyl shootward region ( where cells are not elongated yet ) , middle ( where cell elongation rate is high ) and rootward region ( where cell elongation is slowing down ) ( Chan et al . , 2010; Chan et al . , 2011; Crowell et al . , 2011 ) . In particular , CMTs were shown to be transversely oriented in the inner face of the epidermal cell at most of these stages , promoting the anisotropic expansion of the tissue ( Crowell et al . , 2011 ) , whereas CMTs display a puzzling behavior at the outer epidermal face: they are rotating in the top and middle part , and are longitudinal in the bottom part . Focusing on the bottom part of the hypocotyl , where cell separation was observed in qua1-1 , we observed that CMTs were highly aligned longitudinally in the wild type , when grown in the same conditions as in qua1-1 ( Figure 3H; seedlings grown on 2 . 5% agar medium: Mean θM = 98 ± 6° , Mean anisotropy = 0 . 47 ± 0 . 06 , R = 0 . 98 , n = 182 cells in 12 samples ) , thus following the tensile stress pattern revealed by the cell separations in qua1-1 . Note that similar trends were observed for seedling grown on 1% and 2 . 5% agar medium ( Figure 3—figure supplement 3C , G; seedling grown on 1% agar medium: Mean θM = 93 ± 6° , Mean anisotropy = 0 . 46 ± 0 . 04 , R = 0 . 97 , n = 137 cells in 10 samples ) . In light-grown hypocotyls , CMT orientations seemed random at first sight ( see e . g . the scatter plot of CMT orientations in Figure 5—figure supplement 2D at t = 0 hr ) . Consistently , CMTs in light grown-hypocotyl have also been described to harbor a rotating behavior ( Chan et al . , 2010 ) . Yet , when taking into account the anisotropy level of the CMT arrays ( to put more weight on CMT orientations when anisotropy is high , as done throughout in this article ) , we revealed a significant bias towards longitudinal CMT orientations , that is parallel to the tensile stress pattern revealed by the cell separations in qua1-1 ( Figure 3L , seedling grown on 2 . 5% agar medium: Mean θM = 113 ± 30° , Mean anisotropy = 0 . 25 ± 0 . 10 , R = 0 . 59 , n = 606 cells in 11 samples ) . Note that similar trends were observed for seedling grown on 1% and 2 . 5% agar medium ( Figure 3—figure supplement 3D , H; seedling grown on 1% agar medium: Mean θM = 109 ± 28° , Mean anisotropy = 0 . 21 ± 0 . 08 , R = 0 . 6 , n = 413 cells in seven samples ) . In comparison to dark-grown hypocotyls , the standard deviation was higher , anisotropy and resultant vector length were lower . Note that anisotropy and resultant vector length are describing different properties: anisotropy relates to CMT arrays in individual cells , while vector length relates to the global behavior of CMT arrays in the cell population . Thus lower anisotropy and vector length in light-grown hypocotyls indicates a global reduction of cellular and supracellular CMT alignment . Altogether , these analyses show CMTs globally follow the stress pattern inferred from the cell separations in qua1-1 ( Figure 3K and L ) . To further explore the relation between mechanical stress and CMT orientation in the hypocotyl , we next analyzed the CMT response to ablation in this tissue . Assuming that the epidermis is under tension , ablations should disrupt pre-established stress pattern and lead to circumferential tensile stress pattern and CMT orientations ( Hamant et al . , 2008; Sampathkumar et al . , 2014 ) . As expected , reorientation of CMT arrays was observed in hypocotyls 8 hr after ablation , confirming that CMTs can align along maximal tension in hypocotyls too ( Figure 5D and E ) . Interestingly , the CMT orientation was however not fully circumferential after ablation . In particular , we sometimes observed radial CMT orientations at the opposite edges along the longitudinal axis of the ablation site , matching the growth-derived longitudinal tensile stress pattern in the hypocotyl ( Figure 5—figure supplement 2A and B ) . Conflicts between superposing mechanical stress patterns have been modeled and reported before , notably at the organ-boundary in shoot apical meristems: in that domain , the circumferential CMT orientation after ablation is also mixed , because the boundary is a site of highly anisotropic stresses ( see Figure S7 in [Hamant et al . , 2008] ) . If such a conflict was not present , one would expect a true circumferential CMT orientation ( Figure 5—figure supplement 2C ) . In contrast , our quantifications of the mixed CMT orientations after ablation implies that such a conflict of stress patterns is prominent in hypocotyls ( Figure 5—figure supplement 2D–F ) , further consolidating the presence of longitudinal tensile stress in the hypocotyl epidermis . Overall , CMT orientation and anisotropy correlate well with the tensile stress pattern inferred from the cell separation patterns in qua1-1 . In plants , the outer epidermal wall is inherited from the zygote and its continuity may be one of the key element to propagate tensile stress over several cell files ( Galletti et al . , 2016 ) . Yet , this still remains to be demonstrated . The analysis of CMTs in qua1-1 offers us the unique opportunity to test that hypothesis . We thus investigated whether the supracellular microtubule alignment with maximal tensile stress requires an intact outer wall . We introgressed the GFP-MBD microtubule reporter in the qua1-1 mutant and analyzed microtubule behavior in light-grown hypocotyls , where cracks are mostly superficial early on . Whereas CMTs looked normal within individual qua1-1 cells , they displayed less consistent orientation at the tissue scale in qua1-1 cells than in the wild type ( Figures 3L and 5A ) . When grown on 1% agar medium , for GFP-MBD and qua1-1 GFP-MBD , the mean θM was 109 ± 28° and 107 ± 40° , the mean anisotropy was 0 . 21 ± 0 . 08 and 0 . 21 ± 0 . 09 and R was 0 . 6 and 0 . 36 , respectively ( n = 413 cells in 7 samples and 1019 cells in 16 samples respectively ) . In other words , the distribution of CMT angles was broader in the qua1-1 mutant . Note that this discrepancy was also registered when calculating the CMT orientation resultant vector length for each plant individually ( allowing us to compare cell populations and test whether the difference is significant ) : we found a mean resultant vector length for GFP-MBD samples of 0 . 69 ± 0 . 14 and 0 . 42 ± 0 . 23 for qua1-1 GFP-MBD and the populations were significantly different ( t-test p-value=0 . 012 ) . Our analysis quantifying the coordination of CMTs across the whole tissue shows that cells in qua1-1 hypocotyls are less coordinated across the tissue than in the wild type . The presence of cracks may be the main reason for this response: cracks generate local perturbations in the stress pattern , such that CMTs tend to reorganize around these cracks , as shown for cell ablations ( Figure 5L–N and e . g . [Hamant et al . , 2008; Hervieux et al . , 2016] ) . In turn , the actual cell-to-cell coordination does not seem to be affected when two neighboring cells are still attached . Consistently , the anisotropy of CMTs in qua1-1 GFP-MBD was not decreased compared to the wild type . In many instances the anisotropy of CMT arrays was higher in qua1-1 GFP-MBD and CMTs were somehow oriented following the contours of large gaps , thus apparently following a new tension pattern allowed by the continuity of the cells where they are still adhesive ( Figure 5A , L–N ) . Our analysis thus reveals that local cell separations lead to the loss of the tissue-scale organization of the CMTs that is usually found in the wild-type . Therefore , while these data further support a scenario in which epidermis continuity is required for supracellular CMT behavior , the cell-cell separations are too strong to conclude . When grown on 2 . 5% agar medium , the wild type and qua1-1 mutant hypocotyls exhibited similar CMT behaviors: we found a mean resultant vector length for GFP-MBD ( 2 . 5% agar ) samples of 0 . 69 ± 0 . 15 and 0 . 65 ± 0 . 17 for qua1-1 GFP-MBD ( 2 . 5% agar ) ( n = 606 cells in 11 samples and 865 cells in 15 samples , respectively ) and the populations were not significantly different ( t-test p-value=0 . 556 ) . This is consistent with the phenotypic rescue of the cracks in these conditions , which likely restores the mechanical continuity of the epidermis ( Figure 5B , C ) . To test the contribution of epidermis integrity for supracellular CMT orientation along maximal tensile stress , and building on the observed rescue of CMT behavior on 2 . 5% agar medium , we next modified the mechanical stress pattern in qua1-1 artificially , using growth conditions in which cracks are only starting to appear . As already shown , cortical microtubule orientation became mostly circumferential in wild type hypocotyls 8 hr after ablation ( Figure 5D and E ) . Strikingly , this response was dramatically reduced in qua1-1 ( Figure 5F and G ) . To quantify this response , we calculated the acute angle between the ablation site and the main orientation of the CMT arrays obtained with the FibrilTool macro , for each cell , at t = 0 hr and t = 8 hr after the ablation ( Figure 5—figure supplement 2K ) . We thus obtained angles ranging from 0° to 90° ( 90° corresponding to circumferential CMT orientations around the ablation site , that is parallel to predicted maximal tension ) . At t = 0 hr , the GFP-MBD line exhibited a mean angle of 47 ± 26° showing an homogeneous distribution and no preferred orientation of the microtubules relative to the ablation site ( n = 101 cells in 10 samples; Figure 5H ) . At t = 8 hr the mean angle shifted to 58 ± 25° , and the distribution exhibited a strong skewing towards the 80 to 90° angles ( n = 101 cells in 10 samples; Figure 5I ) , demonstrating a significant reorganization of the CMT arrays around the ablation site ( p-value=0 . 002 ) . The qua1-1 GFP-MBD line however , did not show a significant reorganization of its CMT arrays 8 hr after the ablation ( Figure 5J and K , Mean angle at t = 0h: 45 ± 26° and t = 8h: 50 ± 26° , n = 107 cells in 10 samples; p-value=0 . 186 ) . Whereas the distribution of angles still showed a relative skewing towards 90° , the skewness was much smaller than in the GFP-MBD line: −0 . 23 in qua1-1 GFP-MBD t = 8 hr vs . −0 . 77 in GFP-MBD t = 8 hr ( 0 , 08 in qua1-1 GFP-MBD t = 0 hr and −0 . 04 in GFP-MBD t = 0 hr , Figure 5K ) . In addition a statistical test comparing the skewness of the distribution to a corresponding normal distribution showed that the population of angles are not significantly skewed in GFP-MBD t = 0 hr , qua1-1 GFP-MBD t = 0 hr and qua1-1 GFP-MBD t = 8 hr ( p-value=0 . 858 , 0 . 718 and 0 . 255 respectively ) , while GFP-MBD t = 8 hr shows a significant skewing ( p-value=0 . 002 ) . Overall , these results demonstrate that a discontinuous outer wall hampers the ability of CMTs to align with supracellular maximal tension in the epidermis ( Figure 5O ) . A defining feature of the epidermis in animals is its continuity , which allows a build-up of tension and in turn promotes adhesion and the coordinated behavior of epidermal cells , notably through local cadherin-based cell-cell adhesion ( Galletti et al . , 2016 ) . This loop can be referred as the ‘tension-adhesion feedback loop’ . The conclusions from the present study support this picture in plants too , notably as tension patterns could be revealed by the early cell-cell adhesion defects in qua1-1 and as adhesion defects in qua1-1 hinder the propagation of tensile stress ( Figure 6 ) . The physics of pressure vessel , and more generally of solid mechanics , has been instrumental in the derivation of stress patterns in plants , very much in the spirit of D’Arcy Thompson’s legacy ( Thompson , 1917 ) . Yet , the role of mechanical stress in cell and developmental biology remains a subject of heated debate , notably because forces are invisible and stress patterns can only be predicted . Here we used the simplest possible tool to reveal stress pattern in plants , building on cell adhesion defects and their relation to tension . Our results formally validate several conclusions from previous computational models and stress patterns deduced from ablation experiments . Incidentally , this work also brings further experimental proof of the presence of tissue tension , building on the initial work by Hofmeister ( Hofmeister , 1859 ) and Sachs ( Sachs , 1878 ) . We believe this work can serve as a reference for further studies on cell behavior in the corresponding tissues , notably to relate the behavior of cell effectors to stress patterns . Based on these data , one can also revisit previously published work on microtubule behavior in various organs . For instance , it had been reported that CMTs constantly rotate in young hypocotyl until they reach either their transverse or longitudinal orientation ( Chan et al . , 2007 ) . Although this behavior is still an intriguing phenomenon , it could partly be related to the fact that in this tissue , ( transverse ) shape-derived and ( longitudinal ) growth-derived stresses are in competition . Slight oscillations in growth rate ( Bastien et al . , 2016 ) could periodically affect the ratio of shape- and growth-derived stress and thus trigger this rotating behavior of the microtubules . Another exciting avenue for the future is the analysis of the integration between such tensile stress patterns and other cues such as blue light , and its effect through phototropin and katanin ( Lindeboom et al . , 2013 ) , or hormones ( Vineyard et al . , 2013 ) on microtubule reorientation . By definition , mechanical forces can be propagated at the speed of sound , like acoustic waves . This may provide close to instantaneous coordinating cues to tissues , as proposed in the Drosophila wing disc where cell division stops synchronously ( Shraiman , 2005 ) . Such a mechanism , in turn requires a tightly maintained and controlled cell-to-cell adhesion ( Verger et al . , 2016 ) in order to constantly maintain the mechanical conductivity of the tissue . Yet , the heterogeneous nature of the tissue would add noise to the signal propagation , and so far this question has not been formally addressed in developmental biology . Here we provide evidence that the continuity of the outer cell wall is required for the coordinated response of adjacent cells to mechanical stress . In other words , a cell would experience different tension level and orientation between a situation where it would be separated from its neighbors ( typically , cell shape would be sufficient to prescribe a tensile stress pattern , as in [Sampathkumar et al . , 2014] ) and a situation where it would still adhere to its neighbors ( tensile stress could build up at the outer epidermal wall , to a magnitude higher than cell shape-derived stress , and with a pattern that would depend on tissue shape and differential growth , as in [Hamant et al . , 2008; Louveaux et al . , 2016] ) . This opens exciting prospect for the future , as cell variability and growth heterogeneity is attracting increasing attention because of its , sometimes counterintuitive , instructive function in development ( Hong et al . , 2018 ) . Note that although CMT alignments match maximal tensile stress direction , as predicted from adhesion defects in qua1 , this does not necessarily mean that microtubule , and cellulose , become aligned to promote cell-cell adhesion . In fact , mutants in microtubule dynamics or cellulose deposition have not been reported to exhibit adhesion defects . It is therefore more likely that other factors , and most likely actin through its impact on pectin delivery to cell walls ( Mathur et al . , 2003 ) , plays a critical role in the tension-induced reinforcement of cell-cell adhesion , in parallel to microtubule-driven cell wall reinforcement . Our conclusions also echo the recent analysis of the DEFECTIVE KERNEL 1 ( DEK1 ) protein , which was shown to be required for tensile stress perception via its association with a mechanosensitive Ca2+ channel ( Tran et al . , 2017 ) . Interestingly , DEK1 RNA interference lines exhibit loss of cell adhesion in their epidermis ( Johnson et al . , 2005 ) . Our finding in qua1 thus allows us to revisit these results , suggesting that plant epidermis requires tensile stress perception , and in turn , that cell-cell adhesion allows tensile stress propagation . While the coordinating role of the outer wall is difficult to match with a comparable structure in animals , the basement membrane may in principle have a similar role , given its continuity and key role in both adhesion and mechanotransduction . Based on our results in plants , the analysis of basement membrane continuity , and its disruption , may very well help understand how consistent supracellular epidermal patterns relate to mechanical stress , in parallel to the well-established role of cadherin and stress in cell-cell adhesion and epidermal functions ( Galletti et al . , 2016 ) . The qua1-1 ( WS-4 ) T-DNA insertion line , the GFP-MBD ( WS-4 ) microtubule reporter line the p35S::PMEI5 ( Col-0 ) and the pPDF1::mCit:KA1 ( Col-0 ) L1 expressed plasma membrane marker , were previously reported in ( Bouton et al . , 2002; Marc et al . , 1998; Wolf et al . , 2012; Simon et al . , 2016 ) respectively . The qua1-1 line was genotyped by PCR using the primers described in ( Bouton et al . , 2002 ) and the p35S::PMEI5 homozygous lines were selected based on their strong phenotype ( Wolf et al . , 2012; Müller et al . , 2013 ) . Arabidopsis thaliana seeds were cold treated for 48 hr to synchronize germination . Plants were then grown in a phytotron at 20°C , in a 16 hr light/8 hr dark cycle on solid custom-made Duchefa ‘Arabidopsis’ medium ( DU0742 . 0025 , Duchefa Biochemie ) . Seedling age was counted from the start of light exposure . For dark-grown etiolated hypocotyls , seeds were exposed to light for 4 hr to induce germination . The plates were then wrapped in three layers of aluminum foil to ensure skotomorphogenesis . Naked shoot apical meristems were obtained by adding 10 µM of NPA ( N- ( 1-naphthyl ) phthalamic acid ) in the medium as described in ( Hamant et al . , 2014 ) . For time-lapse images of cell separation dynamics , seedlings were first grown on ‘Arabidopsis’ medium containing 2 . 5% agar; once cotyledons just opened , they were mounted on ‘Arabidopsis’ medium containing 1% agar and imaged for up to 72 hr every 12 hr . During image acquisition , the seedlings were immersed in water supplemented with 1 ml of PPM ( PPM-Plant Preservative Mixture , Kalys ) per liter of medium to prevent contamination . After each acquisition the water was removed and the plants were placed back in a phytotron ( see growth conditions above ) . Water potential of the medium was changed using either higher agar concentration ( ( 1% and 2 . 5% ) [Owens and Wozniak , 1991] ) or increasing Polyethylene Glycol . We used a PEG-infused plates method adapted from ( van der Weele et al . , 2000 ) . Classic 1% agar ‘Arabidopsis’ medium ( as described above ) , as well as liquid ‘Arabidopsis’ medium containing various concentrations of PEG ( PEG20000 , Sigma-Aldrich ) were prepared . The liquid medium was supplemented with 1 ml of PPM ( PPM-Plant Preservative Mixture , Kalys ) per liter of medium to prevent contamination . The liquid medium osmolarity was measured using a cryoscopic osmometer ( Osmomat 030 , Gonotec ) . Solid medium petri dishes were made , let to solidify for about 2 hr , and an equal volume of liquid medium was poored on top . After 24 hr of diffusion , the liquid medium was recovered and its osmolarity was measured again . The increase of osmolarity due to PEG diffusion in the solid medium was deduced from the difference of osmolarity of the liquid medium before and after diffusion . The petri dishes were let to dry for about 2 hr and the seeds were sown . For cell wall staining , plants were immersed in 0 . 2 mg/ml propidium iodide ( PI , Sigma-Aldrich ) for 10 min and washed with water prior to imaging . Ablations were performed as previously described in Uyttewaal et al . , 2012: seedlings were mounted horizontally with 2% low melting agarose ( Sigma-Aldrich , St . Louis , MO , USA ) on ‘Arabidopsis’ medium containing 1% agar and imaged immediately after the ablation and 8 hr after the ablation . The ablations were performed manually with a fine needle ( Minutien pin , 0 . 15 mm rod diameter , 0 . 02 mm tip width , RS-6083–15 , Roboz Surgical Instrument Co . ) ablating approximately five epidermal cells and some cells from the inner layers . Because the size of the ablation can vary from one sample to another , ablations were originally performed on a large number of samples and hypocotyls with comparable number of ablated epidermal cells ( approx . 5 ) were further imaged and analyzed . For imaging , samples were either placed on a solid agar medium and immersed in water , or placed between glass slide and coverslip separated by 400 µm spacers to prevent tissue crushing . Images were acquired using a Leica TCS SP8 confocal microscope . PI excitation was performed using a 552 nm solid-state laser and fluorescence was detected at 600–650 nm . GFP excitation was performed using a 488 nm solid-state laser and fluorescence was detected at 495–535 nm . mCitrine excitation was performed using a 514 nm solid-state laser and fluorescence was detected at 520–555 nm . Stacks of 1024 × 1024 pixels optical section were generated with a Z interval of 0 . 5 µm for GFP-MBD , 1 µm for PI or 0 . 5 µm when GFP-MBD and PI were acquired at the same time and 0 . 25 µm for mCit:KA1 . Stereomicroscopy images were taken using a leica MZ12 stereo microscope with an axiocam ICc5 Zeiss CCD camera . AFM determination of apparent stiffness k and turgor pressure P in cotyledon epidermis was performed as in ( Beauzamy et al . , 2015 ) with modifications . Specifically , the adaxial surface of 3 day old cotyledons was measured . Dissected cotyledons grown on different agar concentrations were mounted with Patafix ( UHU ) and subsequently submerged in water for measurement , whereas whole seedlings grown on different PEG concentrations were mounted in 2% low-melting agarose ( Sigma-Aldrich ) and submerged in liquid Arabidopsis medium ( DU0742 . 0025 , Duchefa Biochemie ) supplemented with D-Mannitol ( Sigma-Aldrich ) to reach the same osmotic pressure of PEG-infused solid medium . Mannitol was used to prevent potential interference of the high viscosity of PEG solutions , and each measurement was performed under 20 min to reduce osmolite uptake . For additional AFM measurements ( Figure 2—figure supplement 1 and Figure 4—figure supplement 1 ) , whole seedlings grown on different agar concentrations were mounted in 2% low melting agarose ( Sigma-Aldrich ) and immersed in liquid Arabidopsis medium or water depending on the experiment . A BioScope Catalyst AFM ( Bruker ) was used for measurement with spherical-tipped AFM cantilevers of 400 nm tip radius and 42 N/m spring constant ( SD-SPHERE-NCH-S-10 , Nanosensors ) . For topography , peak force error and DMT modulus images , PeakForce QNM mode of the acquisition software was used , with peak force frequency at 0 . 25 kHz and peak force set-point at 1 μN for wild-type and 200 nN for qua1-1 due to their innate difference in stiffness . Larger peak force set-point frequently damaged qua1-1 sample surface . 128*128 pixels images of 30*30 μm2 area were recorded at 0 . 1 Hz scan rate . For Young’s modulus , apparent stiffness and turgor pressure measurements , 1 to 2 μm-deep indentations were performed along the topological skeletons of epidermal cells to ensure relative normal contact between the probe and sample surface . At least three indentation positions were chosen for each cell , with each position consecutively indented three times , making at least nine indentation force curves per cell . Cell registration of AFM force curves were performed with the NanoIndentation plugin for ImageJ ( https://fiji . sc/ ) as described in ( Mirabet et al . , 2018 ) . Parameters for turgor deduction were generated as follows . Cell wall elastic modulus E and apparent stiffness k were calculated from each force curve following ( Beauzamy et al . , 2015 ) . Cell surface curvature was estimated from AFM topographic images , with the curvature radii fitted to the long and short axes of smaller cells or along and perpendicular to the most prominent topological skeleton of heavily serrated pavement cells . Turgor pressure was further deduced from each force curve ( 100 iterations ) with the simplified hypothesis that the surface periclinal cell walls of leaf epidermis has constant thickness ( 200 nm ) , and cell-specific turgor pressure is retrieved by averaging all turgor deductions per cell . AFM-measured mechanical properties were used to deduce outer cell wall tension in cotyledon epidermis . Cells were considered as spherical thin-walled pressure vessels , with the stress equals to σ=Pr2twhere P is turgor pressure , r is the vessel radius as the inverse of cell surface mean curvature , and t is assumed cell wall thickness . Since t is assumed constant , σ only depends on turgor pressure and surface topography . The procedure to perform the automated detection of cell separations is described in more detail at Bio-protocol ( Verger et al . , 2018 ) . We developed a semi-automated image analysis pipeline in python language in order to detect and analyze cell separations in a tissue ( Verger and Cerutti , 2018; copy archived at https://github . com/elifesciences-publications/Cell_separation_analysis ) . Input images are 2D Z-projections from confocal Z-stack . The script works by segmenting cell separation based on a threshold detection method of pixels intensity . In the case of clear gaps between the cells ( as in cotyledons ) the pixel intensity is much lower . The threshold allows a segmentation of theses gaps with low intensity pixels . In the case of bright stripes appearing between the cells ( as in stem apices or light-grown hypocotyls ) , an opposite threshold is used , segmenting only the high intensity pixels zones . Because this threshold may vary from one image to the other , it was manually defined using the ImageJ threshold tool before running the script . Running the script then , labels the different zones , measures their areas , performs a principal component analysis of the label in order to determine their main orientations ( principal component , θG ) and assigns an anisotropy and principal angle to each labeled region . The output of the script is an image of every cell separation segmented with a visual representation of their anisotropy , and a polar histogram giving a visual representation of the global result . The python script was developed and run in the TissueLab environment of the OpenAleaLab platform ( [Cerutti et al . , 2017] , https://github . com/VirtualPlants/tissuelab ) and using functions from the python libraries SciPy ( www . scipy . org ) , NumPy ( www . numpy . org ) , Pandas ( pandas . pydata . org ) and Matplotlib ( matplotlib . org , see Figure 1—figure supplement 4 ) . Original images were confocal Z-stacks . We used the MorphoGraphX software ( Barbier de Reuille et al . , 2015 ) to recover only the outer epidermal cortical microtubules signal . Further analysis was performed in 2D using the imageJ morpholibJ library ( Legland et al . , 2016 ) , to segment cells , the analyze particle tool to define ROIs and an automated version the FibrilTool macro ( Boudaoud et al . , 2014 ) to analyze the principal orientation ( θM ) and anisotropy of the CMT arrays ( see Figure 5—figure supplement 1 ) . For the CMT response to ablation , only the cells directly surrounding the ablation were analyzed . Points were manually placed and used to calculate the acute angle between the ablation site and the orientation of the microtubule array for each cell ( see Figure 5—figure supplement 1 ) . For linear data ( comparing gap area , anisotropy ) , classical statistical tests were used . Normality of the samples was tested using Shapiro’s test . If at least one of the sample population did not have a normal distribution , the populations were compared with the non parametric Wilcoxon Rank Sum test . If both samples had normal distributions , their variance were compared using Bartlett’s test . If they were equal , a Student’s t-test was performed; if they were unequal , a Welch’s t-test was performed . For circular ( or directional ) data ( cell separation orientation and CMT orientation ) , different statistics were used . Since in our case a 0° angle is equal to an 180° angle within a semicircle , the mean , standard deviation , variance , and the accompanied statistical test have to take this fact into account and circular statistics were used . Most of our sample did not show a Von Mises distribution ( equivalent of the normal distribution ) , thus only non-parametric tests were used . The Rao’s spacing test was used to determine if the populations of angles were homogeneously distributed , or had a preferred orientation determined by the circular mean . Finally the data form the CMT response to ablation were treated as linear data , since the calculated angle was between 0° and 90° . The symmetry of CMT response distribution was measured by the skewness of the population towards 90° and the significance of the skewness was tested against a normal distribution . All statistical analyses were performed in python using the scipy . stats library ( scipy . org ) for linear data and the pycircstat library ( github . com/circstat/pycircstat ) for circular data .

The parts of a plant that protrude from the ground are constantly shaken by the wind , applying forces to the plant that it must be able to resist . Indeed , mechanical forces are crucial for the development , growth and life of all organisms and can trigger certain behaviours or the production of particular molecules: for example , forces that bend a plant trigger gene activity that ultimately makes the stem more rigid . Mechanical forces can also originate from inside the organism . For example , the epidermal cells that cover the surface of a plant are placed under tension by the cells in the underlying layers of the plant as they grow and expand . The exact pattern of forces in the plant epidermis was not known because they cannot be directly seen , although scientists have tried to map them using theoretical and computational modeling . A mutant form of the Arabidopsis plant is unable to produce some of the molecules that allow epidermal cells to adhere to each other . Verger et al . placed the mutants in different growth conditions that lowered the pressure inside the plant , and consequently reduced the tension on the epidermal cells . This partly restored the ability of epidermal cells to adhere to each other , although gaps remained between cells in regions of the plant that have been predicted to be under high levels of tension . Verger et al . could therefore use the patterns of the gaps to map the forces across the epidermis , opening the path for the study of the role of these forces in plant development . Further experiments showed that cell adhesion defects prevent the epidermal cells from coordinating how they respond to mechanical forces . There is therefore a feedback loop in the plant epidermis: cell-cell connections transmit tension across the epidermis , and , in turn , tension is perceived by the cells to alter the strength of those connections . The results presented by Verger et al . suggest that plants use tension to monitor the adhesion in the cell layer that forms an interface with the environment . Other organisms may use similar processes; this theory is supported by the fact that sheets of animal cells use proteins that are involved in both cell-cell adhesion and the detection of tension . The next challenge is to analyse how tension in the epidermis affects developmental processes and how a plant responds to its environment .

2018

A tension-adhesion feedback loop in plant epidermis

Circadian clocks coordinate 24-hr rhythms of behavior and physiology . In mammals , a master clock residing in the suprachiasmatic nucleus ( SCN ) is reset by the light–dark cycle , while timed food intake is a potent synchronizer of peripheral clocks such as the liver . Alterations in food intake rhythms can uncouple peripheral clocks from the SCN , resulting in internal desynchrony , which promotes obesity and metabolic disorders . Pancreas-derived hormones such as insulin and glucagon have been implicated in signaling mealtime to peripheral clocks . In this study , we identify a novel , more direct pathway of food-driven liver clock resetting involving oxyntomodulin ( OXM ) . In mice , food intake stimulates OXM secretion from the gut , which resets liver transcription rhythms via induction of the core clock genes Per1 and 2 . Inhibition of OXM signaling blocks food-mediated resetting of hepatocyte clocks . These data reveal a direct link between gastric filling with food and circadian rhythm phasing in metabolic tissues . Extended night or rotating shift work is associated with an elevated risk for developing cancer , cardiovascular disease , immune deficiency , mood disorders , and metabolic alterations ( Rosenberg and Doghramji , 2011; Herichova , 2013 ) . One major factor believed to contribute to this adverse health impact of shift work is a disruption of endogenous circadian clocks by mistimed resetting stimuli , so called Zeitgebers , as a consequence of altered sleep/wake schedules . Most organisms have evolved internal timekeepers to anticipate the environmental changes brought about by the Earth's rotation around its axis . In mammals , these so called circadian clocks are based on ubiquitously expressed cellular interlocking transcriptional–translational feedback loops ( TTLs ) of clock genes/proteins ( Albrecht , 2012 ) . In the core TTL , the transcriptional activators circadian locomotor output cycles kaput ( CLOCK ) and brain and muscle ARNT-like 1 ( BMAL1; ARNTL ) regulate the expression of two Cryptochrome ( Cry1/2 ) and three Period ( Per1-3 ) genes . Towards the end of the day , PER and CRY proteins translocate into the nucleus where they inhibit their own abundance via inhibition of CLOCK/BMAL1 . Further accessory loops serve to stabilize this 24-hr feedback rhythm and integrate the clock with cellular processes ( Takahashi et al . , 2008 ) . The clock machinery regulates physiology via orchestration of tissue-specific rhythmic expression of clock output genes ( Yan et al . , 2008 ) . The external light–dark cycle is the most prominent Zeitgeber of the central circadian pacemaker located in the suprachiasmatic nucleus ( SCN ) of the hypothalamus ( Golombek and Rosenstein , 2010 ) . The SCN receives light information from the retina and synchronizes peripheral clocks throughout the body via neuronal and hormonal pathways ( Dibner et al . , 2010 ) . While the SCN itself is largely non-responsive to non-photic timing signals such as food intake , meal timing is an important Zeitgeber for clocks in peripheral tissues ( Stokkan et al . , 2001 ) . If food access is restricted to the normal rest phase of an organism , that is , the night for humans or daytime for nocturnal rodents , peripheral clocks become uncoupled from the SCN and adapt to the timing of food availability ( Damiola et al . , 2000 ) . Shift workers often eat at times when their digestive timing system is poorly prepared for food ( Lowden et al . , 2010 ) . Animal studies suggest that food intake during the normal rest phase promotes obesity ( Arble et al . , 2009; Hatori et al . , 2012 ) and peripheral circadian uncoupling has been suggested to contribute to the development of metabolic disorders in night shift workers ( Antunes et al . , 2010; Barclay et al . , 2012 ) . Various other factors can regulate clock gene expression in peripheral tissues , including glucocorticoids and changes in body temperature or autonomic signaling ( Dibner et al . , 2010 ) . The mechanisms of food-dependent peripheral clock resetting , however , remain poorly understood . Metabolic hormones such as insulin , ghrelin , and glucagon ( GCG ) have been shown to affect circadian rhythms associated with food restriction ( LeSauter et al . , 2009; Tahara et al . , 2011; Chaves et al . , 2014; Sun et al . , 2015 ) . While ghrelin appears to act primarily on the brain , insulin and GCG levels are mainly regulated via blood glucose . However , it was shown that carbohydrate intake alone has only a minor phase resetting capacity , while complex foods show much stronger effects ( Hirao et al . , 2009 ) , indicating that other factors must be involved . Besides the pancreas , other organs—notably including the gastrointestinal tract itself—show acute hormonal responses to fasting or feeding ( Stanley et al . , 2005 ) . This led us to hypothesize that postprandial , gut-derived signals may be implicated in food-driven resetting of peripheral clocks . In a screen using rhythmic liver slice cultures , we identified oxyntomodulin ( OXM ) as a potent resetting signal of liver circadian clocks . OXM is an anorexigenic incretin hormone produced in the gut by prohormone convertase 1/3-driven cleavage of the precursor preproglucagon ( for review see Drucker , 2005 ) . It modulates energy and glucose metabolism by acting on various tissues , including brain , liver , and pancreas ( Baldissera et al . , 1988; Gros et al . , 1993 ) . Since OXM secretion is dependent on food intake , we hypothesized that OXM may directly link food intake to hepatic transcriptional activity by resetting of the liver clock . We screened a commercially available metabolic peptide library ( Obesity Peptide Library , Phoenix Europe GmbH; DE ) for factors capable of resetting luciferase activity rhythms in organotypic liver slice cultures from Per2::LUC circadian reporter mice ( Yoo et al . , 2004 ) . Interestingly , out of 200 peptides applied during the descending phase ( ∼180° , corresponding to the early morning ) of the luciferase activity rhythm , only a few produced marked phase shifts , including three proglucagon-derived peptide ( PGDP ) hormones: exendin-4 , OXM , and GCG ( Figure 1—source data 1 ) . Exendin-4 has been isolated from the salivary gland of the Gila monster , with no analogue in rodents or humans . To compare the effectiveness of mammalian PGDPs in liver clock resetting , we treated slices with increasing doses of OXM , GCG , and the three other commercially available PGDPs , glicentin-related pancreatic polypeptide ( GRPP ) , glucagon-like peptide-1 ( GLP-1 ) , and glucagon-like peptide-2 ( GLP-2 ) ( Figure 1A ) . GLP-1 , GLP-2 , and GRPP ( 0 . 5–450 nM ) had no significant resetting effects on PER2::LUC phase compared to PBS-treatment ( Figure 1B–D ) . GCG resulted in phase delays of up to 3 hr , but only at relatively high concentrations ( Figure 1E ) . In contrast , OXM reset PER2::LUC rhythms in liver slices at much lower doses , resulting in phase delays of up to 8 hr at higher concentrations ( Figure 1F ) . To act as a true Zeitgeber signal one would expect differential OXM effects depending on treatment time , that is , a circadian gating effect . We tested this by applying OXM at different phases of the PER2::LUC rhythm . To validate the setup , slices were treated at different PER2::LUC phases with 100 μM of the glucocorticoid analog dexamethasone ( DEX ) , which was previously shown to reset hepatocyte clocks in vivo ( Balsalobre et al . , 2000 ) . In a phase-dependent manner DEX treatment reset PER2::LUC activity rhythms in slices ( Figure 1—figure supplement 1 ) . Very similar to what had been observed after DEX treatment in animals ( Balsalobre et al . , 2000 ) , application in the first quarter of the PER2::LUC activity rhythm ( 0–90° ) resulted in phase delays , while later treatments produced phase advances ( 100–180° ) or had no marked effect ( around 270° ) . Likewise , OXM effects were phase dependent . Delays were predominantly observed at 90–210° of the PER2::LUC cycle with a maximum around 180° , while only modest phase shifts were seen at 270–360° ( Figure 1G ) . Though GCG also showed potential in resetting liver clock rhythms , OXM emerged as the most potent liver clock synchronizer from our screen . Moreover , contrary to GCG , OXM secretion is directly induced by food consumption in humans ( Le Quellec et al . , 1992 ) , making it an attractive candidate for linking meal timing and clock function . Therefore , we focused on OXM for further analyses . 10 . 7554/eLife . 06253 . 003Figure 1 . Oxyntomodulin ( OXM ) phase- and dose-dependently resets circadian clocks in liver slices . ( A ) Schematic sequence of the proglucagon-derived peptides ( GRPP—glicentin-related pancreatic peptide; GLIC—glicentin; OXM—oxyntomodulin; GCG—glucagon; IP-1—intervening peptide-1; GLP-1—glucagon-like peptide-1; IP-2—intervening peptide-2; GLP-2—glucagon-like peptide-2 ) . ( B–F ) Example luminescence traces and dose-dependent responses for GLP-1 ( B; F ( 6 , 28 ) = 1 . 509 ) , GLP-2 ( C; F ( 6 , 28 ) = 1 . 530 ) , GRPP ( D; F ( 6 , 28 ) = 1 . 151 ) , GCG ( E; F ( 6 , 28 ) = 3 . 569 ) , and OXM ( F; F ( 6 , 28 ) = 8 . 790 ) -induced phase resetting of PER2::LUC rhythms in liver slices treated at 180–200° . Data are presented as mean ± S . E . M . ( n = 5 ) . One-way ANOVA ( F-values with degrees of freedom provided in brackets ) : *p < 0 . 05; **p < 0 . 01; ***p < 0 . 001 . Asterisks indicate significant differences relative to PBS treatment ( white bars ) . ( G ) Phase response curve for OXM-induced phase resetting of PER2::LUC rhythms in liver slices . Circles: raw data of individual slices; dashed line: sine wave regression with harmonics . DOI: http://dx . doi . org/10 . 7554/eLife . 06253 . 00310 . 7554/eLife . 06253 . 004Figure 1—source data 1 . Table of effects of metabolic peptide treatment on PER2::LUC liver slice rhythms . DOI: http://dx . doi . org/10 . 7554/eLife . 06253 . 00410 . 7554/eLife . 06253 . 005Figure 1—figure supplement 1 . Phase response curve for dexamethasone ( DEX ) treatment in Per2::LUC liver slice cultures . Black dots: phase shifts of individual DEX treatments ( 100 µM ) ; dashed line: sine wave regression with first and second order harmonics ( CircWave ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06253 . 005 So far no OXM-specific receptor has been identified; however , OXM can bind to and activate both GCG and GLP-1 receptors ( Jorgensen et al . , 2007 ) . Gcgr transcripts are strongly expressed in the murine liver ( Sinclair et al . , 2008 ) . In contrast , the majority of previous studies failed to detect a full-length Glp1r mRNA in murine hepatocytes ( Campos et al . , 1994; Dunphy et al . , 1998; Panjwani et al . , 2013 ) . We performed RT-PCR analyses for all annotated coding Glp1r exons on cDNA preparations from wild-type mouse livers with pancreas as positive control . Glp1r transcripts were present in pancreas , but undetectable in liver samples ( Figure 2—figure supplement 1 ) , in line with the absence of significant liver clock resetting effects of GLP-1 ( Figure 1B ) and potent resetting of PER2::LUC rhythms by OXM treatment in slices from PER2::LUC x Glp1r−/− mice ( Figure 2A ) . On the other hand , blocking glucagon receptors ( GCGR ) signaling by co-treatment with 2- ( 4-Pyridyl ) -5- ( 4-chlorophenyl ) -3- ( 5-bromo-2-propyloxyphenyl ) pyrrol ( Calbiochem Glucagon Receptor Inhibitor II; GRI-2 ) potently inhibited GCG- and OXM-induced clock resetting in Per2::LUC slices ( Figure 2B ) . 10 . 7554/eLife . 06253 . 006Figure 2 . Glucagon ( GCG ) receptor regulates phase resetting effects of OXM and GCG in Per2::LUC liver slices . ( A ) OXM-induced phase shifts in Per2::LUC and Per2::LUC x Glp1r−/− liver slices . Mann–Whitney test: ##p < 0 . 01 against solvent . ( B ) GCG and OXM-induced phase shifts in Per2::LUC slices are abolished by co-treatment with GRI-2 . One-way ANOVA with Bonferroni post-test: p < 0 . 05; ###p < 0 . 001 against solvent; *p < 0 . 05; ***p < 0 . 001 . Data are presented as mean ± S . E . M . ( n = 8 ) ; F ( 7 , 56 ) = 7 . 314 . ( C ) OXM treatment promotes binding of CREB to CRE elements at the Per1 gene promoter . One-way ANOVA with Bonferroni post-test: ***p < 0 . 001 against 0′ . Data are presented as mean ± S . E . M . ( n = 5; F ( 5 , 24 ) = 22 . 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06253 . 00610 . 7554/eLife . 06253 . 007Figure 2—figure supplement 1 . Absence of Glp1r transcripts in mouse liver . RT-PCR with different primer sets targeting all annotated coding exons of the murine Glp1r gene . Exon 1 was not tested , as it mainly contains non-coding poly-C- and poly-G-rich sequences , which precludes specific primer design . cDNA preparations from wild-type livers were tested ( lane 2 ) . Wild-type pancreas cDNA was chosen as positive ( lane 4 ) and liver cDNA from Glp1r-deficient mice and water as negative controls ( lanes 3 and 5 ) . Lane 1: 100-bp DNA ladder . DOI: http://dx . doi . org/10 . 7554/eLife . 06253 . 007 GCGR is a G protein-coupled receptor that , via protein kinase A , leads to phosphorylation and activation of the transcription factor cyclic adenosine monophosphate ( cAMP ) response element-binding protein ( CREB ) ( Gonzalez and Montminy , 1989; Dalle et al . , 2004 ) . This pathway is reminiscent of the SCN , where nocturnal light exposure induces Per gene transcription via cAMP signaling and CREB activation downstream of the N-methyl-D-aspartate receptor ( Welsh et al . , 2010 ) . To investigate if OXM would impinge on the hepatic clock machinery in a similar way , we treated liver explants with OXM and performed chromatin immunoprecipitation ( ChIP ) analysis to measure CREB binding to cAMP response elements ( CRE ) in the Per1 gene promoter . 30 min after OXM treatment , CREB binding was significantly increased at the Per1 CRE , but not at downstream sequences ( Figure 2C ) . In addition , we analyzed clock gene expression in liver slices at different intervals after treatment with OXM at 180° . Per1 expression was transiently induced 60 min after addition of OXM to the medium , returning back to normal levels after 120 min ( Figure 3A ) . Similarly , Per2 was induced by OXM after 60 min , but mRNA levels remained high even after 120 min ( Figure 3B ) . No significant effect was seen on Bmal1 expression at all time points ( Figure 3C ) . In line with the absence of OXM-induced phase shifts at 360° ( Figure 1G ) , Per1/2 and Bmal1 mRNA levels were unaffected by OXM treatment at this phase ( Figure 3D–F ) . Induction of Per1 and Per2 expression was preserved in Glp1r−/− slices , suggesting that hepatic OXM effects are independent of GLP-1R signaling ( Figure 3G , H ) . 10 . 7554/eLife . 06253 . 008Figure 3 . OXM treatment induces Per1/2 expression in organotypic liver slices . ( A–C ) WT liver slices were treated with OXM ( grey ) or vehicle ( PBS; black ) at 180° and analyzed for clock gene expression of Per1 ( A; factor treatment F ( 1 , 40 ) = 0 . 785; time F ( 3 , 40 ) = 34 . 95; interaction F ( 3 , 40 ) = 16 . 33 ) , Per2 ( B; factor treatment F ( 1 , 40 ) = 24 . 02; time F ( 3 , 40 ) = 29 . 4; interaction F ( 3 , 40 ) = 38 . 38 ) , and Bmal1 ( C; factor treatment F ( 1 , 40 ) = 0 . 108; time F ( 3 , 40 ) = 17 . 39; interaction F ( 3 , 40 ) = 3 . 607 ) by qPCR . ( D–F ) WT liver slices were treated with OXM ( grey ) or PBS ( black ) at 360° and analyzed for expression of Per1 ( D; factor treatment F ( 1 , 40 ) = 1 . 179; time F ( 3 , 40 ) = 36 . 33; interaction F ( 3 , 40 ) = 1 . 349 ) , Per2 ( E; factor treatment F ( 1 , 40 ) = 5 . 757; time F ( 3 , 40 ) = 13 . 57; interaction F ( 3 , 40 ) = 1 . 135 ) , and Bmal1 ( F; factor treatment F ( 1 , 40 ) = 4 . 112; time F ( 3 , 40 ) = 8 . 788; interaction F ( 3 , 40 ) = 0 . 491 ) by qPCR . ( G and H ) OXM-induced Per1/2 expression is retained in Glp1r−/− liver slices . Per1: factor treatment F ( 1 , 42 ) = 8 . 48; time F ( 2 , 42 ) = 10 . 95; interaction F ( 2 , 42 ) = 0 . 525 . Per2: factor treatment F ( 1 , 42 ) = 10 . 5; time F ( 2 , 42 ) = 3 . 662; interaction F ( 2 , 42 ) = 5 . 845 . Glp1r−/− liver slices were treated as described for WT above . Data are presented as mean ± S . E . M . ( n = 6 for WT and 8 for Glp1r−/− ) . Two-way ANOVA with Bonferroni post-test: **p < 0 . 01; ***p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 06253 . 008 To test if the effect of OXM on liver clock gene activity is preserved in vivo , we analyzed hepatic Per1/2 transcription after OXM treatment in wild-type mice . Analogous to what we observed in slices , robust Per1 and Per2 induction was observed after i . v . injection of OXM at Zeitgeber time ( ZT ) 3 ( Figure 4A ) . When animals were treated with OXM at the opposite phase of the LD cycle ( ZT15 ) —a time when nocturnal mice usually eat and , thus , no food-mediated clock shifts would be expected—no induction of Per gene expression was observed ( Figure 4B ) . Of note , in situ hybridization ( ISH ) showed no acute effect of OXM treatment on Per expression in the SCN at ZT3 ( Figure 4C ) , indicating that OXM acts primarily on the liver clock and in line with the observed phase stability of the SCN clock under time-restricted feeding conditions ( Damiola et al . , 2000 ) . 10 . 7554/eLife . 06253 . 009Figure 4 . OXM treatment induces Per1/2 expression and resets the liver circadian clock in vivo . ( A and B ) Hepatic Per gene expression after OXM ( grey ) or vehicle ( PBS; black ) i . v . injection at ZT3 ( A ) and ZT15 ( B ) . ZT3: Per1: factor treatment F ( 1 , 24 ) = 5 . 695; time F ( 2 , 24 ) = 34 . 74; interaction F ( 2 , 24 ) = 4 . 965; Per2: factor treatment F ( 1 , 24 ) = 64 . 84; time F ( 2 , 24 ) = 9 . 381; interaction F ( 2 , 24 ) = 6 . 915 . ZT15: Per1: factor treatment F ( 1 , 12 ) = 1 . 096; time F ( 2 , 12 ) = 0 . 005; interaction F ( 2 , 12 ) = 1 . 367; Per2: factor treatment F ( 1 , 24 ) = 0 . 255; time F ( 2 , 24 ) = 0 . 001; interaction F ( 2 , 24 ) = 1 . 172 . ( C ) Suprachiasmatic nucleus ( SCN ) signal after in situ hybridization ( ISH ) of brain sections with 35S-labelled antisense probes for Per1/2 30 min after OXM/PBS treatment at ZT3 in the same animals used in ( A ) . Left panel: representative autoradiograph scans containing the SCN; right panel: quantification of the ISH . ( D ) Resetting of Per2 and Dbp rhythms in livers of wild-type mice after an i . p . injection of either OXM ( grey ) or vehicle ( PBS; black ) after 12-hr darkness; Per2: factor treatment F ( 1 , 24 ) = 5 . 531; time F ( 5 , 24 ) = 46 . 37; interaction F ( 5 , 24 ) = 18 . 71 . Dbp: factor treatment F ( 1 , 24 ) = 0 . 094; time F ( 5 , 24 ) = 119 . 2; interaction F ( 5 , 24 ) = 38 . 58 . All data are presented as mean ± S . E . M . ( n = 3–5 ) . A , B , and D: two-way ANOVA with Bonferroni post-test: *p < 0 . 05 , **p < 0 . 01; ***p < 0 . 001; C: Mann–Whitney test . DOI: http://dx . doi . org/10 . 7554/eLife . 06253 . 009 To assess OXM effects on liver clock phase , we next treated wild-type mice with either PBS or OXM at the beginning of their rest phase on the first day in constant darkness ( DD ) and under fasting conditions , thus excluding potential confounding effects of light exposure or food intake . Per2 and Dbp expression were determined from liver cDNA preparations at different time points using qPCR . We detected phase delays of Per2 and Dbp mRNA rhythms in livers of OXM-treated mice relative to those in PBS-injected control animals ( Figure 4D ) , in line with the phase-delaying effects of OXM administration in slices at this time ( Figure 1G ) . In addition , Per2 , but not Dbp , rhythms appeared dampened after OXM injection . Tissue clocks regulate local physiology via coordination of transcriptional programs . To test if OXM treatment would impinge on hepatic energy metabolism , we analyzed the expression of important metabolic transcripts after OXM treatment . Similar to what we observed for clock gene activity , transcript profiles of genes involved in liver carbohydrate metabolism were found either phase delayed ( Foxo1 and Pdk4; Figure 5A , B ) and/or dampened ( Foxo1 and Pklr; Figure 5A , C ) . Of note Pepck , which was previously described as a clock output gene in liver ( Lamia et al . , 2008 ) was not rhythmic under these conditions , and OXM had no further effect on Pepck mRNA levels ( Figure 5D ) . Expression levels of the glucose transporter Slc2a2 ( Glut2 ) and the pyruvate transporter Slc16a7 were also dampened or phase-delayed , respectively ( Figure 5E , F ) . 10 . 7554/eLife . 06253 . 010Figure 5 . OXM treatment modulates diurnal expression profile of hepatic genes involved in liver carbohydrate metabolism . ( A–F ) Relative gene expression of Foxo1 ( A; factor treatment F ( 1 , 24 ) = 0 . 001 , time F ( 5 , 24 ) = 5 . 547 , interaction F ( 5 , 24 ) = 11 . 13 ) , Pdk4 ( B; factor treatment F ( 1 , 24 ) = 0 . 197 , time F ( 5 , 24 ) = 3 . 35 , interaction F ( 5 , 24 ) = 3 . 247 ) , Pklr ( C; factor treatment F ( 1 , 24 ) = 11 . 63 , time F ( 5 , 24 ) = 5 . 61 , interaction F ( 5 , 24 ) = 3 . 61 ) , Pepck ( D; factor treatment F ( 1 , 24 ) = 0 . 574 , time F ( 5 , 24 ) = 2 . 043 , interaction F ( 5 , 24 ) = 0 . 299 ) , the glucose transporter Slc2a2 ( E; factor treatment F ( 1 , 24 ) = 2 . 582 , time F ( 5 , 24 ) = 15 . 98 , interaction F ( 5 , 24 ) = 2 . 642 ) and the pyruvate transporter Slc16a7 ( F; factor treatment F ( 1 , 24 ) = 1 . 539 , time F ( 5 , 24 ) = 7 . 472 , interaction F ( 5 , 24 ) = 6 . 586 ) after i . p . administration of either OXM ( grey ) or vehicle ( PBS; black ) after 12 hr in darkness . Data are presented as mean ± S . E . M . ( n = 4 ) . Two-way ANOVA with Bonferroni post-test: *p < 0 . 05 , **p < 0 . 01; ***p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 06253 . 010 In summary , our data so far show that OXM treatment resets liver circadian mRNA rhythms in a phase- and dose-dependent manner , indicating that it may be involved in food-induced resetting of the liver circadian clock and metabolic machinery . In humans , OXM levels in the blood rise in response to food intake ( Le Quellec et al . , 1992 ) . To test whether this effect is conserved in mice , we determined diurnal plasma oxyntomodulin-like immunoreactivity ( OLI ) profiles in mice with ad libitum food access and in fasted animals . In fed mice , OLI levels were elevated during the active dark phase , while under fasting conditions OLI concentrations were constant and consistently low ( Figure 6A ) . In line with this , non-rhythmic Gcg mRNA expression in the gut was observed under fasting conditions ( data not shown ) . These data suggest a link between food intake and OXM secretion . To test this more directly , we used a 12-hr fasting-refeeding paradigm . Mean fasting OLI levels in the early morning ( ZT1 ) were ∼3 . 5 ng/ml , but showed high inter-individual variation ( Figure 6B ) . Upon food intake , a rapid increase ( relative to individual fasting levels ) was observed after 20 min . This effect persisted for more than 1 hr before returning to baseline levels ( Figure 6C ) . Of note , OLI induction after OXM injections were about twofold–threefold higher than what was observed after refeeding ( Figure 6—figure supplement 1 ) . To test if postprandial OLI induction is sufficient to affect liver clock gene expression , we analyzed Per1/2 mRNA levels in livers of wild-type mice after fasting-refeeding . Parallel to the rise in plasma OLI , we observed a transient postprandial increase of hepatic Per1 expression . Per2 expression showed a delayed , but a more persistent induction ( Figure 6D ) . Food-mediated Per activation was partly inhibited by treatment with purified anti-OXM IgG to neutralize the effects of endogenous OXM in wild-type ( Figure 6E ) and in Glp1r−/− mice ( Figure 6F ) . Importantly , the effects of refeeding on insulin , GLP-1 , and GCG plasma levels were not affected by treatment with anti-OXM IgG , suggesting that these peptides are not involved in the activation of food-induced hepatic Per expression ( Figure 6G–I ) . 10 . 7554/eLife . 06253 . 011Figure 6 . Endogenous OXM signaling regulates food intake-mediated resetting of the liver circadian clock . ( A ) Plasma oxyntomodulin-like immunoreactivity ( OLI ) diurnal profiles under ad libitum food and fasting conditions . Data are presented as mean ± S . E . M ( n = 6 ) ; factor time F ( 5 , 60 ) = 0 . 628 , feeding condition F ( 1 , 60 ) = 15 . 37 , interaction F ( 5 , 60 ) = 0 . 638 . Grey shading indicates the dark phase . ( B ) OLI levels show high individual variations in mice after 12 hr of food deprivation ( ZT13-1 ) . ( C ) Plasma OLI ( normalized to individual fasting levels ) after refeeding ( grey line ) or under continuous starving ( black line ) ; factor time F ( 3 , 21 ) = 3 . 544 , feeding condition F ( 1 , 21 ) = 15 . 82 , interaction F ( 3 , 21 ) = 4 . 717 . ( D ) Liver Per1/2 induction following fasting-refeeding determined by qPCR; Per1: factor time F ( 3 , 25 ) = 0 . 454 , feeding condition F ( 1 , 25 ) = 1 . 376 , interaction F ( 3 , 25 ) = 4 . 453; Per2: factor time F ( 3 , 25 ) = 6 . 938 , feeding condition F ( 1 , 25 ) = 38 . 48 , interaction F ( 3 , 25 ) = 3 . 767 . ( E ) WT and ( F ) Glp1r−/− liver Per1/2 expression after fasting-refeeding with control IgG injection ( grey ) or OXM immuno-neutralization by anti-OXM IgG ( aOXM ) injection at ZT0; WT: Per1: F ( 2 , 12 ) = 71 . 76 , Per2: F ( 2 , 12 ) = 47 . 41; Glp1r−/−: Per1 F ( 2 , 12 ) = 11 . 51 , Per2: F ( 2 , 12 ) = 5 . 585 . ( G–I ) Treatment with anti-OXM IgG does not affect postprandial regulation of insulin , GLP-1 , and GCG . Plasma levels of insulin ( G; F ( 2 , 12 ) = 17 . 44 ) , GLP-1 ( H; F ( 2 , 12 ) = 5 . 563 ) , and GCG ( I; F ( 2 , 12 ) = 7 . 128 ) after fasting-refeeding with control IgG injection ( grey ) or OXM immuno-neutralization by anti-OXM IgG ( aOXM ) treatment at ZT0 . One- ( E–I ) or two-way ANOVA ( A , C , D ) with Bonferroni post-test: *p < 0 . 05; **p < 0 . 01; ***p < 0 . 001 against fasted; ##p < 0 . 01; ###p < 0 . 001 against IgG . Data are presented as mean ± S . E . M ( n = 5 ) . ( J and K ) Liver PER2::LUC rhythms after fasting-refeeding with control IgG or αOXM administration . ( J ) Representative luminescence traces . ( K ) Comparison of phases ( second peak in culture ) after refeeding and/or anti-OXM treatment ( Data are presented as mean ± S . E . M ( n = 4 mice per condition , an average of 3 slice preparations of each mouse were used ) ; two-way ANOVA with Bonferroni post-test: *p < 0 . 05 against fasted; factor treatment F ( 1 , 12 ) = 5 . 127 , feeding condition F ( 1 , 12 ) = 13 . 02 , interaction F ( 1 , 12 ) = 5 . 044 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06253 . 01110 . 7554/eLife . 06253 . 012Figure 6—source data 1 . Primer sequences for PCR reactions . DOI: http://dx . doi . org/10 . 7554/eLife . 06253 . 01210 . 7554/eLife . 06253 . 013Figure 6—figure supplement 1 . Time course of OLI plasma levels after OXM injection . Time course of OLI appearance in plasma following i . v . ( 4 μg ) or i . p . ( 25 μg ) injections of OXM . OLI plasma level changes are expressed relative to starving levels ( 0 min ) for each individual . Data are presented as mean ± SEM ( n = 5 ) . 2-way ANOVA with Bonferroni post-test: ***p < 0 . 001; factor treatment F ( 2 , 48 ) = 9 . 014 , feeding condition F ( 3 , 48 ) = 4 . 95 , interaction F ( 6 , 48 ) = 5 . 698 . DOI: http://dx . doi . org/10 . 7554/eLife . 06253 . 013 We next tested if OXM signaling regulates food-mediated phase resetting of the liver clock using Per2::LUC slice preparations . Per2::LUC mice were starved overnight ( ZT12-24/0 ) and treated with either IgG or anti-OXM antibodies prior to re-feeding or extended starving . Animals were sacrificed at ZT4 and liver slices were prepared to determine luciferase rhythm phases . After a 12-hr fast , food intake caused a significant phase delay of PER2::LUC activity in liver slices and this effect was attenuated by OXM neutralization ( Figure 6J , K ) . No effect on PER2::LUC phase was seen after anti-OXM treatment alone ( Figure 6K ) . Together , these data suggest that elevated OXM levels in response to food intake affect hepatic clock gene expression . Neutralization of endogenous OXM signaling inhibits food-induced clock gene induction and rhythm shifts , suggesting that OXM may act as a metabolic synchronizer of hepatic clocks . Food intake resets circadian clocks in peripheral tissues . In consequence , eating during the normal rest phase leads to a state of internal circadian desynchrony which has been suggested to underlie metabolic deregulation under shiftwork conditions ( Vollmers et al . , 2009; Salgado-Delgado et al . , 2010; Barclay et al . , 2012 ) . Different mechanisms of food-related circadian synchronization have been proposed , including blood glucose-responsive peptide hormones ( Dibner et al . , 2010; Tahara et al . , 2011; Chaves et al . , 2014; Sun et al . , 2015 ) . However , food entrainment does not exclusively depend on glucose ( Hirao et al . , 2009 ) , and the molecular mechanisms underlying food entrainment of peripheral tissues remain poorly characterized ( Dibner et al . , 2010 ) . Here , we identified the incretin peptide OXM as a potential direct link between food intake in the gut and resetting of hepatic clock and metabolic gene transcription . Using an organotypic slice culture system , we found that OXM—and , as was shown recently , GCG ( Sun et al . , 2015 ) —treatment can reset PER2::LUC circadian reporter activity rhythm in a dose- and phase-dependent manner ( Figure 1 ) . Consistently , peripheral OXM injections in mice reset clock gene rhythms , as well as genes involved in hepatic carbohydrate regulation , thus impinging on hepatic energy metabolism ( Figures 4 , 5 ) . In contrast , no significant effect of OXM treatment was observed at the level of the SCN pacemaker ( Figure 4 ) . In mice , food intake during day ( i . e . , the normal rest time ) uncouples the liver clock from that in the SCN ( Lamia et al . , 2008 ) , leading to a state of internal circadian desynchrony that is associated with elevated body weight and other metabolic impairments ( Arble et al . , 2009; Hatori et al . , 2012 ) . In the liver , around 10% of the transcriptome is under circadian regulation including many genes involved in energy metabolism ( Akhtar et al . , 2002; Miller et al . , 2007 ) . Genetic ablation of the liver clock abolishes the circadian rhythms of several glucose regulatory genes and results in a perturbed diurnal profile of blood glucose ( Kornmann et al . , 2001; Lamia et al . , 2008 ) . Our experiments suggest that OXM signaling involves activation of GCGR ( Figure 2 ) . This is puzzling given that OXM displays a much greater capacity for clock gene induction in liver slices than GCG itself , despite having a lower binding affinity for GCGR ( LeSauter et al . , 2009 ) . While we cannot conclusively answer this question at the moment , there are two scenarios that might explain our observations . First , GCGR may not be the only receptor involved in OXM-mediated clock resetting . While our data suggest that GLP1R does not play a role in this context ( Figures 2 , 3 ) , an additional OXM receptor has been suggested ( Baldissera et al . , 1988; Sun et al . , 2015 ) , the activation of which may interact with GCGR downstream signaling . Second , an additional signal may be involved that acts synergistically with OXM to activate hepatic Per transcription . One such candidate could be insulin , which is similarly elevated after food intake and has previously been suggested to affect liver clock rhythms ( Yamajuku et al . , 2012; Chaves et al . , 2014; Sun et al . , 2015 ) . In response to food intake OXM levels go up , while GCG plasma concentrations are reduced ( Figure 6 ) , suggesting that postprandial GCGR activation does not depend on GCG itself . Interestingly , GCG/GCGR signaling has recently been implicated in the regulation of hepatic Bmal1 expression in response to prolonged starvation ( Sun et al . , 2015 ) . Sun et al . show that fasting-induced GCG signaling activates Bmal1 transcription via CREB/CRTC2 during the night . As we did not observe any acute effects of OXM treatment on Bmal1 transcription ( Figure 3 ) , these data suggest that GCGR signaling may have different clock targets depending on the time of activation . GCGR activation is known to induce protein kinase A-mediated nuclear translocation and DNA-binding of phosphorylated CREB on target genes ( Gonzalez and Montminy , 1989; Dalle et al . , 2004 ) . In line with this , we show OXM-induced binding of CREB to Per1 promoter CRE motifs and induction of Per clock gene expression ( Figures 2–4 ) . Similar phase-dependencies were observed for Per induction as were seen for clock shifting ( compare Figures 1 , 3 , 4 ) . For various tissues—including the SCN—it has been shown that resetting of the circadian clock involves acute up-regulation of Per1 and Per2 ( Dunlap , 1999; Lowrey and Takahashi , 2004 ) . Per induction was observed after DEX treatment or serum shock in fibroblast and hepatoma cells ( Balsalobre et al . , 1998 , 2000 ) . Light pulses given during the dark phase induce CREB activation and Per expression in the SCN ( Albrecht et al . , 1997; Yan and Silver , 2002 ) . In line with this , we showed Per1 and Per2 induction after OXM treatment . Of note , Per induction in liver slices was weaker than in vivo , suggesting that additional mechanisms may amplify OXM effects in intact animals . Along the same line , OXM-mediated Per induction in vivo appeared to be slightly faster than in slices ( Figures 3 , 4 ) . Importantly , peripheral treatment with OXM also altered the diurnal expression rhythms of genes involved in regulating liver carbohydrate metabolism which may underline the metabolic effects of daytime feeding in rodents ( Figure 5 ) . Similar to the OXM effects on clock gene expression ( Figures 3 , 4 ) , refeeding acutely induces hepatic Per1/2 expression , which has been proposed as an integral part of the food-driven clock-resetting mechanism ( Oike et al . , 2011; Tahara et al . , 2011 ) . In our study , we demonstrated that food intake after overnight fasting stimulated OXM secretion and led to hepatic up-regulation of Per expression , which was blocked by OXM neutralization in the circulation . Accordingly , liver slice cultures from re-fed Per2::LUC reporter mice showed food-dependent phase delays and these effects were attenuated by OXM neutralization ( Figure 6 ) . OXM neutralization does not completely inhibit the effects of refeeding on Per induction and liver clock resetting . This might be due to incomplete neutralization of OXM or the contribution of other food-induced factors as discussed above . Of note , while our data do not provide evidence for a direct involvement of GLP-1 ( R ) , it has been suggested that postprandial GLP-1 signaling may indirectly affect liver clock phase ( Panjwani et al . , 2013 ) . Our data indicate an involvement of OXM signaling in food-driven resetting of the liver circadian clock . Similar to what was previously reported in humans ( Mayo et al . , 2003 ) , we detected elevated OLI levels in fed as compared to starved mice and in response to acute food intake ( Figure 6 ) . The diurnal variability of OLI under ad libitum feeding conditions was moderate , but fasting resulted in an approximate 20% reduction of diurnal OLI secretion . Together this indicates that small meals as consumed during the inactive phase affect OXM secretion , but that substantial food intake is necessary to acutely increase OXM secretion to an extent sufficient to affect the liver clock . Of note , interpretation of plasma OXM blood levels is difficult as many OXM assays exhibit some degree of cross-reactivity with GCG and the closely related glicentin . Glicentin differs from OXM only by a 32-amino acid N-terminal extension ( IP-1; Figure 1 ) and is released from the gut after food intake at approximately similar levels ( Blache et al . , 1988; Tang-Christensen et al . , 2001 ) , thus we cannot exclude the possibility that glicentin may exhibit actions that overlap with those described for OXM . In summary , we show that food intake induces OLI blood levels and hepatic clock resetting in mice . OXM treatment mimics food-mediated clock resetting in slices and in vivo in a time of day-dependent manner . Food is a major regulator of hepatic transcriptome rhythms ( Vollmers et al . , 2009 ) . A role for metabolic hormones , for example , insulin , GCG , or glucocorticoids , in the regulation of peripheral clocks has been suggested ( Balsalobre et al . , 2000; Tahara et al . , 2011; Sun et al . , 2015 ) . However , the mechanisms of food-related synchronization of peripheral clocks and their uncoupling from the SCN under time-restricted food intake rhythms are still poorly understood . Our data suggest that OXM—most likely in concert with other gut- and pancreas-derived hunger- or satiety-signaling peptides—is involved in the detrimental effects of mistimed food intake on metabolic homeostasis ( Arble et al . , 2009; Albrecht , 2012; Hatori et al . , 2012 ) . While interfering with insulin signaling may be clinically problematic because of its potential deleterious effects on glycemia , chronotherapeutic targeting of peripheral incretin signaling may provide an alternative therapeutic strategy against metabolic disorders arising from circadian strain as observed in shift workers or during jetlag . All animal experiments were ethically assessed and licensed by the Office of Consumer Protection and Food Safety of the State of Lower Saxony and in accordance with the German Law of Animal Welfare ( license nos . V312-7224 . 122-4 and 33 . 12-42502-04-12/0893 ) . For all experiments adult wild-type mice male C57BL/6J ( 8–24 weeks old ) were used . If not stated otherwise , mice were provided with food and water ad libitum . To investigate the effect of OXM on gene expression by qPCR or ISH , mice were peripherally treated with OXM ( i . v . 4 µg/mouse; i . p . 25 µg/mouse ) or vehicle ( PBS ) at ZT3 , ZT15 , or 12 hr after light-off . For luminescence measurements adult heterozygous males Per2::LUC ( Yoo et al . , 2004 ) and Per2::LUC x Glp1r−/− were used . Glp1r−/− mice were maintained on a C57B/6J background ( Scrocchi et al . , 1996 ) . All mice were exposed to a 12-hr: 12-hr light–dark cycle with 100 lux in the light phase ( LD12:12 ) . Animals were sacrificed at indicated time points by cervical dislocation . Animals euthanized during the dark phase were handled under red light and eyes were removed before dissection . All tissue samples were collected and immediately snap-frozen on dry ice or liquid nitrogen . For long-term storage tissues were kept at −80°C . In order to test Per induction and phase shifts after refeeding , mice were fasted for one night ( 13 hr , or from ZT12—ZT1 ) and either food deprived until decapitation or refed at ZT1 . The refed mice received either anti-OXM rabbit IgG ( i . p . 50 µg/mouse; T-4800; Bachem , Bubendorf , Switzerland ) or control IgG from unimmunized rabbit serum ( I5006; Sigma-Aldrich , Seelze , Germany ) at ZT1 when food was returned . Luminescence was measured from cultured liver slices of heterozygous Per2::LUC mice as described previously ( Yoo et al . , 2004 ) modified to include the use of culture plate inserts ( Millipore , Billerica , MA ) . Briefly , the median lobe of the liver was isolated and 300-µm slices were prepared using a vibratome ( Campden Instruments , Loughborough , UK ) . The slices were immediately placed onto a culture plate insert in 35-mm petri dishes filled with 1-ml culture medium ( D-MEM with high glucose , w/o L-glutamine and phenol red; Life Technologies , Darmstadt , Germany ) supplemented with 3 mM sodium carbonate ( Sigma–Aldrich ) , 10 mM HEPES buffer , 2 mM L-glutamine , 2% B-27 supplement , 25 U/ml penicillin/streptomycin and 0 . 1 mM D-luciferin ( all Life Technologies ) . Luminescence was measured in a luminometer ( Actimetrics , Evanston , IL ) at 32 . 5°C with 5% CO2 . Analyses were performed using the LumiCycle analysis ( Actimetrics ) and Prism software packages ( GraphPad , La Jolla , CA ) . PER2::LUC activity in slices closely follows a sine wave shape . The intersection of the ascending cross-section of the sine wave with the x-axis was defined as 0°/360° , the peak as 90° . Degrees at the time point of treatment were calculated as follows: Tp [°] = ( ( Tp [hsm] − Pbt [hsm] ) : Pbt ) × 360 + 90 with Tp = treatment phase; ° = degree; hsm = hours after start of measurement; Pbt = peak before treatment . If the result was >360° , the value was subtracted by 360 . Raw data were baseline subtracted with running averages of 24 hr . Peaks were defined as middle time point between two troughs of the sine wave . Period was determined as the time between peaks averaged over 2–3 consecutive cycles . For the duration of treatment , samples were maintained at 32 . 5°C to avoid resetting of clock gene expression rhythms due to temperature changes . Phase shifts were determined by comparing extrapolated peak times from sine wave fits before and after treatment . Unless otherwise stated , peptides used for experiments were dissolved in culture medium and administered at a final concentration of 450 pM . Quantitative real-time PCR ( qPCR ) was performed with a CFX96 thermocycler system ( Bio-Rad , Munich , Germany ) with iQ-SYBR Green SuperMix ( Bio-Rad ) . Relative quantification of expression levels by a modified ΔΔCT calculation was performed as described ( Pfaffl , 2001 ) . ß-Actin was used as a reference gene . Statistical analyses were performed using GraphPad Prism software . Circadian profiles of clock gene expression were normalized against the average values over all time points . Induction analyses were normalized against untreated conditions ( 0 min ) . PCR primer sequences are listed in Figure 6—source data 1 . The Per1 probe corresponds to nucleotides 1 to 619 ( GenBank accession number AF022992 ) and Per2 corresponds to nucleotides 229 to 768 of GenBank AF036893 . PCR products had been cloned into pCR II TOPO vector using TOPO TA Cloning Kit ( Life Technologies ) ( Oster et al . , 2002 ) . Linearization of vectors for in vitro transcription was done with EcoRI . 35S-UTP ( PerkinElmer , Waltham , MA ) labeled RNA probes were prepared using RNA Transcription Kit ( Maxi Script Labeling Kit , Life Technologies ) with T7 or T3 RNA polymerases according to the manufacturer's protocol . 10-µm cryosections were cut using a Leica CM3050 cryostat . Cryosections were fixed in 4% paraformaldehyde , acetylated in acetic anhydride and dehydrated with ethanol . Hybridization was performed over night at 55–58°C . Autoradiographs were analyzed by densitometry ( Bio-Rad GS-800 ) using QuantityOne software ( Bio-Rad ) . Three sections per brain were used and background values were calculated from adjacent tissue areas on the same slide for each section . Measurements from different animals/experiments were compared for statistical analysis using GraphPad Prism ( GraphPad ) . Liver slices were homogenized and immediately cross linked with 1% formaldehyde . Chromatin was sonicated for 15-s on/20-s off cycles for 22 min using a Bioruptor sonicator ( Diagenode , Denville , NJ ) . Samples were incubated overnight at 4°C with CREB antibody ( ab31387 , Abcam , Cambridge , UK ) . After clearing , samples were incubated with A/G agarose beads ( Thermo Scientific , Braunschweig , Germany ) for 1 hr at 4°C followed by repetitive washings . After boiling for 10 min in 10% Chelex ( Bio-Rad ) with Proteinase K ( 150 mg/ml ) , samples were spun down and DNA-containing supernatant was collected for PCR . qPCR was performed as described above , and values were normalized to percentage of input . Primer sequences were: 5′-CAGCTGCCTCGCCCCGCCTC-3′/5′-CCCAAGCAGCCATTGCTCGC-3′ ( Per1 CRE ) and 5′-CCCCGCAGTCCTACGGTGCTG-3′/5′-AAGCCCCCAAACAACTGAAGGT-3′ ( 500 bp downstream sequence ) . Blood collection for radioimmunoassay ( RIA ) was performed at ZT1 after 12 hr fast . Mice were allowed to recover for 3 days , then fasted again followed by treatments ( with or without refeeding ) . Blood was collected from the tail vein at 0 min , 20 min , 60 min , and 120 min after treatment . Plasma concentrations of OLI were determined by RIA ( Phoenix Pharmaceutics , Karlsruhe , DE ) according to manufacturer's protocol modified to use a 50% reaction volume . GLP-1 ( EZGLP-1T-36K , Millipore , Darmstadt , Germany ) , insulin ( Catalog# 90080 , CrystalChem , Downers Grove , IL ) , and GCG plasma levels ( EZGLU-30K , Millipore ) were determined by ELISA according to the manufacturers' protocols . Data were analyzed with GraphPad Prism ( GraphPad ) . Mann–Whitney tests were used for simple comparisons . For dose responses one-way ANOVAs and for two-factor comparisons two-way ANOVAs with Bonferroni post-tests were used . A p-value of less than 0 . 05 was considered significant .

Humans and other animals have adapted their behavior and their biology to the daily cycle of light and dark . Groups of genes are reliably switched on or off at different times of the day , and act as internal , or ‘circadian’ , clocks that help these organisms to stay on a 24-hour cycle . External signals also synchronize the body's internal clocks . For example , sunlight helps synchronize the master clock in the brain , while mealtimes and other cues help other organs keep time . These internal clocks are often disrupted in people who work overnight or on rotating shifts . It is believed that when these individuals wake up or go to sleep at odd times it confuses their circadian clocks , which can be harmful to their health . People who work these unusual hours are at an increased risk of developing cancer , heart disease , obesity , and other disorders that involve problems with metabolism . Eating at odd hours may also throw off the circadian clocks in the digestive system . This may explain why metabolic problems have been linked to working odd hours . Landgraf , Tsang et al . hypothesized that if the hormones produced after eating are released when a person would normally be sleeping , this may desynchronize the circadian clock in organs like the liver . Screening mice and tissue samples from mice for hormones that perturb circadian rhythms showed that a hormone called oxyntomodulin , which is released from the gut after eating , activated important circadian clock genes in mouse livers . The increases in clock gene activation were comparable to those seen in the brain in response to exposure to light . Landgraf , Tsang et al . revealed that the clock-resetting effects of oxyntomodulin were the greatest when animals were exposed to it by eating , or by injections of the hormone , at times when the animals would normally be fasting . The experiments also showed that blocking oxyntomodulin prevented eating at unusual times from interfering with the liver's circadian clocks . The findings may suggest a way to help protect people who work overnight from the harmful health effects linked to perturbed circadian clocks .

2015

Oxyntomodulin regulates resetting of the liver circadian clock by food

Multiple nuclei sharing a common cytoplasm are found in diverse tissues , organisms , and diseases . Yet , multinucleation remains a poorly understood biological property . Cytoplasm sharing invariably involves plasma membrane breaches . In contrast , we discovered cytoplasm sharing without membrane breaching in highly resorptive Drosophila rectal papillae . During a six-hour developmental window , 100 individual papillar cells assemble a multinucleate cytoplasm , allowing passage of proteins of at least 62 kDa throughout papillar tissue . Papillar cytoplasm sharing does not employ canonical mechanisms such as incomplete cytokinesis or muscle fusion pore regulators . Instead , sharing requires gap junction proteins ( normally associated with transport of molecules < 1 kDa ) , which are positioned by membrane remodeling GTPases . Our work reveals a new role for apical membrane remodeling in converting a multicellular epithelium into a giant multinucleate cytoplasm . Throughout the tree of life , there are upper limits to the size of individual cells . This size limitation is imposed by genome content , which impacts biosynthetic capacity and cell growth ( Conlon and Raff , 1999; Mueller , 2015 ) . In diverse tissues and organisms , the existence of ‘giant cells’ is driven by polyploidy , the presence of greater than a diploid genome content ( Van de Peer et al . , 2017; Schoenfelder and Fox , 2015 ) . Purposes of polyploidy across evolution remain largely unknown . However , one potential advantage of a tissue containing few , large polyploid cells vs . numerous , small diploid cells is the ability of cytoplasmic components to move over much larger distances . A common form of polyploidy is multinucleation . Sharing of cytoplasm in a multinucleate tissue or organism is an important and recurring adaptation across evolution . Multinucleate cells can be large , metabolically-active cells with unique shapes and functions ranging from specialized force distribution to tissue barrier preservation . During organismal development , examples of multinucleation include animal skeletal muscle , mammalian osteoclasts , and mammalian syncytial placental trophoblasts ( Deng et al . , 2017; Gerbaud and Pidoux , 2015; Pereira et al . , 2018 ) . Multinucleation also arises in response to tissue stress , such as following injury to the Drosophila abdominal epithelium or the human corneal epithelium ( Losick et al . , 2013; Ikebe et al . , 1986 ) . A commonality of these numerous examples of multinucleation is the ability to exchange , over long distances , cytoplasmic components such as RNA , proteins , and even organelles ( Rustom et al . , 2004; McLean and Cooley , 2013 ) . The cellular mechanisms underlying multinucleation are diverse . During cell division , multinucleation can occur through incomplete cytokinesis , followed by formation of a stable cytoplasmic bridge between nuclei . This process occurs in diverse examples of germ cell development ( Greenbaum et al . , 2011 ) and also in some somatic cells such as the ring canal of the Drosophila ovary ( McLean and Cooley , 2013 ) and the plasmodesmata of plants ( Lůcas and Wolf , 1993 ) . A second major mechanism of multinucleation involves plasma membrane breaches . Such breaches can involve distinct actin-based protrusive structures . Podosome-like structures facilitate multinucleation in Drosophila skeletal muscle and mammalian macrophages ( Faust et al . , 2019; Sens et al . , 2010 ) . While the mechanisms are diverse , one common feature of the above-discussed examples of multinucleation and cytoplasm sharing identified to date are clearly visible plasma membrane disruptions . Here , we report a visual animal-wide screen , using multi-color lineage labeling approaches in the tractable animal model Drosophila melanogaster , for multinucleate tissues that share cytoplasm . We discover cytoplasm sharing in the rectal papilla , a common insect resorptive intestinal epithelium that is critical for maintaining ionic homeostasis ( Wigglesworth , 1932; Cohen et al . , 2020 ) . Likely due to its extreme proximal location in the gut of many insect species , this epithelium is linked to the infiltration of diverse pathogens , such as those involved in African sleeping sickness and also viruses being pursued as insect control measures ( Gu et al . , 2010; Filosa et al . , 2019 ) . Here , we reveal that cytoplasm sharing onset in Drosophila papillae occurs during a short developmental window , indicating robust molecular regulation . We find that papillar cytoplasm sharing requires neither incomplete cytokinesis nor canonical actin-based membrane breach regulators . Using transmission electron microscopy , we further identify that this developmentally programmed process involves extensive remodeling of apical junctions and lateral membranes , but not clearly identifiable plasma membrane breaches . Using genetic screening , we implicate specific regulators of membrane remodeling , notably the GTPase Dynamin/Shibire , in the mechanism of papillar cytoplasmic sharing . From analysis of shibire mutants , we uncover a requirement for gap junction establishment and specific gap junction proteins in papillar cytoplasm sharing . Mutant animals defective in papillar cytoplasm sharing are intolerant of a high-salt diet , indicating a physiological role of long-range cytoplasm movement in this tissue . Unlike all known examples of multinucleation , our results show that cytoplasm sharing in rectal papillae requires developmentally programmed apical membrane remodeling , which creates a giant resorptive epithelial network of 100 nuclei . This tissue represents a new system to investigate the diversity of multicellular tissue organization and mechanisms and functions of cytoplasm sharing . To identify new examples of adult tissues in Drosophila that share cytoplasm , we ubiquitously expressed Cre and UAS-dBrainbow ( Hampel et al . , 2011; Figure 1A ) , a Cre-Lox-based system that randomly labels cells with only one of three fluorescent proteins . We used animals heterozygous for UAS-dBrainbow to ensure single-labeling of cells . We ubiquitously expressed Cre , which does not require heat-shock induction , from early embryonic stages ( before cells endocycle to become polyploid ) . Cre-mediated excision occurs independently of Gal4 expression and Gal80ts repression of dBrainbow . Therefore , we can ensure that multi-labeled cells only arise by cytoplasm sharing between cells not related by cell division or incomplete cytokinesis ( Figure 1B ) . We examined a wide range of tissues ( Figure 1—figure supplement 1A ) . From our screen , we discovered that the rectal papilla is a new example of a tissue with cytoplasm sharing . Adult Drosophila contain four papillae , each with 100 nuclei of genome content between 8 and 16C ( Fox et al . , 2010 ) , that reside in the posterior hindgut ( Figure 1C ) . Each papilla is a polarized epithelial cone with the apical region facing the gut lumen and the basal region surrounding a central canal that connects to the fly’s hemolymph ( Figure 1D ) . The papillar structure supports its function to reabsorb water , ions , and small molecules from the gut lumen and recycle them back to the hemolymph ( Cohen et al . , 2020 ) . Knowing that adult papillar cells share cytoplasm , we next used our dBrainbow system to identify when papillar cells begin to share relative to other developmental events that we previously identified ( Figure 1E ) . Using both fixed and live imaging of whole organs , we found that at 62 hours post-puparium formation ( HPPF ) , each papillar cell contains only one dBrainbow label ( Figure 1F ) . By contrast , at 69HPPF , multi-labeled cells are apparent ( Figure 1F’ , H–H’ ) . We quantitatively measured papillar sharing across the tissue ( Figure 1—figure supplement 1B , Materials and methods ) and found that cytoplasm sharing initiates over a narrow 6 hr period ( 68-74HPPF , Figure 1G ) . Our results suggested that at least RNA and possibly protein passes between papillar cells to facilitate cytoplasm sharing . To directly test if protein is shared , we photo-activated GFP ( GFPPA ) in single adult papillar cells and observed in real time whether GFPPA spreads to adjacent cells . We find the principal papillar cells , but not the secondary cells at the papillar base ( Garayoa et al . , 1999; Figure 1—figure supplement 1C ) , share protein across an area of at least several nuclei ( Figure 1I–I’ ) . We next tested whether a larger protein can be shared between papillar cells . We used rectal papillae RNA-sequencing data ( Leader et al . , 2018 ) to identify proteins that are endogenously expressed , cytoplasmic , and relatively large . We therefore generated flies expressing a UAS-inducible , photoactivatable GFP fused to Glyceraldehyde 3 phosphate dehydrogenase 2 ( UAS-Gapdh2-GFPPA ) . This construct should produce a tagged protein of 62 . 3 kDa . We found that Gapdh2-GFPPA protein is shared between cells , as it never stops at a papillar cell–cell boundary , though it may move at a slower rate than GFPPA ( Figure 1—figure supplement 1D ) . Therefore , proteins as large as ~62 kDa ( the size of GFP-tagged Gapdh2 ) can move across an area covered by multiple papillar nuclei . Additionally , the movement of our Gapdh2 transgenic protein indicates that papillar cells likely share endogenously expressed proteins . These results indicate that papillae undergo a developmentally programmed conversion from 100 individual cells to a single giant multinuclear cytoplasm that shares the products of ~1200 genomes . We next examined whether cytoplasm sharing requires either programmed endocycles or mitoses . We have previously shown that larval papillar cells first undergo endocycles , which increase cellular ploidy . Then , during metamorphosis , pupal papillar cells disassemble polytene chromosomes and undergo polyploid mitotic cycles , which increase cell number ( Fox et al . , 2010; Stormo and Fox , 2016; Stormo and Fox , 2019 ) . Both endocycles and mitoses occur well prior to the start of papillar cytoplasm sharing ( Figure 1E ) . Papillar endocycles require the Anaphase-Promoting Complex/Cyclosome regulator fizzy-related ( fzr ) while the papillar mitoses require Notch signaling ( Schoenfelder et al . , 2014 ) . Knockdown of fzr significantly disrupts cytoplasm sharing ( Figure 1—figure supplement 1E , F , H ) . We hypothesize that endocycles are required for differentiation of the papillae , which later enables these cells to trigger cytoplasm sharing . In contrast , blocking Notch signaling , which initiates papillar mitotic divisions ( Fox et al . , 2010 ) , does not prevent sharing ( Figure 1—figure supplement 1E , G , H ) . Thus , papillar cytoplasm sharing requires developmentally programmed endocycles but not mitotic cycles . As our dBrainbow approach only identifies cytoplasm sharing events that do not involve incomplete division/cytokinesis , we examined whether sharing results from fusion pore formation , as in skeletal muscle . A well-studied model of such cell–cell fusion in Drosophila is myoblast fusion , which requires an actin-based podosome ( Richardson et al . , 2007; Sens et al . , 2010 ) . We conducted a candidate dBrainbow-based RNAi screen ( 77 genes , Figure 2A , Table 1 ) of myoblast fusion regulators and other plasma membrane components . Remarkably , 0/15 myoblast fusion genes from our initial screen regulate papillar cytoplasm sharing ( Figure 2A , Figure 2—figure supplement 1A , Table 1 ) . Furthermore , dominant-negative forms of Rho family GTPases have no impact on dBrainbow labeling ( Figure 2—figure supplement 1B ) , providing additional evidence against actin-based cytoplasm sharing . Instead , we found 8/77 genes , including subunits of the vacuolar H+ ATPase ( Vha16-1 ) , ESCRT-III complex ( Vps2 ) , and exocyst ( Exo84 ) ( Figure 2A ) are required for papillar cytoplasm sharing . Through additional screening , the only myoblast fusion regulator required for papillar cytoplasm sharing is singles bar ( sing ) , a presumed vesicle trafficking gene ( Estrada et al . , 2007; Figure 2—figure supplement 1A ) . Given the enrichment of our candidate screen hits in membrane trafficking and not myoblast fusion , we further explored the role of membrane trafficking in cytoplasm sharing . We conducted two secondary dBrainbow screens to find specific membrane trafficking pathway components that regulate papillar sharing . First , a focused candidate membrane trafficking screen revealed additional components ( 12/36 genes screened , Figure 2B , Table 2 ) including three more vacuolar H+ ATPase subunits , five more exocyst components , and the Dynamin GTPase shibire ( shi ) ( Figure 2B , D , E , H ) . Second , we screened constitutively-active and dominant-negative versions of all 31 Drosophila Rabs . Sharing requires only a small number of Rabs , specifically the ER/Golgi-associated Rab1 , the early endosome-associated Rab5 , and the recycling endosome-associated Rab11 ( Figure 2C , D , F–H ) . Given our identification of the membrane vesicle recycling circuit involving shi , Rab5 , and Rab11 , we focused on these genes . Two unique RNAi lines for each gene show consistent sharing defects , and most of these knockdowns completely recapitulate the pre-sharing state ( Figure 2H ) . Despite exhibiting strong cytoplasm sharing defects , shi , Rab5 , and Rab11 RNAi papillae appear morphologically normal , with only minor cell number decreases ( Figure 2—figure supplement 1C ) . These results suggest that membrane recycling GTPases regulate a specific developmental event associated with cytoplasm sharing , and not papillar morphogenesis . In agreement with these GTPases acting during development , rather than as part of an ongoing transport process , GTPase knockdown after sharing onset does not block cytoplasm sharing ( Figure 2—figure supplement 1D–F ) . Together , our screens reveal that membrane trafficking , particularly Dynamin-mediated endocytosis and early/recycling endosome trafficking , regulates papillar cytoplasmic sharing . To better understand how membrane trafficking GTPases initiate cytoplasm sharing during development , we examined endosome and Shi localization during sharing onset . We imaged a GFP-tagged pan-endosome marker ( myc-2x-FYVE ) , overexpression of which should not alter endosome shape or localization ( Gillooly et al . , 2000; Wucherpfennig et al . , 2003 ) , and a Venus-tagged shi before and after sharing . Endosomes are evenly distributed shortly before sharing , but become highly polarized at the basal membrane around the time of sharing onset ( Figure 3A–A’ , C , Figure 3—figure supplement 1A ) . This basal endosome repositioning requires Shi ( Figure 3B–C , Figure 3—figure supplement 1A ) and the change in endosome localization is attributed to Rab5-positive early endosomes ( Figure 3—figure supplement 1B–B’’ ) . Additionally , Shi localization changes from apical polarization to a uniform distribution during sharing onset ( Figure 3D–E ) . These localization changes indicate that membrane trafficking factors which regulate cytoplasm sharing are highly dynamic during cytoplasm sharing onset . To determine what membrane remodeling events underlie GTPase-dependent cytoplasm sharing , we turned to ultrastructural analysis . Adult ultrastructure and physiology of papillar cells has been examined previously in Drosophila ( Wessing and Eichelberg , 1973 ) and related insects ( Gupta and Berridge , 1966 ) . These cells contain elaborate membrane networks that facilitate selective ion resorption from the gut lumen , facing the apical side of papillar cells , to the hemolymph , facing the basal side . Still , little is known about developmental processes or mechanisms governing the unique papillar cell architecture . We looked for changes in cell–cell junctions and lateral membranes that coincide with cytoplasm sharing , especially to determine if there is a physical membrane breach between cells . We identified several dramatic changes in membrane architecture . First , apical microvilli-like structures form during sharing onset ( Figure 3F–F’’ ) . Just basal to the microvilli , apical cell–cell junctions are straight in early pupal development and compress into a more curving , tortuous morphology around the time of cytoplasm sharing onset ( Figure 3—figure supplement 1C–C’’ ) . One of the most striking changes , coincident with Shi re-localization , is formation of pan-cellular endomembrane stacks surrounding mitochondria . These stacks are likely sites for active ion transport , such as that mediated by the P-type Na+/K+-ATPase , coupled to mitochondria for ATP ( Figure 3G–G’’; Berridge and Gupta , 1967; Patrick et al . , 2006 ) . Thus , massive apical and intracellular plasma membrane reorganization coincides with both cytoplasm sharing and Shi/endosome re-localization . We next assessed whether the extensive membrane remodeling requires Shi , Rab5 , and Rab11 . In shi and Rab5 RNAi animals , microvilli protrude downward , instead of upward ( Figure 3H–J ) . Additionally , apical junctions do not compress as in controls ( Figure 3—figure supplement 1D–F ) . Notably , membrane stacks are greatly reduced ( Figure 3K–M ) . shi RNAi animals exhibit numerous trapped vesicles , consistent with a known role for Dynamin in membrane vesicle severing ( Damke et al . , 1994; Hinshaw and Schmid , 1995; Figure 3L , inset ) . Together , we find that Shi and endosomes extensively remodel membranes during papillar cytoplasm sharing . Our extensive ultrastructural analysis did not reveal any clear breaches in the plasma membrane , despite numerous membrane alterations . Adult papillae exhibit large extracellular spaces between nuclei that eliminate the possibility of cytoplasm sharing throughout much of the lateral membrane ( Figure 3—figure supplement 2A; Wessing and Eichelberg , 1973; Gupta and Berridge , 1966 ) . Instead , through our GTPase knockdown studies , we identified a striking alteration in the apical cell–cell interface that strongly correlates with cytoplasm sharing . Specifically , shi animals frequently lack apical gap junctions ( Figure 3N–O ) ( p<0 . 0001 ) ( Figure 3P , Figure 3—figure supplement 1H–H’’ ) . Upon closer examination of control animal development , we find that apical gap junction-like structures arise at cytoplasm sharing onset . There is almost no gap junction-like structure before cytoplasm sharing ( Figure 4A–B , Figure 2—figure supplement 1A–A’’ ) . Given our electron micrograph results , we determined which innexins , the protein family associated with gap junctions in invertebrates ( Bauer et al . , 2005; Phelan et al . , 1998 ) , are expressed in rectal papillae . From RNA-seq data ( Methods ) , we determined that ogre ( Inx1 ) , Inx2 , and Inx3 are most highly expressed ( Figure 4C ) . This combination of innexins is not unique to rectal papillae; the non-sharing brain and optic lobe ( Figure 4—figure supplement 1A ) also express high levels of all three ( Leader et al . , 2018 ) . We examined localization of Inx3 ( a gap junction component ) ( Curtin et al . , 1999; Richard et al . , 2017 ) , and compared it to a septate junction component , NeurexinIV ( NrxIV ) ( Laprise et al . , 2009 ) . NrxIV localizes similarly both pre and post-sharing onset ( Figure 4D–D’ ) , indicative of persistent septate junctions remaining between papillar cells . In contrast , Inx3 organizes apically only after cytoplasm sharing ( Figure 4E–E’ , Figure 4—figure supplement 1B–B’ ) . Inx3 also does not localize to cell–cell boundaries in shi RNAi animals ( Figure 4C–C' ) . We tested whether innexins are required for cytoplasm sharing . Knocking down these three genes individually causes mild yet significant cytoplasm sharing defects ( Figure 4F ) . However , we see larger defects in animals expressing dominant-negative ogreDN ( Figure 4F–G; Spéder and Brand , 2014 ) , which contains a N-terminal GFP tag that interferes with channel passage . Also , heterozygous animals containing a ten gene-deficiency spanning ogre , Inx2 , and Inx7 have more severe defects ( Figure 4F , Df ( 1 ) BSC867 ) . Finally , we tested whether cytoplasm sharing is essential for normal rectal papillar function . Rectal papillae selectively absorb water and ions from the gut lumen for transport back into the hemolymph , and excrete unwanted lumen contents ( Cohen et al . , 2020 ) . One test of papillar function is viability following the challenge of a high-salt diet ( Bretscher and Fox , 2016; Schoenfelder et al . , 2014 ) . However , with our pan-hindgut driver byn-Gal4 used for all previous experiments , we noted animal lethality with shi , Rab5 , and Rab11 knockdown within a few days on control food . We observed melanization and necrosis throughout the hindgut ( data not shown ) which prevented us from attributing any phenotypes directly to papillar cytoplasm sharing . We therefore identified an alternative driver ( 60H12-Gal4 ) with rectum-specific expression during pupation and adulthood ( Figure 4—figure supplement 1D–D’ ) . We used this driver to express shiDN . These animals display similar sharing defects as we find with byn-Gal4 ( Figure 4—figure supplement 1E–E’’ ) . Reassuringly , 60H12-Gal4 > shiDN animals do not show lethality on a control food diet ( Figure 4H ) allowing us to test rectal papillar physiological function on a high-salt diet . Using either pan-hindgut or papillae-specific knockdown of cytoplasm sharing regulators , we find both shiDN and ogreDN animals are extremely sensitive to the high-salt diet ( mean survival <1 day , Figure 4H ) . These results underscore an important function for gap junction proteins , as well as membrane remodeling by Dynamin/Shibire , in cytoplasm sharing . Our findings identify Drosophila rectal papillae as a new and distinctive example of cytoplasm sharing between multiple nuclei in a simple , genetically tractable system . One defining property of papillar cytoplasm sharing is the lack of an easily observable conduit in the lateral membrane through which cytoplasm can be exchanged . Cytoplasm sharing in a multinucleate tissue/organism frequently involves the creation of a large membrane breach associated with major actin cytoskeleton rearrangement ( Kim et al . , 2015; Deng et al . , 2017; Martin , 2016 ) . However , papillar cytoplasm sharing does not require canonical myoblast fusion regulators nor major actin remodeling factors such as Rho family GTPases . Aside from membrane breaches , other cell types are known to share cytoplasm through the formation of cytoplasmic bridges such as ring canals or plasmodesmata . Such bridge structures assemble as the result of incomplete cytokinesis ( Mahowald , 1971; Lůcas and Wolf , 1993 ) . In contrast , papillar cytoplasm sharing does not require mitosis or cytokinesis , and does not contain intercellular bridge structures visible by electron microscopy . In addition to lacking a large , observable membrane breach , papillar cytoplasm sharing occurs within an intact , polarized epithelium , and apical cell–cell junctions and lateral membranes are retained after the onset of sharing . In contrast , other epithelia known to fuse cytoplasm , such as C . elegans epithelia fused by Epithelial Fusion Failure 1 ( EFF-1 ) , dismantle cell–cell junctions ( Smurova and Podbilewicz , 2016 ) . Further , cells with ring canals retain cell–cell junctions and lateral membranes ( Peifer et al . , 1993 ) . Given the retention of cell junctions and absence of clear intercellular bridges , channels , or breaches in lateral membrane , our data lead us to propose that a specialized function of gap junction proteins facilitates cytoplasm sharing between neighboring cells in an otherwise intact epithelium ( Figure 4I ) . Although gap junctions typically transfer molecules of <1 kDa , elongated proteins up to 18 kDa are observed to pass through certain vertebrate gap junctions ( Cieniewicz and Woodruff , 2010 ) . Alternatively , gap junction-mediated cell to cell communication has been previously implicated in fusion of placental trophoblasts and osteoclasts ( Firth et al . , 1980; Dunk et al . , 2012; Schilling et al . , 2008 ) , so we cannot rule out an indirect role for gap junctions in papillar cells , such as through regulation/recruitment of a fusogenic protein ( Petrany and Millay , 2019 ) . Future work beyond the scope of this study can determine if , for example , papillar gap junctions exhibit a specialized structure to directly facilitate exchange of large cytoplasmic contents . As for the connection between membrane remodeling and gap junction formation , Rab11 has been previously reported to recycle gap junction components in Drosophila brain and mammalian cell culture ( Augustin et al . , 2017 ) . Dynamin2 was also implicated in gap junction plaque internalization in mammalian cells ( Gilleron et al . , 2011 ) . However , neither of these factors has been previously implicated in gap junction establishment . We show that Dynamin is required for gap junction formation in papillar cells . Future studies will determine the exact role of Dynamin in gap junction establishment . Another clue for future study is that papillar cytoplasm sharing is developmentally regulated , occurring over a brief 6 hr window , and requires membrane remodeling by trafficking GTPases and gap junction establishment ( Figure 4I , Figure 4—figure supplement 1H ) . Our results argue that papillar sharing is triggered by a permanent structural rearrangement rather than an active transport mechanism , as the membrane remodelers we identified are required specifically during developmental membrane remodeling . The mechanisms we report here may be relevant to other emerging roles for membrane remodeling and cytoplasm sharing in the literature . Here , we identify a close relationship between the formation of membrane stacks and cytoplasm sharing . Basolateral membrane infoldings to expand cellular surface area are a common feature of absorptive cells ( Pease , 1956 ) . The mammalian kidney tubule cells exhibit similar basolateral membrane extensions to which ion transporters such as the Na+/K+-ATPase are localized ( Maunsbach , 1966; Molitoris et al . , 1992; Avner et al . , 1992; Pease , 1955; Sjöstrand and Rhodin , 1953 ) . Our results suggest that the same membrane remodeling factors that regulate cytoplasm sharing are required for the formation of membrane stacks . To our knowledge , this is the first study to reveal factors involved in basolateral membrane infolding biogenesis . Additionally , our results may also explain other examples of cytoplasm sharing where the underlying mechanism remains to be determined , such as transient cytoplasm sharing in the zebrafish myocardium ( Sawamiphak et al . , 2017 ) . Together , our studies indicate that the Drosophila papillar epithelium represents a distinctive example of cytoplasmic sharing to generate giant multinucleate cells . Our results have several implications for functions and regulation of multinucleation . Here we show that the membrane and junctional changes associated with cytoplasm sharing are required for normal Drosophila rectal papillar function . Papillae in other insects are known to undergo visible movement upon muscle contraction , which may facilitate cytoplasm movement ( Lowne , 1869 ) . Arthropod papillar structures are subject to peristaltic muscle contractions from an extensive musculature ( Rocco et al . , 2017 ) , which aid in both excretion and movement of papillar contents into the hemolymph ( Habas mantel and Mantel , 1968 ) . Further , relative to other hindgut regions , the rectum appears to have specialized innervation ( Cohen et al . , 2020 ) and regulation by the kinin family of neuropeptides , which are hypothesized to provide additional input in to muscle activity in this critical site of reabsorption ( Audsley and Weaver , 2009; Lajevardi and Paluzzi , 2020 ) . We speculate that these muscle contractions aid in vigorous movement of papillar cytoplasm , which includes ions and water taken up from the intestinal lumen . The movement of these papillar contents may facilitate both cytoplasm exchange between papillar cells and the interaction of ions and ion transport machinery with intracellular membrane stacks . This idea is supported by our finding that animals lacking a large common papillar cytoplasm die when fed a high-salt diet . Given the importance of insect papillae in pathogen biology , the knowledge that this common anatomical structure is a shared cytoplasm can impact both human disease intervention and agricultural pest control . Papillae occur in both primitive insect orders such as Zygentoma and Odonata and also in Lepidopterans , Hymenopterans , and Dipterans , the latter of which exhibit the most prominent and elaborate structures ( Palm , 1949 ) . Furthermore , electron micrographs of the hindgut of the mosquito , Aedes aegypti , and the ant , Formica nigricans , show striking ultrastructural similarity to Drosophila , and these studies leave open the possibility that multinucleation may be conserved in insect papillae ( Hopkins , 1967; Wessing and Eichelberg , 1973; Garayoa et al . , 1999 ) . Cytoplasm sharing is a known mechanism that facilitates pathogen spread ( Eugenin et al . , 2009 ) , and papillae are an avenue of entry for numerous pathogens including kinetoplastids and mosquito viruses ( Gu et al . , 2010; Filosa et al . , 2019 ) . Thus , our findings may impact strategies to prevent diseases such as African sleeping sickness , or to target agricultural pests that threaten agricultural production . The sharing of cytoplasm also has the potential to neutralize detrimental genomic imbalances between nuclei caused by aneuploidy . Our prior work ( Schoenfelder et al . , 2014 ) revealed that papillae are highly tolerant of chromosome mis-segregation , and our work here suggests this tolerance may be due in part to neutralization of aneuploidies through cytoplasm sharing . This finding may also be relevant to the study of multinucleate tumors , such as those found in pancreas , bone , and fibrous tissues ( Doane et al . , 2015; Hasegawa et al . , 2017; Mancini et al . , 2017 ) , or to conditions of aberrant organelle inheritance ( Asare et al . , 2017 ) . Finally , we note that our study reveals that , even in a well-studied model organism such as Drosophila , we still have yet to appreciate the full diversity of tissue organization strategies . Our Brainbow-based approach could be applied to other contexts to identify other tissues with cytoplasm sharing , including those with gap junction-dependent but membrane breach-independent cytoplasm sharing . Collectively , our findings highlight the expanding diversity of multicellular tissue organization strategies . Flies were raised at 25°C on standard media ( Archon Scientific , Durham , NC ) unless specified otherwise . See Table 4 for a list of fly stocks used . See Table 3 for a full list of fly lines screened in primary and secondary screens . See Table 5 for panel-specific genotypes . The UAS-Gapdh2-GFPPA construct was generated by gene synthesis ( Twist Biosciences ) . The GFP was placed at the C-terminus with a 12-amino acid fusion linker ( GSAGSAAGSGEF ) ( Waldo et al . , 1999 ) codon-optimized for Drosophila . This insert was then cloned into the pBID-UASC-FG vector modified to remove the FLAG tag and extraneous cloning sites . Transgenic flies were generated at Duke University . brachyenteron ( byn ) -Gal4 was the driver for all UAS transgenes with the exception of the screen in Figure 1—figure supplement 1A , which used tub-Gal4 , and the shi knockdown in Figure 4H , which used 60H12-Gal4 . 60H12-Gal4 expresses only in the papillar cells and not the rest of the hindgut , and use of this driver blocks cytoplasm sharing using UAS-shiDN ( Figure 4—figure supplement 1D–E’’ ) . For all Gal4 experiments , UAS expression was at 29°C , except in Figure 1F–H , where it was at 25°C . If byn-Gal4 expression of a given UAS-transgene was lethal , the experiment was repeated with a temperature-sensitive Gal80ts repressor transgene and animals were kept at 18°C until shifting to 29°C at an experimentally-determined time point that would both result in viable animals and permit time to express the transgene prior to sharing onset . For salt feeding assays , age- and sex-matched siblings were transferred into vials containing 2% NaCl food made with Nutri-Fly MF food base ( Genesee Scientific ) or control food ( Schoenfelder et al . , 2014 ) . Flies were monitored for survival each day for 10 days . For fixed imaging , tissues were dissected in PBS and immediately fixed in 3 . 7% formaldehyde + 0 . 3% Triton-X for 15 min . Immunostaining was performed in 0 . 3% Triton-X with 1% normal goat serum ( Fox et al . , 2010 ) . The following antibodies were used: Rabbit anti-GFP ( Thermo Fisher Scientific , Cat#A11122 , 1:1000 ) , Rat anti-HA ( Roche , Cat#11867423001 , 1:100 ) , Rabbit anti-Inx3 ( generous gift from Reinhard Bauer , 1:75 ) , [Lehmann et al . , 2006] , 488 , 568 , 633 secondary antibodies ( Thermo Fisher Scientific , Alexa Fluor , 1:2000 ) . Tissue was stained with DAPI at 5 μg/ml and mounted in VECTASHIELD Mounting Media on slides . All image analysis was performed using ImageJ and FIJI ( Rueden et al . , 2017; Schindelin et al . , 2012 ) . Statistical analysis was performed in GraphPad Prism 8 . Detailed statistical tests and methods are reported in Table 6 . Some additional methodological details , including animal genotype , applied to only a specific figure panel . Please see Table 6 for this information .

Most cells are self-contained – they have a cell membrane that delimits and therefore defines the cell , separating it from other cells and from its environment . But sometimes several cells interconnect and form collectives so they can pool their internal resources . Some of the best-known examples of this happen in animal muscle cells and in the placenta of mammals . These cell collectives share their cytoplasm – the fluid within the cell membrane that contains the cell organelles – in one of two ways . Cells can either remain linked instead of breaking away when they divide , or they can fuse their membranes with those of their neighbors . Working out how cells link to their neighbors is difficult when so few examples of cytoplasm sharing are available for study . One way to tackle this is to try and find undiscovered cell collectives in an animal that is already heavily studied in the lab , such as the fruit fly Drosophila melanogaster . Peterson et al . used a genetic system that randomly labels each cell of the developing fly with one of three fluorescent proteins . These proteins are big and should not move between cells unless they are sharing their cytoplasm . This means that any cell containing two or more different colors of fluorescent protein must be connected to at least one of its neighbors . The experiment revealed that the cells of the fruit fly rectum share their cytoplasm in a way never seen before . This sharing occurs at a consistent point in the development of the fruit fly and uses a different set of genes to those used by interconnecting cells in mammal muscles and placenta . These genes produce proteins that reshape the membranes of the cells and fit them with gap junctions – tiny pores that cross from one membrane to the next , allowing the passage of very small molecules . In this case , the gap junctions allowed the cells to share molecules much larger than seen before . The result is a giant cell membrane containing the cytoplasm and organelles of more than a hundred individual cells . These findings expand scientists’ understanding of how cells in a tissue can share cytoplasm and resources . They also introduce a new tissue in the fruit fly that can be used in future studies of cytoplasm sharing . Relatives of fruit flies , including fruit pests and mosquitos , have similar cell structure to the fruit fly , which means that further investigations using this system could result in advances in agriculture or human health .

2020

Cytoplasmic sharing through apical membrane remodeling

To understand a visual scene , the brain segregates figures from background by assigning borders to foreground objects . Neurons in primate visual cortex encode which object owns a border ( border ownership ) , but the underlying circuitry is not understood . Here , we used multielectrode probes to record from border ownership-selective units in different layers in macaque visual area V4 to study the laminar organization and timing of border ownership selectivity . We find that border ownership selectivity occurs first in deep layer units , in contrast to spike latency for small stimuli in the classical receptive field . Units on the same penetration typically share the preferred side of border ownership , also across layers , similar to orientation preference . Units are often border ownership-selective for a range of border orientations , where the preferred sides of border ownership are systematically organized in visual space . Together our data reveal a columnar organization of border ownership in V4 where the earliest border ownership signals are not simply inherited from upstream areas , but computed by neurons in deep layers , and may thus be part of signals fed back to upstream cortical areas or the oculomotor system early after stimulus onset . The finding that preferred border ownership is clustered and can cover a wide range of spatially contiguous locations suggests that the asymmetric context integrated by these neurons is provided in a systematically clustered manner , possibly through corticocortical feedback and horizontal connections . One of the deepest and most enduring mysteries of visual perception is how the brain constructs an internal model of the ever-changing world that falls before our eyes . An essential step in this dynamic process of constructive perception is the assignment of border ownership . Consider panel 1 in Figure 1A . The dashed circle indicates the classical receptive field ( cRF ) of a hypothetical neuron , straddling the edge of an object . The portion of the edge falling inside the circle is perceived to be owned by the light gray square on the bottom left of the edge . Now consider panel 2 in Figure 1A . The local edge within the circle is identical to that in the left panel but this local edge is now perceived as owned by the dark gray square to the upper right of the edge . This phenomenon is called border ownership . It was first recognized by the Gestalt psychologists in the early 20th century and is beautifully illustrated in Rubin’s famous face-vase illusion ( Koffka , 1935; Rubin , 1921 ) . Border ownership represents a fundamental computation in visual perception that is thought to be critical to visual scene segmentation and object recognition ( Nakayama et al . , 1995 ) . Von der Heydt et al . discovered neurons in primate visual cortex that are selective for border ownership , most prominently in extrastriate visual areas V2 and V4 ( Zhou et al . , 2000 ) . Though the existence of border ownership-selective neurons has been well established in prior studies using single electrodes ( Hesse and Tsao , 2016; O’Herron and von der Heydt , 2009; Zhang and von der Heydt , 2010 ) , how this selectivity arises from cortical circuits remains unclear ( Grossberg , 2015; von der Heydt , 2015; Wagatsuma and Sakai , 2016; Yazdanbakhsh and Livingstone , 2006 ) . Some authors have proposed a dominant role for feedforward inputs , which carry information from upstream areas ( Sakai et al . , 2012; Sakai and Nishimura , 2006; Supèr et al . , 2010 ) , whereas others posit a central role for horizontal connections and/or cortical feedback ( Craft et al . , 2007; Grossberg , 2015; Hu and Niebur , 2017; Zhang and von der Heydt , 2010; Zhaoping , 2005 ) . These pathways have a distinct laminar organization: feedforward inputs arrive primarily in the granular ( input ) layer , whereas horizontal and corticocortical feedback connections predominantly target superficial and deep layers ( Douglas and Martin , 2004; Rockland and Pandya , 1979; Rockland et al . , 1994; Rockland and Lund , 1983; Ungerleider and Desimone , 1986; Yoshioka et al . , 1992 ) . The laminar timing of border ownership may thus give clues regarding the roles of these pathways in this computation . Here , we used linear multielectrode probes to compare onset times of border ownership selectivity between laminar compartments . It is also unknown how the preferences of border ownership-selective neurons in different layers relate to each other . This is in contrast to orientation preference , which is well known to be spatially organized in the primate brain . Primate areas V1 and V2 contain orientation columns ( Hubel and Livingstone , 1987; Hubel et al . , 1978; Vanduffel et al . , 2002 ) . In V4 , imaging studies indicate that there is at least clustering of orientation preference in superficial layers of V4 ( Ghose and Ts’o , 1997; Roe et al . , 2012; Tanigawa et al . , 2010 ) , although it is unclear whether these clusters are columnar . Nor do we know if border ownership preference is organized in columns . The border ownership of a given border is defined entirely by asymmetries outside the cRF ( as opposed to the border’s orientation ) . A systematic organization of border ownership preference would therefore imply a clustered arrangement of the neural pathways that underlie these asymmetries . Finally , the relation between border ownership selectivity and orientation tuning is unclear . Prior studies have focused on orientation-selective units and tested border ownership at the neuron’s preferred orientation , without examining the relationship between orientation preference and border ownership preference ( Hesse and Tsao , 2016; Zhang and von der Heydt , 2010; Zhou et al . , 2000 ) . One possibility is that border ownership is fundamentally a border property , such that border ownership selectivity is maximal for the preferred border orientation . Another possibility is that border ownership selectivity rather represents a surface signal ( Grossberg , 2015 ) , and may thus be less strictly tied to border orientation and orientation tuning . Here , we addressed these questions using laminar multielectrode probes to record from border ownership-selective units in macaque area V4 across layers . We replaced the native dura with a transparent artificial dura ( AD ) to enable us to reliably position the probe normal to the cortical surface , on the relatively narrow exposed surface of area V4 . We compared the timing and magnitude of border ownership selectivity across laminar compartments . If border ownership selectivity in V4 is inherited from V2 , its dominant source of cortical input ( Markov et al . , 2011 ) , we expect to see it early and prominently in neurons in the granular layer , which is the main target of this projection ( Rockland , 1992; Gattass et al . , 1997 ) . Next , we compared the functional organization of border ownership preference across layers , to test whether the preferred side of ownership is shared . Finally , we examined the relationship between border ownership preference and border orientation preference . Figure 2A shows the time course of the responses evoked by the preferred ( solid red line ) and non-preferred ( dashed blue line ) side of border ownership , averaged over all well-isolated units that are selective for border ownership . Consistent with prior studies ( Zhou et al . , 2000 ) , we observe that border ownership selectivity ( difference in response to the preferred and the non-preferred side ) occurs early after onset of the stimulus-evoked response . The asterisk indicates when the difference between these functions first becomes statistically significant ( 56 . 5 ms after stimulus onset; sign rank test p < 0 . 05 for 20 adjacent ms ) . When evaluating these functions in each laminar compartment we observe that these functions diverge substantially early after onset in deep cortical layers ( Figure 2D; significant at 58 . 8 ms ) , but later in superficial ( Figure 2B; significant at 85 . 8 ms ) and granular layers ( Figure 2C; significant at 67 . 3 ms ) . This definition of latency depends on sample size , but a subsampling analysis shows that differences in sample size between layers do not explain the shorter latency for deep layers ( Figure 2—figure supplement 1 ) . To compare the time course of border ownership modulation between layers , we defined the BOI function B for each laminar compartment asB ( t ) =Rpref ( t ) −Rnon−pref ( t ) Rpref ( t ) +Rnon−pref ( t ) where Rpref ( t ) and Rnon-pref ( t ) are , respectively , the evoked response functions to the preferred and the non-preferred sides of border ownership ( red and blue dashed functions plotted in Figure 2A–D ) . B ( t ) is plotted for each laminar compartment in Figure 2E , confirming that border ownership modulation rises earlier in the deep layers than in the other laminar compartments . We quantified the difference by defining latencies on these functions as the crossing with a threshold defined from the null distribution obtained by shuffling the stimulus labels ( see Methods ) . Latency was significantly shorter for deep layer units ( 75 . 8 ms , 95% CI [68 . 4 85 . 2] ) than for granular layer units ( 94 . 7 ms , 95% CI [82 . 2 105 . 7] , bootstrap procedure [see Methods] p = 0 . 006 ) and for superficial layer units ( 97 . 7 ms , 95% CI [78 . 0 103 . 7] , bootstrap procedure p = 0 . 018 ) . The same was true when well-isolated units and multiunits were pooled ( deep: 78 . 2 ms , 95% CI [73 . 3 88 . 2] , n = 210; granular: 94 . 0 ms , 95% CI [87 . 9 106 . 5] , n = 167; superficial: 100 . 7 ms , 95% CI [85 . 1 104 . 6] , n = 140; deep vs . granular: p = 0 . 007; deep vs . superficial: p = 0 . 009 ) . To verify the robustness of the temporal differences between layers , we also evaluated the time course of border ownership selectivity using another method , by evaluating border ownership reliability ( BOR ) ( Figure 2F; introduced by Zhou et al . , 2000; detailed in Methods ) . Briefly , this metric quantifies the reliability of border ownership tuning when comparing spike counts between single trials to stimuli with opposite border ownership . Reliability values correspond to the proportion of such single trial comparisons for which the spike count is highest for the border ownership condition that is preferred across trials . We computed BOR in 100-ms sliding windows ( Figure 2F; latency defined similarly as in Figure 2E , using right edge of the analysis window ) . Again we find that BOR rises earlier in the deep layers ( 89 . 9 ms , 95% CI [81 . 5 99 . 9] ) than in the granular ( 105 . 4 ms , 95% CI [94 . 9 114 . 6]; bootstrap procedure p = 0 . 015 ) and superficial layers ( 109 . 4 , 95% CI [98 . 9 124 . 9]; p = 0 . 006 ) . After border ownership selectivity has been established , the average border ownership index tends to be higher in the deep compartment ( Figure 2E; BOI between 200 and 500 ms after stimulus onset: mean ± standard error of the mean [SEM] , 0 . 50 ± 0 . 02 ) than in the granular ( 0 . 43 ± 0 . 02 ) and superficial compartment ( 0 . 45 ± 0 . 03 ) , but the differences do not reach statistical significance ( deep vs . granular: Wilcoxon rank sum test p = 0 . 051; deep vs . superficial: p = 0 . 22 ) . BOR saturates around 0 . 85 in all three compartments ( Figure 2F ) . Together , these data indicate that border ownership selectivity does not occur first in granular layer units but instead in deep layer units . To test whether the short latency in deep layers is specific for border ownership or a general feature of this laminar circuit , we performed two additional analyses . First , we evaluated the latency of spikes evoked by small ring stimuli in the cRF . For these responses we find , in contrast to border ownership selectivity , that the latency is shorter in the granular layer compared to the deep layers and superficial layers ( Figure 3; latency defined as crossing of the functions with a threshold value a third of the way from baseline to peak ) . Note that these responses are derived from the same stimuli used to compute the CSD maps , but represent a different signal ( spiking responses as opposed to the current sink–source patterns from local field potentials used to define the laminar compartments ) . Second , we evaluated the responses to the border ownership stimuli for another type of selectivity , contrast polarity . This refers to the relative luminance contrast across the edge , that is the difference between panels 1 and 3 ( or between panels 2 and 4 ) in Figure 1A . For the contrast polarity index functions ( Figure 2—figure supplement 2 ) , we do not find an earlier rise in selectivity in the deep layers , even though they are derived from the same units as in Figure 2 . Together , these data indicate that the earlier selectivity in deep layers compared to granular and superficial layers is specific for border ownership . How does the preferred side of border ownership compare between units recorded in a column of cortex ? For a given edge , for example vertical , there are two possibilities for preferred side of border ownership: left or right from the edge . Border ownership for such an edge has been assumed to be represented by the activity of two oppositely tuned subpopulations ( green and purple in Figure 4A , top; Craft et al . , 2007 ) . Our data show indeed that these two subpopulations exist in similar proportions ( Figure 4A , top ) . The same is true for units that encode border ownership for horizontal edges ( Figure 4A , bottom ) , and when we express the preferred side of border ownership not relative to the screen , but relative to the fixation point ( Figure 4—figure supplement 1 ) . This indicates that for a given edge , the two possible sides of border ownership are encoded by distinct subpopulations of neurons that are similar in size . These two subpopulations could be mixed in a salt-and-pepper pattern , or they could be clustered according to their preferred side of border ownership ( Figure 4B ) . Figure 4C shows the signed BOI for all well-isolated units and multiunit clusters selective for border ownership for two example penetrations . The sign of BOI indicates which side of the border is preferred by each unit , as indicated by the cartoons above the panels . For both of these penetrations , all border ownership units recorded on the same probe prefer the same side of border ownership . In the population of all penetrations with at least four border ownership units , the proportion of units that share the same preferred side of border ownership is significantly higher than chance ( well-isolated units: 74 . 4% , randomization test p = 0 . 017 , 23 penetrations with 110 units; including multiunit clusters: 71 . 7% , p = 0 . 005 , 47 penetrations with 288 units ) . Also in the subgroup of penetrations with border ownership units spanning from superficial to deep layers , we find significant clustering ( Figure 4D; well-isolated units: 81 . 1% , randomization test p = 0 . 010 , 6 penetrations with 30 units; including multiunit clusters: 73 . 7% , p = 0 . 003 , 20 penetrations with 136 units ) . Together , these data indicate that border ownership preference is organized in spatial clusters that span cortical layers in a columnar fashion . Border ownership acts on edges that are oriented . Similar to our finding that preferred border ownership is clustered , several imaging studies have shown that preferred orientation is spatially clustered in V4 in domains ( e . g . , Li et al . , 2013; Tanigawa et al . , 2010 ) , but it is not known whether these form columns . To test whether orientation preference in V4 is shared in clusters that extend vertically across layers , we determined orientation tuning for the units in our sample from responses to luminance contrast edges centered on the cRF ( independent data set from the border ownership stimuli , see Methods ) . Our data confirm clustering of preferred orientation and reveal that these clusters span across laminar compartments in V4 ( Figure 5 ) . Each polar plot in Figure 5 shows a penetration with at least four orientation-selective well-isolated units or multiunit clusters . Solid vectors indicate the preferred orientation of each orientation-selective unit as the resultant vector of the responses to edges of different orientations , and color indicates laminar compartment ( open symbols indicate multiunit clusters ) . The preferred orientation for a penetration was then calculated as the resultant vector across the vectors of all orientation-selective units in that penetration , and its angle is indicated by the blue dashed line . The significance of clustering of orientation preference was assessed by comparing the magnitude of this resultant vector against a null distribution generated by randomizing the preferred orientation for each orientation-selective unit in the penetration and performing the same calculation . For each of these six penetrations , this resultant was significantly larger than expected from the null distribution ( p < 0 . 05 ) . This was true for 9 out of 18 penetrations with at least four orientation-selective well-isolated units distributed over all three laminar compartments ( for 17 out of 34 penetrations when well-isolated units and multiunit clusters were pooled ) . These data suggest that orientation domains in V4 are columnar . Having found that both border ownership preference and orientation preference are organized as columnar clusters in V4 , we next asked what the relation was between orientation selectivity and border ownership selectivity . We find that border ownership selectivity is significantly more common in units that are selective for orientation than in those that are not , but it is also surprisingly common in units that are not selective for orientation ( Table 1 , respectively , 53 . 3% and 36 . 8% , Chi square = 9 . 51 , p = 0 . 002; including multiunit clusters: respectively , 45 . 9% and 31 . 3% , Chi square = 19 . 5 , p = 0 . 00001 ) . Three example orientation-selective well-isolated units are shown in Figure 6A . These polar plots are organized according to the location of the square object relative to the cRF , for different orientations of the central edge ( the edge that runs through the cRF ) . This is indicated by the stimulus cartoons around the plot . For opposite locations on the polar plot , the central edge has thus the same orientation , but is owned by a square positioned on opposite sides of the cRF ( opposite border ownership ) . The BOI for each tested orientation of the central edge is shown by a red vector ( filled red circles indicate statistically significant border ownership selectivity ) . The amplitude of this vector corresponds to |BOI| , and the polarity of the vector points toward the square location that corresponds to the preferred side of border ownership for that orientation of the central edge . For example , in case of a vertical central edge , the third unit ( cyan triangle ) prefers that that edge is owned by a square on the right ( if it would have preferred the vertical central edge to be owned by a square on the left , the red vector that points down would have pointed up ) . The black line indicates the preferred orientation of a luminance contrast edge in the cRF for each unit . For the left unit in Figure 6A ( red star ) , this preferred edge orientation matches with the orientation of the central edge for which border ownership selectivity is significant ( filled red circle ) . This is the expected scenario: border ownership has been assumed to act on edges at the preferred orientation , and prior studies have used edges at the preferred orientation to test border ownership selectivity ( Zhou et al . , 2000 ) . At the population level , the average orientation of edges with border ownership selectivity is indeed mildly biased toward the preferred orientation for edges . This is subtle but can be seen as the tendency of the data points in Figure 6C to appear close to the identity lines . Note that both plotted variables in Figure 6C are periodic and two periods are shown for each ( thus each data point is plotted four times , and the identity line appears three times; the gray square outlines an area corresponding to one period for both variables ) . This bias is easier to see when the closest distance of the data points to the identity lines is plotted ( Figure 6D ) . This distance is on average distance 27 . 6° , significantly smaller than chance ( randomization test [see Materials and methods] p = 0 . 018; n = 68 well-isolated units tested with at least four orientations; including multiunit clusters: p = 0 . 017; n = 159 ) . Orientation-tuned units thus tend to be border ownership-selective for edges with an orientation near their preferred orientation , but this relation is surprisingly subtle . Some units only have border ownership selectivity for orientations that are far from the identity line , such as the unit indicated with the blue square in Figure 6A ( middle panel , corresponding to the same symbol in Figure 6C ) . This unit shows , paradoxically , significant border ownership selectivity for edge orientations that are nearly orthogonal to the preferred edge orientation . This is true for two edge orientations ( filled red circles ) , suggesting that this misalignment is systematic . We ruled out that this is related to a difference in orientation preference for isolated edges versus edges that are part of squares by comparing orientation tuning for both data sets . Figure 6—figure supplement 1A shows that the preferred orientations for both stimuli match very well for this unit ( black line vs . cyan line ) , and this is true for the population as well ( Figure 6—figure supplement 1B ) . Furthermore , we find that border ownership-selective units often show border ownership selectivity to edges that maximally differ in orientation , that is orthogonal edges . An example unit is shown in the right panel in Figure 6A ( cyan triangle ) . This is true for 29 . 7 % of border ownership-selective well-isolated units ( n = 182; 24 . 4 % including multiunits , n = 353 ) that were tested with sets of squares at orthogonal angles . In such cases , the preferred side of border ownership for a central edge with an orientation in between those orthogonal orientations could in theory be spatially discontiguous ( Figure 6B , left panel ) or contiguous ( Figure 6B , middle panel ) with the preferred sides of border ownership for the pair of orthogonal orientations . We find that for all 24 out of 24 well-isolated units that were selective for the border ownership of squares at orthogonal and intermediate orientations , the preferred side of border ownership for these intermediate orientations was spatially contiguous ( such as the example in Figure 6B , right panel; including multiunits , this is true for 35 out of 35 units ) . Those units thus have a wide but spatially contiguous area of preferred sides of border ownership . For example , the unit in Figure 6B ( right panel ) has an area of preferred sides of border ownership that spans 120° ( cyan double arrow ) . A span of at least this width occurs for 18 . 0 % of border ownership units ( Figure 6E; 15 . 4 % including multiunit clusters ) . Given that about half of the neurons responsive to edges in V2 are selective for border ownership ( Zhou et al . , 2000 ) , neurons in V4 could simply inherit this property from V2 through the feedforward projection from V2 to V4 . This projection is the most important source of feedforward input to V4 ( Markov et al . , 2011 ) , and targets mostly layer 4 ( granular layer ) in V4 ( Gattass et al . , 1997; Rockland , 1992 ) . Our data show that spike latency for stimuli in the cRF is shortest in the granular layer , which is similar to what has been described in other areas in macaque visual cortex ( V1: e . g . , Nowak et al . , 1995; MT: Raiguel et al . , 1999 ) , and consistent with work by others in awake V4 ( Lu et al . , 2018 ) . Note that this may not be universal across areas: in anesthetized V2 spike latency has been reported to be longer in layer 4 than in infragranular layers ( Nowak et al . , 1995 ) . If this feedforward projection would thus simply provide the earliest border ownership signals in V4 , we would thus detect border ownership selectivity in the granular layer at least as early as in the other compartments , or earlier . While we do observe such a pattern for responses evoked by flashed stimuli appearing within the cRF , and also for the emergence of selectivity for contrast polarity ( Figure 2—figure supplement 2 ) , we do not observe this pattern for selectivity to border ownership . Instead , we find that deep layer neurons compute border ownership selectivity significantly earlier than neurons in the granular layer and in the superficial layers . One might argue that these data are consistent with V2 afferents preferentially targeting apical dendrites of deep layer neurons that extend in the input layer . But in that case , we would expect shorter spike latencies for deep layer neurons than for granular layer neurons irrespective of the cue or stimulus . This is , again , not what we find: spike latencies for responses evoked by small stimuli in the cRF are shorter in the granular layer than in the deep layers , and also for contrast polarity we do not observe the earliest selectivity in deep layers in V4 . Another possibility is that deep layer neurons perform this computation earliest solely by relying on an integration of feedforward signals from the V2-to-V4 projection relayed by multiple granular layer neurons . We think this is unlikely . First , this would mean that those relaying granular layer neurons do not receive sufficient contextual information to become selective for border ownership as fast as the deep layer neurons that integrate their output , despite the substantial input that granular layer neurons receive from other granular layer neurons ( Xu et al . , 2016 ) . Second , while deep layer neurons certainly receive input from granular layer neurons , the densest projections from granular layer neurons typically target superficial layers ( e . g . , macaque V1: ; Callaway , 1998; rat barrel cortex: Lübke et al . , 2000 ) . If that projection leads to the first computation of border ownership signals , one would thus expect to observe these signals in superficial layer neurons as fast as in deep layer neurons , in contrast with our data . We think that the earlier computation of border ownership selectivity by neurons in deep layers is more likely a consequence of their unique properties . Scanning laser photostimulation studies in primary visual cortex showed that layer 5 neurons receive significant input from nearly all cortical layers , regardless of cell type , as opposed to neurons in other layers ( Briggs and Callaway , 2005; Xu et al . , 2016 ) . Deep layers of the cortex , including in primates , include tall pyramidal cells whose apical dendrites reach up to layer 1 , which could thus directly sample and integrate afferent information that arrives in a wide range of layers ( Binzegger et al . , 2004; Callaway , 1998; Lund and Boothe , 1975; Markov et al . , 2014; Zarrinpar and Callaway , 2016 ) . Such neurons thus seem well suited to integrate information provided by feedforward input with contextual information provided through corticocortical feedback and horizontal connections ( Harris and Shepherd , 2015 ) . Corticocortical feedback in V4 terminates densely in all layers except layer 4 ( Markov et al . , 2011; Rockland et al . , 1994 ) . Intra-areal horizontal connections are prominent in layer 2/3 and in layer 5 ( Lund et al . , 1993; Yoshioka et al . , 1992; Douglas and Martin , 2007 ) . This may set these deep layer neurons up to be able to integrate the border in the cRF ( light gray arrow in Figure 7 ) with visual context from a wide region of space ( dark gray arrows in Figure 7 ) , which is required to compute border ownership selectivity . Indeed , studies in V2 showed that border ownership selectivity does not rely on a small number of localized object features but instead occurs through integration of extraclassical stimulus features over large areas of visual space ( Zhang and von der Heydt , 2010 ) . Specialized intrinsic properties could further assist in this integration , such as the calcium spikes in apical dendrites of layer five neurons that can amplify the effects of feedback inputs ( Takahashi et al . , 2016 ) . It would be interesting to test whether deep layer neurons in other areas , for example in V2 , also display early border ownership signals ( although obtaining penetrations orthogonal to the cortical surface in macaque V2 is challenging , given that only a small part of this area is exposed on the lateral convexity of the brain ) . Perhaps such contextual integration is a general computation performed by deep layer neurons in different cortical areas . In addition , as discussed below , deep layer neurons in V2 may inherit early border ownership signals from V4 . Our findings are consistent with the concept of the grouping cell model ( von der Heydt , 2015 ) . This model proposes that border ownership is a modulation of feature neurons by an external grouping signal that acts through feedback . The grouping signal is supposed to represent a ‘proto-object’ , a computational structure that ( 1 ) links the visual feature signals , ( 2 ) is being remapped across eye movements , and ( 3 ) serves object-selective attention . Compared to the large number of studies of feature selectivity in visual cortex , the question of the representation of object structure has prompted relatively few studies . It remains unclear where and how such grouping signals are generated ( Zhu et al . , 2020 ) . It is also possible that these deep layer neurons in V4 , rather than computing border ownership selectivity de novo from an external grouping signal , inherit this selectivity from higher areas through corticocortical feedback . The source of feedback , whether it provides border ownership signals or grouping signals , would need to have a spike latency that is short enough to explain the early border ownership signals in deep layers in V4 . Because of shorter latencies , dorsal stream areas such as LIP may therefore be more likely sources of such feedback than inferotemporal cortex , where shape-selective signals occur after 100 ms ( Brincat and Connor , 2006; Bullier , 2001; see also discussion in Zhu et al . , 2020 ) . Our conclusion that the earliest border ownership signals in V4 are not inherited from V2 does not imply that the projection from V2 to V4 does not carry any border ownership signals . Indeed , since about half of V2 neurons are selective for border ownership ( Zhou et al . , 2000 ) , this feedforward input most likely contributes to the border ownership signals in V4 later in the response . As argued in the next section , these border ownership signals in V2 may have been sculpted by border ownership-selective feedback from deep layers in V4 . That V4 does not simply inherit its earliest border ownership signals from V2 is also consistent with a study in V4 on shape encoding that compared responses to identical shapes in the presence or absence of an occluder ( Bushnell et al . , 2011 ) . When the preferred shapes had sharp convexities and concavities , responses were suppressed if those features were rendered incidentally by an adjacent occluder . Suppression latency was relatively short ( 63 ms after stimulus onset in the population , and for an example neuron as short as 46 ms ) . Because this is shorter than the latency of border ownership signals reported in V2 ( under somewhat different stimulus consitions: ~68 ms; Zhou et al . , 2000 ) , the suppression effect is likely not due to border ownership signals that are inherited from V2 . A caveat when interpreting this suppression as border ownership selectivity is that both the shape and the occluder fell in the cRF . Thus , while the suppression may be due to the change in border ownership , it may instead be due to local contour geometry at the T-junctions in the cRF . In any case , this suppression of incidental shapes in V4 could contribute to border ownership signals in V2 in the presence of occluders , through feedback . A computational model has shown that contextual information carried by myelinated feedback afferents could indeed arrive fast enough to result in border ownership selectivity within ~20 ms after response onset ( Craft et al . , 2007 ) . An alternative mechanism based solely on horizontal fibers is unlikely , given the slow conduction velocity of these fibers ( Girard et al . , 2001; Zhang and von der Heydt , 2010 ) , although , as discussed by Zhaoping , 2005 , it should be noted that data on horizontal fibers are scarce . That said , horizontal fibers could still provide part of the required contextual information: since these fibers are limited in length , their role could be to provide information from the portions of the stimulus closest to the cRF ( reminiscent to their proposed role in surround suppression in V1 [Angelucci et al . , 2017] ) . This would explain why the effect of near corners of squares tends to occur later than that of far corners of squares ( Zhang and von der Heydt , 2010 ) . Once border ownership selectivity has been established in V4 deep layers , these signals may contribute to visual processing along different pathways ( black arrows in Figure 7 ) . First , V4 is the dominant source of cortical feedback to V2 , and 75 % of these feedback neurons are located in deep layers in V4 ( Markov et al . , 2014 ) . Early border ownership signals in deep layers of V4 may thus sculpt border ownership selectivity in V2 . Second , V4 deep layers include neurons that project to the superior colliculus ( Fries , 1984; Gattass et al . , 2014 ) . The superior colliculus may thus receive early border ownership signals from V4 , which could contribute to planning upcoming saccades . Saccades target objects more frequently than background ( Rothkopf et al . , 2007 ) . Early border ownership signals in deep layer neurons that project to the superior colliculus could serve to facilitate the rapid foveation of objects . Our data show that these signals are established in deep layers well before 100 ms after stimulus onset . They are thus computed quickly enough to allow time for saccade planning within the intersaccadic interval , which is typically on the order of 200–300 ms ( Otero-Millan et al . , 2008 ) . This proposed early role for deep layer neurons in supporting saccadic decisions is consistent with recent findings showing that deep layer V4 neurons encode more information about the direction of planned eye movements than do superficial neurons ( Pettine et al . , 2019 ) . Border ownership signals may thus play a role beyond visual perception , perhaps including a role in guiding rapid oculomotor behavior . Finally , deep layer neurons may convey border ownership signals to neurons in more superficial layers , which would be consistent with the canonical laminar circuit ( Douglas and Martin , 2004 ) . Our results suggest a potential strategy to study the flow of these signals within single trials . We find that border ownership preference is shared across layers in a columnar fashion . Border ownership is presumably evaluated in the brain by comparing the activity between neurons from such oppositely tuned border ownership columns . In future work , it would thus be interesting to record from such columns simultaneously , and use directionality measures such as Granger causality ( e . g . , Ferro et al . , 2021 ) to study the interlaminar flow of border ownership signals within single trials . Prior studies on border ownership only evaluated border ownership selectivity at the preferred orientation of each neuron ( Hesse and Tsao , 2016; Zhang and von der Heydt , 2010; Zhou et al . , 2000 ) . In our data , there is indeed a bias toward border ownership selectivity for edges near the preferred orientation . However , we find that there is substantial scatter ( Figure 6C and D ) . This does not seem to stem from measurement noise . First , there is a good match between preferred orientations estimated from two separate data sets: one based on recordings made with isolated edges , the other using edges that are part of squares ( Figure 6—figure supplement 1 ) , suggesting that our estimates of orientation preference are highly reliable , with relatively tight bounds . Second , for some units , border ownership selectivity , surprisingly , systematically only occurs for edges with orientations that are nearly orthogonal to the preferred orientation ( Figure 6A , middle ) . Third , almost a third of units that are selective for border ownership to an edge with a given orientation are also border ownership-selective to an edge that is orthogonal to that orientation . This reflects systematic tuning , because the preferred sides of border ownership in such cases form a single contiguous area in retinotopic space ( Figure 6B ) . Finally , a substantial number of units that are selective for border ownership are not selective for orientation ( Table 1 ) . Together , these data indicate that the relation between orientation and border ownership is much richer than has previously been appreciated . Encoding of orientation and of border ownership may represent separate axes of representation . This is not inconsistent with the idea that border ownership assignment represents a surface signal rather than a property tied to a border ( Grossberg , 2015; Nakayama et al . , 1995; Peterson and Skow , 2008 ) . Functional clustering is a recurring theme in primate cortical visual areas ( Hubel and Livingstone , 1987 ) , and has been reported for several modalities in V4 including orientation , hue , direction of motion , spatial frequency , and recently curvature ( Hu et al . , 2020; Jiang et al . , 2021; Li et al . , 2013; Liu et al . , 2020; Lu et al . , 2018; Roe et al . , 2012; Tang et al . , 2020; Tanigawa et al . , 2010 ) . We find here that border ownership preference is preserved across layers within a vertical penetration , indicating that this modality is also organized in a columnar fashion . Importantly though , border ownership is of a fundamentally different nature than these other modalities: the border ownership of an edge is computed based on stimulus features falling outside the cRF . Combined with our finding that early border ownership signals are computed by deep layer neurons rather than being passed on from upstream areas , this columnar organization has important implications for the functional anatomy . It indicates that there is a systematic asymmetric arrangement of the extraclassical contextual information , which , as argued above , may be provided through horizontal fibers and corticocortical feedback . For example , consider a cluster in which neurons prefer that a vertical edge belongs to an object on the left ( left column in Figure 7 , symbol below left column indicates preferred border ownership ) . The neurons in this column thus need to receive asymmetric contextual synaptic information favoring the presence of an object to the left of the edge ( symbol above the left column in Figure 7 ) . Neurons in another cluster ( right column in Figure 7 ) will instead prefer that the same edge is part of an object on the right , and these neurons will thus necessarily receive asymmetric contextual information of the opposite polarity ( symbol above right column ) . The clustered architecture of preferred border ownership therefore requires clusters of asymmetric contextual information in cortex , such that opposite polarities of contextual information occur in distinct and complementary clusters . The present data indicate that border ownership selectivity initially does not arrive through the feedforward pathway . They thus suggest that the substrate of these clusters consists of clustered asymmetries in the retinotopic information carried by afferents from horizontal fibers and cortical feedback . We obtained recordings in two male rhesus macaques ( Macaca mulatta ) , age 13 years ( animal Z ) and age 15 years ( animal D ) . Both animals had no prior experimental history and were housed in separate cages in a primate room with up to six animals of the same species . This study was performed in accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health . All procedures were approved by the Institutional Animal Care and Use Committee of the Salk Institute for Biological Studies ( protocol 14-00014 ) . Surgical procedures have been described before ( Nandy et al . , 2017 ) . In brief , in a first surgical session a titanium recording chamber was installed in a craniotomy over the prelunate gyrus , according to stereotactic coordinates derived from anatomical MRI scans from each animal ( left hemisphere in animal Z , right hemisphere in animal D ) . In a second surgical session , the dura mater within the chamber was removed , and replaced with a silicone-based optically clear artificial dura , establishing an optical window over dorsal area V4 . At the beginning of a recording session , a sterile insert consisting of a metal ring covered with a plastic membrane on the bottom was lowered in the chamber . The membrane was perforated to allow insertion of probes . The function of the insert is to stabilize the recording site from cardiopulmonary pulsations . A linear multielectrode probe ( 32-channel single-shaft acute probes , 100 µm electrode pitch [ATLAS Neuroengineering , Leuven , Belgium] ) was mounted on the chamber using a hydraulic microdrive on an adjustable x–y stage ( MO-972A , Narashige , Japan ) . The probe was then lowered through the artificial dura over the prelunate gyrus , positioned orthogonally relative to the cortical surface under visual guidance ( Zeiss microscope ) . While monitoring the voltage signals from the electrodes for multiunit activity , the probe was lowered to penetrate the cortical surface . The probe was advanced until multiunit activity was visible on the deepest ~2600 µm of the probe . Then , the probe was retracted typically by several 100 µm to ease dimpling of the cortex . Between recording sessions , probe position was varied ( RF eccentricity median 4 . 87 degrees of visual angle ( dva ) , interquartile range 2 . 51 dva ) . Neural signals were recorded extracellularly , filtered and saved using Intan hardware ( RHD2132 amplifier chip and RHD2000 amplifier evaluation system , Intan Technologies LLC , Los Angeles , USA ) controlled by a Windows computer . Visual stimuli were presented using a LED projector , back-projected on a rear-projection screen that was positioned at a distance of 52 cm from the animal’s eyes ( PROPixx , VPixx Technologies , Saint-Bruno , Canada ) . The MonkeyLogic software package developed in MATLAB ( https://www . brown . edu/Research/monkeylogic/; https://monkeylogic . nimh . nih . gov/; Hwang et al . , 2019 ) was used for stimulus presentation , behavioral control and recording of eye position . A photodiode was used to measure stimulus timing . Eye position was continuously monitored with an infrared eye tracking system ( ISCAN model ETL-200 , Woburn , MA ) and eye traces were saved using MonkeyLogic . Trials were aborted if eye position deviated from the fixation point ( threshold typically 1 dva radius ) . At the beginning of each recording session , RF mapping data were obtained using a subspace reverse correlation approach ( Nandy et al . , 2017 ) . Stimuli consisted of static modified Gabors , constructed using square-wave instead of sinusoidal gratings , and dark gray rings ( 80 % luminance contrast , diameter 2 dva , thickness 0 . 25 dva ) . Grating spatial frequency and phase were such that a single contrast edge was visible centrally in the Gaussian window ( grating parameters: 6 orientations , 2 contrast polarities , typically 80 % luminance contrast , one half was one of seven colors or grayscale , the other half was always grayscale; window: FWHM 2 dva ) . The stimuli were presented every 50 or 60 ms while the animal maintained fixation . Each stimulus appeared at a random location selected from a grid sized 25 × 25 dva with 1 dva spacing centered at coordinates [7 . 5 dva; 7 . 5 dva] in the appropriate visual quadrant . During the recording session , high-gamma filtered voltage waveforms in response to these stimuli were analyzed to estimate the retinotopy of the probe position , and choose location and size for the border ownership stimuli . Detailed RFs were calculated for each unit offline after spike sorting and used to verify the proper position of the border ownership stimuli . A CSD mapping procedure on evoked local field potentials ( LFP ) was used to estimate the laminar position of recorded channels ( Nandy et al . , 2017 ) . Briefly , animals maintained fixation while dark gray ring stimuli were flashed ( 32 ms stimulus duration , 94 % luminance contrast , sized and positioned to fall within the cRF of the probe position ) . The CSD was calculated as the second spatial derivative of the stimulus-triggered LFP ( filtered between 3 . 3 and 88 Hz ) and visualized as spatial maps after smoothing using bicubic ( 2D ) interpolation ( Figure 1—figure supplement 1; MATLAB function interp2 with option cubic , the spatial dimension was interpolated at a resolution of 10 µm ) , but the laminar analysis did not critically depend on this particular method of smoothing ( Figure 2—figure supplement 3 ) . Red regions depict current sinks , blue regions depict current sources . As described in more detail in Results , we observed a consistent pattern between different penetrations , strikingly similar to the current sink–source maps reported by other laboratories in behaving macaques beyond V1 ( V4: Pettine et al . , 2019; V4: Lu et al . , 2018; area 36: Takeuchi et al . , 2011 ) . Through histological verification Takeuchi et al . , 2011 found that the prominent current sink with the shortest latency corresponded to the position of the granular layer . We therefore identified this current sink ( current sink indicated by white star , between dashed and dotted white lines in Figure 1F and G; Figure 1—figure supplements 1 and 2 ) as the granular layer . For each unit , the positions of the electrode contacts ( using the five contacts surrounding the one with the largest spike waveform ) were weighed by the average peak-to-trough amplitude of the unit’s spike waveform on these contacts , to assign the unit to the depth where its spike waveform is largest . By comparing this position with the range of contacts in the granular layer , we could locate units to superficial , granular , or deep layers . Seven penetrations where the CSD map could not be interpreted were excluded from the laminar analyses , and we also required that all units included in the laminar analyses were located ≤2 mm from the most superficial channel with multiunit activity , to minimize the risk of including white matter activity . We evaluated the orthogonality of penetrations by comparing the receptive field ( RF ) positions for neural activity recorded on different electrode contacts on the probe . To obtain an estimate for these RF positions , for the population of neurons in the vicinity of each electrode contact , we derived multiunit activity as the amplitude of the envelope from the signal on each electrode contact , by band-pass filtering the recorded signal between 500 Hz and 5 kHz , rectifying it and then low-pass filtering it at 200 Hz ( adopting the approach from Self et al . , 2019 ) . RF contours at z = 3 were then computed on the z-scored root mean square level of this analog signal , using the same procedure as for the spiking data ( see Receptive field mapping ) . For each contour , the RF center was defined as the centroid of the region defined by the contour ( using MATLAB function centroid ) . A line was then fitted in three dimensions through these RF centers ( stacked according to the position of the electrode contact from which they were computed ) from a range of electrode contacts covering 2 mm , starting with the most superficial contact that recorded multiunit activity . The variation between RF positions along the probe was then computed as D , the distance in the azimuth × elevation plane between the ends of this line , per mm depth ( D is indicated for the vertical positions of RF contours shown in Figure 1H , I and Figure 1—figure supplement 1C , and the population data are described in Results ) . A data set for orientation tuning was obtained using luminance contrast edges similar to those used for RF mapping , but with a circular window , sized and positioned such that the stimulus covered and was centered on the estimated aggregate cRF of the probe ( 12 orientations; 2 contrast polarities; 54 % luminance contrast; 200 ms stimulus duration ) . From the online analysis on the RF mapping data , position , size , and color of the border ownership stimulus set were chosen . An isoluminant square was positioned on an isoluminant background , as in prior studies on border ownership ( Zhou et al . , 2000 ) . There were four basic conditions ( Figure 1A and C ) , consisting of a factorial combination of two square positions and two contrast polarities . Note that this results in two pairs of stimuli , where both scenes in a pair have identical stimulus information inside the cRF ( dotted black circles in Figure 1A and C ) . The two luminance areas in each scene were either both grays or a combination of gray and a color , and luminance contrast was 54% , as in prior studies ( Zhou et al . , 2000 ) . Square sizes were between 12 × 12 and 20 × 20 dva . At the beginning of each trial , a small light gray fixation point ( 0 . 2 × 0 . 2 dva , 80 % luminance contrast ) was presented on a blank gray screen ( with luminance set at the geometric mean of the luminances in the border ownership stimulus ) . After the animal maintained fixation for 400 ms , the border ownership stimulus appeared for 500 ms . Then , for a separate project , the stimulus was replaced with a stimulus in which the central edge was prolonged to cover the entire screen ( with identical colors and luminances ) and the other parts of the square removed , for another 1000 ms . Spikes elicited during that time window were not included in the analyses in this paper . The animal received a juice reward if it maintained fixation throughout the trial . Depending on recording time , data were obtained for different orientations or positions . Conditions were played pseudorandomly in counterbalanced blocks such that each condition was played once before repeating conditions . Typically 8–10 repetitions were obtained per condition . Often these stimuli were played at a few different orientations and/or positions , in order to increase the likelihood of proper stimulus placement for most units recorded on the probe ( because the precise cRF for each unit was only available offline , after spike sorting ) . On some days , the trials analyzed here were randomly interdigitated with similar trials using stimuli defined for other projects . Some of these trials – not analyzed here – included a condition in which the animal had to saccade to a new fixation position if the fixation point moved . Data were analyzed in MATLAB ( MathWorks , Natick , MA ) . Circular statistics were computed using the MATLAB toolbox CircStat ( Berens , 2009 ) . Statistical tests for the different analyses are described in detail below . Statistical significance was defined as p < 0 . 05 . The data were sorted offline using SpyKING CIRCUS ( Yger et al . , 2018 ) . The clusters resulting from the automatic sorting step were curated manually using the MATLAB GUI provided by the SpyKING CIRCUS software . Well-isolated units were identified based on a well-defined refractory period in the interspike-interval histogram . Multiunit clusters included in the analysis had to pass a criterion for the signal-to-noise ratio: peak-to-peak amplitude of the average waveform had to exceed five times the standard deviation of the signal 5 ms prior to the peak ( similar to Kashkoush et al . , 2019 ) , after high-pass filtering the data ( 1-pole butterworth filter , cutoff 300 Hz , implemented using functions butter and filtfilt in MATLAB ) . To determine the cRF , spikes were counted in a window [30 100] ms after each stimulus onset . The resulting mean counts per stimulus position were transformed to z-scores by first subtracting the mean of and then dividing by the standard deviation of spike counts occurring in a window of the same size preceding that stimulus position ( Keliris et al . , 2019 ) . Using the stimulus positions , z-scores were transformed to a spatial map , which was smoothed with a Gaussian filter ( MATLAB function imgaussfilt with σ = 1 ) . The outline of the cRF was defined as the contour at z = 3 on this smoothed map ( calculated using MATLAB function contourc ) . Border ownership responses were obtained by recording evoked responses to stimuli as in Figure 1A and C ( see Border ownership stimuli above ) . A unit’s response was evaluated for border ownership selectivity if it passed the following inclusion criteria: ( 1 ) average spike rate was ≥1 spike/s for at least one of the four conditions; ( 2 ) the evoked spike count for at least one of the four conditions was significantly different from that recorded prior to all trials across conditions ( two-sided Wilcoxon rank sum test [MATLAB function ranksum] with Bonferroni correction ) ; ( 3 ) at least six trials per condition were available; ( 4 ) the central edge of the squares in the border ownership stimulus intersected the cRF; ( 5 ) the distance between any part of the cRF contour and any part of the noncentral edges of the squares in the border ownership stimulus was ≥1 dva . Data sets obeying these inclusion criteria were candidate data sets for border ownership selectivity . The border ownership index ( BOI ) ( Zhou et al . , 2000 ) was then calculated for these data sets , which is defined asBOI=R1+R3-R2+R4R1+R3+R2+R4 where Ri represents the average spike rate in the window [50 500] ms after stimulus onset for condition i ( numbering as in Figure 1A and C ) . Statistical significance of border ownership selectivity was evaluated using a permutation test: a null distribution was created by shuffling the border ownership stimulus labels 10 , 000 times , separately for each luminance contrast pair , and the p value was estimated as the fraction of |BOIshuffled| that was at least as large as |BOI| . A data set was defined to be border ownership-selective if p < 0 . 05 ( with Bonferroni correction if multiple orientations or positions were available for the same unit ) , and a unit was defined to be border ownership-selective if it had at least one border ownership-selective data set . For the time course of evoked activity of border ownership-selective units ( Figure 2A–D ) , each unit contributed one data set , that is the border ownership-selective data set – as defined above – for which |BOI| was maximal . For each of these , the time courses of the responses to the preferred side of border ownership ( side resulting in the highest spike rate ) and to the non-preferred side of border ownership ( side resulting in the lowest spike rate ) were calculated separately ( respectively , solid red lines and dashed blue lines Figure 2A–D ) as follows . Spike trains were rounded to 0 . 1 ms resolution and convolved with a postsynaptic kernel K ( t ) ( Thompson et al . , 1996 ) Kt=1-e-tτg∙e-tτd where τg = 1 ms and τd = 20 ms . The resulting traces were averaged per condition , and then across both contrast polarities . These average traces were normalized for each unit’s evoked response by dividing them by the average value across conditions in the window [50 500] ms after stimulus onset . The mean ± SEM of the resulting functions across units are shown in Figure 2A–D . The latency of statistical difference between the functions to the preferred and the non-preferred functions ( asterisks in Figure 2A–D; Figure 2—figure supplement 1 ) was defined as the first time after which the functions differed statistically ( Wilcoxon sign rank test p < 0 . 05 ) for 20 adjacent milliseconds . BOI functions ( Figure 2E ) were defined as the difference between the response function for the preferred side of border ownership and the function for the non-preferred side of border ownership , divided by their sum . Latency of these functions was defined as the earliest crossing of a fixed threshold that was followed by values above the threshold for 20 consecutive milliseconds . The threshold used was derived from shuffled data , by shuffling the stimulus labels for each laminar compartment ( 1000 shuffles ) and finding the lowest value for which <1% of shuffles resulted in a defined latency . Since these values depend on sample size , the highest value across compartments was used as threshold , so that the different functions could be timed using the same threshold . Confidence intervals ( 95% ) were calculated on these latencies using a bootstrap approach with the bias corrected and accelerated percentile method ( MATLAB function bootci , 2000 bootstraps ) . The latencies were statistically compared between laminar compartments using a bootstrap approach similar to other studies ( Self et al . , 2019 ) , by computing the difference in latency between bootstrap samples of different compartments and estimating p as the fraction of samples on which the difference was ≤0 ( one-sided test ) . Spike response functions to small rings in the cRF ( Figure 3 ) were computed similarly as the response functions to border ownership stimuli ( Figure 2A–D ) . Responses were normalized by dividing them by the peak response for each unit . Latency was defined using a threshold a third of the way from baseline to peak ( 0 . 435 ) using the highest baseline across compartments , and confidence intervals and statistical tests were computed in the same way as for the BOI functions . We also evaluated the time course of border ownership selectivity using the metric border ownership reliability ( BOR; Zhou et al . , 2000 ) , which reflects the trial-to-trial reliability of encoding border ownership . This metric was computed for the border ownership-selective units in each laminar compartment , in 100 ms sliding windows ( 1 ms steps ) . For each unit , 10 , 000 sets of four spike trains were generated , where each set contained one random spike train from each of the four conditions in Figure 1A and C . For each window position , spikes were counted for each spike train in each set . BOR for a unit at a particular window position was defined asBOR=∑jAj∑jAj+∑jBj where j corresponds to the index of all spike train sets for the unit . A ( j ) and B ( j ) indicate whether the sign of the spike count difference between border ownership conditions for spike train set j is , respectively , the same or opposite compared to the unit’s preferred side of border ownership:{A ( j ) =1ifsgn[ ( Cj , 1+Cj , 3 ) − ( Cj , 2+Cj , 4 ) ]=SA ( j ) =0ifsgn[ ( Cj , 1+Cj , 3 ) − ( Cj , 2+Cj , 4 ) ]≠S{B ( j ) =1ifsgn[ ( Cj , 1+Cj , 3 ) − ( Cj , 2+Cj , 4 ) ]=−SB ( j ) =0ifsgn[ ( Cj , 1+Cj , 3 ) − ( Cj , 2+Cj , 4 ) ]≠−SS=sgn[ ( R1+R3 ) − ( R2+R4 ) ] where Cj , i represents the window spike count for condition i in spike train set j , Ri is the average spike rate for condition i ( for the interval [50 500] ms after stimulus onset ) , and sgn is the sign function . For each window position for each unit , BOR was only computed if there were at least 10 spikes across conditions . For every window position the mean was calculated across units per laminar compartment , resulting in the BOR functions shown in Figure 2F . The abscissa in Figure 2F corresponds to the position of the right edge of the window , that is it indicates the latest spike times that may have determined BOR for that window position ( and thus represents a conservative estimate for the latency ) . Definition of latency , computation of confidence intervals , and statistical tests used to compare latencies between layers were similar to those for the BOI functions . For a given penetration , for each orientation , the border ownership-selective data set with the highest |BOI| was selected for each unit . Then from this group of data sets the largest subgroup that shared the same edge orientation and position was retained for analysis . Each unit could thus maximally contribute one data set . The preferred side of border ownership was then determined for all these data sets ( example penetrations are shown in Figure 4C ) . The proportion of units sharing the most common preferred side was calculated for each penetration . The average of these proportions across penetrations ( Ppref ) was compared with a null distribution generated by randomly assigning the preferred side of border ownership for each data set ( 2000 randomizations; that is , a binomial process with chance of success = 0 . 5 ) . The p value was estimated as the fraction of the null distribution for which Ppref was at least as large as the actual data . Orientation selectivity was evaluated using a Kruskal–Wallis test ( MATLAB function kruskalwallis ) on the evoked spike counts in the analysis window ( between 30 ms and 200 after stimulus onset ) from the orientation tuning data set , using orientation as the grouping variable ( Pettine et al . , 2019 ) , and defined as p < 0 . 05 . All units for which the center of the contrast edge in the orientation data set was positioned in the cRF , and the z-scored spike rate for the orientation with the highest rate was ≥3 were included . Z-scores were calculated by first subtracting the mean of and then dividing by the standard deviation of the spike rate in a window preceding the first stimulus in the trial , equal in duration to the analysis window . For each orientation-selective unit i , the response to each orientation j is summarized as R→j , which has a direction 2j ( because orientation has a period of 180° ) and magnitude equal to the spike rate . The resultant O→i ( vectors shown as solid lines in Figure 5 ) for all R→j is then calculated . The magnitude of orientation selectivity and the preferred orientation of unit i were defined , respectively , as the magnitude and as the direction divided by 2 ( because of the definition of direction of R→j ) of O→i . For each penetration , the aggregate preferred orientation was defined as the direction divided by 2 of the resultant vector O→ of all O→i on that penetration ( direction of O→ is indicated by blue dashed lines in Figure 5 ) . Statistical significance of the aggregate preferred orientation was assessed by randomizing the directions of O→i ( 2000 randomizations ) , calculating the null distribution O→shuffled , and estimating the p value as the fraction of O→shuffled for which the magnitude was at least as large as that of O→ . The independence between the fractions of , respectively , orientation-selective and border ownership-selective units ( Table 1 ) was assessed using a Chi-square test ( MATLAB function crosstab ) . For units that were border ownership-selective to multiple orientations , the circular mean of border ownership-selective orientations ( ordinate in Figure 6C ) was calculated as the direction divided by 2 of the resultant vector of all vectors B→j that have a direction 2 j and a magnitude equal to |BOI| , for all border orientations j . The shortest distance of the data points in Figure 6C to the identity line ( Figure 6D ) was analyzed by comparing the mean to a null distribution generated by shuffling the values for the preferred edge orientation and the circular mean of border ownership-selective orientations , and calculating the mean of the shortest distance to the identity line ( 2000 shuffles ) . The p value was estimated as the fraction of the null distribution that was as small or smaller than the observed value .

To understand a visual scene , the brain needs to identify objects and distinguish them from background . A border marks the transition from object to background , but to differentiate which side of the border belongs to the object and which to background , the brain must integrate information across space . An early signature of this computation is that brain cells signal which side of a border is ‘owned’ by an object , also known as border ownership . But how the brain computes border ownership remains unknown . The optic nerve is a cable-like group of nerve cells that transmits information from the eye to the brain’s visual processing areas and into the visual cortex . This flow of information is often described as traveling in a feedforward direction , away from the eyes to progressively more specialized areas in the visual cortex . However , there are also numerous feedback connections in the brain , running backward from more specialized to less specialized cortical areas . To better understand the role of these feedforward and feedback circuits in the visual processing of object borders , Franken and Reynolds made use of their stereotyped projection patterns across the cortex layers . Feedforward connections terminate in the middle layers of a cortical area , whereas feedback connections terminate in upper and lower layers . Since time is required for information to traverse the cortical layers , dissecting the timing of border ownership signals may reveal if border ownership is computed in a feedforward or feedback manner . To find out more , electrodes were used to record neural activity in the upper , middle and lower layers of the visual cortex of two rhesus monkeys as they were presented with a set of abstract scenes composed of simple shapes on a background . This revealed that cells signaling border ownership in deep layers of the cortex did so before the signals appeared in the middle layer . This suggests that feedback rather than feedforward is required to compute border ownership . Moreover , Franken and Reynolds found evidence that cells that prefer the same side of border ownership are clustered in columns , showing how these neural circuits are organized within the visual cortex . In summary , Franken and Reynolds found that the circuits of the primate brain that compute border ownership occur as columns , in which cells in deep layers signal border ownership first , suggesting that border ownership relies on feedback from more specialized areas . A better understanding of how feedback in the brain works to process visual information helps us appreciate what happens when these systems are impaired .

2021

Columnar processing of border ownership in primate visual cortex

Recent bacterial ( meta ) genome sequencing efforts suggest the existence of an enormous untapped reservoir of natural-product-encoding biosynthetic gene clusters in the environment . Here we use the pyro-sequencing of PCR amplicons derived from both nonribosomal peptide adenylation domains and polyketide ketosynthase domains to compare biosynthetic diversity in soil microbiomes from around the globe . We see large differences in domain populations from all except the most proximal and biome-similar samples , suggesting that most microbiomes will encode largely distinct collections of bacterial secondary metabolites . Our data indicate a correlation between two factors , geographic distance and biome-type , and the biosynthetic diversity found in soil environments . By assigning reads to known gene clusters we identify hotspots of biomedically relevant biosynthetic diversity . These observations not only provide new insights into the natural world , they also provide a road map for guiding future natural products discovery efforts . Soil-dwelling bacteria produce many of the most important members of our pharmacy , including the majority of our antibiotics as well as many of the cytotoxic compounds used in the treatment of cancers ( Cragg and Newman , 2013 ) . The traditional approach for characterizing the biosynthetic potential of environmental bacteria has been to examine metabolites produced by bacteria grown in monoculture in the lab . However , it is now clear that this simple approach has provided access to only a small fraction of the global microbiome's biosynthetic potential ( Rappe and Giovannoni , 2003; Gilbert and Dupont , 2011; Rajendhran and Gunasekaran , 2011 ) . In most environments uncultured bacteria outnumber their cultured counterparts by more than two orders of magnitude , and among the small fraction of bacteria that has been cultured ( Torsvik et al . , 1990 , 1998 ) , only a small subset of gene clusters found in these organisms is generally expressed in common fermentation broths ( Bentley et al . , 2002; Ikeda et al . , 2003 ) . The direct extraction and subsequent sequencing of DNA from environmental samples using metagenomic methods provides a means of seeing this ‘biosynthetic dark matter’ for the first time . Unfortunately , the genomic complexity of most metagenomes limits the use of the shotgun-sequencing and assembly approaches ( Iverson et al . , 2012; Howe et al . , 2014 ) that are now routinely used to study individual microbial genomes ( Donadio et al . , 2007; Cimermancic et al . , 2014 ) . Although bacterial natural products represent an amazing diversity of chemical structures , the majority of bacterial secondary metabolites , including most clinically useful microbial metabolites , arise from a very small number of common biosynthetic themes ( e . g . , polyketides , ribosomal peptides , non-ribosomal peptides , terpenes , etc ) ( Dewick , 2002 ) . Because of the functional conservation of enzymes used by these common systems , degenerate primers targeting the most common biosynthetic domains provide a means to broadly study gene cluster diversity in the uncultured majority in a way similar to what is now regularly done for bacterial species diversity using 16S rRNA gene sequences . Here we use this approach to conduct the first global examination of non-ribosomal peptide synthetase ( NRPS ) adenylation domain ( AD ) and polyketide synthase ( PKS ) ketosynthase ( KS ) domain biosynthetic diversity in soil environments . We chose to explore NRPS and PKS biosynthesis because the highly modular nature of these biosynthetic systems has provided a template for the production of a wide variety of gene clusters that give rise to a correspondingly diverse chemical repertoire , including many of the most clinically useful microbial metabolites ( Cragg and Newman , 2013 ) . With the help of a citizen science effort ( www . drugsfromdirt . org ) , soil samples were collected from five continents ( North America , South America , Africa , Asia , Australia ) and several oceanic islands ( Hawaii , Dominican Republic ) , covering biomes that include multiple rainforests , temperate forests , deserts and coastal sediments . DNA was extracted directly from these soils as previously described ( Brady , 2007 ) and 96 samples were chosen for analysis of NRPS/PKS diversity using 454 pyro-sequencing of AD and KS domain PCR amplicons . Samples were chosen on the basis of DNA quality and biome diversity; raw sequence reads from these samples were combined with existing amplicon datasets derived from other biomes using the same DNA isolation , PCR and sequencing protocols ( Charlop-Powers et al . , 2014 ) . The entire dataset representing 185 biomes was clustered into operational taxonomic units ( OTUs ) at a sequence distance of 5% . Despite millions of unique sequencing reads yielding a predicted Chao1 OTU estimate of greater than 350 , 000 for each domain , rarefaction analysis suggests that we have not yet saturated the sequence space of either domain ( Figure 1A , C ) . 10 . 7554/eLife . 05048 . 003Figure 1 . Global abundance and comparative distribution of AD/KS sequences . The global abundance ( A and C ) , sample-to-sample variation ( B and D ) , and geographic distribution ( E , F , G , and H ) of adenylation domains ( AD ) and ketosynthase domains ( KS ) were assessed by pyro-sequencing of amplicons generated using degenerate primers targeting AD and KS domains found in 185 soils/sediments from around the world . ( A and C ) Global AD ( A ) or KS ( C ) domain diversity estimates were obtained by rarefying the global OTU table ( de novo clustering at 95% ) for AD and KS sequences and calculating the average Chao1 diversity metric at each sampling depth . ( B and D ) The ecological distance ( i . e . , Jaccard dissimilarity ) between AD ( B ) or KS ( D ) domain populations sequenced from each metagenome was determined as a function of the great circle distance between sample collection sites ( km ) . Insets show local relationships ( <500 km ) in more detail . ( E and F ) All sample collection sites are shown on each world map and lines are used to connect sample sites that share at least the indicated fraction ( 3% , 10% ) of AD ( E ) or KS ( F ) OTUs . ( G and H ) Biome-specific relationships within domain OTU populations sequenced from geographically proximal samples assessed by Jaccard similarity . Samples were collected from ( G ) Atlantic forest , saline or cerrado environments or from the ( H ) New Mexican desert topsoils or hot springs sediments . DOI: http://dx . doi . org/10 . 7554/eLife . 05048 . 003 The first question we sought to address with this data was how biosynthetic sequence composition varies by geographic distance . To do this we calculated the pairwise Jaccard distances between AD/KS sequence sets derived from each sampling site and used these metrics to compare samples . The Jaccard distance , a widely used metric for comparing the fraction of shared OTUs between samples , was chosen over alternative metrics due to its simplicity and to the lack of a comprehensive reference phylogenetic tree for AD and KS domains as exists for 16S analyses . Most Jaccard distances were found to be quite small ( <3% ) , indicating large differences in secondary metabolite gene sequence composition between almost all sample collection sites ( Figure 1B , D ) . Although the OTU overlap between our individual experimental samples is generally small , these relationships allow us to begin to develop a picture of how biosynthetic diversity varies globally . On a global level , the strongest biosynthetic sequence composition relationships are seen between samples collected in close physical proximity to one another ( Figure 1B , D , E , F ) as opposed to between samples from similar biomes in different geographic locations . For example , at a cutoff of even as low as 3% shared KS or AD OTUs , essentially all inter-sample relationships are observed between immediate geographic neighbors and not similar biomes in different global locations ( Figure 1E , F ) . This likely explains the limited inter-sample relationships we observe between samples from the Eastern hemisphere as most samples from this part of the world were collected from sites at a significant geographic distance from one another . The only exception is the set of soil samples from South Africa , of which a number were collected in relatively close geographic proximity . These samples exhibit similar pairwise Jaccard metrics to those observed between geographically proximal samples collected in the Western hemisphere ( Figure 1E , F ) . Although differences in biosynthetic composition of microbiomes appear to depend at least in part on the geographic distance between samples , our data suggests that change in the biome type is an important additional factor for the differentiation of biosynthetic diversity on a more local level ( Figure 1G , H ) . For example , at a cutoff of 3% shared OTUs , essentially all inter-sample relationships are observed between immediate geographic neighbors; when this is raised to 10% shared OTUs ( Figure 1E , F ) , relationships are only seen between nearby samples belonging to the same biome . This phenomenon is highlighted by the two examples shown in Figure 1G , H . In the first example , Brazilian soils were collected from Atlantic rainforest , saline or cerrado ( savanna-like ) sites located only a few miles from one another . Our AD and KS data show these sample are ( i ) distinct from other globally distributed samples , ( ii ) most strongly related to the samples from the same Brazilian biome and ( iii ) only distantly related to the samples from other Brazilian biomes . In the second example , a sample collected from a New Mexican hot spring where the soil is heated continuously by subterranean water is compared with samples derived from the dry soils of the surrounding environment . Once again our amplicon data show that these samples are ( i ) distinct from other globally distributed samples , ( ii ) most strongly related to other samples from the same biome and ( iii ) only distantly related to samples from other nearby biomes . Although it is possible that at a much greater sampling depth all AD and KS domains will be found at all sites as predicted by Baas-Becking's ‘everything is everywhere but the environment selects’ hypothesis of global microbial distribution ( O'Malley , 2007; de Wit and Bouvier , 2006 ) , our PCR-based data suggest that both geography and ecology play a role in determining the major biosynthetic components of a microbiome . The vast majority of AD and KS domain sequences coming from environmental DNA ( eDNA ) are only distantly related to functionally characterized NRP/PK gene clusters , precluding precise predictions about the specific natural products encoded by the gene clusters from which most amplicons arise . However , in cases where eDNA sequence tags show high sequence similarity to domains found in functionally characterized gene clusters , this information can be used to predict the presence of specific gene cluster families within a specific microbiome . This type of phylogenetic analysis is the basis of the recently developed eSNaPD program , a BLAST-based algorithm for classifying the gene cluster families that are associated with eDNA-derived sequence tags ( Owen et al . , 2013; Reddy et al . , 2014 ) . When an eDNA sequence tag clades with , but is not identical to , a reference sequence in an eSNaPD-type analysis , it is considered to be indicative of the presence of a gene cluster that encodes a congener ( i . e . , a derivative ) of the metabolite encoded by the reference cluster . Interestingly , eSNaPD analysis of the data from all sites reveals two distinct types of biomedically relevant natural product gene cluster ‘hot spots’ within our data ( Figure 2A , B , D ) . These include ‘specific gene cluster hotspots’ and ‘gene cluster family hotspots’ . Metagenomes from ‘specific gene cluster hotspots’ are predicted to be enriched for a gene cluster that encodes a congener of the target natural product , while metagenomes from ‘gene cluster family hotspots’ are predicted to encode multiple congeners related to the target natural product . Figure 2A shows several of the strongest examples of ‘specific gene cluster hotspots’ where reads falling into an OTU related to a specific biomedically relevant gene cluster or gene cluster family are disproportionately represented in the sequence data from individual microbiomes . These examples highlight the different enrichment patterns that we observe in the environment—hotspots are either local in nature , consisting of only one or two samples containing sequence reads mapping to the target ( epoxomycin , oocydin ) ; regional ( tiacumicinB ) ; or global with punctuated increases in diversity ( glycopeptides ) . We would predict ‘specific gene cluster hotspots’ ( Figure 2D ) are naturally enriched for bacteria that encode congeners of the biomedically relevant target metabolites , thereby potentially simplifying the discovery of new congeners . Figure 2B shows examples of ‘gene cluster family hotspots’ , where metagenomes having a disproportionately high number of OTUs mapping to a specific biomedically relevant target molecule family ( e . g . , nocardicin , rifamycin , bleomycin , and daptomycin families are shown ) are highlighted . This analysis identifies specific sample sites , from among those surveyed , that are predicted to contain the most diverse collection of gene clusters associated with a target molecule of interest ( Figure 2B ) . Both types of hotspots should represent productive starting points for future natural product discovery efforts aimed at expanding the structural diversity and potential utility of specific biomedically relevant natural product families . 10 . 7554/eLife . 05048 . 004Figure 2 . Biomedically relevant natural product hotspots and diversity . Hotspot analysis of natural product biosynthetic diversity to identify samples with a high total proportion of reads corresponding to a natural product family of interest ( A and D ) , the maximum unique OTUs corresponding to a natural product family of interest ( B and D ) , or the estimated sample biodiversity ( C and D ) . In A and B samples are arranged by longitude and hemisphere as is shown in the Sample Key . ( A ) For each sample , sequence reads assigned by eSNaPD are expressed as a percentage of total reads obtained for that sample . A sample is designated a hotspot if more than one percent ( 0 . 01; horizontal line ) of its reads map to a specific gene cluster . Fractional observance data for five representative gene clusters or gene cluster families ( zorbamycin , oocydin , tiacumicinB , epoxomicin , glycopeptides ) that show significant sample dependent difference in read frequency are shown . ( B ) Hotspots of elevated gene cluster family diversity can be identified by determining the number of unique OTUs occurring in each sample that , by eSNaPD , map to a natural product gene cluster of interest . Sample specific OTU counts for nocardicin , rifamycin , bleomycin , and daptomycin clusters are shown . Samples containing greater than 50% of the maximum observed OTU value are colored and mapped in ( C ) . OTU diversity measurements do not predict the abundance of a specific cluster in a metagenome [as predicted in ( A ) ] , but instead are used to identify locations where the largest number of congener-encoding clusters may be found . These sites are predicted to be most useful for increasing the structural diversity and therefore potential clinical utility of these medically important families of natural products . ( C ) Estimated diversity of AD/KS reads by sample . AD and KS OTU tables were combined and for each sample the Chao1 diversity metric was calculated at 5000 reads , providing a baseline metric for comparing sample biosynthetic diversity . The average number of unique OTUs observed over 10 rarefactions analyses is shown ( also see Supplementary file 7 ) . ( D ) Hotspot map of samples identified in A , B and C . ( E ) Representative structures of target molecule families highlighted in A and B . DOI: http://dx . doi . org/10 . 7554/eLife . 05048 . 004 Biosynthetic domain sequence tag data are not only useful for pinpointing environments that are rich in specific biosynthetic targets of interest but also as a metric for natural product biosynthetic diversity in general . As only a small fraction ( 5–10% ) of total AD and KS sequences can be confidently assigned by the eSNaPD algorithm , samples showing the largest collection of unique OTUs ( at a common sequencing depth ) might be expected to contain the most diverse collection of novel biosynthetic gene clusters ( Figure 2C ) and therefore be the most productive sites to target for future novel molecule discovery efforts . Once normalized for sequencing depth , the number of unique KS and AD sequence tags observed per collection site differs by almost an order of magnitude between environments ( Figure 2C ) , with the most diverse samples mapping to Atlantic forest and Desert environments ( Figure 2C , D teal spots , Supplementary file 7 ) . The development of cost effective high-throughput DNA sequencing methodologies and powerful biosynthesis focused bioinformatics algorithms allow for the direct interrogation and systematic mapping of global microbial biosynthetic diversity . Our analyses of 100s of distinct soil microbiomes suggests that geographic distance and local environment play important roles in the sample-to-sample differences we detected in biosynthetic gene populations . As variations in biosynthetic gene content are expected to correlate with variations in the small-molecule producing capabilities of a microbiome , the broader implication of these observations from a drug discovery perspective is that the dominant biosynthetic systems of geographically distinct soil microbiomes are expected to encode orthogonal , largely unexplored collections of natural products . Taken together , our biosynthetic domain hotspot and OTU diversity analyses represent a starting point in the creation of a global natural products atlas that will use sequence data to guide natural product discovery in the future . Based on the historical success of natural products as therapeutics , microbial ‘biosynthetic dark matter’ is likely to hold enormous biomedical potential . The key will be learning how to harvest molecules encoded by the biosynthetic diversity we are now able to find through sequencing . Soil from the top 6 inches of earth was collected at unique locations in the continental United States , China , Brazil , Alaska , Hawaii , Costa Rica ( Brady and Clardy , 2004 ) , Ecuador , the Dominican Republic , Australia , Tanzania and South Africa . The full sample table is available in Supplementary file1 . To reduce the potential for cross contamination , DNA was extracted from soil using a simplified version of our previously published DNA isolation protocol ( Brady , 2007; Reddy et al . , 2012 ) . The modified protocol was as follows: 250 grams of each soil sample was incubated at 70°C in 150 ml of lysis buffer ( 2% sodium dodecyl sulfate [wt/vol] , 100 mM Tris–HCl , 100 mM EDTA , 1 . 5 M NaCl , 1% cetyl trimethyl-ammonium bromide [wt/vol] ) for 2 hr . Large particulates were then removed by centrifugation ( 4000×g , 30 min ) , and crude eDNA was precipitated from the resulting supernatant with the addition of 0 . 6 vol of isopropyl alcohol . Precipitated DNA was collected by centrifugation ( 4000×g , 30 min ) , washed with 70% ethanol and resuspended in a minimum volume of TE ( 10 mM Tris , 1 mM EDTA [pH 8] ) . Crude environmental DNA was passed through two rounds of column purification using the PowerClean system ( MO BIO , Carlsbad , California ) . Purified environmental DNA was then diluted to 30 ng/µl and archived for use in PCR reactions . Degenerate primers targeting conserved regions of AD [A3F ( 5′-GCSTACSYSATSTACACSTCSGG ) and A7R ( 5′-SASGTCVCCSGTSCGGTA ) ( Ayuso-Sacido and Genilloud , 2005 ) ] and KS [degKS2F . i ( 5′-GCIATGGAYCCICARCARMGIVT ) and degKS2R . i ( 5′-GTICCIGTICCRTGISCYTCIAC ) ( Schirmer et al . , 2005 ) ] domains were used to amplify gene fragments from crude eDNA . Forward primers were designed to contain a 454 sequencing primer ( CGTATCGCCTCCCTCGCGCCATCAG ) followed by a unique 8 bp barcode that allowed simultaneous sequencing of up to 96 different AD- or KS- samples in a single GS-FLX Titanium region . PCR reaction consisted of 25 µl of FailSafe PCR Buffer G ( Epicentre , Madison , Wisconsin ) , 1 µl recombinant Taq Polymerase ( Bulldog Bio , Portsmouth , New Hampshire ) , 1 . 25 µl of each primer ( 100 mM ) , 14 . 5 µl of water and 6 . 5 µl of purified eDNA . PCR conditions for AD domain primers were as follows: 95°C for 4 min followed by 40 cycles of 94°C for 0 . 5 min , 67 . 5°C for 0 . 5 min , 72°C for 1 min and finally 72°C for 5 min . PCR conditions for KS domain primers were as follows: 95°C for 4 min followed by 40 cycles of 54°C for 40 s , 56 . 3°C for 40 s , 72°C for 75 s and finally 72°C for 5 min . PCR reactions were examined by 2% agarose gel electrophoresis to determine the concentration and purity of each amplicon . Amplicons were pooled in equal molar ratios , gel purified using the Invitrogen eGel system and DNA of the appropriate size was recovered using Agencourt Ampure XP beads ( Beckman Coulter , Brea , California ) . Amplicons were sequenced using the 454 GS-FLX Titanium platform . Raw flowgram files from 454's shotgun processing routine were used for downstream analysis . Raw reads were assigned to samples using the unique primer barcodes and filtered by quality ( 50 bp rolling window PHRED cutoff of 20 ) using Qiime ( version 1 . 6 ) ( Caporaso et al . , 2010 ) . USEARCH ( version 7 ) , which implements the improved UPARSE clustering algorithm ( Edgar , 2013 ) , was used to remove Chimeric sequences with the default 1 . 9 value of the de novo chimera detection tool . UPARSE clustering requires all sequences to be of the same length . In an effort to balance read quality and abundance with the ability to phylogenetically discriminate gene clusters we used 419 bp as our read length cutoff . The trimmed fasta file was then clustered to 5% to compensate for sequencing error and natural polymorphism that is often observed in gene clusters found in natural bacterial populations . Clustering proceeded as per the USEARCH manual by clustering at a distance of 3% and using representative sequences from each cluster to cluster again at 5% . The resulting ‘5%’ AD and KS OTU tables were used for all subsequent rarefaction and diversity analyses . Read and OTU counts available in Supplementary file 2 . To assess global AD and KS diversity in our sample set we sought to assess the global number of AD and KS domains we might expect to see if all of our data had been generated from a single sample . To do this , all reads assigned to an OTU were consolidated to generate a single-column OTU table where each row contains the sum of all sequences assigned to that OTU from any of the 185 samples . To assess the global diversity we subsampled this table at multiple depths using Qiime ( Caporaso et al . , 2010 ) and used the Chao1 formula to estimate the expected number of OTUs at this depth . This rarefaction analysis was performed ten times at each subsampling depth ( Figure 1A , C; Supplementary files 3 , 4 ) and the curves were fit to the data using the following equation: y = 1 + log ( x ) + log ( x^2 ) + log ( x^3 ) where x is the read value and y is the Chao1 diversity . Ecological distances are calculated using the Jaccard [1 − ( OTUA&B ) / ( OTUA + OTUB − OTUA&B ) ] or inverse Jaccard metric ( Oksanen et al . , 2013 ) and geographic distances were calculated using great circle ( spherical ) distance derived from the latitude/longitude values of each set of points ( Bivand and Pebesma , 2005 ) ( Supplementary file 5 ) . Pairwise ecological and geographic distances were used to create Figure 1B , D . Network plots of subsamples ( Figure 1G , H ) were generated using Phyloseq ( McMurdie and Holmes , 2013 ) to calculate the intersample Jaccard distance . As expected , the strongest relationships are observed between sample proximity controls where soils were collected approximately 10 meters from one another and processed independently , demonstrating that closely related samples do in fact group together in our analysis pipeline . AD and KS amplicon reads were assigned to known biosynthetic gene clusters using the eSNaPD algorithm at an e-value cutoff of 10−45 ( Reddy et al . , 2014 ) . At this threshold eSNaPD has been used to successfully assign-and-recover gene clusters that encode congeners of multiple natural product families using only the sequence from a single domain amplicon ( Owen et al . , 2013; Chang and Brady , 2014; Kang and Brady , 2014 ) . NRPS/PKS clusters typically have multiple KS or AD domains . Hits to all domains in a cluster were aggregated in our analyses . Data for eSNaPD hits broken down by sample and molecule are included as Supplementary file 6 . AD and KS OTU tables were analyzed for the presence of eSNaPD hits . For each sample the abundance of each eSNaPD hit ( i . e . , a particular molecule ) was calculated as either a percentage of total reads ( Figure 2A , C ) or as the total number of unique OTUs assigned to the molecule that were found in that sample ( Figure 2B , C ) , or as the total number of OTUs mapped to a molecule in each sample . In the read-based hotspot analysis , the number of reads assigned by eSNaPD to a specific gene cluster is expressed as a fraction of total per sample reads: ( reads-to-cluster-of-interest ) /total sample reads ) . In the OTU-based hotspot analysis we calculated the number of unique eSNaPD assigned OTUs found in each sample that map to a specific gene cluster . The full eSNaPD dataset is available in Supplementary file 6 . To compare global biosynthetic diversity of each sample , the AD and KS OTU tables were combined and for each sample they were subsampled ten times to a depth of 5000 reads . The Chao1 diversity metric was calculated for each sample and the average was used to compare the expected biodiversity in different samples at the same sampling depth ( Figure 1C , Supplementary file 7 ) .

Many of the most useful medicinal drugs—including antibiotics and cancer drugs—are derived from bacteria living in the soil that produce these chemicals as part of their natural life cycle . Many of these chemicals have been found by culturing bacteria in the laboratory , but this approach is limited because it only provides access to the chemicals produced by the small fraction of bacteria species that we can culture in this way . Also , many bacteria do not produce as many different chemicals when they are grown under these artificial conditions , instead of their natural environment . This suggests that bacteria living in the environment are likely to provide an additional source of new chemicals that could have medicinal benefits . Here , Charlop-Powers et al . tackle this issue by employing a high-throughput genetic method for assessing the potential of soil-dwelling bacteria to make compounds with biological activity . They extracted DNA directly from soil samples collected from five continents , in part through the efforts of a citizen-science project called ‘Drugs from Dirt’ ( drugsfromdirt . org ) . These samples came from many different environments , including rainforests , deserts , and coastal sediments . After extracting the DNA from the soil samples , Charlop-Powers et al . focused on sequencing the genes that encode enzymes called NRPS and PKS . These enzymes are involved in the production of a range of diverse compounds , including many clinically useful antibiotics . By comparing the sequences of the genes found in the different soils , it was possible to estimate how common the genes were in each sample , and also to compare the collections of genes found in different soil types . This comparison revealed that the DNA sequences of the genes encoding NRPS and PKS vary widely among the soil samples , except for samples that came from similar environments in close proximity to each other . These findings show that populations of soil-dwelling bacteria living in different locations are likely to produce related , but different and largely unexplored , natural compounds that could have the potential to be used in drug therapies or in other industries .

2015

Global biogeographic sampling of bacterial secondary metabolism

The replicative DNA polymerase PolIIIα from Escherichia coli is a uniquely fast and processive enzyme . For its activity it relies on the DNA sliding clamp β , the proofreading exonuclease ε and the C-terminal domain of the clamp loader subunit τ . Due to the dynamic nature of the four-protein complex it has long been refractory to structural characterization . Here we present the 8 Å resolution cryo-electron microscopy structures of DNA-bound and DNA-free states of the PolIII-clamp-exonuclease-τc complex . The structures show how the polymerase is tethered to the DNA through multiple contacts with the clamp and exonuclease . A novel contact between the polymerase and clamp is made in the DNA bound state , facilitated by a large movement of the polymerase tail domain and τc . These structures provide crucial insights into the organization of the catalytic core of the replisome and form an important step towards determining the structure of the complete holoenzyme . In Escherichia coli , DNA replication is highly efficient with speeds of up 600–1000 nucleotides per second ( Mok and Marians , 1987; Mcinerney et al . , 2007 ) , >100 , 000 basepairs ( bp ) synthesized per binding event ( Yao et al . , 2009 ) , and an error rate of ∼1 per million ( Bloom et al . , 1997 ) . Importantly , DNA replication is greatly complicated by the antiparallel orientation of the two DNA strands that need to be replicated simultaneously . To do so , DNA replication is performed by a large multi-protein complex termed the DNA polymerase III holoenzyme that synthesizes the leading strand in a continuous manner , while the lagging strand is synthesized in short fragments of ∼1000 bp . The holoenzyme is composed of 10 subunits ( α , β , ε , θ , δ , δ' , γ , τ , χ , ψ ) , that together with the helicase DnaB and the RNA primase DnaG form the replisome with a combined molecular weight of 1 MDa . The replisome can be divided into three functional subcomplexes that together catalyze a series of events . The helicase DnaB separates the two DNA strands ( Mok and Marians , 1987 ) and transiently associates with the RNA primase DnaG that synthesizes short RNA primers required for DNA synthesis at the lagging strand ( Wu et al . , 1992 ) . The clamp loader subcomplex ( δ , δ' , γ , τ , χ , ψ ) loads the DNA sliding clamp β , the processivity factor for the DNA polymerase , onto the DNA ( Stukenberg et al . , 1991 ) . It furthermore connects the leading and lagging strand polymerases via its τ subunits ( McHenry , 1982; Onrust et al . , 1995 ) . Finally , DNA synthesis is performed by the polymerase subcomplex that contains the DNA polymerase III α ( PolIIIα ) , the DNA sliding clamp β , the proofreading exonuclease ε , and the C-terminal domain of the clamp loader subunit τ . The activity of PolIIIα is poor in isolation ( Maki and Kornberg , 1985 ) and is greatly enhanced by its associated proteins . For error-free DNA synthesis the polymerase relies on the exonuclease ε that removes any misincorporated bases and decreases the error rate of DNA replication by 1–2 orders of magnitude ( Scheuermann et al . , 1983; Lancy et al . , 1989 ) . In addition , the exonuclease strengthens the interactions between the polymerase and clamp as it binds both proteins simultaneously ( Toste Rêgo et al . , 2013; Jergic et al . , 2013 ) . For processivity , PolIIIα binds to the DNA sliding clamp ( β subunit ) ( Stukenberg et al . , 1991 ) . At the leading strand this interaction is stable and results in DNA segments of >100 . 000 bp synthesized per binding event ( Yao et al . , 2009 ) . At the lagging strand in contrast , DNA synthesis is discontinuous , with an averaged length of 1000 bp synthesized per fragment , depending on the frequency of the RNA primase activity ( Wu , Zechner and Marians , 1992 ) . This therefore requires repeated binding and release of the polymerase and clamp . Finally , the C-terminal domain of τ ( τc ) acts as a 'processivity switch' for the polymerase to enable repeated binding and release at the lagging strand ( Leu et al . , 2003; Georgescu et al . , 2009 ) . How this tetrameric complex of PolIIIα-clamp-exonuclease-τc assembles and how it is repeatedly loaded and released during lagging strand synthesis is poorly understood . The structures of the helicase-primase subcomplex ( Bailey et al . , 2007; Wang et al . , 2008 ) and the clamp loader subcomplex ( Jeruzalmi et al . , 2001; Simonetta et al . , 2009 ) have been known for some time . The structure of the PolIIIα-clamp-exonuclease-τc complex on the other hand has remained elusive due to its dynamic nature that forms a significant hurdle for structure determination . To overcome this , we have used a combination of site directed mutagenesis and computational classification of different structural states to determine the cryo-EM structures of the complex in both a DNA-bound and a DNA-free state to 8 Å resolution . The well defined features of the cryo-EM maps enable the unambiguous fitting of the crystal structures of the individual proteins , revealing the unique interactions between the four proteins and DNA . In the DNA-bound complex , the polymerase is tethered to the DNA through multiple contacts with the clamp . The interaction with the clamp is further stabilized by the exonuclease that is wedged between the two proteins and forms a second , indirect interaction between polymerase and clamp . Strikingly , a large conformational change in the polymerase switches its tail domain from interacting with the clamp in the DNA-bound structure , to more than 30 Å away from the clamp in the DNA-free structure . Finally , the processivity switch τc binds the tail of the polymerase and appears to sequester the polymerase tail away from the clamp in the DNA-free structure . Hence , our structures provide crucial insights into the regulation of the replicative DNA polymerase PolIIIα by its associated proteins clamp , exonuclease and τc . They furthermore form a crucial step towards determining the structure of the complete DNA polymerase III holoenzyme . The interaction between PolIIIα and the clamp is weak , in the order of 1 μM ( Toste Rêgo et al . , 2013 ) , and is not sufficient to maintain an intact complex at the low concentrations used for cryo-EM . Therefore , to stabilize the complex we altered the sequences of the clamp binding motifs of PolIIIα and the exonuclease to increase the affinity for the clamp . For this we used sequences derived from the translesion DNA polymerase UmuC and the DNA replication initiation factor Hda that out of a panel of 15 peptide sequences were the most potent inhibitors of the interaction between the polymerase and clamp ( Wijffels et al . , 2004 ) ( see Materials and methods for more details ) . The obtained complex is >100 fold more stable than the wild-type complex ( Figure 1—figure supplement 1A ) This stabilized complex of PolIIIα , clamp and exonuclease was used together with τ500 ( the polymerase-binding domain of τ: residues 500–643 ) and a 25 base pair ( bp ) DNA substrate to prepare samples for cryo-EM ( Figure 1—figure supplement 2A , B ) . Three structurally distinct groups of particles could be identified from a single data set ( 63 , 215 particles ) . Two of these represent the PolIIIα-clamp-exonuclease-τ500 with and without DNA bound ( Figure 1 , Videos 1 and 2 ) . The third class contains DNA too , but in this complex the tail domain of the polymerase and τ500 are not visible due to structural heterogeneity . The DNA-bound ( 5663 particles ) and DNA-free ( 16 , 970 particles ) structures were refined to 8 . 0 and 8 . 3 Å resolution , respectively ( see Figure 1—figure supplement 2 for details ) . The remaining particles ( 40 , 582 ) were classified into the third class in which the tail domain is not visible . Due to the larger number of particles , this structure was refined to 7 . 3 Å resolution . As this structure is otherwise identical to the complete DNA-bound complex , it will not be discussed further . 10 . 7554/eLife . 11134 . 003Figure 1 . Cryo-EM structures of the E . coli PolIIIα-clamp-exonuclease-τ500 complex . ( A ) Surface representation of the three structures , shown at 5 σ . Left to right: DNA-free , DNA-bound , and DNA-bound without tail . Colors indicate the position of the different proteins ( B ) Individual structures of PolIIIα , clamp , exonuclease , and τ500 fitted into the cryo-EM map ( shown in grey at 5 σ ) ( C ) Detailed views of the cryo-EM map ( shown in grey mesh at 6 σ ) . Left panel: exit channel of the clamp in the DNA-free structure showing the ‘DNA-free’ map . Middle panel: bottom view of the polymerase showing the ‘DNA-free’ map . Right panel: detail of the DNA showing the ‘DNA-bound , no tail’ map . See also Videos 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 11134 . 00310 . 7554/eLife . 11134 . 004Figure 1—figure supplement 1 . Characterization of improved clamp binding mutants . ( A ) Gel filtration analysis of the wild-type PolIIIα-clamp-exonuclease complex ( top panel ) and the PolIIIαQLDLF-clamp-exonucleaseQLSLPL complex ( lower panel ) . The wild-type complex dissociates at lower protein concentrations , while the stabilized complex remains intact even at 0 . 1 μM . DOI: http://dx . doi . org/10 . 7554/eLife . 11134 . 00410 . 7554/eLife . 11134 . 005Figure 1—figure supplement 2 . Microscopy data analysis and validation . ( A ) Typical micrograph of the PolIIIα-clamp-exonuclease-τ500-DNA complex . ( B ) 2D class averages derived from the final 63 , 215 particle dataset ( C ) Fourier shell correlation for the DNA-free and DNA-bound models . In solid lines , the correlation between two independently refined halves of the data is indicated ( gold-standard FSC ) . Estimated resolution at a correlation of 0 . 143 is 8 . 3 Å and 8 . 0 Å for the DNA-free and DNA-bound complex , respectively . In dashed lines , the correlation between the rigid-body docked models and their respective maps is indicated . ( D ) 3D model reconstruction . An initial model was obtained using Eman2 and subsequently classified into six 3D classes . Two of the 3D classes were merged into the ‘DNA-free’ map ( 16 , 970 particles ) and one of these ( 5663 particles ) was used for the ‘DNA-bound’ map . The remaining three classes were merged into the ‘DNA-bound , no tail’ map ( 40 , 582 particles ) and further refined in Relion , resulting in three structurally distinct models . ( E ) Orientational distribution for particles of the DNA-free complex . The circle represents a flattened sphere plotted using Lambert equal area projection with the pole at the center and the equator at the outer rim of the circle . The radius indicates the tilt angle and the azimuth indicates the rotation or direction of the tilt . ( F ) Same for the DNA-bound complex ( G ) Tilt pair validation using 267 particle pairs that were selected from 20 image pairs collected at 0 and 20° tilt angle of the sample stage . The angular difference between the same particle collected from the two images is displayed . The black cross indicates the expected angular difference between pairs . DOI: http://dx . doi . org/10 . 7554/eLife . 11134 . 00510 . 7554/eLife . 11134 . 006Figure 1—figure supplement 3 . Rigid body movements in PolIIIα . ( A ) Domain definitions used for the rigid body fitting of the PolIIIα structure into the cryo-EM maps . Domain boundaries are: PHP ( residues 1–280 ) , palm-fingers ( residues 281–432 + 510–810 ) , thumb ( residues 433–509 ) , tip-of-fingers ( residues 811–928 ) and C-terminal tail ( residues 929–1160 ) . ( B and C ) Comparison of crystal structure of E . coli PolIIIα ( shown in grey ) and PolIIIα as fitted into the cryo-EM maps ( the tail of the polymerase is omitted for clarity ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11134 . 006 The cryo-EM maps enable the unambiguous fitting of the high-resolution structures of the individual subunitsinto the cryo-EM maps ( Figure 1B , C ) . No conformational changes were required for the fitting of the clamp , exonuclease or τ500 , while the polymerase was divided into five domains that were independently fitted into density as rigid bodies ( see Figure 1—figure supplement 3 ) . None of the loops were modified , with the exception of the clamp binding motifs of the polymerase and exonuclease that were modeled after existing crystal structures of clamp-bound peptides from Pol II and Pol IV ( Georgescu et al . , 2008a; Bunting et al . , 2003 ) . B-form DNA was used for the DNA substrate , except for the last four base pairs that deviate from B-form DNA and were modeled after the DNA substrate from the Thermus aquaticus ( Taq ) PolIIIα crystal structure ( Wing et al . , 2008 ) . We describe the DNA-free complex first ( Figure 2 ) . The overall conformation of PolIIIα resembles that of the X-ray structure of E . coli and Taq PolIIIα ( Lamers et al . , 2006; Bailey et al . , 2006 ) and reveals only a ∼15° rotation of the fingers domain between the two structures ( Figure 1—figure supplement 3 ) . PolIIIα interacts with the clamp through the internal clamp binding motif ( residues 920–924 ) ( Dohrmann and McHenry , 2005; Toste Rêgo et al . , 2013 ) that binds in the canonical binding pocket of the clamp ( Figure 2B ) . Immediately after the clamp binding motif the density for the polymerase disappears , and resumes ∼10 residues later , just before the oligonucleotide/oligosaccharide binding ( OB ) domain , indicating that this region of the polymerase is flexible ( Figure 2A , left and middle panel ) . 10 . 7554/eLife . 11134 . 009Figure 2 . Multiple contacts between the subunits hold the complex together . ( A ) Three different views of the DNA-free complex of PolIIIα-clamp-exonuclease-τ500 showing extensive contacts between the polymerase and other subunits . Missing loops in PolIIIα ( residues 927–936 ) and exonuclease ( residues 190–207 ) are shown in dots . Dashed boxes indicate views shown in panels B-D . ( B ) Modified clamp binding motif of PolIIIα ( QLDLF: shown in sticks ) modeled into the binding pocket of the clamp . ( C ) Modified clamp binding motif of the exonuclease ( QLSLPL: shown in sticks ) modeled into the second binding pocket of the dimeric clamp . ( D ) τ500 simultaneously binds the fingers and tail domain of the polymerase . The C-terminal residues of τ500 ( residues 622–643: not modeled ) bind to the tail of the polymerase , while the globular domain of τ500 binds to the polymerase fingers domain ( see Figure 2—figure supplement 2 for more details ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11134 . 00910 . 7554/eLife . 11134 . 010Figure 2—figure supplement 1 . Previously determined cross-links fit accurately with the cryo-EM model . ( A ) Model of the polymerase-clamp-exonuclease complex based on chemical cross-links reported in ( Toste Rêgo et al . , 2013 ) . Magenta dashed lines: polymerase-clamp cross-links . Cyan dashed lines: polymerase-exonuclease cross-links . Black dashed lines: clamp-exonuclease cross-links . ( B ) Same cross-links mapped onto the DNA-free cryo-EM model . DOI: http://dx . doi . org/10 . 7554/eLife . 11134 . 01010 . 7554/eLife . 11134 . 011Figure 2—figure supplement 2 . Details of the interactions between τ500 and the PolIIIα fingers domain . ( A ) Three orthogonal views of the fit of τ500 into the cryo-EM density . Dashed box in left panel indicates view shown in panel B . ( B ) Detailed view of the τ500 - PolIIIα fingers domain interaction . Contact regions at the interface are indicated with thick coil in red ( τ500: residues 530–535 and residues 562–566 ) and blue ( PolIIIα: residues 657–667 ) DOI: http://dx . doi . org/10 . 7554/eLife . 11134 . 01110 . 7554/eLife . 11134 . 012Figure 2—figure supplement 3 . Comparison of τ binding in E . coli and Taq PolIIIα . ( A , B ) DNA-free and DNA-bound E . coli PolIIIα-τ500 . The clamp and exonuclease are omitted for clarity . ( C ) Taq PolIIIα-τc ( Liu et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11134 . 012 On the other side of the complex , across the opening of the clamp , the PHP domain of the polymerase comes close to , but makes no contacts with the clamp ( Figure 2A , left panel ) . Instead , the exonuclease is wedged between the clamp and the thumb domain of PolIIIα ( Figure 2A , right panel ) . The catalytic domain of the exonuclease is in direct contact with the polymerase thumb domain whereas the contact with the clamp is mediated via a canonical clamp binding motif that is located immediately downstream of the catalytic domain ( Toste Rêgo et al . , 2013; Jergic et al . , 2013 ) . This clamp binding motif is bound to the pocket of the clamp in a manner similar to the polymerase in the other half of the clamp ( cf . Figure 2B , C ) and hence both pockets of the dimeric clamp are occupied in the ternary complex . Downstream of the clamp binding motif the tail of the exonuclease follows the contours of the polymerase PHP domain , where it is disordered for a stretch of ∼15 residues that were shown to be mobile by NMR ( Ozawa et al . , 2013 ) . Finally , the C-terminal helix of the exonuclease that mediates most of the binding to the polymerase ( Ozawa et al . , 2013 ) packs tightly against the PHP domain of PolIIIα ( Figure 2A , left and right panel ) , similar to the crystal structure of the PolIIIα-PHP domain and C-terminal helix of the exonuclease ( Ozawa et al . , 2013 ) . Hence , in the ternary complex the exonuclease simultaneously binds the polymerase and clamp . By doing so it strengthens the association between the two proteins ( Toste Rêgo et al . , 2013 ) , which is required for processive DNA synthesis ( Jergic et al . , 2013 ) . Previously , we built an approximate model for the polymerase , clamp and exonuclease complex , using distance restraints provided by chemical cross-linking coupled to mass-spectrometry ( Toste Rêgo et al . , 2013 ) . When we map the same cross-links onto the cryo-EM model we find an improved fit of the cross-links with the model , due to the conformational changes in the complex as well as the detailed information about the interactions between the proteins that could not be modeled based on the cross-links alone ( Figure 2—figure supplement 1 ) . The NMR structure of residues 500–621 of τ ( Su et al . , 2007 ) can be fitted into the density between the tail and fingers domain of the polymerase ( Figure 2D and Figure 2—figure supplement 2A ) . The C-terminal end of τ is in contact with the tail of the polymerase . Unfortunately , the last 23 residues of τ that bind the polymerase ( Jergic et al . , 2007 ) are not part of the NMR structure and are therefore not present in our model . A second contact is found between the globular domain of τ500 and the fingers domain of the polymerase . This contact is mediated by residues 530–535 and 562–566 of τ500 and residues 657–667 of PolIIIα ( Figure 2—figure supplement 2B ) . The position of τ500 in this complex is different from the position of the C-terminal domain of τ in complex with Taq PolIIIα , where it is located at the opposite side of the polymerase tail ( Figure 2—figure supplement 3 ) ( Liu et al . , 2013 ) . It must be noted though that the C-terminal domain of τ from E . coli and Taq share no sequence or structural homology and therefore engage with the polymerase in different ways . In the DNA-bound complex ( Figure 3 ) , the entire length of the 25 base pair duplex is in contact with protein ( Figure 3A ) . The position of the DNA is similar to that of the DNA in the crystal structure of Taq PolIIIα and Geobacillus kaustophilus PolC ( Wing et al . , 2008; Evans et al . , 2008 ) ( Figure 3—figure supplement 1 ) . No density is observed for the 4 nucleotide ( nt ) single stranded overhang on the template strand indicating that this part of the DNA is flexible . In the complex , all contacts to the DNA are mediated by the thumb , palm and fingers domains of the polymerase and the inner surface of the clamp . No contacts to the DNA are made by the polymerase OB domain , the exonuclease , or τ500 . The most extensive DNA contacts occur at the primer 3’ end in polymerase active site where the thumb , palm and fingers domain of the polymerase contact the first 9 base pairs of the DNA duplex . It is also here that the only non-backbone contact is made by a loop of the thumb ( residues 464–470 ) , which is inserted into the major groove of the DNA ( Figure 3B ) . 10 . 7554/eLife . 11134 . 013Figure 3 . The DNA has extensive contacts with PolIIIα and clamp . ( A ) Overview of the DNA-bound complex . The N-termini of the two helices that point at the DNA backbone are colored in blue . Potential DNA interacting side chains are shown in sticks . The tail of PolIIIα , the exonuclease and τ500 are omitted for clarity . Arrow indicates viewpoint in panel B ( B ) Polymerase active site , with the DNA held between thumb , palm and fingers domain . Polymerase active site residues are indicated with magenta spheres . Arrow indicates viewpoint in panel C ( C ) DNA interactions downstream of the active site . The OB domain is positioned on top of the DNA but does not make any contacts with it . ( D ) DNA exit channel in the clamp with positively charged residues within 10 Å of the DNA indicated in magenta sticks . Note that the positions of the side chains have not been refined and should be seen as approximate positions . ( E ) In the DNA-bound complex , the clamp is at a ∼80° angle from the DNA . A dashed line indicates the position of the clamp alone bound to a DNA substrate . ( Georgescu et al . , 2008a ) . The other subunits ( PolIIIα , exonuclease , τ500 ) are shown in light grey for clarity . DOI: http://dx . doi . org/10 . 7554/eLife . 11134 . 01310 . 7554/eLife . 11134 . 014Figure 3—figure supplement 1 . Comparison of DNA binding by C family DNA polymerases . ( A ) E . coli PolIIIα , ( B ) T . aquaticus PolIIIα ( Wing et al . , 2008 ) , ( C ) G . kaustophilus PolC ( Evans et al . , 2008 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11134 . 01410 . 7554/eLife . 11134 . 015Figure 3—figure supplement 2 . Pol IIIα has more extensive DNA interactions than other DNA polymerases . ( A ) Left panel: Electro-mobility shift assay with the E . coli DNA polymerases Pol I ( Klenow fragment ) , Pol II , Pol IIIα , and Pol IV . At 2 . 5 μM Pol I , Pol II , and Pol IV retain DNA , whereas Pol IIIα does not . The more intense bands for Pol I and Pol II are caused by protein induced fluoresence enhancement ( PIFE ) ( Hwang et al . , 2011 ) . Right panel: SDS-page analysis of the same samples used for the electro-mobility shift assay ( proteins stained with coomassie blue ) . Molecular weights: Pol I ( Klenow fragment ) 70 kDa , Pol II 90 kDa , Pol IIIα 130 kDa , Pol IV 40 kDa . ( B ) Structural comparison of Pol I ( PDB code: 1QTM [Li et al . , 1999] ) , Pol II ( PDB code: 3K57 [Wang and Yang , 2009] ) , PolIIIα ( this work ) , and Pol IV ( PDB code: 4IRD [Sharma et al . , 2013] ) . Polymerases were aligned on the 3’ end of the primer , indicated by the solid line . For Pol I , Pol II , and Pol IV , the clamp was modeled based on predicted clamp interacting motifs in the respective polymerases . The distance measured in base pairs between the 3’ end of the primer and the opening of the clamp is indicated on top of the structures , together with the rate of DNA synthesis of each polymerase . ( C ) Detailed view of the interaction of the polymerase thumb domains and the DNA . All polymerase thumb domains interact with the backbone of the minor groove . Only Pol IIIα inserts a loop into the major groove of the DNA . DOI: http://dx . doi . org/10 . 7554/eLife . 11134 . 015 Away from the active site , the tip of the fingers domain ( i . e . little finger [Lamers et al . , 2006] or β binding domain [Bailey et al . , 2006] ) , makes additional contacts to the DNA . Here , several positively charged residues ( K831 , K872 , R876 , R877 , K881 ) as well as the positive charge of the helix dipole of two helices ( residues 842–856 and 875–886 ) are pointed towards the DNA backbone ( Figure 3A ) . At this position , the OB domain is in close proximity of the DNA but makes no contacts with it ( Figure 3C ) . Instead , the OB domain forms a bridge between the PolIIIα fingers domain , thumb domain , and the exonuclease . Furthermore , while the isolated OB domain has been shown to bind to ssDNA ( Georgescu et al . , 2009 ) , in this complex the OB domain is ∼40 Å away from the ssDNA template . The DNA furthermore interacts with the clamp that surrounds the DNA like a nut around a bolt ( Figure 3D ) . Several non-specific contacts are made to the backbone of the DNA providing an electrostatic cushion for the DNA to pass through as it leaves the complex . Compared to the crystal structure of the isolated clamp bound to DNA ( Georgescu et al . , 2008a ) the clamp is rotated by ∼20° , resulting in an almost perpendicular orientation ( ∼80° ) with respect to the DNA . In the 19 Å , negative stain EM structure of Pyrococcus furiosus PolB , the only other known structure of a DNA polymerase in complex with clamp and DNA , the DNA runs straight through the clamp as well ( Mayanagi et al . , 2011 ) . PolIIIα is an extremely fast DNA polymerase with DNA synthesis speeds of up to 600–1000 nt/s ( Mok and Marians , 1987; Mcinerney et al . , 2007 ) . In contrast , the E . coli DNA polymerases Pol I , Pol II , and Pol IV have synthesis speeds of 15 , 10 , and 1 nt/s , respectively ( Schwartz and Quake , 2009; Indiani et al . , 2009 ) . Furthermore , in isolation PolIIIα has a surprisingly low affinity for DNA when compared to the other E . coli DNA polymerase ( Figure 3—figure supplement 2A ) . Because of its weak binding to DNA , PolIIIα must therefore have developed a different way to keep itself correctly positioned on the DNA during rapid DNA synthesis . To this effect , PolIIIα may have evolved its uniquely long fingers domain that is more than twice as long as the other E . coli DNA polymerases ( Figure 3—figure supplement 2B ) . These polymerases have considerably smaller fingers domains , use shorter regions of DNA contacts , and have shorter predicted distances between the polymerase active site and the clamp ( Figure 3—figure supplement 2B ) . In PolIIIα , the number of base pairs between the 3’ end of the primer and the opening of the clamp is 22 , while the predicted number of base pairs for Pol I , Pol II , and Pol IV is ∼15 . This unusually long DNA-protein contact appears to be well suited to accurately position the DNA without requiring tight binding that could slow down the translocation of the DNA . At the same time , the sequence-independent backbone contacts and the perfectly straight B-form DNA may facilitate the rapid exit of the DNA from the active site . The active site itself is wrapped tightly around the DNA where PolIIIα is the only polymerase that inserts a loop of the thumb domain into the major groove of the DNA , while the thumb domains of Pol I , Pol II , and Pol IV only have backbone contacts with the DNA ( Figure 3—figure supplement 2C ) . Hence , it seems plausible that this combination of unique contacts with the DNA may have evolved to support the high speeds of DNA synthesis by PolIIIα without compromising accuracy . To enable the many contacts with the DNA , the complex undergoes extensive conformational changes from the DNA-free to the DNA-bound state . Most prominent is a ∼35° rotation of the polymerase tail and τ500 , which move from a position over the polymerase active site to a position adjacent to the sliding clamp ( Figures 1 , 4 and Video 3 ) . The simultaneous movement of the polymerase tail and τc results in a 70 Å displacement of the globular domain of τ500 . The tail of PolIIIα consists of the OB domain ( residues 960–1071 ) and the C-terminal τ-binding domain ( residues 1079–1160 ) . Together with τ500 they form a rigid structure that shows few changes between the DNA free and DNA bound structure , indicating that the interaction between τ500 and the tail of PolIIIα must be stable ( Figure 4C ) . As a result of the repositioning of the polymerase tail , the contact between τ500 and the fingers domain of the polymerase is broken and a new contact between the OB domain and the clamp is forged ( Figure 4B and Video 3 ) . The OB domain makes two new contacts with the clamp via a short helix ( 1035–1043 ) and a long protruding loop ( 1003–1013 ) ( Figure 4D , E ) . These motifs contact the clamp at loops 24–28 and 275–278 , respectively . Hence in the DNA-bound complex , the polymerase has three points of contact to the clamp: one via the canonical clamp binding motif ( residues 920–925: Figure 2B ) ; one indirectly via the exonuclease ( Figure 2C ) ; and one contact via the OB domain ( Figure 4D , E ) . Previously , it has been shown that a triple mutation in OB domain result in reduced DNA synthesis ( Georgescu et al . , 2009 ) which was attributed to the loss of ssDNA binding by the OB domain . However , in our structure the OB domain is far away ( ∼40 Å ) from the ssDNA overhang of the template strand . Instead , the mutations ( R1004S , K1009S , R1010S ) are located at the interface between the OB domain and the clamp ( Figure 4E ) and therefore could weaken the interaction between the polymerase and clamp , providing an alternative explanation for the reduced DNA synthesis . 10 . 7554/eLife . 11134 . 016Figure 4 . DNA binding induces large conformational changes in the polymerase . ( A ) Clamp binding by PolIIIα in the DNA-free complex . Arrows indicate movement of the PolIIIα tail ( see also Video 3 ) . ( B ) Clamp binding by PolIIIα in the DNA-bound complex . Dashed boxes indicate views shown in panel D and E ( C ) Comparison of the PolIIIα-tail - τ500 interaction in the DNA-free ( in grey ) and DNA-bound structure . ( D and E ) Detailed view of the clamp - PolIIIα OB domain interaction . Interacting regions at the interface are indicated in thick coil in magenta ( clamp: residues 24–24 and residues 275–278 ) and red ( PolIIIα-OB domain: residues 1035–1043 and residues 1003–1013 ) . Residues mutated in Georgescu et al ( Georgescu et al . , 2009 ) are shown in sticks and labeled with outlined boxes . Note that the positions of the side chains have not been refined and should be seen as approximate positions . DOI: http://dx . doi . org/10 . 7554/eLife . 11134 . 016 The E . coli replisome consists of 12 different proteins that can be divided into three subcomplexes: the helicase-primase complex , the clamp loader complex , and the PolIIIα-clamp-exonuclease-τc complex . The structures of two of the three subcomplexes have been determined previously: the helicase-primase complex ( Bailey et al . , 2007; Wang et al . , 2008 ) , and the clamp loader complex ( Jeruzalmi et al . , 2001; Simonetta et al . , 2009 ) . The structure of the PolIIIα-clamp-exonuclease-τ500 complex on the other hand has remained elusive due to its dynamic nature . The cryo-EM structures of the PolIIIα-clamp-exonuclease-τ500 complex presented in this work finally reveal the nature of the interactions in the ternary complex and are a crucial step forward towards determining the structure of the complete bacterial replisome . Our cryo-EM structures furthermore provide a crucial insight into the structural organization of the replicative DNA polymerase and its associated proteins clamp , exonuclease and τ500 . They show how the clamp and exonuclease tether the polymerase to the DNA through multiple contacts . Importantly , they also reveal a large conformational change where the tail of the polymerase moves from interacting with the clamp in the DNA-bound state , to a position 35 Å away from the clamp in the DNA-free state . What may be the role for such a conformational change ? On the lagging strand , the polymerase repositions to a newly primed site every ∼1000 bp . To do so , the polymerase needs to release both clamp and DNA . We propose that the switch-like movement of the polymerase tail may play a part in the release and consequent repositioning of the polymerase at the end of the Okazaki fragment . A hypothetical model describing how this could work is presented in Figure 5 . During DNA synthesis , the tail of the polymerase is bound to the clamp , stabilizing the interaction between polymerase and clamp ( marked with ‘1’ in Figure 5 ) . This confirmation may be further stabilized by the presence of a DNA binding region immediately upstream of τ500 ( marked with ‘2’ ) ( Jergic et al . , 2007 ) . Upon encounter of a release signal , τ500 rebind to the polymerase fingers domain ( marked with ‘3’ ) thus sequestering the polymerase tail away from the clamp ( marked with ‘4’ ) and initiating the dissociation of the polymerase from clamp and DNA . What could serve as the release trigger ? Two non-exclusive models have been proposed ( Li and Marians , 2000 ) . In the ‘collision' model , the encounter with the dsDNA of the previous Okazaki fragment induces the release of the polymerase . In support of this model , it has been shown that a decreasing gap size between the 3' terminus of the lagging strand and the 5' end of the previously synthesized Okazaki fragment promotes release of the polymerase ( Leu et al . , 2003; Georgescu et al . , 2009; Dohrmann et al . , 2011 ) . Two possible sensors for the decreasing gap on the lagging strand have been suggested . The ssDNA binding properties of the OB domain in the tail of the polymerase has been proposed to play a role in the sensing of the ssDNA vs dsDNA ( Georgescu et al . , 2009 ) . Yet our cryo-EM models show that the OB domain is ∼40 Å away from the ssDNA and that it is involved in binding to the sliding clamp . Alternatively , it has been proposed that the C-terminal fragment of τ may act as the sensor as it is required to release the polymerase from a decreasing gap size ( Leu et al . , 2003 ) . Indeed , it was found that the region in τ , immediately upstream of τ500 , has DNA binding affinity ( Jergic et al . , 2007 ) . 10 . 7554/eLife . 11134 . 018Figure 5 . Schematic representation for a possible role of the conformational changes in the polymerase . During processive DNA synthesis , the tail of the polymerase is attached to the clamp ( indicated with ‘1’ ) and pulls τ500 away from the polymerase fingers domain . This conformation may be further stabilized by the presence of a DNA binding region immediately upstream of τ500 ( indicated with ‘2’; see text for more details ) . Upon encounter of a release trigger , the contact between τ500 and the polymerase fingers domain is restored ( indicated with ‘3’ ) , and the contact between the clamp and polymerase tail is broken ( indicated with ‘4’ ) . The release trigger may either come from a collision with the previous Okazaki fragment ( indicated with ‘Collision’ ) , or a signal from other replisome components via the flexible linker of τ ( indicated with ‘Signaling’ ) . Once the polymerase-tail clamp contact has been broken , the two remaining contacts between the clamp and polymerase-exonuclease are not enough to keep the polymerase bound to the clamp . The polymerase is released from clamp and DNA and can be repositioned to a newly primed site to reinitiate DNA synthesis . DOI: http://dx . doi . org/10 . 7554/eLife . 11134 . 018 Contesting the collision model is the observation that the release of the polymerase by a decreasing gap size is too slow ( t1/2 = 110 s ) for the frequency at which the lagging strand polymerase is re-positioned ( every 1–2 s ) ( Dohrmann et al . , 2011 ) , suggesting that additional or alternative release factors are required . The alternative ‘signaling’ model therefore proposes that the trigger comes from one of the other components of the replisome such as the RNA primase DnaG , based on the observation that the increased primase concentration results in shorter lagging strand fragments ( Wu , Zechner and Marians , 1992; Li and Marians , 2000 ) . However , it was recently shown that the presence of a primer alone is sufficient to induce release at the lagging strand and that activity of the primase is not required ( Yuan and McHenry , 2014 ) . How the presence of the RNA primer is signaled to the polymerase remains unclear . Yet , the fact that the τ protein is both part of the clamp loader complex that positions clamps onto the primer and simultaneously binds the polymerase makes this a suitable conveyor of the signal . While our structures do not discriminate between the type of release trigger for the lagging strand polymerase , they do now provide the means to test the precise workings of the molecular switch that enables the release of the polymerase . All chemicals and oligonucleotides were purchased from Sigma-Aldrich ( Gillingham , United Kingdom ) and chromatography columns from GE healthcare ( Little Chalfont , United Kingdom ) . To increase binding to the clamp , amino acid residues 920–924 of E . coli PolIIIα were changed by site directed mutagenesis from QADMF to QLDLF , while in the exonuclease residues 182–187 were changed from QTSMAF to QLSLPL , based on sequences described in ( Wijffels et al . , 2004 ) . All proteins were expressed in E . coli ( DE3 ) BL21 . PolIIIα , clamp and exonuclease were expressed and purified as described before ( Toste Rêgo et al . , 2013 ) . τ500 was purified by Histrap HP column , Resource S column , and a Superdex 75 gel filtration column . His-tags were removed by proteolytic cleavage with human rhinovirus 3C protease . Proteins were flash frozen in liquid nitrogen and stored at -80° . Proteins were analyzed by gel filtration using a 2 . 4 mL Superdex 200 Increase column ( GE healthcare ) in 25 mM Hepes pH 7 . 5 , 150 mM NaCl , and 2 mM DTT . PolIIIα-clamp-exonuclease complex was assembled at 10 , 1 , and 0 . 1 μM and 50 μL injected onto the column . A DNA substrate identical to the substrate used for the cryo-EM samples was used , with the exception of a 6-carboxyfluorescein ( 6-FAM ) at the 5' end of the primer strand and a phosphorothioate link at the 3' terminal bond to prevent exonuclease digestion . 5 nM DNA was incubated with 2 . 5 μM polymerase ( E . coli Pol I ( Klenow fragment ) , Pol II , Pol IIIα , or Pol IV ) for 10 min at room temperature . Reaction mixtures contained 20 mM Tris pH 7 . 5 , 4% glycerol , 5 mM DTT , 40 μg/ml BSA , and 40 mM Potassium Glutamate . Half of the sample was separated on a native 6% acrylamide gel and imaged on a Typhoon laser scanner ( GE Healthcare ) . The remaining half of the sample was analyzed on a denaturing 4–12% SDS acrylamide gel and stained with Coomassie blue . The PolIIIα-clamp-exonuclease-τ500 protein complex was assembled from the individual components to a final concentration of 15 μM and purified on a 2 . 4 mL Superdex 200 Increase gel filtration column ( GE Healthcare ) in 25 mM Hepes pH 7 . 5 , 50 mM Potassium Glutamate , 3 mM Magnesium Acetate , and 2 mM DTT . The peak fraction ( ∼4 μM ) was retrieved and incubated for 5 min with 20 μM of a 25 bp DNA substrate with a 4 nt overhang ( template: 5′-TCAGGAGTCCTTCGTCCTAGTACTACTCC-3′ , primer: 5′-GGAGTAGTACTAGGACGAAGGACTC-3′ ) for 5 min at room temperature . Subsequently , 0 . 1 volume of 0 . 05% ( V/V ) Tween 20 was added and incubated for another 5 min before the samples were pipetted onto glow-discharged holey carbon cryo-EM grids ( Quantifoil Cu R1 . 2/1 . 3 ) , and frozen in liquid ethane using a Vitrobot ( FEI , Hillsboro , OR ) . All data was collected using a Titan Krios electron microscope ( FEI ) operated at 300 kV equipped with a K2 summit direct electron detector ( Gatan , Pleasaston , CA ) . Although this detector was mounted after a Gatan Imaging Filter ( GIF ) , the filter was not used to remove any inelastic scattering . Images were collected in single-electron counting mode at a calibrated magnification of 28 . 571x ( 1 . 76 Å/pixel ) , using a flux of 2 e/Å2/sec and a total dose of 40 e/Å2 over a total of 20 frames . Frames were aligned and averaged using whole-image movement correction using MOTIONCORR ( Li et al . , 2013 ) . Contrast transfer function parameters were calculated using CTFFIND3 ( Mindell and Grigorieff , 2003 ) . All subsequent particle picking and data processing was performed using Relion-1 . 3 ( Scheres , 2012 ) , with the exception of the generation of the initial model , which was done using Eman2 ( Tang et al . , 2007 ) . A total of 1350 micrographs were recorded from which >550 , 000 particles were picked automatically in Relion . After 2D classification , a large number of spurious particles as well as particles that show free polymerase or free clamp were removed , yielding a dataset of ∼90 , 000 particles . After 3D classification a another ∼27 , 000 were removed to yield a final dataset of 63 , 215 particles . From these , six 3D classes were calculated that were subsequently merged into the final three 3D classes of 'DNA-free' ( 16 , 970 particles ) , 'DNA-bound' ( 5663 particles ) and 'DNA-bound , no tail' ( 40 , 582 particles ) . Particle-based movement correction and per-frame B-factor weighting to account for radiation damage and unresolved particle movement was performed in the later stages of refinement using the particle polishing option in Relion ( Scheres , 2014 ) . Reported resolutions are based on the gold-standard FSC-0 . 143 criterion ( Scheres and Chen , 2012 ) and FSC-curves were corrected for the convolution effects of a soft mask using high-resolution noise-substitution ( Chen et al . , 2013 ) . All density maps were sharpened by applying a negative B-factor that was estimated using automated procedures ( Rosenthal and Henderson , 2003 ) . We believe that the resolution of these reconstructions is limited by both the relatively small size of the complex ( 250 kDa ) , which hampers accurate alignment and classification , and the inherent flexibility of this four-protein and DNA complex . Still , the maps are of excellent quality , with individual helices , β-sheets , and loops clearly visible in the map ( Figure 1C ) . Individual crystal or NMR structures were manually placed into the cryo-EM map in PyMOL ( Schrödinger , LLC 2010 ) and subsequently rigid-body fitted into the density using Coot ( Emsley et al . , 2010 ) . PDB codes of the fitted structures are: PolIIIα: 2HNH ( Lamers et al . , 2006 ) , clamp: 2POL ( Kong et al . , 1992 ) , exonuclease: 1J54 ( Hamdan , et al . , 2002 ) , τ500: 2AYA ( Su et al . , 2007 ) . The C-terminal tail of Eco PolIIIα that is lacking in the crystal structure ( 2HNH ) was modeled as described in ( Toste Rêgo et al . , 2013 ) . The PolIIIα structure was divided into five domains that were further fitted independently into density as rigid bodies ( see Figure 1—figure supplement 3B , C ) . These domains were: PHP ( residues 1–280 ) , palm-fingers ( residues 281–432 + 510–810 ) , thumb ( residues 433–509 ) , tip-of-fingers ( residues 811–928 ) and C-terminal tail ( residues 929–1160 ) . Clamp binding motifs of PolIIIα and exonuclease were manually built into the clamp in Coot guided by the crystal structures of clamp-bound peptides from Pol II and Pol IV , ( Bunting et al . , 2003; Georgescu et al . , 2008b; Jeruzalmi et al . , 2001 ) . The DNA substrate was generated with Coot , and the last four base pairs of the DNA were adjusted guided by the DNA from Taq Pol IIIα ( Wing et al . , 2008 ) . The following crystal structures of C family DNA polymerases were used to compare DNA binding and τ binding . DNA bound Taq PolIIIα ( PDB code: 3E0D [Wing et al . , 2008] ) , τ bound Taq PolIIIα ( PDB code: 4IQJ [Liu et al . , 2013] ) , DNA bound G . kaustophilus PolC ( PDB code: 3F2B [Evans et al . , 2008] ) . Crystal structures of bacterial DNA polymerases in complex with DNA were used to compare the distance between the polymerase active site and the opening to the clamp . The following structures were used: T . aquaticus DNA Pol I ( PDB code: 1QTM [Li et al . , 1999] ) , E . coli Pol II ( PDB code: 3K57 [Wang and Yang , 2009] ) , E . coli PolIIIα ( this work ) , and E . coli Pol IV ( PDB code: 4IRD [Sharma et al . , 2013] ) . For the structures of Pol I , Pol II , and Pol IV , the sliding clamp ( PDB code: 2POL [Kong et al . , 1992] ) was manually placed close to the clamp binding sequences in the different polymerases , taking care not to cause any clashes with other parts of the polymerase .

DNA replication is complicated because the two strands that form its “double helix” structure run in opposite directions and need to be replicated at the same time . One of the new strands , the leading strand , is built continuously . While the other strand , called the lagging strand , is made in stretches that are about 1000-times shorter and run in the opposite direction from the leading strand . This means that the enzyme that builds the new strands of DNA ( called DNA polymerase ) must be repeatedly released and repositioned when it builds the lagging strand , however it is not fully understood how this achieved . Fernandez-Leiro , Conrad et al . have now used a technique called cryo-electron microscopy to reveal the three-dimensional structure of a DNA polymerase from a bacterium called Escherichia coli complete with other associated factors and a DNA molecule . These factors include: the “sliding clamp” that allows the polymerase to slide along the DNA; the “proofreading exonuclease” that removes mistakes in the newly built DNA strand , and the “processivity switch Tau” that is needed for the repeated release and repositioning of the polymerase at the lagging strand . These structures show how the polymerase is bound to the DNA by multiple interactions with the sliding clamp and exonuclease . Fernandez-Leiro , Conrad et al . also solved the structure of the same proteins but without the DNA molecule . This revealed a large structural change between the DNA-bound and DNA-free states , which provides some clues as to how the polymerase can be quickly released from the DNA during the repeated cycles of DNA synthesis at the lagging strand . Further research is now needed to uncover what signals trigger this release of the DNA polymerase .

2015

cryo-EM structures of the E. coli replicative DNA polymerase reveal its dynamic interactions with the DNA sliding clamp, exonuclease and τ

Choice confidence , an individual’s internal estimate of judgment accuracy , plays a critical role in adaptive behaviour , yet its neural representations during decision formation remain underexplored . Here , we recorded simultaneous EEG-fMRI while participants performed a direction discrimination task and rated their confidence on each trial . Using multivariate single-trial discriminant analysis of the EEG , we identified a stimulus-independent component encoding confidence , which appeared prior to subjects’ explicit choice and confidence report , and was consistent with a confidence measure predicted by an accumulation-to-bound model of decision-making . Importantly , trial-to-trial variability in this electrophysiologically-derived confidence signal was uniquely associated with fMRI responses in the ventromedial prefrontal cortex ( VMPFC ) , a region not typically associated with confidence for perceptual decisions . Furthermore , activity in the VMPFC was functionally coupled with regions of the frontal cortex linked to perceptual decision-making and metacognition . Our results suggest that the VMPFC holds an early confidence representation arising from decision dynamics , preceding and potentially informing metacognitive evaluation . Our everyday lives involve situations where we must make judgments based on noisy or incomplete sensory information – for example deciding whether crossing the street on a foggy morning , in poor visibility , is safe . Being able to rely on an internal estimate of whether our perceptual judgments are accurate is fundamental to adaptive behaviour and accordingly , recent years have seen a growing interest in understanding the neural basis of confidence judgments . Within the perceptual decision making field , several studies have sought to characterise the neural correlates of confidence during metacognitive evaluation ( i . e . , while subjects actively judge their performance following a choice ) , revealing the functional involvement of frontal networks , in particular the lateral anterior and anterior cingulate prefrontal cortices ( Fleming et al . , 2012; Hilgenstock et al . , 2014; Morales et al . , 2018 ) . Concurrently , psychophysiological work in humans and non-human primates using time-resolved measurements has shown that confidence encoding can also be observed at earlier stages , and as early as the decision process itself ( Kiani and Shadlen , 2009; Zizlsperger et al . , 2014; Gherman and Philiastides , 2015 ) . In line with these latter observations , recent fMRI studies have reported confidence-related signals nearer the time of decision ( e . g . , during perceptual stimulation ) in regions such as the striatum ( Hebart et al . , 2016 ) , dorsomedial prefrontal cortex ( Heereman et al . , 2015 ) , cingulate and insular cortices ( Paul et al . , 2015 ) , and other areas of the prefrontal , parietal , and occipital cortices ( Heereman et al . , 2015; Paul et al . , 2015 ) . Interestingly , confidence-related processing has also been reported in the ventromedial prefrontal cortex ( VMPFC ) during value-based decisions and various ratings tasks ( De Martino et al . , 2013; Lebreton et al . , 2015 ) , however the extent to which this region is additionally involved in perceptual judgments relying on temporal integration of sensory evidence remains unclear . Importantly , the studies above suggest that confidence is likely to involve a temporal progression of neural events requiring the involvement of multiple networks , as opposed to a single event or quantity . Identifying neural confidence representations that arise early in the decision process ( e . g . , prior to metacognitive report or as early as the choice itself ) is an important prerequisite in understanding the broader confidence-related dynamics , as these signals may provide the basis for higher-order and more deliberate processes such as metacognitive appraisal . Nevertheless , efforts to characterise early confidence representations in the human brain have been limited . One potential limitation in previous approaches to studying the neural representations of confidence is the exclusive reliance on correlations with behavioural measures , most commonly in the form of subjective ratings given by participants after the decision ( Grimaldi et al . , 2015 ) . However , theoretical and empirical work suggests that post-decisional metacognitive reports may be affected by processes occurring after termination of the initial decision ( Resulaj et al . , 2009; Pleskac and Busemeyer , 2010; Fleming et al . , 2015; Moran et al . , 2015; Murphy et al . , 2015; Yu et al . , 2015; Navajas et al . , 2016; van den Berg et al . , 2016; Fleming and Daw , 2017 ) , such as integration of existing information , processing of novel information arriving post-decisionally , or decay ( Moran et al . , 2015 ) , and may consequently be only partly reflective of early confidence-related states . Here we aimed to derive a more faithful representation of these early confidence signals using EEG , and exploit the trial-by-trial variability in these signals to build parametric EEG-informed fMRI predictors , thus providing a starting point to a more comprehensive spatiotemporal account of decision confidence . We hypothesised that using an electrophysiologically-derived ( i . e . , endogenous ) representation of confidence to detect associated fMRI responses would provide not only a more temporally precise , but also a more accurate spatial representation of confidence around the time of decision . To test this hypothesis , we collected simultaneous EEG-fMRI data while participants performed a random-dot direction discrimination task and rated their confidence in each choice . Using a multivariate single-trial classifier to discriminate between High vs . Low confidence trials in the EEG data , we extracted an early , stimulus-independent discriminant component appearing prior to participants’ behavioural response . These early representations of confidence correlated across subjects with measures of confidence predicted by an accumulation-to-bound model of decision making . We then used the trial-to-trial variability in the resulting confidence signal as a predictor for the fMRI response , revealing a positive correlation within a region of the VMPFC not commonly associated with confidence for perceptual decisions . Crucially , activation of this region was unique to our EEG-informed fMRI predictor ( i . e . , additional to those detected with a conventional fMRI regressor , which relied solely on participants’ post-decisional confidence reports ) . Furthermore , a functional connectivity analysis revealed a link between the activation in the VMPFC , and regions of the prefrontal cortex involved in perceptual decision making and metacognition . Subjects ( N = 24 ) performed a speeded perceptual discrimination task whereby they were asked to judge the motion direction of random dot kinematograms ( left vs . right ) , and rate their confidence in each choice on a 9-point scale ( Figure 1A ) . Stimulus difficulty ( i . e . , motion coherence ) was held constant across all trials , at individually determined psychophysical thresholds . We found that on average , subjects indicated their direction decision 994 ms ( SD = 172 ms ) after stimulus onset and performed correctly on 75% ( SD = 5 . 2% ) of the trials . In providing behavioural confidence reports , subjects tended to employ the entire rating scale , showing that subjective confidence varied from trial-to-trial despite perceptual evidence remaining constant throughout the task ( Figure 1B ) . As a general measure of validity of subjects’ confidence reports , we first examined the relationship with behavioural task performance . Specifically , confidence is largely known to scale positively with decision accuracy and negatively with response time ( Vickers and Packer , 1982; Baranski and Petrusic , 1998 ) , though this relationship is not perfect , and is subject to individual differences ( Baranski and Petrusic , 1994; Fleming et al . , 2010; Fleming and Dolan , 2012 ) . As expected , we found a positive correlation with accuracy ( subject-averaged R = 0 . 30; one-sample t-test , t ( 23 ) = 13 . 9 , p<0 . 001 ) ( Figure 1C ) , and a negative correlation with response time ( subject-averaged R = −0 . 27; one-sample t-test , t ( 23 ) = −7 . 8 , p<0 . 001 ) ( Figure 1D ) . Thus , subjects’ confidence ratings were generally reflective of their performance on the perceptual decision task . Next , we asked whether the observed variability in subjects’ confidence reports could be explained by sustained fluctuations in attention ( i . e . , spanning multiple trials ) . We reasoned that decreases in attention may be reflected as serial correlations in confidence ratings across trials . To test this possibility , we performed a serial autocorrelation regression analysis on a single subject basis , which predicted confidence ratings on the current trial from ratings given on the immediately preceding five trials . On average , this model accounted for only a minimal fraction of the variance in confidence ratings ( subject-averaged R2 = 0 . 07 ) . Finally , we sought to rule out the possibility that trial-to-trial variability in confidence could be explained by potential subtle differences in low-level physical properties of the stimulus that may go beyond motion coherence ( e . g . , location and/or timing of individual dots ) . To this end , we compared subjects’ confidence reports on the two experimental blocks ( consisting of identical sequences of random-dot kinematograms ) , and found no significant correlation between these ( subject-averaged R = 0 . 02 , one-sample t-test , p=0 . 44 ) . Taken together , these results support the hypothesis that subjects’ reports reflected internal fluctuations in their sense of confidence , which are largely unaccounted for by external factors . To identify confidence-related signals in the EEG data , we first separated trials into three confidence groups ( Low , Medium , and High ) on the basis of subjects’ confidence ratings . We then conducted a single-trial multivariate classifier analysis ( Parra et al . , 2005; Sajda et al . , 2009 ) on the stimulus-locked EEG data , designed to estimate linear spatial weightings of the EEG sensors ( i . e . , spatial projections ) discriminating between Low- vs . High-confidence trials ( see Materials and methods ) . Applying the estimated electrode weights to single-trial data produced a measurement of the discriminating component amplitudes ( henceforth yCONF ) , which represent the distance of individual trials from the discriminating hyperplane , and which we treat as a surrogate for the neural confidence of the decision . Note that even though participants’ post-decision ratings may not form an entirely faithful representation of earlier confidence signals , they can nevertheless be used to separate trials into broad confidence groups for training the classifier and estimating the relevant discrimination weights at the time of decision . Data from individual trials , including those not originally used in the discrimination analysis , were subsequently subjected through these electrode weights to obtain a trial-specific graded measure of internal confidence . In other words , these electrophysiologically-derived confidence measures depart from their behavioural counterparts in that they contain trial-to-trial information from the neural generator giving rise to the relevant discriminating components . As such , these estimates can potentially offer additional insight into the internal processes that underlie confidence at these early stages of the decision . To quantify the discriminator's performance over time we used the area under a receiver operating characteristic curve ( i . e . , Az value ) with a leave-one-out trial cross validation approach to control for overfitting ( see Materials and methods ) . We found that discrimination performance ( Az ) between the two confidence trial groups peaked , on average , 708 ms after stimulus onset ( SD = 162 ms , Figure 2A; see Figure 2—figure supplement 1 for Az locked to the time of rating ) . To visualise the spatial extent of this confidence component , we computed a forward model of the discriminating activity ( Materials and methods ) , which can be represented as a scalp map ( Figure 2A ) . Importantly , both the temporal profile and electrode distribution of confidence-related discriminating activity were consistent with our previous work ( Gherman and Philiastides , 2015 ) where we used stand-alone EEG to identify time-resolved signatures of confidence during a face vs . car visual categorisation task . Together these observations are an indication that the temporal dynamics of decision confidence can be reliably captured using EEG data acquired inside the MR scanner , and that these early confidence-related signals may generalise across tasks . To provide additional support linking this discriminating component to choice confidence , we considered the Medium-confidence trials . Importantly , these trials can be regarded as ‘unseen’ data , as they are independent from those used to train the classifier . We subjected these trials through the same neural generators ( i . e . , spatial projections ) estimated during discrimination of High- vs . Low-confidence trials and , as expected from a graded quantity , found that the mean component amplitudes for Medium-confidence trials were situated between , and significantly different from , those in the High- and Low-confidence trial groups ( both p<0 . 001 , Figure 2B ) . To ensure these results were not due to overfitting , we also repeated the above comparisons using fully out-of-sample discriminant component amplitudes obtained from our leave-one-out cross-validation procedure ( see Materials and methods ) , and found that differences remained significant ( both p<0 . 001 , Figure 2—figure supplement 2 ) We next examined the relationship between the confidence-discriminating component and objective performance on the perceptual discrimination task . We found that component amplitudes were positively correlated with decision accuracy ( one-sample t-test on logistic regression coefficients , t ( 23 ) =8 . 6 , p<0 . 001 , Figure 2C ) , and were consistently higher for correct vs . incorrect responses across subjects ( t ( 23 ) =7 . 58 , p<0 . 001 , Figure 2D ) , in line with the well-established relationship between confidence and accuracy . To rule out the possibility that the modulation of discriminant component amplitude by confidence was purely explained by objective performance , we compared component amplitudes for Medium-confidence against High-/Low-confidence using only trials associated with correct responses , and showed that differences between these trial groups remained significant ( both p<0 . 001 , Figure 2E ) . The same pattern was found when repeating the analysis separately on error trials ( both p<0 . 001 ) . These results indicate that the confidence-related neural component can be dissociated from objective performance , as might be expected from previous reports ( Lau and Passingham , 2006; Rounis et al . , 2010; Komura et al . , 2013; Lak et al . , 2014; Fleming and Daw , 2017 ) . As the duration of the visual motion stimulus varied across trials in our task ( i . e . , remained on until subjects made a motor response on the perceptual task ) another potential concern might be that the variability in the EEG-derived confidence signatures we identified here could be explained by these stimulus-related factors . We reasoned that if that were the case , we might expect high correlation between stimulus duration and discriminant component amplitudes . However , we found that this correlation was weak ( subject-averaged R = - . 15 ) , suggesting that our classification results could not have been solely driven by this factor . Finally , we addressed the possibility that the observed variability in the confidence discriminating component could be attributed to sustained fluctuations in attention , by conducting a serial autocorrelation analysis which predicted component amplitudes on a given trial from those on the preceding five trials ( separately for each subject ) . As before , we expected that if attentional fluctuations are driving the variability in our EEG-derived confidence measures , component amplitudes on a given trial would be reliably predicted by those observed in the immediately preceding trials . We found that this model only explained a small fraction of the variance in component amplitudes ( subject-averaged R2 = 0 . 03 ) . We also assessed the influence of a neural signal known to correlate with attention ( Thut et al . , 2006 ) and predict visual discrimination ( van Dijk et al . , 2008 ) , namely occipitoparietal prestimulus alpha power . To do this , we separated trials into High vs . Low alpha power groups , individually for each subject , and compared the corresponding average discriminant component amplitudes . We found that these did not differ significantly between the two groups ( paired t-test , p=0 . 19 , Figure 2F ) . Note that variability in the confidence discriminant component was also independent of stimulus difficulty , as this was held constant across all trials . In line with this , discriminant component amplitudes for the two identical-stimulus experimental blocks were not significantly correlated ( subject-averaged R = 0 . 02; one-sample t-test , p=0 . 39 ) . We next sought to identify potential influences of neural confidence signals on decision-related behaviour . In particular , there is evidence that confidence , as reflected in behavioural ( Braun et al . , 2018 ) and physiological ( Urai et al . , 2017 ) correlates , can play a role in the modulation of history-dependent choice biases . Here , we tested whether the strength of our EEG-derived confidence signals ( i . e . , confidence discriminant component amplitude yCONF ) on a given trial might influence the probability to repeat a choice on the immediately subsequent trial ( PREPEAT ) . While we observed no overall significant links between yCONF and subsequent choice behaviour when considering the entire data set , we found a positive relationship between yCONF and PREPEAT if stimulus motion on the immediately subsequent trial was in the same direction as in the current trial ( F ( 2 , 46 ) =5 . 89 , p= . 005 , with post-hoc tests showing a significant difference in PREPEAT following Low vs . High yCONF trials , p= . 015 , Bonferroni corrected ) , as shown in Figure 2G . Thus , stronger confidence signals were associated with an increased tendency to repeat the previous choice . In contrast , we did not find any modulatory effect of yCONF on choice repetition/alternation behaviour when motion on the current trial was in the opposite direction from that of the previous trial . Thus , choices were only affected by previous confidence when no global change in motion direction had occurred from one trial to the next . Interestingly , this dependence of confidence-related repetition bias on stimulus identity points to a mechanism by which the representation of confidence interacts with a putative process of ( subliminal ) stimulus-consistency detection ( distinguishable from the decision process itself ) on the subsequent trial , to influence the decision and/or behaviour . To seek preliminary insight into how our confidence-related EEG measure relates to the decision formation process , we compared our neural signals with a measure of confidence derived from a dynamic model of decision making . Namely , we fitted subjects’ behavioural data ( i . e . , accuracy and response time ) with an adapted version of the race model ( Vickers , 1979; Vickers and Packer , 1982; De Martino et al . , 2013 ) ( see Materials and methods ) . This class of models describes the decision process as a stochastic accumulation of perceptual evidence over time by independent signals representing the possible choices ( Figure 3A ) . The decision terminates when one of the accumulators reaches a fixed threshold , with choice being determined by the winning accumulator . Importantly , confidence for binary choices can be estimated in these models as the absolute distance ( Δe ) between the states of the two accumulators at the time of decision ( i . e . , ‘balance of evidence’ hypothesis ) . Overall , we found that this model provided a good fit to the behavioural data ( Accuracy: R = 0 . 76 , p<0 . 001 , Figure 3B; RT: subject-averaged R = 0 . 965 , all p<=0 . 0016 , see Figure 3—figure supplement 1 for individual subject fits ) . We illustrate model fits to response time data in Figure 3C ( see Figure 3—figure supplement 2 for individual subject fits ) , whereby response time distributions for correct and error trials are summarised separately using five quantile estimates of the associated cumulative distribution functions ( Forstmann et al . , 2008 ) . Here , we were interested in how our neural measures of confidence ( EEG-derived discriminant component yCONF ) compared against the confidence estimates predicted by the decision model ( Δe ) , at the subject group level . To this end , we computed the mean difference in confidence ( as reflected by yCONF and Δe , respectively ) between correct and error trials , separately for each subject , and tested the extent to which these quantities were correlated across participants . This relative measure , which captured the relationship between confidence and choice accuracy , also ensured that comparisons across subjects remained meaningful after averaging across trials . We found a significant positive correlation ( i . e . , subjects who showed stronger difference in yCONF between correct and error trials also showed a higher difference in Δe , R= . 48 , p= . 019 , robust correlation coefficient obtained using the percentage bend correlation analysis ( Wilcox , 1994 ) ; see Figure 3D ) , opening the possibility that neural confidence signals might be informed by a process similar to the race-like dynamic implemented by the current model . We sought to further clarify the link between model-derived confidence estimates ( Δe ) , early neural signatures of confidence ( yCONF ) , and subjects’ behavioural reports during the rating phase of the trial ( Ratings ) , by performing an exploratory mediation analysis on these measures . We hypothesised that yCONF may be informed by quantities equivalent to Δe , and in turn influence the confidence estimates reflected in post-choice reports . Thus , we tested whether yCONF may act as a statistical mediator on the link between Δe and Ratings . As with our previous analysis linking yCONF and Δe ( Figure 3D ) , we first computed the mean difference between correct and error trials for each of the three variables of interest , to produce comparable measures across subjects ( i . e . , by removing potentially task-irrelevant individual differences in the trial-averaged scores , such as rating biases ) . These quantities ( henceforth referred to as ΔeDIFF , yCONF_DIFF , and RatingsDIFF ) were then submitted to the mediation analysis . Specifically , we defined a three-variable path model ( Wager et al . , 2008 ) with ΔeDIFF as the predictor variable , RatingsDIFF as the dependent variable , and yCONF_DIFF as the mediator ( Materials and methods ) . In line with our prediction , we found that: 1 ) ΔeDIFF was a significant predictor of yCONF_DIFF ( p= . 01 ) , 2 ) yCONF_DIFF reliably predicted RatingsDIFF after accounting for the effect of predictor ΔeDIFF ( p< . 001 ) , and 3 ) the indirect effect of yCONF_DIFF , defined as the coefficient product of effects 1 ) and 2 ) , was also significant ( p= . 004 ) . While the across-subject nature of the analysis calls for caution in interpreting the results , these observations are consistent with the possibility that yCONF reflects a ( potentially noisy ) readout of decision-related balance of evidence ( as modelled by Δe ) , and informs eventual confidence reports . We sought primarily to identify fMRI activations correlating uniquely with the endogenous signatures of confidence at the time of the perceptual decision , as obtained from our EEG discrimination analysis . In particular , we were interested in confidence-related variability in the fMRI response that might be over and above what can be inferred from behavioural confidence reports alone . To this end , we constructed a general linear model ( GLM; see Materials and methods ) of the fMRI using an EEG-derived regressor for confidence ( yCONF ) together with additional regressors accounting for variance related to subjects’ behavioural confidence reports ( i . e . , ratings ) , and other potentially confounding factors ( task performance , response time , attention , and visual stimulation ) . fMRI correlates of behavioural confidence reports . We first investigated the activation patterns associated with confidence ratings during the perceptual decision phase of the trial ( Figure 4A ) , defined as the time window beginning at the onset of the random-dot stimulus ( and ending prior to the onset of the confidence rating prompt ) . The coordinates of all activations are listed in Supplementary Table 1 ( Supplementary file 1 ) . We found that the BOLD response increased with reported confidence in the striatum , lateral orbitofrontal cortex ( OFC ) , the ventral anterior cingulate cortex ( ACC ) – areas thought to play a role in human valuation and reward ( O'Doherty , 2004; Rushworth et al . , 2007; Grabenhorst and Rolls , 2011 ) – as well as the right anterior middle frontal gyrus , amygdala/hippocampus , and visual association areas . Overall , these activations appear consistent with findings from previous studies that have identified spatial correlates of decision confidence ( Rolls et al . , 2010; De Martino et al . , 2013; Heereman et al . , 2015; Hebart et al . , 2016 ) . Negative activations ( i . e . , regions showing increasing BOLD response with decreasing reported confidence ) were found in the right supplementary motor area , dorsomedial prefrontal cortex , right inferior frontal gyrus ( IFG ) , anterior insula/frontal operculum , in line with previous reports of decision uncertainty near the time of decision ( Heereman et al . , 2015; Hebart et al . , 2016 ) . During the metacognitive report stage of the trial ( i . e . , 'rating phase' , defined as the time window beginning at the onset of the confidence prompt; Figure 4B ) , we found negative correlations with confidence ratings in extended networks ( Supplementary Table 2; Supplementary file 1 ) which included regions of the rostrolateral prefrontal cortex ( bilateral , right lateralised ) , middle frontal gyrus , superior frontal gyrus ( extending along the cortical midline and into the medial prefrontal cortex ) , orbital regions of the IFG , angular gyrus , precuneus , posterior cingulate cortex ( PCC ) , and regions of the occipital and middle temporal cortices . These activations are largely in line with research on the spatial correlates of choice uncertainty ( Grinband et al . , 2006; Fleming et al . , 2012; ) and metacognitive evaluation ( Fleming et al . , 2010; Molenberghs et al . , 2016 ) . Finally , positive correlations were observed in the striatum and amygdala/hippocampus , as well as motor cortices . fMRI correlates of EEG-derived confidence signals . To identify potential brain regions encoding early representations of confidence as captured by our confidence-discriminating EEG component , we turned to the parametric EEG-derived fMRI regressor ( i . e . , yCONF regressor ) , which captured the inherent single-trial variability in these signals . Our approach therefore allowed us to model the fMRI response using time-resolved neural signatures of confidence , which were specific to each subject . Crucially , as these measures captured the variability in the neural representation of confidence near the time of the perceptual decision itself ( i . e . , prior to behavioural response ) , they may be better suited for spatially characterising confidence during this time window compared to the behavioural confidence reports obtained later on in the trial ( as the latter may be more reflective of confidence-related information arriving post-decisionally ) . Note that these signals were only moderately correlated with reported confidence ( subject-averaged R= . 39 , SD= . 07 ) , and thus could potentially provide additional explanatory power in our fMRI model . This EEG-informed fMRI analysis revealed a large cluster in the ventromedial prefrontal cortex ( VMPFC , peak MNI coordinates [−8 40 – 14] ) , extending into the subcallosal region and ventral striatum , and a smaller cluster in the right precentral gyrus ( peak MNI coordinates [30 -20 64] ) , where the BOLD response correlated positively with the EEG-derived confidence discriminating component ( Figure 5 ) . The VMPFC has been linked to confidence-related processes in value-based , as well as other complex decisions ( De Martino et al . , 2013; Lebreton et al . , 2015 ) , however this region is not typically associated with confidence in perceptual decisions ( though see Heereman et al . , 2015; Fleming et al . , 2018 ) . Note also that , as regression parameter estimates resulting from standard GLM analysis reflect variability unique to each regressor ( i . e . , disregarding common variability ) ( Mumford et al . 2015 ) , the correlation we observed with the EEG-derived yCONF regressor in the VMPFC during the perceptual decision period is over and above what can be explained by behavioural confidence ratings alone ( i . e . , the RatingsDEC regressor , Figure 4A ) . Consistent with this , correlation of the RatingsDEC regressor with activity in the relevant VMPFC cluster ( including in a supplementary GLM analysis whereby the yCONF regressor was removed ) failed to pass statistical thresholding and would have therefore been missed using behavioural ratings alone . Interestingly , the scalp map associated with our confidence discriminating EEG component showed a diffused topography including contributions from several centroparietal electrode sites . One possibility is that the observed spatial pattern reflects sources of shared variance between the EEG component and confidence ratings themselves ( which was otherwise controlled for in our original fMRI analysis ) . To test this , we ran a separate control GLM analysis where the confidence ratings regressor ( RatingsDEC ) was removed , and found that with this model the yCONF regressor explained additional variability of the BOLD signal within several regions , including precuneus/PCC regions of the parietal cortex ( Figure 5—figure supplement 1 ) . Notably , activity in these regions has been previously shown to scale with confidence ( De Martino et al . , 2013; White et al . , 2014 ) and hypothesised to play a role in metacognition ( McCurdy et al . , 2013 ) . In a separate analysis , we also explored BOLD signal correlations with the yCONF regressor locked to the confidence rating stage ( as part of a GLM model which only included regressors at the time of rating ) . We found no correlation with yCONF in the VMPFC , suggesting confidence-related activation in this region was specific to the earlier stages of the decision . Clusters showing positive correlation with yCONF were found in the ( bilateral ) motor cortex , left planum temporale , putamen/pallidum , and lateral occipital cortex ( Figure 5—figure supplement 2 ) . Suggestive mainly of motor-related processes , these activations may have been partially confounded by repeated movement ( i . e . , button pushes ) during the rating stage of the trial . More speculatively , confidence representations may be present within motor regions , in line with the idea that decision-related information 'leaks' into the motor systems that support relevant action ( Gold and Shadlen , 2000; Song and Nakayama , 2009 ) . We found no clusters showing negative correlation with yCONF at this stage of the trial . Having identified the VMPFC as uniquely encoding a confidence signal early on in the trial ( i . e . , near the time of the perceptual decision ) , we next sought to explore potential functional interactions of this region with the rest of the brain ( for instance , with networks involved in perceptual decision making and/or post-decision metacognitive processes ) . To this end , we conducted a whole-brain PPI analysis ( see Materials and methods ) , whereby we searched for areas showing increased correlation of their BOLD response with that of a VMPFC seed , during the perceptual decision phase of the trial ( i . e . , defined here as the trial-by-trial time window between the onset of the motion stimulus and subject’s explicit commitment to choice ) . Based on existing literature showing negative BOLD correlations with confidence ratings in regions recruited post-decisionally ( e . g . , during explicit metacognitive report ) , such as the anterior prefrontal cortex ( Fleming et al . , 2012; Hilgenstock et al . , 2014; Morales et al . , 2018 ) , we expected that increased functional connectivity of such regions with the VMPFC would be reflected in stronger negative correlation in our PPI . Similarly , we hypothesised that fMRI activity in regions encoding the perceptual decision would also correlate negatively with confidence/VMPFC activation , in line with the idea that easier ( and thus more confident ) decisions are characterised by faster evidence accumulation to threshold ( Shadlen and Newsome , 2001 ) and weaker fMRI signal in reaction time tasks ( Ho et al . , 2009; Kayser et al . , 2010; Liu and Pleskac , 2011; Filimon et al . , 2013; Pisauro et al . , 2017 ) . Accordingly , we expected that if such regions increased their functional connectivity with the VMPFC during the decision , this would manifest as stronger negative correlation in the PPI analysis . We found that clusters in the bilateral orbitofrontal cortex ( OFC; peak MNI: [16 18 -16] and [−28 28–20] ) , left anterior prefrontal cortex ( aPFC; peak MNI: [−40 46 4] ) , and right dorsolateral prefrontal cortex ( dlPFC; peak MNI: [48 22 30] ) ( Figure 6 ) showed increased negative correlation with VMPFC activation during the perceptual decision . Interestingly , regions in the aPFC and dlPFC in particular have been previously linked to perceptual decision making ( Noppeney et al . , 2010; Liu and Pleskac , 2011; Philiastides et al . , 2011; Filimon et al . , 2013 ) , as well as post-decisional confidence-related processes ( Fleming et al . , 2012; Hilgenstock et al . , 2014; Morales et al . , 2018 ) and metacognition ( Fleming et al . , 2010; Rounis et al . , 2010; McCurdy et al . , 2013 ) . Here , we used a simultaneous EEG-fMRI approach to investigate the neural correlates of confidence during perceptual decisions . Our method capitalised on the unique explanatory power of time-resolved , internal measures of confidence to identify associated responses in the fMRI , allowing for a more precise spatiotemporal characterisation of confidence than if relying solely on behavioural measures . We found that BOLD response in the VMPFC was uniquely explained by the single-trial variability in an early , EEG-derived neural signature of confidence occurring prior to subjects’ behavioural expression of response . This activity was additional to what could be explained by subjects’ behavioural reports alone . Our results provide empirical support for the involvement of the VMPFC in confidence of perceptual decisions , and suggest that this region may support an early readout of confidence ( i . e . , at , or near , the time of decision ) preceding explicit choice or metacognitive evaluation . We first showed that our EEG results - namely the temporal and spatial profile of the confidence-discriminating activity - were consistent with our previous work ( Gherman and Philiastides , 2015 ) where we used a different perceptual task involving face vs . car visual categorisations , indicating that these confidence-related signals may generalise across a broader range of tasks . Interestingly , the spatial topography associated with this activity appears consistent with centroparietal scalp projections arising from signals culminating near the decision ( O'Connell et al . , 2012; Kelly and O'Connell , 2013; Philiastides et al . , 2014 ) . While the spatial limitation of EEG precludes conclusive interpretations based on this similarity , this pattern could potentially reflect a mixture of decision- and confidence-related signals , in line with the evidence that suggests these quantities may unfold together around the decision process itself ( Kiani and Shadlen , 2009; Gherman and Philiastides , 2015; van den Berg et al . , 2016; Dotan et al . , 2018 ) . Signals such as the centroparietal positivity ( CPP ) ( O'Connell et al . , 2012 ) and/or related P300 may themselves hold information about confidence as suggested by electrophysiological work ( Boldt and Yeung , 2015 ) ( see also ( Urai and Pfeffer , 2014; Twomey et al . , 2015 ) for brief discussions ) . Further , our fMRI data revealed activation patterns suggesting that distinct neural networks carry information about confidence during perceptual decision vs . explicit confidence reporting stages of the trial , respectively . Indeed , it seems plausible that qualitatively distinct representations of confidence may be encoded at different times relative to the decision process . In particular , activations during the decision phase of the trial such as the VMPFC or anterior cingulate cortex , are in line with a more automatic encoding of confidence , i . e . , in the absence of explicit confidence report ( Lebreton et al . , 2015; Bang and Fleming , 2018 ) . In line with this idea , we also observed activations in regions associated with the human reward/valuation system , such as the striatum and orbitofrontal cortex . In contrast , regions showing correlation with confidence during the confidence rating stage , in particular the anterior prefrontal cortex , have been previously associated with explicit metacognitive judgment/report ( Fleming et al . , 2012; Morales et al . , 2018 ) , potentially serving a role in higher-order monitoring and confidence communication . We presented several findings that sought to further clarify the nature and role of the early confidence signals observed in the EEG data , as well as their relationship with the perceptual decision and metacognition . Our computational modelling approach provided preliminary insight into the potential decision dynamics that might inform early confidence . Namely , we showed that these neural signals were consistent with predictions from a dynamic model of decision that quantifies confidence as the difference in accumulated evidence in favour of the possible choice alternatives , at the termination of the decision process . A possible interpretation is that the early confidence representations reflect a readout of this difference ( for instance , by a distinct system than the one supporting the perceptual choice itself ) . In other words , early confidence representations could be informed by , yet be distinct from , the quantities reflected in the model-derived confidence , in line with a dissociation between the information supporting the decision vs . confidence . Our exploratory mediation analysis is in agreement with this interpretation , suggesting that EEG-derived confidence representations can be thought of as a statistical mediator between model-derived confidence measures ( reflecting the balance of accumulated evidence at the time of decision ) and confidence ratings . In another exploratory analysis that aimed to better understand the potential impact of neural confidence signals on subsequent behaviour , we found that stronger signal amplitude increased the likelihood of repeating a choice on the subsequent trial , when the motion direction of the stimulus was consistent with that of the previous trial . Interestingly however , we did not observe this effect when subsequent motion was in the opposite direction . This dependence of the confidence-related choice repetition bias on stimulus identity is counterintuitive yet intriguing , as it points to a process that detects stimulus consistency ( i . e . , independently of the decision process itself ) , which interacts with representations of previous confidence to alter decision/behaviour ( e . g . , through selective re-weighting of evidence ) . While our current decision model cannot account for this confidence-driven trial-to-trial dependence , future computational developments may help reconcile these observations with formal models of decision and confidence . Our main fMRI finding , linking early confidence representations with VMPFC activity suggests partial independence of these signals from decision centres . Specifically , as the VMPFC is not typically known to support perceptual decision processes , it seems more plausible that the confidence signals we observe here represent a ( potentially noisy ) readout of confidence-related information . In line with this , computational and neurobiological accounts of confidence processing have proposed architectures by which a first-level form of confidence in a decision emerges as a natural property of the neural processes that support the decision , and in turn is read out ( i . e . , summarised ) by separate higher-order monitoring network ( s ) ( Insabato et al . , 2010; Meyniel et al . , 2015; Pouget et al . , 2016 ) . The timing of our EEG-derived confidence representations arising in close temporal proximity to the decision ( but prior to commitment to a motor response ) further endorse the hypothesis that the VMPFC may encode an automatic readout of confidence ( Lebreton et al . , 2015 ) in decision making , or early ( and automatic ) ‘feeling of rightness’ ( Hebscher and Gilboa , 2016 ) in memory judgments . While dedicated research will be necessary to establish the functional role of these early signals , fast pre-response confidence signals could be necessary to regulate the link between decision and impending action , for example with low confidence signalling the need for additional evidence ( Desender et al . , 2018 ) . Consistent with a role in providing a confidence readout , recent work suggests the VMPFC may encode confidence in a task-independent and possibly domain-general manner . Specifically , several functional neuroimaging studies have shown positive modulation of VMPFC activation by confidence , across a range of decision making tasks ( Rolls et al . , 2010; De Martino et al . , 2013; Heereman et al . , 2015; Lebreton et al . , 2015; Fleming et al . , 2018 ) . Notably , one study showed that fMRI activation in the VMPFC was modulated by confidence across four different tasks involving both value-based and non-value based rating judgments ( Lebreton et al . , 2015 ) . Furthermore , evidence from memory-related decision making research appears to also implicate the VMPFC in confidence processing ( Hebscher and Gilboa , 2016 ) . An outstanding question is whether , and how , the early confidence signals we identified in the VMPFC might further contribute to post-decisional metacognitive signals and eventual confidence reports . It has been long proposed that metacognitive evaluation relies on additional processes taking place post-decisionally ( Pleskac and Busemeyer , 2010; Moran et al . , 2015; Yu et al . , 2015 ) . For instance , recent evidence suggests that choice itself ( and corresponding motor-related activity ) affects confidence ( Fleming et al . , 2015; Gajdos et al . , 2018 ) and may help calibrate metacognitive reports ( Siedlecka et al . , 2016; Fleming and Daw , 2017 ) . The early confidence signals in the VMPFC could serve as one of multiple inputs to networks supporting retrospective metacognitive processes , e . g . , anterior prefrontal regions ( Fleming et al . , 2012 ) . Interestingly , our functional connectivity analysis revealed a strengthening of the link between the VMPFC and frontal areas ( notably the aPFC and dlPFC ) during the perceptual decision stage of the trial . While the functional significance of these connections remains to be determined , previous involvement of these regions in perceptual decision making and metacognition makes them likely candidates for providing or receiving input to/from the VMPFC within a confidence-related network . The observation that the VMPFC , a region known for its involvement in choice-related subjective valuation ( Philiastides et al . , 2010; Rangel and Hare , 2010; Bartra et al . , 2013; Pisauro et al . , 2017 ) encodes confidence signals during perceptual decisions raises an interesting possibility for interpreting our results . Our behavioural paradigm did not involve any explicit reward/feedback manipulation and accordingly , the observed confidence-related activation cannot be interpreted as an externally driven value signal . Instead , as has been suggested previously ( Barron et al . , 2015; Lebreton et al . , 2015 ) , a likely explanation is that as an internal measure of performance accuracy , confidence is inherently valuable . Such a signal may represent implicit reward and possibly act as a teaching signal ( Daniel and Pollmann , 2012; Guggenmos et al . , 2016; Hebart et al . , 2016; Lak et al . , 2017 ) to drive learning . In line with this interpretation , recent work suggests that confidence may be used in the computation of prediction errors ( i . e . , the difference between expected and currently experienced reward ) ( Lak et al . , 2017; Colizoli et al . , 2018 ) , thus guiding a reinforcement-based learning mechanism . Relatedly , confidence prediction error ( the difference between expected and experienced confidence ) has been hypothesised to act as a teaching signal and guide learning in the absence of feedback . In particular , regions in the human mesolimbic dopamine system , namely the striatum and ventral tegmental area , have been shown to encode both anticipation and prediction error related to decision confidence , in the absence of feedback ( Guggenmos et al . , 2016 ) , similarly to what is typically observed during reinforcement learning tasks where feedback is explicit ( Preuschoff et al . , 2006; Fouragnan et al . , 2015; Fouragnan et al . , 2017; Fouragnan et al . , 2018 ) . Importantly , these effects were predictive of subjects’ perceptual learning efficiency . Thus , confidence in valuation/reward networks could be propagated back to the decision systems to optimize the dynamics of the decision process , possibly by means of a reinforcement-learning mechanism . At the neural level , this could be implemented through a mechanism of strengthening or weakening information processing pathways that result in high and low confidence , respectively ( Guggenmos and Sterzer , 2017 ) . Though testing this hypothesis extends beyond the scope of the current study , we might expect that fluctuations in expected vs . actual confidence signals observed in our data have a similar influence on learning ( e . g . , perceptual learning ( Law and Gold , 2009; Kahnt et al . , 2011; Diaz et al . , 2017 ) . In conclusion , we showed that by employing a simultaneous EEG-fMRI approach , we were able to localise an early representation of confidence in the brain with higher spatiotemporal precision than allowed by fMRI alone . In doing so , we provided novel empirical evidence for the encoding of a generalised confidence readout signal in the VMPFC preceding explicit metacognitive report . Our findings provide a starting point for further investigations into the neural dynamics of confidence formation in the human brain and its interaction with other cognitive processes such as learning , and the decision itself . Thirty subjects participated in the simultaneous EEG-fMRI experiment . Four were subsequently removed from the analysis due to near chance ( n = 3 ) and near ceiling ( n = 1 ) performance , respectively , on the perceptual discrimination task . Additionally , one subject was excluded whose confidence reports covered only a limited fraction of the provided rating scale , thus yielding an insufficient number of trials to be used in the EEG discrimination analysis ( see below ) . Finally , one subject had to be removed due to poor ( chance ) performance of the EEG decoder . All results presented here are based on the remaining 24 subjects ( age range 20 – 32 years ) . All were right-handed , had normal or corrected to normal vision , and reported no history of neurological problems . The study was approved by the College of Science and Engineering Ethics Committee at the University of Glasgow ( CSE01355 ) and informed consent was obtained from all participants . While we conducted no explicit power analysis for determining sample size , note that our EEG analysis was performed on individual subjects using cross validation , such that in estimating our electrophysiologically-derived measure of confidence , each subject became their own replication unit ( Smith and Little , 2018 ) . All stimuli were created and presented using the PsychoPy software ( Peirce , 2007 ) . They were displayed via an LCD projector ( frame rate = 60 Hz ) on a screen placed at the rear opening of the bore of the MRI scanner , and viewed through a mirror mounted on the head coil ( distance to screen = 95 cm ) . Stimuli consisted of random dot kinematograms ( Newsome and Pare , 1988 ) , whereby a proportion of the dots moved coherently to one direction ( left vs . right ) , while the remainder of the dots moved at random . Specifically , each stimulus consisted of a dynamic field of white dots ( number of dots = 150; dot diameter = 0 . 1 degrees of visual angle , dva; dot life time = 4 frames; dot speed = 6 dva/s ) , displayed centrally on a grey background through a circular aperture ( diameter = 6 dva ) . Task difficulty was controlled by manipulating the proportion of dots moving coherently in the same direction ( i . e . , motion coherence ) . We aimed to maintain overall performance on the main perceptual decision task consistent across subjects ( i . e . , near perceptual threshold , at approximately 75% correct ) . For this reason , task difficulty was calibrated individually for each subject on the basis of a separate training session , prior to the day of the main experiment . EEG data was collected using an MR-compatible EEG amplifier system ( Brain Products , Germany ) . Continuous EEG data was recorded using the Brain Vision Recorder software ( Brain Products , Germany ) at a sampling rate of 5000 Hz . We used 64 Ag/AgCl scalp electrodes positioned according to the 10 – 20 system , and one nasion electrode . Reference and ground electrodes were embedded in the EEG cap and were located along the midline , between electrodes Fpz and Fz , and between electrodes Pz and Oz , respectively . Each electrode had in-line 10 kOhm surface-mount resistors to ensure subject safety . Input impedance was adjusted to < 25 kOhm for all electrodes . Acquisition of the EEG data was synchronized with the MR data acquisition ( Syncbox , Brain Products , Germany ) , and MR-scanner triggers were collected separately to enable offline removal of MR gradient artifacts from the EEG signal . Scanner trigger pulses were lengthened to 50μs using a built-in pulse stretcher , to facilitate accurate capture by the recording software . Experimental event markers ( including participants’ responses ) were synchronized , and recorded simultaneously , with the EEG data . Preprocessing of the EEG signals was performed using Matlab ( Mathworks , Natick , MA ) . EEG signals recorded inside an MR scanner are contaminated with gradient artifacts and ballistocardiogram ( BCG ) artifacts due to magnetic induction on the EEG leads . To correct for gradient-related artifacts , we constructed average artifact templates from sets of 80 consecutive functional volumes centred on each volume of interest , and subtracted these from the EEG signal . This process was repeated for each functional volume in our dataset . Additionally , a 12 ms median filter was applied in order to remove any residual spike artifacts . Further , we corrected for standard EEG artifacts and applied a 0 . 5 – 40 Hz band-pass filter in order to remove slow DC drifts and high frequency noise . All data were downsampled to 1000 Hz . To remove eye movement artifacts , subjects performed an eye movement calibration task prior to the main experiment ( with the MRI scanner turned off , to avoid gradient artifacts ) , during which they were instructed to blink repeatedly several times while a central fixation cross was displayed in the centre of the computer screen , and to make lateral and vertical saccades according to the position of the fixation cross . We recorded the timing of these visual cues and used principal component analysis to identify linear components associated with blinks and saccades , which were subsequently removed from the EEG data ( Parra et al . , 2005 ) . Next , we corrected for cardiac-related ( i . e . , ballistocardiogram , BCG ) artifacts . As these share frequency content with the EEG , they are more challenging to remove . To minimise loss of signal power in the underlying EEG signal , we adopted a conservative approach by only removing a small number of subject-specific BCG components , using principal component analysis . We relied on the single-trial classifiers to identify discriminating components that are likely to be orthogonal to the BCG . BCG principal components were extracted from the data after the data were first low-pass filtered at 4 Hz to extract the signal within the frequency range where BCG artifacts are observed . Subject-specific principal components were then determined ( average number of components across subjects: 1 . 8 ) . The sensor weightings corresponding to those components were projected onto the broadband data and subtracted out . Finally , data were baseline corrected by removing the average signal during the 100 ms prestimulus interval . To increase statistical power of the EEG data analysis , trials were separated into three confidence groups ( Low , Medium , High ) , on the basis of the original 9-point confidence rating scale . Specifically , we isolated High- and Low-confidence trials by pooling across each subject’s three highest and three lowest ratings , respectively . To ensure robustness of our single trial EEG analysis , we imposed a minimum limit of 50 trials per confidence trial group . For those data sets where subjects had an insufficient number of trials in the extreme ends of the confidence scale , neighbouring confidence bins were included to meet this limit . We used a single-trial multivariate discriminant analysis , combined with a sliding window approach ( Parra et al . , 2005; Sajda et al . , 2009 ) to discriminate between High- and Low-confidence trials in the stimulus-locked EEG data . This method aims to estimate , for predefined time windows of interest , an optimal combination of EEG sensor linear weights ( i . e . , a spatial filter ) which , applied to the multichannel EEG data , yields a one-dimensional projection ( i . e . , a 'discriminant component' ) that maximally discriminates between the two conditions of interest . Importantly , unlike univariate trial-average approaches for event-related potential analysis , this method spatially integrates information across the multidimensional sensor space , thus increasing signal-to-noise ratio whilst simultaneously preserving the trial-by-trial variability in the signal , which may contain task-relevant information . In our data , we identified confidence-related discriminating components , y ( t ) , by applying a spatial weighting vector w to our multidimensional EEG data x ( t ) , as follows: ( 1 ) yt=wTxt= ∑i=1Dwixi ( t ) where D represents the number of channels , indexed by i , and T indicates the transpose of the matrix . To estimate the optimal discriminating spatial weighting vector w , we used logistic regression and a reweighted least squares algorithm ( Jordan and Jacobs , 1994 ) . We applied this method to identify w for short ( 60 ms ) overlapping time windows centred at 10 ms-interval time points , between -100 and 1000 ms relative to the onset of the random dot stimulus ( i . e . , the perceptual decision phase of the trial ) . This procedure was repeated for each subject and time window . Applied to an individual trial , spatial filters ( w ) obtained this way produce a measurement of the discriminant component amplitude for that trial . In separating the High and Low trial groups , the discriminator was designed to map the component amplitudes for one condition to positive values and those of the other condition to negative values . Here , we mapped the High confidence trials to positive values and the Low confidence trials to negative values , however note that this mapping is arbitrary . To quantify the performance of the discriminator for each time window , we computed the area under a receiver operating characteristic ( ROC ) curve ( i . e . , the Az value ) , using a leave-one-out cross-validation procedure ( Duda et al . , 2001 ) . Specifically , for every iteration , we used N-1 trials to estimate a spatial filter ( w ) , which was then applied to the remaining trial to obtain out-of-sample discriminant component amplitudes ( y ) for High- and Low-confidence trials and compute the Az . Note that these out-of-sample y values were highly correlated with the y values resulting from the original High- vs . Low-confidence discrimination described above ( subject-averaged R= . 93 ) . We determined significance thresholds for the discriminator performance using a bootstrap analysis whereby trial labels were randomised and submitted to a leave-one-out test . This randomisation procedure was repeated 500 times , producing a probability distribution for Az , which we used as reference to estimate the Az value leading to a significance level of p<0 . 01 . Given the linearity of our model we also computed scalp projections of the discriminating components resulting from Equation 1 by estimating a forward model for each component: ( 2 ) a= X yyTywhere the EEG data ( X ) and discriminating components ( y ) are now in a matrix and vector notation , respectively , for convenience ( i . e . , both X and y now contain a time dimension ) . Equation 2 describes the electrical coupling of the discriminating component y that explains most of the activity in X . Strong coupling indicates low attenuation of the component y and can be visualised as the intensity of vector a . We calculated prestimulus alpha power ( 8 – 12 Hz ) in the 400 ms epoch beginning at −500 ms relative to the onset of the random dot stimulus . To do this , we used the multitaper method ( Mitra and Pesaran , 1999 ) as implemented in the FieldTrip toolbox for Matlab ( http://www . ru . nl/neuroimaging/fieldtrip ) . Specifically , for each epoch data were tapered using discrete prolate spheroidal sequences ( two tapers for each epoch; frequency smoothing of ± 4 Hz ) and Fourier transformed . Resulting frequency representations were averaged across tapers and frequencies . Single-trial power estimates were then extracted from the occipitoparietal sensor with the highest overall alpha power and baseline normalised through conversion to decibel units ( dB ) . To test whether fluctuations in the confidence-discriminating component amplitudes , yCONF , were predictive of the probability to repeat a choice on the immediately subsequent trial , PREPEAT ) , we divided yCONF into 3 equal bins ( Low , Medium , and High ) , separately for each subject , and compared the corresponding PREPEAT across subjects , using a one-way repeated measures ANOVA . To ensure that any observed modulation of PREPEAT by yCONF was independent of the correlation of yCONF with accuracy on the current trial ( s ) , we first equalised the number of correct and error trials within each yCONF bin . Specifically , for each subject , we removed either exclusively correct or error trials ( depending on which of the two was in excess ) via random selection from 500 permutations of the trial set . We report results based on the average yCONF values obtained with this procedure ( see Results ) . We modelled the perceptual decision process using a variant of the original race model of decision making ( Vickers , 1979; Vickers and Packer , 1982; De Martino et al . , 2013 ) . Specifically , each decision was represented as a race-to-threshold between two independent accumulating signals - variables L and R - which collected evidence in favour of the left and right choices , respectively . At each time step of the accumulation ( time increment = 1 ms ) , the two variables were updated separately with an evidence sample s ( t ) extracted randomly from normal distributions with mean μ and standard deviation σ , s ( t ) =N ( μ , σ ) , such that: ( 3 ) L ( t+1 ) =L ( t ) +sL ( t ) R ( t+1 ) =R ( t ) +sR ( t ) Here , we assumed that evidence samples for the two possible choices are drawn from distributions with identical variances but distinct means , whereby the mean of the distribution is dependent on the identity of the presented stimulus . For instance , a leftward motion stimulus would be associated with a larger distribution mean ( and thus on average faster rate of evidence accumulation ) in the left ( stimulus-congruent ) than right ( stimulus-incongruent ) accumulator . We defined the mean of the distribution associated with the stimulus-congruent accumulator as μcongr=0 . 1 ( arbitrary units ) , and that of the stimulus-incongruent accumulator as μincongr=μcongr/r , where r is a free parameter in the model . For each simulated trial , evidence accumulation for the two accumulator variables began at 0 and progressed towards a fixed decision threshold θ , with choice being determined by the first accumulator to reach this threshold . Finally , response time was defined as the time taken to reach the decision threshold plus a non-decision time ( nDT ) accounting for early visual encoding and motor preparation processes . We fitted the model to each subject’s response time data , using a maximum likelihood function ( as in Pisauro et al . , 2017 ) . Namely , we combined RTs for correct and incorrect trials into a single distribution by mirroring the distribution of incorrect trials at the 0 point on the time axis , and thus transforming all error RTs into negative values . We compared resulting distributions and mean choice accuracies obtained from behavioural data vs . model simulations . The log likelihood function was estimated according to: ( 4 ) LL ~ logKSRTdata , RTmodel+logexp-Accuracydata-Accuracymodel 0 . 12 KS represents the estimated probability that two independent samples ( here , behavioural vs . simulated RTs ) come from populations with the same distribution , as inferred with the two-sample Kolmogorov-Smirnov test ( implemented in Matlab function kstest2 ) . For each subject , the free model parameters were iteratively adjusted to maximise the LL . This was done by performing a grid search through a fixed range of values ( σ=[ . 6:0 . 1:1] , θ=[55:7:97] , nDT=[250:50:450] , r=[1 . 2:0 . 05:1 . 6] ) , determined after an initial exploratory search which sought to identify parameter ranges that generated plausible behavioural measures ( RT and accuracy ) ( i . e . , comparable to those observed in subjects’ behaviour ) . For each set of parameters , we simulated 500 trials and recorded mean choice accuracy , RT , and confidence ( Δe ) . To assess the quality of the model fits , we computed the correlation between observed vs . model-predicted behaviour ( namely response time quantiles for correct and error responses , as well as mean choice accuracy ) , using the robust percentage bend correlation analysis ( Wilcox , 1994 ) . To examine the relationship between model-derived confidence estimates ( Δe ) , neural confidence signals ( yCONF ) , and behavioural confidence reports ( Ratings ) , we performed an exploratory mediation analysis ( M3 toolbox for Matlab; Wager , 2018 http://wagerlab . colorado . edu/tools ) on these measures . A mediation analysis aims to identify whether the link between a predictor variable ( here , Δe ) and an outcome ( Ratings ) can be explained , fully or partially , by the indirect effect of a mediator variable ( yCONF ) . For each of the three variables of interest , we computed the mean difference between correct and error trials , and resulting values ( ΔeDIFF , yCONF_DIFF , and RatingsDIFF , respectively ) were subjected to the mediation analysis . To establish significance of the mediator effect of yCONF_DIFF , three conditions must be met 1 ) ΔeDIFF reliably predicts yCONF_DIFF , 2 ) yCONF_DIFF reliably predicts RatingsDIFF when the effect of ΔeDIFFis accounted for , and ( 3 ) a significant indirect effect of yCONF_DIFF , defined as the coefficient product of effects ( 1 ) and ( 2 ) , can be observed . We established coefficient significance in the three models using a 5000 sample bootstrap test ( Wager et al . , 2008 ) . Imaging was performed at the Centre for Cognitive Neuroimaging , Glasgow , using a 3-Tesla Siemens TIM Trio MRI scanner ( Siemens , Erlangen , Germany ) with a 12-channel head coil . Cushions were placed around the head to minimize head motion . We recorded two experimental runs of 794 whole-brain volumes each , corresponding to the two blocks of trials in the main experimental task . Functional volumes were acquired using a T2*-weighted gradient echo , echo-planar imaging sequence ( 32 interleaved slices , gap: 0 . 3 mm , voxel size: 3 × 3 × 3 mm , matrix size: 70 × 70 , FOV: 210 mm , TE: 30 ms , TR: 2000 ms , flip angle: 80° ) . Additionally , a high-resolution anatomical volume was acquired at the end of the experimental session using a T1-weighted sequence ( 192 slices , gap: 0 . 5 mm , voxel size: 1 × 1 × 1 mm , matrix size: 256 × 256 , FOV: 256 mm , TE: 2300 ms , TR: 2 . 96 ms , flip angle: 9° ) , which served as anatomical reference for the functional scans . The first 10 volumes prior to task onset were discarded from each fMRI run to ensure a steady-state MR signal . Additionally , 13 volumes were discarded from the post-task period at the end of each block . The remaining 771 volumes were used for statistical analyses . Pre-processing of the MRI data was performed using the FEAT tool of the FSL software ( FMRIB Software Library , http://www . fmrib . ox . ac . uk/fsl ) and included slice-timing correction , high-pass filtering ( >100 s ) , and spatial smoothing ( with a Gaussian kernel of 8 mm full width at half maximum ) , and head motion correction ( using the MCFLIRT tool ) . The motion correction preprocessing step generated motion parameters which were subsequently included as regressors of no interest in the general linear model ( GLM ) analysis ( see fMRI analysis below ) . Brain extraction of the structural and functional images was performed using the Brain Extraction tool ( BET ) . Registration of EPI images to standard space ( Montreal Neurological Institute , MNI ) was performed using the Non-linear Image Registration Tool with a 10 mm warp resolution . The registration procedure involved transforming the EPI images into an individual’s high-resolution space ( with a linear , boundary-based registration algorithm [Greve and Fischl , 2009] ) prior to transforming to standard space . Registration outcome was visually checked for each subject to ensure correct alignment . Whole-brain statistical analyses of functional data were conducted using a general linear model ( GLM ) approach , as implemented in FSL ( FEAT tool ) : ( 5 ) Y=βX+ ε= β1X1+ β2X2+…+ βnXn+ εwhere Y represents the BOLD response time series for a given voxel , structured as a T×1 ( T time samples ) column vector , and X represents the T×N ( N regressors ) design matrix , with each column representing one of the psychological regressors ( see GLM analysis below for details ) , convolved with a canonical hemodynamic response function ( double-gamma function ) . β represents the parameter estimates ( i . e . , regressor betas ) resulting from the GLM analysis in the form of a N × 1 column vector . Lastly , ε is a T × 1 column vector of residual error terms . A first-level analysis was performed to analyse each subject’s individual runs . These were then combined at the subject-level using a second-level analysis ( fixed effects ) . Finally , a third-level mixed-effects model ( FLAME 1 ) was used to combine data across all subjects . With the combined EEG-fMRI approach , we sought to identify confidence-related activation in the fMRI surpassing what could be explained by the relevant behavioural predictors alone . In particular , we looked for brain regions where BOLD responses correlated with the confidence-discriminating component derived from the EEG analysis . Our primary motivation behind this approach was the hypothesis that endogenous trial-by-trial variability in the confidence discriminating EEG component ( near the time of perceptual decision , and prior to behavioural response ) would be more reflective of early internal representations of confidence at the single-trial level , compared to the metacognitive reports which are provided post-decisionally and therefore likely to be subjected to additional processes . We predicted that the simultaneous EEG-fMRI approach would enable identification of latent brain states that might remain unobserved with a conventional analysis approach . To this end , we extracted trial-by-trial amplitudes of yt ( resulting from Eq . 1 ) at the time window of maximum confidence discrimination , and used these to build a BOLD predictor ( i . e . , the yCONF regressor ) . Importantly , to avoid possible confounding effects of motor preparation/response , the time of this component was determined on a subject-specific basis , by only considering the period prior to the behavioural choice ( mean peak discrimination time = 708 ms from stimulus onset , SD=162 ms ) . Thus , on average this was selected 287ms ( SD=171 ms ) prior to each subject’s mean response time . To ensure our results were not affected by potential overfitting during the estimation of y , we conducted a control GLM analysis whereby the yCONF regressor was built using fully out-of-sample y values resulting from our leave-one-out cross-validation procedure detailed above ( Figure 5—figure supplement 5 ) . Note that the trial-by-trial variability in our EEG component amplitudes is driven mostly by cortical regions found in close proximity to the recording sensors and to a lesser extent by distant ( e . g . , subcortical ) structures . Nonetheless , an advantage of our EEG-informed fMRI predictors is that they can also reveal relevant fMRI activations within deeper structures , provided that their BOLD activity covaries with that of the cortical sources of our EEG signal . We designed our GLM model to account for variance in the BOLD signal at two key stages of the trial , namely the perceptual decision period ( beginning at the onset of the random dot visual stimulus ) and the metacognitive evaluation/rating ( beginning at the onset of the rating scale display ) , respectively . A total of 10 regressors were included in the model . Our primary predictor of interest was the EEG-derived endogenous measure of confidence ( yCONF regressor ) . We modelled this as a stick function ( duration = 0 . 1 s ) locked to the stimulus onset , with event amplitudes parametrically modulated by the trial-to-trial variability in the confidence discriminating componentyt . To ensure variance explained by this regressor was unique ( i . e . , not explained by subjects’ behavioural reports ) , we included a second regressor whose event amplitudes were parametrically modulated by confidence ratings , and which was otherwise identical to the yCONF regressor ( i . e . , RatingsDEC regressor , duration = 0 . 1 s , locked to stimulus onset ) . Importantly , yCONF amplitudes were only moderately correlated with behavioural confidence ratings , thus allowing us to exploit additional explanatory power inherent to this regressor . Other regressors of no interest for the perceptual decision stage included: one regressor parametrically modulated by prestimulus alpha power in the EEG signal ( to control for potential attentional baseline effects ) , one categorical regressor ( 1/0 ) accounting for variability in response accuracy , and one unmodulated regressor ( all event amplitudes set to ( 1 ) modelling stimulus-related visual responses of no interest across both valid and non-valid ( missed ) trials ( all event durations = 0 . 1 s , locked to stimulus onset ) . To control for motor preparation/response , we also included a parametric regressor modulated by subjects’ reaction time on the direction discrimination task ( duration = 0 . 1 s , locked to the time of behavioural response ) . Note that including an additional unmodulated regressor locked to the time of the behavioural response did not alter our results . Additionally , locked to the onset of the metacognitive rating period , we included one parametric regressor ( duration = 0 . 1 s ) with event amplitudes modulated by subjects’ confidence ratings , one boxcar regressor with duration equivalent to subjects’ active behavioural engagement in confidence rating ( to minimise effects relating to motor processes ) , and one unmodulated regressor ( duration = 0 . 1 s ) . Lastly , we included one categorical boxcar regressor ( 1/0 ) to model non-task activation ( i . e . , rest breaks within each run ) . Motion correction parameters obtained from fMRI preprocessing were entered as additional covariates of no interest . As we included two rating-modulated regressors in our model , which were identical except for their onset times ( i . e . , decision and rating phases , respectively ) , we sought to ensure that these were not highly correlated . We computed the correlation between the convolved regressors , separately for each subject and experimental run ( mean R = - . 13; Figure 5—figure supplement 3 ) . Additionally , we conducted two separate control GLM analyses whereby only the regressors pertaining to one trial phase ( i . e . , decision or rating , respectively ) were included at a time . This allowed us to further validate our results , to ensure they remained unaffected by potential correlations between regressors at the two stages of the trial ( Figure 5—figure supplement 4 ) . Finally , we also assessed the correlations between all regressors by computing the variance inflation factors ( VIF ) for the regressors in our model . We found that mean VIF = 3 . 57 ( ±1 . 83 ) , with multicollinearity typically being considered high if VIF > 5 – 10 . To estimate a significance threshold for our fMRI statistical maps whilst correcting for multiple comparisons , we performed a nonparametric permutation analysis that took into account the a priori statistics of the trial-to-trial variability in our primary regressor of interest ( yCONF ) , in a way that trades off cluster size and maximum voxel Z-score ( Debettencourt et al . , 2011 ) . For each resampled iteration , we maintained the onset and duration of the regressor identical , whilst shuffling amplitude values across trials , runs and subjects . Thus , the resulting regressors for each subject were different as they were constructed from a random sequence of regressor amplitude events . This procedure was repeated 200 times . For each of the 200 resampled iterations , we performed a full 3-level analysis ( run , subject , and group ) . Our design matrix included the same regressors of non-interest used in all our GLM analysis . This allowed us to construct the null hypothesis H0 , and establish a threshold on cluster size and Z-score based on the cluster outputs from the permuted parametric regressors . Specifically , we extracted cluster sizes from all activations exceeding a minimal cluster size ( 5 voxels ) and Z-score ( 2 . 57 per voxel ) for positive correlations with the permuted parametric regressors . Finally , we examined the distribution of cluster sizes ( number of voxels ) for the permuted data and found that the largest 5% of cluster sizes exceeded 162 voxels . We therefore used these results to derive a corrected threshold for our statistical maps , which we then applied to the clusters observed in the original data ( that is , Z=2 . 57 , minimum cluster size of 162 voxels , corrected at p=0 . 05 ) . We conducted a psychophysiological ( PPI ) analysis to explore potential functional connectivity between the region of the VMPFC found to uniquely explain trial-to-trial variability in our electrophysiologically-derived measures of confidence , and the rest of the brain , during the perceptual decision phase of the trial . To carry out the PPI analysis , we first extracted the time-series data from the seed region . Specifically , we identified the cluster of interest at the group level ( i . e . , in standard space ) by applying the cluster correction procedure described in the previous section . Using this as a template , we constructed subject-specific masks of the voxels exhibiting the strongest correlation with the VMPFC region of interest , and back-projected these into the functional space of each individual . Resulting masks were used to compute average time-series data , separately for each subject and functional run , which subsequently served as the physiological regressor ( s ) in the PPI model . To carry out the PPI analysis , we performed a new GLM analysis . This included the following regressors , locked to the time of stimulus onset: ( 1 ) an unmodulated regressor ( all event amplitudes set to 1 ) , ( 2 ) the physiological regressor ( time course of the VMPFC seed ) , ( 3 ) the psychological regressor ( a boxcar function with event amplitudes set to one and duration parametrically modulated by trial-specific decision times ( i . e . , interval between stimulus presentation and behavioural response on the perceptual task ) , and ( 4 ) the interaction regressor . Additionally , motion parameters estimated during registration ( see preprocessing step ) were included as regressors of no interest . The statistical output from the interaction regressor thus reveals regions of the brain where correlation with the BOLD signal in the VMPFC is stronger during the perceptual decision than the rest of the trial . Importantly , this represents variance additional to that explained by the psychological and physiological regressors alone . Correction for multiple comparisons was performed on the whole brain using the outcome of the resampling procedure as described earlier . To illustrate the activation time course within the VMPFC cluster identified with our EEG-informed fMRI analysis , we first extracted the average BOLD response time-series from this region , separately for each subject and functional run ( as detailed in the previous section ) . We aligned our data to the onset of the random-dot stimulus , by approximating to the time of the nearest fMRI volume , and defined the temporal window of interest as the -4 s to 10 s interval relative to stimulus onset . We proceeded to separate trials into three bins according to the magnitude of the confidence discriminating component yCONF ( i . e . , Low , Medium , and High yCONF ) , and computed the respective percent signal change as follows: ( 6 ) %BOLD Changej ( t ) = BOLDj ( t ) − BOLDjbaselineBOLD¯where j represents the trial index , BOLDt represents the stimulus-locked data at time point t , andBOLDbaseline is the mean baseline data , with the baseline window defined as the 4 s interval prior to stimulus onset . Finally , BOLD¯ is the average signal across the entire functional run . Resulting signals were averaged across trials , runs , and subjects .

While waiting to cross the road on a foggy morning , you see a shape in the distance that appears to be an approaching car . How do you decide if it is safe to cross ? We often have to make important decisions about the world based on imperfect information . What guides our subsequent actions in these situations is a sense of accuracy , or confidence , that we associate with our initial judgments . You would not step off the kerb if you were only 10% confident the car was a safe distance away . But how , when , and where in the brain does such confidence emerge ? Gherman and Philiastides examined how brain activity relates to confidence during the early stages of decision-making , that is , before people have explicitly committed to a particular choice . Healthy volunteers were asked to judge the direction in which dots were moving across a screen . They then had to rate how confident they were in their decision . Two techniques – EEG and fMRI – tracked their brain activity during the task . EEG uses scalp electrodes to reveal when and how electrical activity is changing inside the brain , while fMRI , a type of brain scan , shows where these changes in brain activity occur . Used together , the two techniques provide a greater understanding of brain activity than either used alone . Activity in multiple regions of the brain correlated with confidence at different stages of the task . Certain brain networks showed confidence-related activity while the volunteers tried to judge the direction of movement , and others were engaged when volunteers made their confidence ratings . However , activity in only one area reliably indicated how confident the volunteers felt before they had made their choice . This area , the ventromedial prefrontal cortex , also helps process rewards . This suggests that feelings of confidence early in the decision-making process could guide our behaviour by virtue of being rewarding . Many brain disorders – including depression , schizophrenia and Parkinson's disease – compromise decision-making . Patients show changes in accuracy , response times , and in their ability to accurately evaluate their decisions . The methods used in the current study could help reveal the neural changes that cause these impairments . This could lead to new methods to diagnose and predict cognitive deficits , and new ways to treat them at an earlier stage .

2018

Human VMPFC encodes early signatures of confidence in perceptual decisions

Symbiotic microbes impact the function and development of the central nervous system ( CNS ) ; however , little is known about the contribution of the microbiota during viral-induced neurologic damage . We identify that commensals aid in host defense following infection with a neurotropic virus through enhancing microglia function . Germfree mice or animals that receive antibiotics are unable to control viral replication within the brain leading to increased paralysis . Microglia derived from germfree or antibiotic-treated animals cannot stimulate viral-specific immunity and microglia depletion leads to worsened demyelination . Oral administration of toll-like receptor ( TLR ) ligands to virally infected germfree mice limits neurologic damage . Homeostatic activation of microglia is dependent on intrinsic signaling through TLR4 , as disruption of TLR4 within microglia , but not the entire CNS ( excluding microglia ) , leads to increased viral-induced clinical disease . This work demonstrates that gut immune-stimulatory products can influence microglia function to prevent CNS damage following viral infection . Viral infection of the CNS can cause permanent neurologic damage and psychiatric disorders ( Bergmann et al . , 2006; Koyuncu et al . , 2013; van den Pol , 2009 ) . Indeed , multiple viruses including Zika and West Nile virus have been associated with Guillain-Barre syndrome and memory deficits , respectively ( Carson et al . , 2006; Parra et al . , 2016; Vasek et al . , 2016 ) . Moreover , while Parkinson’s disease , multiple sclerosis , Alzheimer’s disease , narcolepsy and other chronic neurologic diseases have unknown etiologies , initiation by CNS viral infection has been implicated ( Itzhaki et al . , 2016; Kakalacheva et al . , 2011; Stoessl , 1999; Tesoriero et al . , 2016; Virtanen and Jacobson , 2012 ) . Viral infection of the CNS presents unique challenges to the immune system with regard to controlling and eliminating the invading pathogen . Resident cells of the CNS are often targets for infection and are critical in induction of distinct pro-inflammatory signatures that function to attract virus-specific lymphocytes into the CNS . Consequently , a significant hurdle encountered by infiltrating antigen-specific lymphocytes is to eliminate virus from infected cells while limiting damage that may have long-term detrimental consequences to the host . Therefore , defining antiviral host defense mechanisms within the CNS might highlight novel therapeutic interventions to prevent or treat many of these viral-induced neurologic disorders . Microglia are resident tissue phagocytes of the CNS that function in host defense and tissue homeostasis . Recent studies have shown that despite their location within the CNS , microglia are impacted by the gut microbiota ( Erny et al . , 2015; Matcovitch-Natan et al . , 2016; Sampson et al . , 2016 ) . This is consistent with the known function of the microbiota on both immune cell composition and functionality ( Belkaid and Hand , 2014; Hooper et al . , 2012 ) . Several studies have shown that resident commensal species can instruct the development of regulatory and inflammatory immune cell subsets that aid to limit viral replication at mucosal sites ( Abt et al . , 2012; Robinson and Pfeiffer , 2014 ) . However , while the microbiota inhibits infection by some viruses , it can also enhance the infectivity of other viruses ( Robinson and Pfeiffer , 2014 ) . Thus , the contribution of the microbiota during viral infection may be distinct for each virus and might be determined by the location of infection . Despite advancements in this area , it remains unclear if microbial commensals impact either host defense and/or disease in response to CNS viral infection . To characterize the role of the microbiota in a model of viral-induced neurologic disease , 6- to 8-week-old C57BL/6 specific-pathogen-free ( SPF ) or germ-free ( GF ) mice were infected intracranially ( i . c . ) with the neurotropic JHM strain of mouse hepatitis virus ( JHMV ) . JHMV infection results in an acute encephalomyelitis characterized by viral replication in astrocytes , microglia , and oligodendrocytes ( Bergmann et al . , 2006 ) ( Greenberg et al . , 2014 ) ( Blanc et al . , 2014; Carbajal et al . , 2010; Marro et al . , 2016; Schrauf et al . , 2008; Stiles et al . , 2006; Chen et al . , 2014 ) . Virus-specific T cells control viral replication through secretion of anti-viral cytokines and cytolytic activity . Sterile immunity is not achieved and virus persists within white matter tracts in surviving mice that subsequently develop an immune-mediated demyelinating disease ( Bender and Weiss , 2010; Bergmann et al . , 2006; Cheng et al . , 2018; Lane and Hosking , 2010; Libbey et al . , 2014 ) . As is typical in this model , infected SPF mice reach peak clinical disease by day 10–11 post-infection ( p . i . ) followed by a reduction of symptoms that plateaus around day 15–16 p . i . In contrast , GF mice developed worsened clinical disease that peaks at day 13 and is maintained through 21 days p . i . ( Figure 1A ) . While GF mice had slightly lower titers of virus within the brain at 7 days p . i . , control of viral replication was impaired at 14-days p . i . and virus was still present in GF mice at 21 days p . i . , a time point where SPF animals have almost undetectable levels of infectious viral particles ( Figure 1B ) . Importantly , GF mice displayed enhanced neurologic damage , as there was significantly more demyelination in the spinal cord when compared to SPF animals ( Figure 1C , D ) . Immune cells traffic to the CNS in response to JHMV infection and viral clearance is dependent on both CD4 and CD8 T cells . Fewer total cells were recovered from the CNS of GF mice during peak viral infection ( day 7 ) ( Figure 1E ) . Moreover , there was a reduced frequency of total CNS-infiltrating CD4+ and CD8+ T cells ( Figure 1F , G ) and fewer viral-specific CD4+ and CD8+ T cells in GF mice at day 7 p . i . ( Figure 1H , I ) . These differences are diminished at a chronic timepoint ( day 21 ) , as reductions in only total CD8+ and viral-specific CD8+ T cells are observed ( Figure 1G–I ) . Inflammation is controlled by the presence of CD4+ , FoxP3+ T regulatory ( Tregs ) cells and previous reports have shown a critical role for these cells in controlling neuroinflammation in response to JHMV infection ( Anghelina et al . , 2009 ) . GF mice have increased CNS infiltrating Tregs during the acute phase of infection ( 7 p . i ) ; however , there is a paucity of Tregs in infected GF mice during the chronic demyelinating phase of disease ( Figure 1J ) . The gut microbiota can be significantly reduced in the presence of antibiotics and is often used to validate findings in GF mice . To this end , progeny from breeder pairs of SPF mice that were orally treated with antibiotics , and remained on antibiotics after weaning , were subsequently infected i . c . with JHMV at 6–7 weeks of age . The same antibiotic regimen was performed previously , and qRT-PCR of bacterial 16S sequences demonstrated a reduction of fecal bacteria by approximately 100-fold ( Soto et al . , 2017 ) . Similar to what was observed in GF mice , lifelong antibiotic-treated infected animals developed worsened chronic disease when compared to animals with an intact microbiota ( Figure 1—figure supplement 1A ) and were unable to control viral replication ( Figure 1—figure supplement 1B ) . These data provide further support for the microbiota in functioning to coordinate disease following CNS viral infection . Numerous reports have demonstrated a critical role for the microbiota in maturation of the immune response ( Belkaid and Hand , 2014; Hooper et al . , 2012 ) . In particular , commensal microbes promote CD4+ T helper cell responses both within and outside of the intestine ( Atarashi et al . , 2011; Ivanov et al . , 2009; Lee et al . , 2011; Mazmanian et al . , 2005 ) . To determine whether increased disease in JHMV-infected GF mice is due to T cell-intrinsic or-extrinsic effects , either CD4+ or CD8+ T cells from immunized SPF or GF mice were adoptively transferred into JHMV infected Rag1-/- mice . SPF or GF mice were infected intraperitoneally ( i . p . ) with mouse hepatitis virus strain DM ( MHV-DM ) to generate virus-specific T cells ( Bergmann et al . , 2004; Glass and Lane , 2003 ) . Splenic CD4+ or CD8+ T cells were purified and stained with tetramers to enumerate viral-specific T cells; subsequently , equal numbers of viral-specific T cells were separately transferred into JHMV infected Rag1-/- mice ( Figure 2A ) . Consistent with published reports using other viruses , we observed slight systemic immune deficiencies in the response to i . p . infection with MHV-DM in GF mice in both total T cells and virus-specific T cells; only the number of viral-specific CD8+ T cells reached statistical significance ( Figure 2—figure supplement 1A–D ) ( Abt et al . , 2012; Ichinohe et al . , 2011 ) . Importantly , there were no differences observed in the recruitment of either CD4+ or CD8+ T cells derived from either GF or SPF mice to the CNS of JHMV-infected Rag1-/- mice , nor were there differences in CNS viral burden ( Figure 2B–G ) . Collectively , these data show that neither CD4+ or CD8+ T cells from GF mice are intrinsically deficient in their ability to respond to JHMV or traffic to the CNS . Thus , these data suggest that the microbiota exerts effects on alternate cell types to protect from viral infection within the CNS . Microglia are the resident innate immune cells of the brain and are important for diverse processes including maintaining neuronal homeostasis , regulating brain development , and mounting immunity against pathogenic organisms . Commensal microbes are necessary for proper microglia maturation , with gene expression and morphological analysis suggesting that microglia isolated from GF mice have an immature phenotype ( Erny et al . , 2015; Matcovitch-Natan et al . , 2016; Sampson et al . , 2016 ) . Yet , the effect of commensal microbes on microglial responses to CNS viral infection has yet to be well-tested . Flow cytometric analysis from homogenized brains followed by percoll gradient purification did not demonstrate significant differences in the frequencies of microglia from GF and SPF mice under homeostatic conditions ( Figure 3—figure supplement 1A–C ) . One role for microglia is to activate antigen-specific T cell responses through presentation of antigens via MHC molecules ( Mack et al . , 2003 ) . We identified a lower frequency of MHC class II + microglia in uninfected GF mice ( Figure 3A , B ) . Disruption to the microbiota with antibiotics decreases both the percentage of total microglia and the number of MHC expressing microglia in uninfected animals ( Figure 3C–F ) . Presentation of antigens requires co-stimulatory molecules such as CD86 and CD40 which are significantly reduced in expression on microglia from uninfected GF mice ( Figure 3G and Figure 3—figure supplement 1H ) . CIITA is a transcription factor that is critical to the expression of MHCII expression ( Mach et al . , 1996 ) . Expression of Ciita mRNA transcripts is significantly reduced in microglia isolated from GF or antibiotic-treated mice compared to SPF mice , further corroborating a reduction in the antigen presentation machinery in the absence of microbes ( Figure 3H , I ) ( Butovsky et al . , 2014 ) . Three days post-infection , antigen presentation markers were also reduced in antibiotic-treated mice ( Figure 3—figure supplement 1D-G ) , suggesting a weakened ability to activate T lymphocytes . To functionally validate these findings , we measured CD8+ T cell proliferation during coculture with microglia from either SPF or antibiotic-treated mice . Corroborating the reduced antigen presentation markers on microglia from antibiotic-treated mice , the ability of these microglia to induce T cell proliferation was reduced when presenting the immunodominant JHMV peptide S510–518 ( Figure 3J ) . Microglia are considered to be the first line of defense during a CNS infection . Indeed , recent studies have demonstrated that a reduction in microglia leads to a defect in control of JHMV replication within the CNS ( Wheeler et al . , 2018 ) . However , the effect of microglia depletion on neurologic damage and demyelination remains unclear . To address this , we reduced the number of microglia using a colony-stimulating factor one receptor ( CSFR1 ) inhibitor drug , PLX-3397 ( Elmore et al . , 2014 ) . Administration of PLX-3397 1 week prior to JHMV infection of SPF mice resulted in a marked reduction in microglia ( CD45lo , CD11b+ , F4/80+ cells as determined by flow cytometric staining ) in both number and percentage at days 7 and 14 p . i . ( Figure 4A , B ) . Recent studies have demonstrated that treatment of West Nile Virus-infected mice with a CSF1R inhibitor results in limited viral control associated with impaired activation of antigen presenting cells ( APCs ) in both blood and draining lymph nodes arguing that in addition to targeting microglia other immune cells are also depleted leading to a potentially immunosuppressed state ( Funk and Klein , 2019 ) . Similar to this study , we did not observe a reduction in either splenic ( CD45+ , CD11b+ ) or CNS-infiltrating macrophage ( CD45hi , CD11b+ ) populations ( Figure 4—figure supplement 1A; Figure 4—figure supplement 1B; Figure 4—figure supplement 1C ) . We observed enhanced mortality in JHMV-infected mice following microglia reduction that correlated with an impaired ability to control viral replication within the CNS ( Figure 4C , D ) . Most importantly , microglia depletion leads to increased demyelination compared to control mice ( Figure 4E , F ) . Similar to what we observe in GF mice , both CD4+ and CD8+ T cell responses are deficient in the absence of microglia and IFN-γ production from these cells is impaired ( Figure 4G–l and Figure 4—figure supplement 1D; Figure 4—figure supplement 1E; Figure 4—figure supplement 1F; Figure 4—figure supplement 1G; Figure 4—figure supplement 1H ) . Collectively , these data support a role for microglia for limiting neurologic damage . Previous reports have identified that short chain fatty acids ( SCFAs ) are able to influence the morphology and transcriptome of microglia ( Erny et al . , 2015 ) ; however , the microbiota is also a rich source of microbial ligands that can bind to Toll-like receptors ( TLRs ) that are known to influence systemic immune responses ( Kubinak and Round , 2012 ) . Furthermore , the microbiota is also known to induce several cytokines that have the potential to influence microglial function ( Pinteaux et al . , 2002 ) . To identify relevant microbial products that might directly activate microglia , we screened various SCFAs , TLR ligands and innate cytokines on BV2 cells to evaluate induction of MHC expression . The BV2 cell line was originally developed by infecting primary microglial cell cultures with a -raf/v-myc oncogene carrying retrovirus ( J2 ) ( Blasi et al . , 1990 ) and has been used by numerous investigators as an effective alternative model system for primary microglia cultures to study various cell biology aspects ( Henn et al . , 2009 ) . While SCFAs were able to induce MHCII expression at very high concentrations , lower levels of most SCFAs and cytokines associated with the innate immune response , for example IL-1 , IL-18 and IL-33 , did not stimulate MHCI and II expression in BV2 cells ( Figure 5A , B and Figure 5—figure supplement 1A–C ) . Butyrate was able to stimulate a small , but significant , increase in MHC expression; however , the highest levels of MHCI and II upregulation were achieved by bacterial-associated TLR ligands ( Figure 5A , B and Figure 5—figure supplement 1C ) . LPS ( and to a lesser extent , the TLR1/2 heterodimer ligand Pam3CysK4 ) significantly upregulated both the frequency of MHCII expressing cells and the total amount of MHCI per cell ( Figure 5A , B and Figure 5—figure supplement 1C ) . These data suggest that TLR ligands from the microbiota are able to prime microglia for antigen presentation . To test if microbial ligands could increase antigen presentation machinery on microglia in vivo , GF mice were orally fed either the TLR4 ligand LPS or LPS in combination with the TLR1/2 ligand Pam3CysK4 to replicate the natural route of exposure to microbiota-produced ligands . Animals were fed TLR ligands for 2 weeks , and subsequently infected with JHMV . The level of TLR ligands fed in these experiments does not lead to excessive levels of TLRs in the serum and does not exceed those found in SPF animals ( Figure 5—figure supplement 2A , B ) . Feeding GF mice bacterial cell wall components resulted in reduced paralysis and lower viral titers at day 9 p . i . ( Figure 5C , D ) . Consistent with this , increases in cell number , CD4 and CD8 T cells in the CNS of TLR ligand fed animals was also observed ( Figure 5—figure supplement 2C–H ) . Feeding GF animals LPS alone , led to similar increases in cell number in the CNS of infected mice ( Figure 5E ) , more MHCII expressing microglia and increased CD4 T cell trafficking to the CNS during infection ( Figure 5F , G ) . Interestingly , feeding mice both ligands increased numbers and frequency of CD4 and number of CD8 T cells in the CNS ( Figure 5—figure supplement 2D–H ) , while LPS only increased total numbers of CD4 T cells ( Figure 5G and data not shown ) . Feeding mice LPS alone , however , increased MHCII expression on microglia during disease ( Figure 5F ) , but this was not observed when feeding both LPS and Pam3Cysk4 ( data not shown ) . These results demonstrate that TLR ligands derived from the gut microbiota are sufficient to influence microglial activation and protection in response to viral-induced neurologic disease . Our results demonstrate that gut microbiota products influence priming of resident cells of the CNS through TLR-dependent signals . Feeding TLR ligands to GF mice suggests that these microbial products are sufficient to decrease neurologic disease , and our results show that TLR4 exposure may play a more prominent role than TLR2 signaling in this process . A previous report demonstrated that there is no difference in the morphology of microglia from TLR deficient mice ( Erny et al . , 2015 ) , but antigen presentation in these mice was not evaluated in these studies . Therefore , we infected TLR2-/- and TLR4-/- animals with JHMV and assayed for neurologic damage . Compared to WT mice , TLR2-/- animals had no difference in mortality ( not shown ) or clinical presentation of disease ( Figure 6—figure supplement 1A ) . TLR4-/- mice had significantly worsened disease symptoms that were similar to GF mice ( Figure 6A ) . Importantly , TLR4-/- animals had significantly increased demyelination , indicating that TLR4 signals are important for protection from neurologic damage ( Figure 6B , C ) . Consistent with this , microglia , but not macrophages , from TLR4-/- animals express lower levels of MHCII under homeostatic conditions , prior to infection ( Figure 6D , E and Figure 6—figure supplement 1B ) and lower total numbers of MHCII expressing microglia ( and total microglia ) during JHMV infection when compared to WT mice ( Figure 6F and G ) . JHMV is unable to directly activate TLR4 ( Figure 6—figure supplement 1C ) , further arguing that microbiota-based signals function to prime microglia against CNS infection . Collectively , these results argue that brain resident immune cells are specifically influenced by TLR4 signals to protect from viral-induced CNS damage . Collectively , these data argue that gut microbial products are important for priming microglia within the CNS . However , these experiments cannot differentiate between TLR4 signals exerting activity within the gut or by directly priming microglia themselves . Indeed , we and others have detected TLR4 agonists within the blood of mice and humans , suggesting that microbial products could be circulating within the blood and thus be available to prime cells at sites distant from mucosal tissue ( Clarke et al . , 2010; Soto et al . , 2017 ) . One study , however , has suggested that LPS is unable to cross the blood-brain barrier; therefore , it is unclear how TLR4 signals derived from the gut microbiota could influence microglia within the CNS ( Banks and Robinson , 2010 ) . To initially address this , we performed a bone marrow chimera experiment . Radiation of animals followed by a bone marrow transplant readily replaces cells of the hematopoietic system in most compartments; however , microglia are known to be radio-resistant and therefore , during these experiments the majority of microglia are still from the recipient animal . Based on this , we performed bone marrow chimeras where we transplanted either WT or TLR4-/- bone marrow into WT recipients . Similarly to our previous results , we find that TLR4-/- animals develop significantly worsened clinical disease compared to WT animals ( Figure 7A ) . Moreover , animals receiving either WT or TLR4-/- bone marrow , where most of the systemic hematopoietic system is replaced by TLR4 deficient cells , still develop disease similar to WT animals ( Figure 7A ) . To address this using a genetic approach , we crossed TLR4-floxed animals to either a nestin-Cre ( CNSΔTLR4 ) driver or an inducible CX3CR1CreER driver ( CX3CR1ΔTLR4 ) . Nestin is expressed in neural progenitor cells that give rise to neurons , oligodendrocytes and astrocytes of the CNS; however , its expression in microglia has been contentious . To verify that Nestin is not expressed in microglia , we crossed the nestin-Cre driver to a Rosa-lox-stop-lox-eYFP animal , allowing us to fate map nestin ( and thus YFP ) expressing cells . While we could readily detect YFP expression within the CNS , we were unable to detect YFP+ cells in the microglia fraction ( Figure 7—figure supplement 1A ) , further corroborating that nestin-Cre will not drive recombination in microglia . CX3CR1 is expressed in microglia , but also in other immune cells such as dendritic cells in the gut; however , since microglia are long-lived , the use of an inducible system allows for depletion of a gene within microglia , while other immune cells that might be affected by the tamoxifen treatment will rapidly turn over and revert back to a WT status ( Goldmann et al . , 2013 ) . This is currently the most widely used method to knock genes out within microglia . We verified that this treatment lead to disruption of TLR4 within microglia ( Figure 7—figure supplement 1B ) . Based on this , CX3CR1ΔTLR4 animals were tamoxifen treated and subsequently infected 1 month post-tamoxifen treatment . While infection of CNSΔTLR4 animals did not demonstrate a difference in clinical disease compared to WT animals ( Figure 7B ) , disruption of TLR4 specifically within microglia , lead to worsened paralysis ( Figure 7C ) . Collectively , these data demonstrate that gut microbial signals can directly influence microglia function to prevent viral-induced neurologic diseases . Consistent with the known contribution of the microbiota to immune system development , studies performed almost 50 years ago suggested that resident commensals aid in protection from influenza virus ( Dolowy and Muldoon , 1964 ) . However , more recent studies have uncovered that other viruses take advantage of the microbiota to enhance infectivity ( Robinson and Pfeiffer , 2014 ) . Many human neurotropic viral pathogens are associated with the development of neurologic dysfunction ( Koyuncu et al . , 2013 ) , yet little is known about the functional contribution of bacterial symbionts in host defense and disease following infection of the CNS . We demonstrate that products from the microbiota are sufficient to prime microglia within the CNS that aid in control of viral replication through microglia-intrinsic TLR4 signaling . Reductions in microbial stimulation through increased antibiotic use and sanitation , has been suggested to contribute to the significant increase in obesity , autoimmunity and even some neurologic disorders in the western world ( Blaser , 2016; Strachan , 2000 ) . Supporting this hypothesis , our findings demonstrate that loss of immune stimulation by the gut microbiota leads to failure to control viral replication within the CNS leading to enhanced neuropathy . Thus , loss of protective resident microbes can lead to CNS dysfunction . There is an emerging interest in how gut commensals can influence diverse processes within the nervous system ( Fung et al . , 2017 ) . The microbiota influences serotonin production , microglia gene expression patterns , and disease pathology in a mouse model of Parkinson’s disease ( Erny et al . , 2015; Reigstad et al . , 2015; Sampson et al . , 2016; Yano et al . , 2015 ) , yet little is known about the mechanisms by which the microbiota govern these processes . Recent studies have demonstrated that SCFAs are able to influence the gene expression pattern and activation state of microglia ( Erny et al . , 2015 ) ; however , it is likely that other molecules exist that could influence the maturation of the CNS . We identify that bacterial cell wall products , such as LPS , are sufficient to prime microglia for antigen presentation to effectively clear virus . While microglia have long been known to express TLRs , this family of receptors has been primarily thought to function only during infection . Here , we show that the microbiota regulates microglia function through TLR4 , priming these cells to respond to infection . Microglia develop early in embryogenesis from yolk sac progenitors; however , in contrast to macrophages , microglia are long-lived without any significant input from circulating blood cells ( Prinz et al . , 2014 ) . How TLR ligands signal to cells within the CNS is unclear , however , there is evidence that gut microbial products are found circulating within the blood and could reach the CNS through this route ( Clarke et al . , 2010 ) . This requires further investigation as some reports have shown that LPS is not able to penetrate the blood–brain barrier . While our data demonstrate that TLR4 signaling by microglia is , in part , responsible for orchestrating microglia activation , the possibility exists that gut microbiota signals can be transmitted to the CNS from the enteric nervous system . Indeed , enteroendocrine cells have been reported to contain neuropods that are directly linked to neuronal cells and are able to transmit signals to the CNS ( Bohórquez et al . , 2015 ) . It should also be noted that many of our experiments do not differentiate effects of the gastrointestinal microbiota from commensals occupying other niches . Our data from orally administered LPS likely limits effects to gastrointestinal exposure , but oronasalpharyngeal and pneumonic exposure may be occurring as well . Relatedly , differences observed between feeding mice LPS alone , and in combination with Pam3CysK4 suggest that more research could be performed investigating the interaction between multiple TLR ligands on microglia function and response to infection . Moreover , we cannot completely rule out a contribution from gut-resident CX3CR1 cells to this phenotype or other migrating DC populations . Recent studies have demonstrated that cells from the gut can migrate to the brain and there exist a population of long-lived CX3CR1 cells within the gut ( Shaw et al . , 2018 ) . As these gut cells are still radio-sensitive ( microglia are not ) , our bone marrow chimera studies argue , that these cells likely only play a minor role , if any . Future investigations using animal models that allow tracking of specific populations from the gut will aid in parsing out these details . Therefore , additional work remains to be performed to identify how gut microbes influence distal CNS processes . The clinical implications of this work further support that maintenance of a complex microbial community is important to protect from diseases that plague the western world . Indeed , there are several examples of cell wall molecules from the microbiota that are sufficient to treat disease in various animal models ( An et al . , 2014; Mazmanian et al . , 2008; Vatanen et al . , 2016; Wang et al . , 2014 ) . Our findings support that immunostimulatory LPS is one of these microbial cell wall products . There are several different types of LPS that are derived from the microbiota , indeed LPS from Bacteriodes is less stimulatory than LPS derived from E . coli ( Vatanen et al . , 2016 ) ; therefore , the composition of the microbiota and the type of immune stimulatory products being delivered by the microbiota likely dictate maturation of microglia within the CNS . Additionally , fatty acids that can be produced by the microbiota are also capable of binding TLR4 , and might constitute other relevant agonists ( Wong et al . , 2009 ) . As microglia are also known to influence the function of neurons , these results suggest that microbial control of microglia could influence neuronal activity and function ( Parkhurst et al . , 2013 ) . Much remains to be explored regarding the mechanisms by which gut microbes influence the brain; however , our findings highlight that appropriate microbial stimulation is critical to steady state development of the CNS and optimum neurologic health . SPF C57BL/6J , birth antibiotic C57BL/6J , Rag1-/- , nestin-Cre , Il1r-/- , Il18-/- , Tlr4-/- , TLR4-floxed , Cx3xrCreER , and Tlr2-/- mice were originally purchased from Jackson Labs and reared in the SPF facility at the University of Utah . Germfree C57BL/6J mice were reared in gnotobiotic chambers in the germfree facility at the University of Utah . Microbial sterility is determined every 3 weeks by plating and PCR . Mice used in this study were 6–10 weeks old at time of infection . For microglia depletion , PLX-3397 ( formulated in an AIN-76A ) was prepared by Research Diets and provided by Plexxikon Inc SPF mice were provided PLX-3397 in their chow beginning 7 days prior to viral infection . Mice were chosen at random to receive either PLX-3397 or control chow . All experiments were performed in accordance to federal regulations as well as the guidelines for animal use set forth by the University of Utah Institutional Animal Care and Use Committee . Mice were intra-cranially infected with 150–600 plaque forming units ( PFU ) of the neurotropic JHMV strain J2 . 2v-1 suspended in 30 μL of PBS and clinical disease monitored using an established scoring sysem previously described ( Blanc et al . , 2014; Carbajal et al . , 2010; Chen et al . , 2014; Greenberg et al . , 2014; Marro et al . , 2016; Schrauf et al . , 2008; Stiles et al . , 2006 ) . Germ-free mice were maintained on antibiotics ( see below ) during JHMV infection . For transfer of virus-specific T cells , mice were intraperitoneally ( i . p . ) infected with 2 . 5 × 105 PFU the DM strain of MHV ( Stiles et al . , 2006; Bergmann et al . , 2004; Glass and Lane , 2003; Dickey et al . , 2016; Glass et al . , 2004; Glass and Lane , 2003; Plaisted et al . , 2014 ) . At day 7 post-infection ( p . i . ) , spleens were collected and CD4 +or CD8+ T cells were enriched using magnetic based separation columns ( MACS Miltenyi Biotec ) . The frequencies of virus-specific T cells were determined using CD4+ and CD8+ immunodominant tetramers ( PE-conjugated tetramers I-Ab/M133–147 and Db/S510–518 , respectively ) at 8 μg/ml ( NIH Tetramer Core Facility ) ( Stiles et al . , 2006 ) , JI; Chen et al . ( 2014 ) , Stem Cell Reports ) ; 2 . 5 × 105 virus-specific cells were transferred via retro-orbital injection into SPF Rag1-/- recipient animals 3 days p . i . and brains from recipient mice were collected 7 days post-transfer to determine brain viral titers ( Dickey et al . , 2016; Stiles et al . , 2006 ) . Rag1-/- mice were randomly chosen to receive either SPF or GF T cells . Viral titers within the brains were determined on the DBT astrocytoma line using previously defined protocols . After animal sacrifice and perfusion with ice cold PBS , half of the brain of mice was collected and homogenized at defined times p . i . and leukocytes enriched on a Percoll gradient as previously described ( Blanc et al . , 2014 ) . Cells were stained using the following antibodies: CD4 ( GK1 . 5 , Biolegend ) , CD3 ( 145–2 C11 , Biolegend ) , IFN-γ ( XMG1 . 2 , Biolegend ) , FoxP3 ( FJK-16s , eBioscience ) , IL-10 ( JES5-16E3 , Biolegend ) , CD8 ( 53-6 . 7 , Biolegend ) , CD45 ( 30-F11 , Biolegend ) , F4/80 ( BM8 , Biolegend ) , CD11b ( M1/70 , Biolegend ) , CD40 ( HM40-3 , Biolegend ) , CD86 ( GL-1 , Biolegend ) , MHCI ( M1/42 , Biolegend ) , and MHCII ( M5/114 . 15 . 2 , Biolegend ) . Virus-specific T cells were determined via flow cytometric analysis through either intracellular staining for IFN-γ or defined tetramers specific for immunodominant CD4 +and CD8+T cell-specific viral epitopes ( Chen et al . , 2014; Stiles et al . , 2006 ) . Samples were analyzed using BD LSRFortessa and FACSDiva software and data was measured using FlowJo . RNA was collected from cells using a Ribozol ( Amresco ) extraction method . RNA was treated with DNaseI ( Sigma-Aldrich ) per supplier protocol . cDNA was synthesized using qScript cDNA SuperMix ( Quantigene ) . Amplification of cDNA was measured using Lightcycler 480 SYBR Green I Master ( Roche ) with the CT method and normalized to L32 . Primers used were L32F 5-AAGCGAAACTGGCGGAAAC-3 , L32R 5-TAACCGATGTTGGGCATCAG-3 , CIITAF 5-ACACCTGGACCTGGACTCAC-3 , CIITAR 5-GCTCTTGGCTCCTTTGTCAC-3 , TGF-BF 5-CTGCTGAGGCTCAAGTTAAAAGTG-3 , TGF-BR 5-CAGCCGGTTGCTGAGGTA-3 . JHMV-F 5-CGAGCCGTAGCATGTTTATCTA-3 , JHMVR 5-CGCATACACGCAATTGAACATA-3 . TLR4F 5-TGGGTCAAGGAACAGAAGCAGT-3 TLR4R 5-AATCCAACACTAAGGAGGTATTCA-3 . Enrichment of CD11b + cells was performed by subjecting single-cell suspension of brain homogenates ( as performed in CNS Cell Isolation and Flow Cytometry ) to CD11b+ ( microglia ) Microbead MACS purification ( Miltenyi Biotec ) according to manufacture instruction . After animal sacrifice at defined times p . i . and leukocytes enriched on a Percoll gradient as previously described ( Blanc et al . , 2014 ) . Cell were sorted at the University of Utah Flow Cytometry Core based using the following markers , CD45lo , CD11b+ , F4/80+ . Microglia were then plated in a 96-well flat bottom plate coated with Poly-D-Lysine ( Sigma Aldrich ) and incubated with the S510 peptide overnight . Spleens were collected from MHV-DM infected SPF mice 7 D . P . I . , viral-specific CD8 +T cells were sorted at the University of Utah flow cytometry core . T cells were stained using Far Red Cell Trace or CFSE ( Thermo Fisher ) , CD8 +T cells were then co-cultured with microglia for 72 hr , samples were analyzed using BD LSRFortessa and FACSDiva software and data was measured using FlowJo . Experimental mice were sacrificed at defined times points and the length of spinal cord extending from thoracic vertebrate 6–10 was cryoprotected in 30% sucrose , cut into 1 mm transverse blocks and processed so as to preserve the craniocaudal sequence/orientation and subsequently embedded in O . C . T . ( VWR , Radnor , PA ) . Eight-μ M-thick coronal sections were cut and sections were stained with hematoxylin/eosin ( H &E ) in combination with luxol fast blue ( LFB ) and between 4 and 8 sections/mouse analyzed . Areas of total white matter and demyelinated white matter were determined with Image J Software and demyelination was scored as a percentage of total demyelination from spinal cord sections analyzed as previously described ( Blanc et al . , 2015; Blanc et al . , 2014; Dickey et al . , 2016 ) . Breeder pairs of Adult SPF mice were treated with an antibiotic cocktail of Ampicillin , Erythromycin , Neomycin , and Gentamycin ( 0 . 5 g/L ) supplemented with Equal ( 4 g/L ) , progeny of these breeder pairs were also kept on the same antibiotic cocktail after weaning . For TLR ligand experiments , Pam3CSK4 ( 10 μg/mL ) ( Invivogen ) and LPS ( 10 μg/mL ) ( Sigma-Aldrich ) were added to the drinking water of germfree mice for 2 weeks prior to infection . Germfree mice were chosen at random to receive TLR ligands or to serve as controls . The murine BV-2 cell line was acquired from the ATCC and was mycoplasma-free grown in RPMI and supplemented with SCFAs [Acetate , Butyrate , Propionate ( 0 . 1 μg/mL ) , Sigma Aldrich] TLR ligands [LPS , Pam3CysK4 , FSL-1 and Poly I:C ( 0 . 1 μg/mL ) , Invivogen] and recombinant mouse cytokines [IL-1 IL-18 , and IL-33 ( 0 . 25 μg/mL ) , Biolegend] for 24 hr . At age of weaning ( 3 weeks ) , mice were orally gavaged with 0 . 2 mg tamoxifen per g body weight once daily for 5 days as previously described ( Fonseca et al Journal of Neuroinflammation 201714:48 ) . Mice were given 1 month after tamoxifen administration to allow for myeloid cell turnover before infection with JHMV . HEK-Blue mTLR2 or mTLR4 cells ( InvivoGen ) were used according to manufacturer’s instruction to determine serum concentrations of TLR2 and TLR4 and if JHMV signals through TLR4 . After passaging according to manufacturer instruction , a flat-bottom 96-well plate was loaded with 25 , 000 cells per well . Then , 20 μL of PBS , mouse serum , varying dilutions of JHMV or serial dilutions of 0 . 1 μg/ml LPS added to each well . Cells were incubated for 6 hr , spun down and supernatant isolated . 20 μL of supernatant was then incubated with 180 μL of QUANTI-Blue solution ( InvivoGen ) for 15 min . Optical density at 620 nm was then measured using a Biotek Synergy H1M plate reader . No a priori power analyses were used to determine sample size . Sample size was determined by availability of germ-free mice and age/sex matched controls ( when applicable ) . The sample sizes reported here are similar to those in previous literature utilizing the JHMV model ( Ireland et al . , 2008 ) . Sample sizes are provided within the legend of each figure . Pair-wise comparison of experimental groups were performed with an unpaired two-tailed Student’ t-test or one-way ANOVA or Mann-Whitney test for viral titers . JHMV disease curves were analyzed using two-way ANOVA or area under the curve analysis . Outlier testing was performed using Grubb’s test for outliers . If an outlier was detected ( p<0 . 05 ) , that data point was omitted from any analysis . All statistics were performed using Prism six software ( GraphPad Software ) .

Trillions of bacteria , fungi and viruses live inside us , forming what is known as our microbiota . Far from causing problems , these microbes benefit our health in many ways . Most of our microbiota lives in our gut , yet there is increasing evidence that it can influence how our central nervous system works . In particular , these communities of microbes could have a role in multiple sclerosis , a disease that emerges when the immune system attacks the insulating sheath which protects neurons , slowly leading to paralysis . What causes multiple sclerosis is still unknown , but scientists believe that a viral infection could trigger the condition . In the gut , the microbiota helps the immune system to fight off harmful microbes . It is still unclear whether it performs the same role in the central nervous system , and if it can participate in diseases where the immune system harms nerve tissues . Previous studies in mice have looked into how gut microbes influence the development of illnesses similar to multiple sclerosis , but they did not use the type of live viral infection that is thought to trigger the condition . In rodents , a strain of mouse hepatitis virus ( or MHV ) causes symptoms similar to the ones observed in patients with multiple sclerosis: the animals become paralyzed and their neurons’ protective sheaths get damaged . Brown , Soto et al . compared how mice that have their normal microbes , that were raised to be free of microbes , or that were given antibiotics responded to this virus . Animals that were germfree or had received antibiotics had weakened immune responses and failed to clear MHV . These mice also showed worse paralysis . Further experiments revealed that gut microbes protected against paralysis by switching on a cascade of molecular events in a specific type of immune cell in the nervous system . These findings suggest that in the central nervous system , the microbiota is critical to quickly clear viruses and to stop symptoms associated with multiple sclerosis from emerging . Our own genetic background , but also lifestyle changes such as diets , antibiotics or sanitation can influence our microbiota . In parallel , in the past decades there has been an increase in the number of diseases , such as multiple sclerosis , in which the immune system turns against the body . The work by Brown , Soto et al . therefore emphasizes the need to maintain a diverse microbial community . For example , species of gut bacteria should be replaced or maintained after antibiotic treatments . However , future work is necessary to understand which of these microbes are protective , and whether they operate during specific timeframes .

2019

The microbiota protects from viral-induced neurologic damage through microglia-intrinsic TLR signaling

Dnmt1 is critical for immediate postnatal intestinal development , but is not required for the survival of the adult intestinal epithelium , the only rapidly dividing somatic tissue for which this has been shown . Acute Dnmt1 deletion elicits dramatic hypomethylation and genomic instability . Recovery of DNA methylation state and intestinal health is dependent on the de novo methyltransferase Dnmt3b . Ablation of both Dnmt1 and Dnmt3b in the intestinal epithelium is lethal , while deletion of either Dnmt1 or Dnmt3b has no effect on survival . These results demonstrate that Dnmt1 and Dnmt3b cooperate to maintain DNA methylation and genomic integrity in the intestinal epithelium . DNA methylation patterns are established by DNA methyltransferase enzymes ( Dnmts ) , for which two categories have been defined based on in vitro assays . The ‘de novo’ methyltransferases , Dnmt3a and Dnm3b , establish novel patterns of DNA methylation , and prefer to bind unmethylated DNA in vitro ( Okano et al . , 1998 ) . The ‘maintenance’ methyltransferase , Dnmt1 , has a high affinity for hemi-methylated DNA in vitro , and preserves DNA methylation in replicating cells ( Bestor , 1992; Leonhardt et al . , 1992 ) . Dnmts are crucial for embryonic development in mice , as mice null for Dnmt1 or Dnmt3b arrest at mid-gestation , and Dnmt3a null mice die in the first few weeks of life ( Li et al . , 1992; Okano et al . , 1999 ) . Although DNA methylation is not necessary for murine embryonic stem ( ES ) cell growth , the differentiation of Dnmt1-hypomorphic and Dnmt3a; Dnmt3b-mutant ES cells is severely impaired ( Chen et al . , 2003; Jackson et al . , 2004; Lei et al . , 1996; Tsumura et al . , 2006 ) . These results indicate an important role for DNA methylation and Dnmts in directing cell differentiation processes . One of the primary consequences of hypomethylation is increased DNA damage and genomic instability . Global hypomethylation in mice results in chromosome duplications and invasive T-cell lymphomas at four months of age ( Gaudet et al . , 2003 ) , and in mouse ES cells , loss of Dnmt1 also causes global hypomethylation and increased mutation rates ( Chen et al . , 1998 ) . In the HCT116 colorectal cancer cell line , ablation of the catalytically active DNMT1 results in cell cycle arrest and apoptosis due to increased chromosomal instability ( Chen et al . , 2007; Spada et al . , 2007 ) . In mouse embryonic fibroblasts , ablation of either Dnmt1 ( Jackson-Grusby et al . , 2001 ) or Dnmt3b ( Dodge et al . , 2005 ) causes gradual hypomethylation , deregulated gene expression , and cell death . Dnmt1 and DNA methylation are also required for viability in most proliferating somatic cell populations , including human skin cells ( Sen et al . , 2010 ) , mouse embryonic fibroblasts ( Jackson-Grusby et al . , 2001 ) , and neuronal ( Fan et al . , 2001 ) and pancreatic ( Georgia et al . , 2013 ) progenitor cells . Interestingly , Dnmt1 is not required for adult intestinal stem cell survival ( Sheaffer et al . , 2014 ) . The mature intestinal epithelium is a single cell layer lining the lumen of the intestine , structured into finger-like protrusions , designated ‘villi’ , and invaginations into the underlying mesenchymal tissue , termed ‘crypts . ’ Intestinal stem cells are located in the crypt and respond to multiple signaling pathways that control proliferation and differentiation ( Elliott and Kaestner , 2015 ) . Stem cells give rise to rapidly dividing transit-amplifying cells , which move in ordered cohorts up the crypt-villus axis . As cells migrate out the crypt , they differentiate into one of several distinct cell lineages , a process that is largely dependent on levels of Notch signaling . Loss of Dnmt1 in the adult mouse intestinal epithelium causes hypomethylation of regulatory regions associated with several intestinal stem cell genes , resulting in inappropriate gene expression during differentiation , and expansion of the crypt zone ( Sheaffer et al . , 2014 ) . In contrast , ablation of Dnmt1 during intestinal crypt development causes hypomethylation , DNA damage , and apoptosis of epithelial cells , resulting in increased perinatal lethality ( Elliott et al . , 2015 ) . Previous studies did not investigate the requirement for Dnmt1 in maintaining global DNA methylation or preserving genomic stability in the mature intestine . Thus , the mechanism behind preservation of the Dnmt1-mutant adult intestinal epithelium , the only rapidly dividing somatic tissue known to survive without Dnmt1 , is not known . To determine the mechanism underlying Dnmt1 mutant intestinal survival , we employed tissue-specific , inducible mouse models and analyzed the effects immediately after Dnmt1 deletion in the adult intestinal epithelium . Ablation of Dnmt1 caused an acute phenotype characterized by weight loss , global DNA hypomethylation , genome instability , and apoptosis . Strikingly , animals returned to baseline DNA methylation levels within two months of Dnmt1 deletion , indicating recovery by a de novo methyltransferase . We demonstrate that the de novo methyltransferase Dnmt3b is upregulated following loss of Dnmt1 , and essential for epithelial survival in the Dnmt1 mutant intestine . Our results implicate a role for DNA methylation , maintained by both Dnmt1 and Dnmt3b , in protecting genomic stability in intestinal epithelial homeostasis . These data are the first to show that Dnmt3b can function in maintenance DNA methylation in vivo . To determine the primary effects of Dnmt1 deletion in the adult intestinal epithelium , we employed an inducible , intestinal epithelial-specific gene ablation model . The Dnmt1loxP/loxP;Villin-CreERT2 mice ( Dnmt1 mutants ) and their Dnmt1loxP/loxP siblings ( controls ) were tamoxifen-treated at four weeks of age to induce Cre recombinase activity ( El Marjou et al . , 2004; Jackson-Grusby et al . , 2001 ) . Although Dnmt1 mutants lost a significant amount of weight in the two weeks following Cre induction , mice recovered by day 17 post-tamoxifen treatment , and survived at rates identical to controls ( Figure 1A–B ) . To elucidate the mechanism underlying acute weight loss , we isolated Dnmt1 mutant and control intestines one week following tamoxifen treatment . 10 . 7554/eLife . 12975 . 003Figure 1 . Dnmt1 ablation results in genomic instability and apoptosis one week following tamoxifen treatment . ( A , B ) Dnmt1loxP/loxP ( control , n=5 ) and Dnmt1loxP/loxP;Villin-CreERT2 ( Dnmt1 mutant , n=9 ) mice were tamoxifen treated at four weeks of age , and weighed every day over a 17-day period . Dnmt1 mutants lost a significant amount of weight by day 11 , but recovered to near-starting weight by day 16 . *p<0 . 05 , Student’s t-test . ( B ) All Dnmt1 mutants survive the 17-day time-course described in ( A ) , similar to controls . ( C–D ) Hematoxylin and Eosin staining of control and Dnmt1 mutant intestines . One week post-ablation , Dnmt1 mutants exhibit loss of crypt integrity , vacuolization of the epithelium , and an increase in crypt fission ( D ) compared to controls ( C ) . ( E–F ) Immunohistochemistry confirms loss of Dnmt1 protein in mutants one week following tamoxifen treatment ( F ) relative to control intestine ( E ) . ( G-H ) Immunofluorescent staining for Ki67 ( red ) , which marks proliferating cells , and γH2AX ( green ) , which marks DNA double-strand breaks as a marker of chromosomal instability . One week following Dnmt1 ablation , mutant crypts display increased levels of γH2AX foci ( H , yellow arrows ) relative to controls ( G ) . Dnmt1 mutants also display slightly enlarged crypts ( Ki67 in H versus G ) , as described previously ( Sheaffer et al . , 2014 ) . ( I–J ) Immunofluorescent TUNEL staining ( red ) , which marks apoptotic nuclei , and E-cadherin ( green ) , to outline the intestinal epithelium . One week after tamoxifen treatment , Dnmt1 mutants display increased crypt cell apoptosis ( J , yellow arrows ) compared to controls ( I ) . ( K–L ) Crypts were isolated from paraffin-embedded tissue by laser capture microdissection , and the methylation levels of LINE1 loci and the imprinting control region of H19 were determined by targeted bisulfite sequencing . One week after tamoxifen treatment , methylation of LINE1 ( K ) and H19 ( L ) are significantly decreased in Dnmt1 mutants compared to controls ( n=4 per genotype ) . ***p<0 . 001 , *p<0 . 05 , Student’s t-test . For data and p values , refer to Figure 1—source data 1 . Error bars represent S . E . M . Scale bars are 50 μm . For all staining , n=3 biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 12975 . 00310 . 7554/eLife . 12975 . 004Figure 1—source data 1 . Contains mouse weight/survival data in Figure 1A–B , targeted bisulfite sequencing data in Figure 1K–L . DOI: http://dx . doi . org/10 . 7554/eLife . 12975 . 004 One-week post-tamoxifen treatment , Dnmt1 mutants exhibited multiple abnormalities in intestinal epithelial morphology , with partial loss of epithelial integrity and a high frequency of crypt fission ( Figure 1C–D ) . Dnmt1 loss was confirmed on the protein level by immunohistochemistry ( Figure 1E–F ) , and the Dnmt1 mutant epithelium displayed a slight expansion of the proliferative crypt zone ( Figure 1G–H ) , as reported previously ( Sheaffer et al . , 2014 ) . However , Dnmt1 mutant epithelia also exhibited regions that lacked crypts and/or villi , juxtaposed with hyperplastic crypts that replaced damaged tissue ( Figure 1D ) . Since Dnmt1 ablation results in increased double stranded breaks and apoptosis in the neonatal intestine ( Elliott and Kaestner , 2015 ) , we investigated if the Dnmt1 deficient adult intestine also displayed altered genomic stability . As a general marker of chromosomal instability , we stained for γH2AX , which labels DNA double strand break foci , and is an indicator of the DNA damage response . We observed an increase in γH2AX foci in the crypts of Dnmt1 mutants compared to controls , which displayed minimal DNA double strand breaks ( Figure 1G–H ) . We also performed TUNEL staining to identify apoptotic nuclei , and found that the Dnmt1 mutant intestine exhibited increased crypt cell apoptosis relative to the control intestinal epithelium , which did not contain any apoptotic nuclei ( Figure 1I–J ) . To determine global DNA methylation levels , we isolated mutant and control intestinal crypt epithelium by laser capture microdissection , and performed targeted bisulfite-sequencing of the repetitive LINE1 loci . LINE1 retrotransposons account for approximately 20% of the mouse genome ( Waterston et al . , 2002 ) , and are a representative of genome-wide DNA methylation levels ( Lane et al . , 2003; Yang et al . , 2004 ) . The LINE1 repeats were significantly demethylated in Dnmt1 mutant crypts , with methylation reduced to approximately 50% of controls at each CpG analyzed ( Figure 1K ) . We also performed bisulfite sequencing of the H19 imprinting control region to analyze maintenance methylation , and found that the region was slightly demethylated relative to controls , although this difference was only significant when comparing the methylation of the entire sequenced region ( Figure 1L ) . Overall , these results suggest a phenotype in which loss of Dnmt1 results in hypomethylation of LINE1 repeats and the H19 imprinting control region , increased DNA damage , and apoptosis . Interestingly , Dnmt1loxP/loxP;Villin-CreERT2 mutant mice survive at rates comparable to controls ( Figure 1B ) , indicating that Dnmt1 is not required for continued intestinal maintenance in the adult mouse . To determine the long-term effects of Dnmt1 ablation , we harvested Dnmt1 mutant and sibling control intestines two months following tamoxifen injection . We confirmed that Dnmt1 deletion had been maintained in mutant epithelia ( Figure 2C–D ) , but found that Dnmt1 mutant intestinal epithelial morphology ( Figure 1A–B ) and proliferation ( Figure 1E–F ) were comparable to controls . In addition , levels of DNA damage and apoptosis , as indicated by γH2AX and TUNEL staining , respectively , were similar to control mice ( Figure 2E–H ) . We isolated crypt cells from two-month Dnmt1 mutants and controls by laser-capture microdissection , and performed targeted bisulfite sequencing for LINE1 and H19 , as described above . LINE1 methylation levels remained significantly decreased compared to controls ( Figure 2I ) , but demethylation was not as severe compared to Dnmt1 mutants at one-week post-tamoxifen treatment ( compare Figure 2I to 1K ) . Strikingly , methylation at the H19 imprinting control region had been fully restored ( Figure 2J ) . These results implicate a mechanism that compensates for loss of Dnmt1 and leads to recovery of intestinal epithelial DNA methylation and genomic integrity . 10 . 7554/eLife . 12975 . 005Figure 2 . The Dnmt1 mutant intestinal epithelium recovers with time . Dnmt1loxP/loxP ( control ) and Dnmt1loxP/loxP;Villin-CreERT2 ( Dnmt1 mutant ) mice were tamoxifen treated at four weeks of age , and intestines were harvested two months later for immunostaining and DNA methylation analysis . ( A–B ) Hematoxylin and eosin staining revealed that two months following tamoxifen treatment , the Dnmt1 deficient epithelium appears similar to controls ( B versus A ) . ( C–D ) Epithelial Dnmt1 deletion is maintained in Dnmt1 mutants two months after tamoxifen injections ( D versus C ) . ( E–F ) Immunofluorescent staining for Ki67 ( red ) , which marks proliferating cells , and γH2AX ( green ) , which marks DNA double-strand breaks as a marker of chromosomal instability . By two months post-Dnmt1 deletion , the mutant epithelium has returned to baseline levels of DNA damage ( F versus E control ) . ( G–H ) TUNEL staining ( red ) , which marks apoptotic nuclei , and immunostaining for E-cadherin ( green ) , to outline the intestinal epithelium . Two months following tamoxifen injection , Dnmt1 mutants appear similar to controls and display no apoptosis in the epithelium ( H versus G , respectively ) . ( I , J ) Crypts were isolated from paraffin-embedded tissue by laser capture microdissection , and the methylation levels of LINE1 loci and the imprinting control region of H19 were determined by targeted bisulfite sequencing . Two months following tamoxifen injection , Dnmt1 mutants have mostly regained methylation at both the LINE1 ( E ) and H19 ( F ) loci , and are comparable to controls ( n=5–6 per genotype ) . However , the slight demethylation across the entire LINE1 loci is significantly decreased compared to controls . For data and p values per CpG , refer to Figure 2—source data 1 . Error bars represent S . E . M . Scale bars are 50 μm . For all staining , n=3 biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 12975 . 00510 . 7554/eLife . 12975 . 006Figure 2—source data 1 . Contains targeted bisulfite sequencing data presented in Figure 2I–J . DOI: http://dx . doi . org/10 . 7554/eLife . 12975 . 006 We surmised that the de novo methyltransferases might compensate for loss of Dnmt1 , and performed qRT-PCR and immunofluorescent staining for both Dnmt3a and Dnmt3b in mutant and control intestines harvested one week following tamoxifen treatment . We confirmed loss of Dnmt1 mRNA in mutant crypts by qRT-PCR , but observed no changes in Dnmt3a transcript or protein expression ( Figure 3A–C ) . In addition , simultaneous loss of both Dnmt1 and Dnmt3a did not alter or exaggerate the Dnmt1 mutant phenotype , or result in decreased viability ( Figure 3—figure supplement 1 ) . Dnmt1loxP/loxP;Dnmt3aloxP/loxP;Villin-CreERT2 mice ( Dnmt1;Dnmt3a mutants ) displayed acute changes in cell death , and LINE1 and H19 methylation , identical to those seen in Dnmt1 mutants ( Figure 3—figure supplements 1G-H and 2A–B ) . Furthermore , Dnmt1;Dnmt3a mutants do not exhibit increased γH2AX foci compared to single Dnmt1 deficient mice ( Figure 3—figure supplement 1E–F ) . Combined with its unchanged protein and mRNA expression , we concluded that Dnmt3a is not required for the recovery of Dnmt1 mutant epithelia . 10 . 7554/eLife . 12975 . 007Figure 3 . Dnmt3b is upregulated following Dnmt1 ablation . Dnmt1loxP/loxP ( control ) and Dnmt1loxP/loxP;Villin-CreERT2 ( Dnmt1 mutant ) intestines were harvested one week following tamoxifen treatment for gene expression and immunostaining analysis . ( A ) qRT-PCR comparing the relative gene expression levels of Dnmt1 , Dnmt3a , and Dnmt3b in the jejunum of tamoxifen-treated controls and Dnmt1 mutants ( n=3–4 per genotype ) . Compared to controls , Dnmt1 mutants express significantly lower levels of Dnmt1 , while Dnmt3b expression is significantly increased . Gene expression was calculated relative to the geometric mean of TBP and β-actin . p<0 . 01 , Student’s t-test . For data and p-values , refer to Figure 3—source data 1 . ( B–C ) Controls ( B ) and Dnmt1 mutants ( C ) display similar levels of Dnmt3a protein ( green ) . ( D–E ) Dnmt1 mutants ( E ) display elevated Dnmt3b protein in intestinal crypts , compared to controls ( E ) . Error bars represent S . E . M . Scale bars are 50 μm . For all staining , n=3 biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 12975 . 00710 . 7554/eLife . 12975 . 008Figure 3—source data 1 . Contains qPCR data and analysis shown in Figure 3A . DOI: http://dx . doi . org/10 . 7554/eLife . 12975 . 00810 . 7554/eLife . 12975 . 009Figure 3—source data 2 . Contains targeted bisulfite sequencing data presented in Figure 3—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 12975 . 00910 . 7554/eLife . 12975 . 010Figure 3—figure supplement 1 . Deletion of Dnmt3a in addition to Dnmt1 causes no additive effects on epithelial proliferation , genome stability or cell death within one week . ( A , B ) Dnmt1loxP/loxP; Dnmt3aloxP/loxP ( A , control ) and Dnmt1loxP/loxP;Dnmt3aloxP/loxP;Villin-CreERT2 ( B , Dnmt1;Dnmt3a mutant ) intestinal epithelium one week after tamoxifen injection . Immunohistochemistry for Dnmt1 confirms loss of protein in tamoxifen-treated Dnmt1;Dnmt3a mutant relative to controls . ( C , D ) Immunofluoresence for Dnmt3a confirms loss of protein in tamoxifen-treated Dnmt1;Dnmt3a mutant intestinal epithelium ( D ) compared to control ( C ) . ( E , F ) Co-staining for Ki67 ( red ) , a marker of proliferation , and γH2AX ( green ) , which marks DNA double-stranded breaks . Control ( E ) has minimal γH2AX foci . Dnmt1;Dnmt3a mutants ( F ) have increased proliferation , similar to what is seen in Dnmt1 mutants . Contrary to Dnmt1 deletion , Dnmt1;Dnmt3a mutants do not display increased γH2AX foci one week after tamoxifen injection . ( G , H ) TUNEL staining to detect apoptosis ( red ) with E-cadherin ( green ) to outline the epithelium . Controls ( G ) display no apoptotic nuclei in crypt cells . Dnmt1;Dnmt3a mutants ( H ) display a slight increase in apoptosis , similar to what is observed in Dnmt1 mutant crypts one week after tamoxifen injection . All scale bars are 50 μm . For all staining , n=3 biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 12975 . 01010 . 7554/eLife . 12975 . 011Figure 3—figure supplement 2 . Ablation of Dnmt3a and Dnmt1 induces genome demethylation at LINE1 and H19 loci . Intestines from Dnmt1loxP/loxP;Dnmt3aloxP/loxP ( control , n=2 ) and Dnmt1loxP/loxP;Dnmt3aloxP/loxP;VilllinCreERT2 ( Dnmt1;Dnmt3a mutant , n=3 ) mice were harvested one week after tamoxifen treatment . Crypt cells were isolated by laser capture microdissection and the methylation levels of LINE1 loci and the imprinting control region of H19 were determined by targeted bisulfite sequencing . ( A , B ) Methylation of LINE1 ( A ) and H19 ( B ) are significantly decreased in Dnmt1;Dnmt3a mutants one week after tamoxifen treatment . In all graphs , error bars are S . E . M . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , Student’s t-test . For data and p-values per CpG , refer to Figure 3—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 12975 . 011 In contrast , Dnmt3b mRNA expression was significantly increased in the Dnmt1 mutant crypts compared to control crypt cells ( Figure 3A ) . In agreement with our qRT-PCR data , we discovered an increase in Dnmt3b protein levels in the mutant intestinal epithelium while expression levels in the lamina propria were unchanged ( Figure 3D–E ) , suggesting that Dnmt3b is upregulated within one week of Dnmt1 ablation . Overall , these data suggest a mechanism in which Dnmt3b is activated to counteract the loss of Dnmt1 in the intestinal epithelium . We next aimed to test this proposed compensation mechanism using mouse genetics . To directly test the requirement for Dnmt3b in maintaining DNA methylation in the Dnmt1 mutant intestinal epithelium , we bred the Dnmt3bloxP/loxPallele onto the mutant genotype ( Lin et al . , 2006 ) , producing Dnmt1loxP/loxP;Dnmt3bloxP/loxP;Villin-CreERT2 , along with Dnmt1loxP/loxP;Dnmt3bloxP/loxP siblings as controls . To assess the overall requirement for Dnmt3b in Dnmt1 mutant survival , we injected tamoxifen into groups of Dnmt1loxP/loxP;Dnmt3bloxP/loxP;Villin-Cre-ERT2 mutants ( Dnmt1;Dnmt3b mutants ) and littermate controls at four weeks of age , and weighed mice each day following CreERT2 induction . 60% of Dnmt1;Dnmt3b mutant mice ( n=10 ) became severely morbid within two weeks following tamoxifen administration and had to be euthanized ( Figure 4A–C ) . Dnmt1;Dnmt3b mutants lost significantly more weight compared to Dnmt1 mutants , which contributed to the increased lethality observed in double mutant mice ( Figure 4B–C ) . In contrast , 94% of Dnmt1 mutant mice ( n=16 ) survived to 17 days following tamoxifen injection , confirming that loss of Dnmt1 alone is non-lethal in the mature intestinal epithelium ( Figure 4A ) . Dnmt1;Dnmt3b mutant mice that survived contained intestinal epithelium positive for Dnmt3b ( Figure 4—figure supplement 1A–B ) resulting from inefficient Cre-mediated gene ablation followed by expansion of “escaper” crypts , consistent with our hypothesis that Dnmt3b is required to preserve epithelial integrity in the absence of Dnmt1 . 10 . 7554/eLife . 12975 . 012Figure 4 . Loss of both Dnmt1 and Dnmt3b in the intestinal epithelium results in decreased survival . ( A ) Percent survival of Dnmt1loxP/loxP;Dnmt3bloxP/loxP ( control , n=8 ) , Dnmt1loxP/loxP;Villin-CreERT2 ( Dnmt1 mutant , n=16 ) , and Dnmt1loxP/loxP;Dnmt3bloxP/loxP;Villin-CreERT2 ( Dnmt1;Dnmt3b mutant , n=10 ) . All mice were tamoxifen treated at four weeks of age , and followed 17 days to record weight and survival . Significantly fewer Dnmt1;Dnmt3b mutants survive , compared to both controls and Dnmt1-deficient mice . ***p<0 . 001 , Log-rank test . ( B ) Percent of initial weight each day following tamoxifen treatment in ( A ) . Gradual weight loss is observed in both Dnmt1 mutants ( white diamonds ) and Dnmt1;Dnmt3b mutants ( grey triangles ) . ( C ) Statistical comparison of weight loss between controls , Dnmt1 mutants , and Dnmt1;Dnmt3b mutants . At day 10 post-tamoxifen treatment , both mutant groups have lost a significant amount of weight relative to controls . Dnmt3b;Dnmt1 mutants have also lost significantly more weight relative to Dnmt1 mutants . At day 15 , both mutant genotypes weigh significantly less than controls . **p<0 . 01 , ***p<0 . 001 , one-way ANOVA . For all data and p-values , refer to Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12975 . 01210 . 7554/eLife . 12975 . 013Figure 4—source data 1 . Contains mouse weight and survival data analysis in Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 12975 . 01310 . 7554/eLife . 12975 . 014Figure 4—figure supplement 1 . Dnmt1;Dnmt3b mutant intestinal epithelia contain Dnmt3b+ escaper crypts , which do not display DNA damage . Intestines were harvested from mutant Dnmt1loxP/loxP;Dnmt3bloxP/loxP;Villin-CreERT2 ( Dnmt1;Dnmt3b mutant ) and sibling Dnmt1loxP/loxP;Dnmt3bloxP/loxPcontrols one week post-tamoxifen treatment . Immunofluorescent staining was performed for Dnmt3b ( red ) , γH2AX ( green ) to visualize DNA double strand breaks , and DAPI ( blue ) to stain nuclei . The majority of mutant crypts are null for Dnmt3b relative to control ( yellow arrows in B pointing out Dnmt3b-null crypts ) . However , some mutants contain Dnmt3b-positive crypts , which escaped Cre-loxP recombination ( white outline in B ) . The escaper crypts do not display elevated levels of DNA damage compared to the Dnmt3b-ablated epithelium ( yellow arrow points out γH2AX staining in a Dnmt3b-null crypt ) . Scale bars are 50 μm . For all staining , n=3 biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 12975 . 014 Next , we isolated small intestines from Dnmt1;Dnmt3b mutants and sibling controls one week after tamoxifen administration for DNA methylation and immunostaining analysis . We employed laser-capture microdissection to isolate one-week Dnmt1;Dnmt3b mutant and control crypt cells for DNA methylation analysis . As expected , Dnmt1;Dnmt3b mutants were significantly demethylated at the LINE1 loci compared to controls ( Figure 5A ) . Although the Dnmt1;Dnmt3b mutants displayed decreased methylation levels relative to Dnmt1 mutants , this difference was not statistically significant across the entire region ( Figure 5A ) . The H19 imprinting control region was slightly demethylated in compared Dnmt1;Dnmt3b mutants to controls , but was only significant when comparing the entire sequenced region ( Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 12975 . 015Figure 5 . Simultaneous loss of Dnmt1 and Dnmt3b results in acute genomic instability , increased apoptosis , and genome demethylation . ( A ) Dnmt1loxP/loxP;Dnmt3bloxP/loxP ( control ) , Dnmt1loxP/loxP;Villin-CreERT2 ( Dnmt1 mutants ) and Dnmt1loxP/loxP;Dnmt3bloxP/loxP;Villin-CreERT2 ( Dnmt1;Dnmt3b mutant ) intestines were harvested one week following tamoxifen treatment for DNA methylation analysis . Crypts were isolated from paraffin-embedded tissue by laser capture microdissection , and the methylation levels of LINE1 loci were determined by targeted bisulfite sequencing . LINE1 methylation is significantly decreased in both Dnmt1 and Dnmt1;Dnmt3b mutants compared ( n=4–6 per genotype ) . Error bars are S . E . M . **p<0 . 01 , one-way ANOVA . For data and p-values per CpG , refer to Figure 5—source data 1 . ( B–C ) Hematoxylin and eosin staining of Dnmt1;Dnmt3b mutants ( C ) compared to controls ( B ) . Double mutants display severe crypt and villus loss compared to controls . ( D-G ) Immunostaining confirms Dnmt1 and Dnmt3b protein loss in the Dnmt1;Dnmt3b mutant intestinal epithelium ( E , G ) compared to control ( D , F ) . ( H , I ) Immunofluorescent staining for Ki67 ( red ) , which marks proliferating cells , and γH2AX ( green ) , which marks DNA double-strand break loci as a marker of genome instability . One week following tamoxifen injection , Dnmt1;Dnmt3b mutants ( I ) display decreased proliferation and increased DNA damage compared to controls ( H ) . ( J , K ) Immunofluorescent TUNEL staining ( red ) , which marks apoptotic nuclei , and E-cadherin ( green ) , to outline the intestinal epithelium . One week after tamoxifen treatment , Dnmt1;Dnmt3b mutants ( K ) display increased crypt cell apoptosis relative to controls ( J ) . All scale bars are 50 μm . For all staining , n=3 biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 12975 . 01510 . 7554/eLife . 12975 . 016Figure 5—source data 1 . Contains targeted bisulfite sequencing data shown in Figure 5A . DOI: http://dx . doi . org/10 . 7554/eLife . 12975 . 01610 . 7554/eLife . 12975 . 017Figure 5—source data 2 . Contains targeted bisulfite sequencing data presented in Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12975 . 01710 . 7554/eLife . 12975 . 018Figure 5—source data 3 . Contains targeted bisulfite sequencing data shown in Figure 5—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 12975 . 01810 . 7554/eLife . 12975 . 019Figure 5—figure supplement 1 . Ablation of Dnmt3b and Dnmt1 induces genome demethylation at H19 loci . ( A ) Dnmt1loxP/loxP;Dnmt3bloxP/loxP ( control ) , and Dnmt1loxP/loxP;Dnmt3bloxP/loxP;Villin-CreERT2 ( Dnmt1;Dnmt3b mutant ) intestines were harvested one week following tamoxifen treatment for DNA methylation analysis . Crypts were isolated from paraffin-embedded tissue by laser capture microdissection , and the methylation levels of the H19 imprinting control region was determined by targeted bisulfite sequencing . H19 methylation is significantly decreased in the Dnmt1;Dnmt3b mutants , relative to controls , when comparing averages of the entire sequenced region ( n=4–6 per genotype ) . Error bars are S . E . M . *p<0 . 05 , Student’s t-test . For data and p-values per each CpG , refer to Figure 5—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 12975 . 01910 . 7554/eLife . 12975 . 020Figure 5—figure supplement 2 . Dnmt3b deletion has no effect on epithelial proliferation , genome stability , or cell death within one week . ( A , B ) Dnmt1loxP/loxP; Dnmt3bloxP/loxP ( A , control ) and Dnmt1loxP/+;Dnmt3bloxP/loxP;Villin-CreERT2 ( B , Dnmt3b mutant ) intestinal epithelium one week after tamoxifen injection . Hematoxylin and Eosin staining demonstrates no difference in epithelial integrity in the Dnmt3b mutant ( B ) versus control ( A ) . ( C , D ) Immunofluoresence for Dnmt3b confirms loss of protein in tamoxifen-treated Dnmt3b mutant ( D ) compared to control ( C ) . ( E , F ) Co-staining for Ki67 ( red ) , a marker of proliferation , and γH2AX ( green ) , which marks DNA double-stranded breaks . Both control ( E ) and Dnmt3b-deficient ( F ) epithelia have minimal γH2AX foci and display no difference in proliferation . ( G , H ) TUNEL staining to detect apoptosis ( red ) with E-cadherin ( green ) to outline the epithelium . No apoptotic nuclei is found in control ( G ) or Dnmt3b mutant ( H ) crypt cells . All scale bars are 50 μm . For all staining , n=3 biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 12975 . 02010 . 7554/eLife . 12975 . 021Figure 5—figure supplement 3 . Ablation of Dnmt3b alone is not sufficient to induce genome demethylation . Intestines from Dnmt3bloxP/loxP ( control , n=4 ) and Dnmt3bloxP/loxPVillin-CreERT2 ( Dnmt3b mutant , n=2 ) mice were harvested one week after tamoxifen treatment . Crypt cells were isolated by laser capture microdissection and the methylation levels of LINE1 loci and the imprinting control region of H19 were determined by targeted bisulfite sequencing . ( A , B ) Methylation of LINE1 ( A ) and H19 ( B ) are similar in Dnmt3b mutants compared to controls . In all graphs , error bars are S . E . M . For data and p-values per CpG , refer to Figure 5—source data 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 12975 . 021 Histological examination revealed a grossly abnormal epithelium in Dnmt1;Dnmt3b mutantmice , with many areas lacking villi and/or crypts completely ( Figure 5B–C ) . We performed immunohistochemistry for Dnmt1 and Dnmt3b to confirm loss of both proteins in the majority of the epithelium ( Figure 5D–G ) . Differentiation in the double mutant intestine was largely unaffected , and exhibited normal distribution of goblet cells , Paneth cells , and enterocytes ( data not shown ) . Unlike the situation in Dnmt1 mutants , many crypts in double Dnmt1;Dnmt3b mutants were completely Ki67-negative , and harbored extensive DNA damage as indicated by γH2AX foci ( Figure 5H–I ) . TUNEL staining revealed increased crypt cell apoptosis in Dnmt1;Dnmt3b mutants compared to sibling controls ( Figure 5J–K ) . Overall , the Dnmt1;Dnmt3b double mutants displayed increased phenotypic severity compared to Dnmt1 single mutants , characterized by hypomethylation , DNA damage , and cell death . To confirm that Dnmt3b ablation alone does not replicate the Dnmt1;Dnmt3b double mutant phenotype , we also analyzed the intestine of Dnmt1loxP/+;Dnmt3bloxP/loxP;Villin-CreERT2 ( Dnmt3b mutant ) mice . Dnmt3b deletion was confirmed by immunoflourescent staining , and we proceeded to further histological analysis ( Figure 5—figure supplement 2 ) . Loss of Dnmt3b had no affect on intestinal crypt-villus architecture ( Figure 5—figure supplement 2A–B ) , and immunostaining for cell proliferation , DNA damage , and apoptosis was similar to controls ( Figure 5—figure supplement 2E–H ) . Furthermore , methylation of both the LINE1 repetitive loci and the H19 imprinting control region in Dnmt3b mutant crypt cells was equivalent to controls ( Figure 5—figure supplement 3A–B ) . These results demonstrate that Dnmt3b alone is not required for intestinal homeostasis . DNA methylation has been linked to genomic instability in multiple contexts , in both cell lines and in disease . Dnmt1 hypomorphic mice exhibit increased chromosomal duplications and rearrangements , and develop invasive T-cell lymphomas at approximately four months of age ( Gaudet et al . , 2003 ) . It is important to note that these effects are not restricted to Dnmt1 deficiency . Loss of Dnmt3b also induces hypomethylation and chromosomal instability in mouse embryonic fibroblasts ( Dodge et al . , 2005 ) , suggesting that both Dnmt1 and Dnmt3b are crucial for maintaining DNA methylation and preserving genome integrity . In some cases , de novo methyltransferases are essential for methylation of certain elements or enhancers , and cannot be compensated for by Dnmt1 . For example , hematopoietic stem cells ( HSCs ) require Dnmt3a for normal self-renewal and differentiation processes ( Challen et al . , 2012 ) . Loss of Dnmt3a in HSCs causes demethylation at essential stem cell genes , inducing hyper-proliferation and reducing differentiation rates ( Challen et al . , 2012 ) . Dnmt3b also contributes to silencing of germline genes in somatic cells ( Velasco et al . , 2010 ) , and maintenance methylation in ES cells ( Chen et al . , 2003; Liang et al . , 2002 ) . Our work adds to this body of evidence that implicates a crucial role for DNA methylation in maintaining genome stability . Given that methylation of LINE elements is significantly reduced , it is tempting to speculate that reactivation of retrotransposition might be a contributing factor to genome instability . Aberrant DNA methylation and genome instability correlate in a number of gastrointestinal pathologies , including inflammatory bowl disease ( IBD ) and colitis-associated cancer ( Hartnett and Egan , 2012 ) . The phenotype presented in our Dnmt1;Dnmt3b double mutant mice is reminiscent of mouse models of IBD , in which altered epithelial barrier function leads to increased immune cell recruitment and chronic inflammation in the gastrointestinal tract ( Wirtz and Neurath , 2007 ) . Indeed , a recent study in zebrafish demonstrated that loss of DNA methylation at the tumor necrosis factor alpha ( tnf-a ) promoter prompted increased tnf-a expression in the gut epithelium , leading to elevated apoptosis , barrier dysfunction , and immune cell localization ( Marjoram et al . , 2015 ) . In a chemically induced mouse model of IBD , inhibition of DNA methylation aggravated the inflammatory response , suggesting DNA methylation acts to protect against inflammation and IBD ( Kominsky et al . , 2011 ) . Our data support the hypothesis that DNA methylation supports intestinal epithelial homeostasis and helps to maintain crypt architecture . Current dogma holds that deletion of Dnmt1 is lethal in all dividing somatic cells ( Liao et al . , 2015 ) ; conversely , we find that the rapidly dividing intestinal epithelium can survive without Dnmt1 . Following acute loss of Dnmt1 , Dnmt3b expression is induced , and methylation of repetitive elements is restored . However , if Dnmt3b is ablated concurrently with Dnmt1 , restoration of DNA methylation is prevented , resulting in massive DNA damage , and cell death . It is important to note that our data does not provide direct evidence for methyltransferase activity of Dnmt3b at hemimethylated CpG sites , which are targeted by Dnmt1 . Indeed , it is also possible that Dnmt3b is compensating for the loss of Dnmt1 via repeated cycles of de novo methylation , in which Dnmt3b remethylates regions that are demethylated following Dnmt1 deletion after each cycle of DNA replication . Interestingly , we find that Dnmt3a cannot compensate adequately for the loss of both Dnmt1 and Dnmt3b , nor is Dnmt3a required in the absence of Dnmt1 . Indeed , we see that the pressure to maintain Dnmt1 and/or Dnmt3b expression is so high that Dnmt3b+ escaper cells proliferate in excess to recover the intestinal epithelium in a small subset of mutants ( Figure 4—figure supplement 1 ) . Although the basal levels of Dnmt3a mRNA in the control intestinal epithelium are much higher than those of Dnmt3b or Dnmt1 ( Figure 3A ) , Dnmt3a does not appear to be necessary to maintain intestinal homeostasis or DNA methylation patterns . This may be due to a number of factors , such as differing abilities of the methyltransferases to interact with cofactors and epigenetic complexes . Both Dnmt1 and Dnmt3b interact with the polycomb group repression complexes ( PRC1 and PRC2 ) , and regulate distinct sites in colorectal cancer development ( Jin et al . , 2009 ) . Furthermore , different methyltransferases are required in distinct ways in certain tissue and cell types . For example , in the hematopoietic system , loss of Dnmt3a leads to elevated proliferation rates and cancer development , while loss of Dnmt3b has little to no effect on hematopoietic stem cell function ( Challen et al . , 2012; 2014 ) . In conclusion , we show that Dnmt1 and Dnmt3b cooperate to maintain methylation in the adult mouse intestinal epithelium . Loss of Dnmt1 results in short-term hypomethylation , genomic instability , and apoptosis , followed by recovery over time . Dnmt3b is upregulated in response to deletion of Dnmt1 in the adult intestine , and is required to recover DNA methylation and epithelial integrity . These results provide the first example of a rapidly dividing somatic tissue that can survive in the absence of Dnmt1 , and suggest that the strict division of the Dnmt enzymes into ‘de novo’ and ‘maintenance’ methyltransferases might not fully represent the situation in vivo . Dnmt1loxP/loxP and Dnmt3bloxP/loxP mice were provided by Rudolf Jaenisch ( Jackson-Grusby et al . , 2001; Lin et al . , 2006 ) . Dnmt3aloxP/loxP mice were provided by En Li ( Kaneda et al . , 2004 ) , and Villin-CreERT2 mice were received from Sylvie Robine ( El Marjou et al . , 2004 ) . For Dnmt1 and Dnmt3b deletion experiments , Cre-recombination was induced by three daily intraperitoneal injections of 1 . 6 mg tamoxifen ( Sigma-Aldrich , St . Louis , MO ) in an ethanol/sunflower oil mixture . In all experiments , littermate controls without the Villin-CreERT2 transgene were also tamoxifen treated . All procedures involving mice were conducted in accordance with approved Institutional Animal Care and Use Committee protocols . Tissues were isolated and fixed using 4% paraformaldehyde in PBS and then embedded in paraffin . Antigen retrieval was performed using the 2100 Antigen-Retriever in Buffer A ( Electron Microscopy Sciences , Hatfield , PA ) and standard immunostaining procedures were performed for Dnmt1 ( Santa Cruz ) , Dnmt3a ( Santa Cruz Biotechnology , Dallas , TX ) , Dnmt3b ( Imgenex , San Diego , CA ) , E-Cadherin ( BD Biosciences , San Jose , CA ) , Ki67 ( BD Biosciences ) , and γH2AX ( Cell Signaling Technology , Berverly , MA ) . TUNEL staining was performed using TUNEL Label and Enzyme ( Roche , Indianapolis , IN ) and AlexaFluor 555-aha-dUTP ( Molecular Probes , Eugene , OR ) . All microscopy was performed on a Nikon Eclipse 80i ( Tokyo , Japan ) . For all immunofluorescence and immunohistochemistry staining , n=3 biological replicates per genotype , per timepoint . Crypt cell DNA was collected using a Leica LMD7000 Laser Microdissection microscope ( Wetzlar , Germany ) and the Arcturus PicoPure DNA isolation kit ( Applied Biosystems , Carlsbad , CA ) . Intestines were gently scraped to remove villi , and treated with EDTA to isolate crypt cells . RNA was extracted using the Trizol RNA isolation protocol ( Invitrogen , Carlsbad , CA ) , followed by RNA cleanup using the RNeasy Mini Kit ( Qiagen , Hilden , Germany ) . mRNA expression was determined using quantitative RT-PCR , as described previously ( Gupta et al . , 2007 ) . The SYBR green qPCR master mix ( Agilent , Santa Clara , CA ) was used in all qPCR reactions , and the fold change was calculated relative to the geometric mean of Tbp and β-Actin , using the △CT method . The method of normalizing to the geometric mean of a set of reference genes has been described previously ( Vandesompele et al . , 2002 ) . Primer sets can be found in Table 1 . 10 . 7554/eLife . 12975 . 022Table 1 . qRT-PCR primer sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 12975 . 022GeneForward 5'-3'Reverse 5'-3'Beta-actinGAAGTGTGACGTTGACATCCGGTCAGCAATGCCTGGGTACATTBPCCCCTTGTACCCTTCACCAATGAAGCTGCGGTACAATTCCAGDnmt1CTTCACCTAGTTCCGTGGCTACCCTCTTCCGACTCTTCCTTDnmt3aGCACCAGGGAAAGATCATGTCAATGGAGAGGTCATTGCAGDnmt3bGGATGTTCGAGAATGTTGTGGGTGAGCAGCAGACACCTTGA 100 ng of mouse genomic DNA was bisulfite converted using the Epitect bisulfite kit ( Qiagen ) . Template DNA was amplified using KAPA HIFI Uracel+ ( KAPA Biosystems , Wilmington , MA ) with primers directed to the LINE1 and H19 regions ( Table 2 ) . Sequencing libraries were prepared and analyzed using the BiSPCR2 strategy , described previously ( Bernstein et al . , 2015 ) . 10 . 7554/eLife . 12975 . 023Table 2 . Bisulfite sequencing primer sets . DOI: http://dx . doi . org/10 . 7554/eLife . 12975 . 023GeneSequence ( 5'-3' ) H19 PCR#1 ForwardACACTCTTTCCCTACACGACGCTCTTCCGATCTGTTTGTTGAATTAGTTGTGGGGTTTATAH19 PCR#1 ReverseGTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTTAAAAAAAAAAACTCAATCAATTACAATCCLINE1 PCR#1 ForwardACACTCTTTCCCTACACGACGCTCTTCCGATCTGTTAGAGAATTTGATAGTTTTTGGAATAGGLINE1 PCR#1 ReverseGTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTCCAAAACAAAACCTTTCTCAAACACTATATUnique Barcode PCR#2 ForwardAATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACUnique Barcode PCR#2 ReverseCAAGCAGAAGACGGCATACGAGATCGTGATGTGACTGGAGTTCAGACGTGTRed text , Illumina adapter sequence; blue text , Unique Illumina Sequencing Barcodes , 1–48 . Where indicated , GraphPad Prism 6 ( La Jolla , CA ) was employed to calculate log-rank test and ANOVA statistics .

Genes in a cell can be switched on or off at different times depending on the cell’s requirements . Small chemical groups can be attached to the gene’s DNA , which dictates whether it is activated or inactivated . For example , methyl groups can be attached to DNA by enzymes known as DNA methytransferases in a process called DNA methylation . In mammals , the pattern of DNA methylation changes in a highly regulated manner as the embryo develops . Mice that lack enzymes called DNA methytransferase 1 or DNA methytransferase 3b ( shortened to Dnmt1 or Dnmt3b ) die before they are born . It is also widely believed that Dnmt1 is needed to preserve DNA methylation patterns in dividing cells , and this enzyme is often called a ‘maintenance’ methyltransferase because it maintains the DNA methylation pattern in newly formed cells . Epithelial cells that form an animal’s intestine are constantly being produced from dividing cells found in folds of the intestine called crypts . These cells are among the most rapidly dividing cells in animals , and are replaced every three to five days throughout adult life . Recently in 2015 , researchers deleted the gene for Dnmt1 in intestinal cells in adult mice . This caused a reduction in DNA methylation and led to abnormal gene activation . However , the crypt cells were still able to form new cells to renew the intestine . Elliott , Sheaffer and Kaestner – who were all involved in the previous study – have now explored why intestinal cells in adult mice can survive without Dnmt1 . First , the experiments showed that the mutant mice recovered their normal levels DNA methylation within two months of the gene deletion . Further analysis uncovered that this recovery was due to the fact that Dnmt3b became activated in Dnmt1-lacking cells , and then re-methylated the DNA . Deleting the genes for both Dnmt1 and Dnmt3b led to loss of DNA methylation and many of the adult mice died prematurely . Previously , it was thought that Dnmt3b only acted to establish new DNA methylation patterns , but these latest findings suggest that this enzyme can act a ‘maintenance’ methyltransferase as well . Together the findings also reveal that Dnmt1 and Dnmt3b cooperate to maintain methylation in the epithelial cells in the intestines of adult mice . Further work could next investigate if this is also the case for other tissues in the body .

2016

The ‘de novo’ DNA methyltransferase Dnmt3b compensates the Dnmt1-deficient intestinal epithelium

Acid-base conditions modify artery tone and tissue perfusion but the involved vascular-sensing mechanisms and disease consequences remain unclear . We experimentally investigated transgenic mice and performed genetic studies in a UK-based human cohort . We show that endothelial cells express the putative HCO3–-sensor receptor-type tyrosine-protein phosphatase RPTPγ , which enhances endothelial intracellular Ca2+-responses in resistance arteries and facilitates endothelium-dependent vasorelaxation only when CO2/HCO3– is present . Consistent with waning RPTPγ-dependent vasorelaxation at low [HCO3–] , RPTPγ limits increases in cerebral perfusion during neuronal activity and augments decreases in cerebral perfusion during hyperventilation . RPTPγ does not influence resting blood pressure but amplifies hyperventilation-induced blood pressure elevations . Loss-of-function variants in PTPRG , encoding RPTPγ , are associated with increased risk of cerebral infarction , heart attack , and reduced cardiac ejection fraction . We conclude that PTPRG is an ischemia susceptibility locus; and RPTPγ-dependent sensing of HCO3– adjusts endothelium-mediated vasorelaxation , microvascular perfusion , and blood pressure during acid-base disturbances and altered tissue metabolism . Inadequate tissue perfusion relative to metabolic demand is a fundamental pathophysiological cause of acute ( e . g . myocardial infarction and stroke ) and chronic ( e . g . heart failure and neurodegenerative disorders ) cardiovascular disease . Newer treatment options improve the prognosis for some of these debilitating conditions but we need alternative therapeutic strategies in order to minimize ischemia-related morbidity and mortality . Sensing of the chemical environment in the wall of arteries coordinates local blood flow to meet the oxidative metabolic demand and counteract ischemia . Despite their obvious clinical importance , the cellular and molecular mechanisms responsible for sensing metabolic disturbances and responding to inadequate perfusion are not well understood ( Boedtkjer , 2018 ) . Local acidification is an important signal of insufficient nutrient delivery and waste product elimination . Although functional effects of H+ in the vascular wall have long been appreciated ( Boedtkjer , 2018; Gaskell , 1880; Boedtkjer et al . , 2016a; Boedtkjer and Aalkjaer , 2012; Boedtkjer and Aalkjaer , 2013a; Boedtkjer and Aalkjaer , 2013b ) , the consequences of associated changes in CO2/HCO3– buffer composition have received far less attention . CO2 and HCO3– constitute an important buffer pair that minimizes acute changes in pH ( Roos and Boron , 1981 ) , improves spatial H+ mobility ( Spitzer et al . , 2002; Boedtkjer et al . , 2016b ) , and provides substrate for acid-base transporters in cell membranes ( Aalkjaer and Hughes , 1991; Boedtkjer et al . , 2006 ) . Evidence supports that cells also possess sensors that respond to changes in [HCO3–] in the intracellular and extracellular space ( Zhou et al . , 2016; Boedtkjer et al . , 2016c; Chen et al . , 2000 ) . Interventions—such as NaHCO3 supplementation—that modify systemic acid-base status are available ( Voss et al . , 2020 ) but expected to carry considerable adverse effects and be too general for cardiovascular therapy . The identification of HCO3–-related proteins ( e . g . carbonic anhydrases , Na+ , HCO3–-cotransporters , Cl–/HCO3–-exchangers , and HCO3–-sensors ) opens largely unexplored avenues for pharmacological treatment . Because the vasculature of vital organs—particularly the brain and heart—responds to metabolic deregulation and is sensitive to acid-base disturbances , it is likely that therapy targeting HCO3–-related proteins will be able to modify myocardial and cerebral perfusion preferentially in areas of unmet metabolic demand . Through mechanisms that require the transmembrane Receptor Protein Tyrosine Phosphatase ( RPTP ) γ , isolated decreases in extracellular [HCO3–] enhance HCO3– reabsorption in the renal proximal tubule ( Zhou et al . , 2016 ) and contractions of basilar arteries ( Boedtkjer et al . , 2016c ) when pCO2 and pH are kept constant using out-of-equilibrium technology ( Zhao et al . , 1995 ) . The extracellular domain of RPTPγ resembles the active site of the carbonic anhydrases ( Barnea et al . , 1993 ) and therefore likely binds HCO3– . However , the carbonic anhydrase-like domain of RPTPγ lacks the histidine residues considered essential for catalyzing equilibration of the reaction: CO2+H2O⇄HCO3−+H+ ( Zhou et al . , 2016 ) . Instead , HCO3– may induce a dimerization-dependent auto-inhibitory response in RPTPγ and thereby regulate signaling via the intracellular tyrosine phosphatase domains ( Barr et al . , 2009 ) . Apart from these structural observations , we do not currently understand the cellular and molecular mechanisms for impact of HCO3– and RPTPγ on resistance artery function , blood pressure , and control of tissue perfusion . Although CO2 has no direct net effect on cerebrovascular tone ( Boedtkjer et al . , 2016c ) , respiratory changes leading to hyper- or hypocapnia will lead to secondary rises or falls in [HCO3–] and pH that in turn influence artery contractions and tissue perfusion . In the present study , we demonstrate—based on experimental investigations in transgenic mice—that RPTPγ ( a ) facilitates endothelium-dependent relaxation of resistance arteries through mechanisms regulated by extracellular HCO3– and ( b ) adjusts microvascular perfusion and blood pressure during increased tissue metabolism and acid-base disturbances . Through translational genetic studies in the UK Biobank cohort ( Bycroft et al . , 2018 ) , we further substantiate the importance of the identified mechanisms for human cardiovascular health as we demonstrate that predicted loss-of-function variants in PTPRG , encoding RPTPγ , are associated with human ischemic vascular disease in the heart and brain . We first evaluated promoter activity for Ptprg by β-galactosidase staining of mice with a promoterless LacZ insert under transcriptional control of the Ptprg promoter . We found signs of prominent Ptprg transcriptional activity in the endothelium of mouse basilar , middle cerebral , and coronary arteries ( Figure 1A ) . We also saw signs of Ptprg transcriptional activity in pulmonary and skeletal muscle arteries , whereas transcription of Ptprg appeared lower in the aorta and mesenteric arteries ( Figure 1A ) . Since promoter activity is no direct measure of expression—and cytosolic dilution of the chromogenic reaction product lowers the sensitivity of the β-galactosidase reporter assay—we followed up with quantitative RT-PCR analyses that confirmed substantial steady-state levels of Ptprg mRNA in basilar , middle cerebral , coronary , and pulmonary arteries and identified clearly detectable levels in the aorta , mesenteric , and skeletal muscle arteries ( Figure 1B ) . We next explored the endothelial expression pattern of Ptprg based on single-cell transcriptomic data from healthy mice ( Kalucka et al . , 2020 ) . These data verify Ptprg expression in the vascular endothelium of the brain , heart , lung , skeletal muscle , and small intestine ( Figure 1C , D ) . The data further show that the endothelial Ptprg expression extends from the arterial vasculature , across capillaries , and into the veins ( Figure 1D , E ) . Taken together , we identify widespread endothelial Ptprg expression with prominent levels in the cerebral and coronary vasculature where metabolites , including acid-base equivalents , strongly influence arterial tone ( Gaskell , 1880; Boedtkjer et al . , 2016c; Wahl et al . , 1970 ) . Isolated small arteries pre-contracted with the thromboxane A2 analog U46619 relaxed concentration-dependently in response to the classical endothelium-dependent agonist acetylcholine ( Figure 2 ) . In order to evaluate a distinct additional endothelial signaling pathway ( D'Andrea et al . , 1998 ) capable of inducing potent endothelium-dependent vasorelaxation of cerebral and mesenteric small arteries ( Boedtkjer et al . , 2013a; McNeish et al . , 2005; Bucci et al . , 2005 ) , we also tested the effect of the proteinase-activated receptor ( PAR ) 2 agonist SLIGRL-amide ( Figure 3 ) . Under physiological conditions with CO2/HCO3– present in the buffer solutions , acetylcholine- and SLIGRL-amide-induced vasorelaxation of basilar ( Figures 2A–C and 3A–C ) , mesenteric ( Figures 2I and 3I ) , and coronary ( Figures 2M and 3L ) arteries from RPTPγ knockout ( KO ) mice was reduced compared to arteries from wild type ( WT ) mice . We observed this difference between arteries from RPTPγ KO and WT mice whether experiments in presence of CO2/HCO3– were conducted with ( Figures 2B and 3B ) or without ( Figures 2C and 3C ) the non-physiological buffer HEPES . Addition of an artificial buffer is necessary for pH control in absence of CO2/HCO3–; and we included HEPES also in the CO2/HCO3–-containing solutions in order to separate the vascular effects of CO2/HCO3– omission from potential effects of adding HEPES ( Altura et al . , 1980 ) . Acetylcholine-induced vasorelaxation of basilar arteries from RPTPγ KO mice was suppressed in presence of CO2/HCO3– mostly because of attenuated NO-dependent ( i . e . N-nitro-L-arginine methyl ester ( L-NAME ) -sensitive ) signaling ( Figure 2D–G ) . In contrast , the diminished vasorelaxation of basilar arteries in response to SLIGRL-amide ( Figure 3D–F ) and of mesenteric and coronary arteries in response to acetylcholine ( Figure 2J–L and Figure 2N–P ) and SLIGRL-amide ( Figure 3J , K , M , N ) was mostly explained by smaller endothelium-dependent hyperpolarization ( EDH ) -type responses that were L-NAME-insensitive but inhibited by apamin and TRAM-34 . Consistent with these findings , L-NAME-insensitive smooth muscle hyperpolarization was significantly attenuated in basilar arteries from RPTPγ KO mice when elicited by SLIGRL-amide ( Figure 3G , H ) but not when elicited by acetylcholine ( Figure 2H ) . In order to evaluate whether RPTPγ-dependent signaling is influenced by HCO3– in the vascular wall , we next repeated studies of endothelium-dependent vasorelaxation using nominally CO2/HCO3–-free , HEPES-buffered solutions ( Figure 4 ) . Under these conditions , we saw no RPTPγ-dependent differences in overall vasorelaxant function of basilar or mesenteric arteries in response to acetylcholine ( Figure 4A , B , G ) or SLIGRL-amide ( Figure 4K , L , P ) . The relative contribution of NO-dependent signaling and EDH-type responses to vasorelaxation was also not significantly affected by RPTPγ KO in basilar ( Figure 4C–F ) or mesenteric ( Figure 4H–J ) arteries in response to acetylcholine . Whereas the dependency on NO-signaling vs EDH-type vasorelaxation in response to SLIGRL-amide was likewise not affected in mesenteric arteries ( Figure 4Q , R ) , we saw indications of an apparent relative increase in NO-mediated vasorelaxation when basilar arteries from RPTPγ KO mice were stimulated with SLIGRL-amide in absence of CO2/HCO3– ( Figure 4M–O ) . This unexpected change in the relative importance of the underlying vasorelaxant mechanisms in basilar arteries could be a compensation secondary to the predominant inhibition of NO-dependent vasorelaxation in these arteries in presence of CO2/HCO3– ( Figure 2F ) and hence presumably in vivo . We next explored the mechanism whereby RPTPγ influences resistance artery function ( Figure 5 ) . We first evaluated whether the difference in vasorelaxation between arteries from RPTPγ KO and WT mice depends on a change in expression of the endothelial NO-synthase ( eNOS ) . Based on immunoblotting of basilar ( Figure 5A ) and mesenteric ( Figure 5D ) arteries , we found equal eNOS expression between arteries from RPTPγ KO and WT mice . We also evaluated whether the smooth muscle vasorelaxant responses to exogenous NO differ between arteries from WT and RPTPγ KO mice . In presence of CO2/HCO3– , spermine NONOate and S-nitroso-N-acetylpenicillamine ( SNAP ) produced equivalent vasorelaxation in basilar and mesenteric arteries from RPTPγ KO and WT mice ( Figure 5B , C , E , F ) . Together , our findings support that RPTPγ exerts acute influences on endothelial signaling pathways regulating the activity of endothelium-dependent vasorelaxation in response to [HCO3–] . This interpretation is in agreement with the near-complete normalization of vasorelaxation observed in arteries from RPTPγ KO mice upon omission of CO2/HCO3– ( Figure 4 ) . We then determined whether endothelial intracellular Ca2+ signals differ between arteries from RPTPγ KO and WT mice ( Figure 5G–N ) . Loading of isolated arteries with Ca2+-sensitive fluorophores through luminal perfusion ensured endothelium-specific fluorescence signals ( Figure 5G , K ) . In the presence of CO2/HCO3– , application of acetylcholine elevated intracellular [Ca2+] substantially more in endothelial cells of basilar and mesenteric arteries from WT mice than in arteries from RPTPγ KO mice ( Figure 5H , J , L , N ) . In contrast , we saw no significant difference between the intracellular Ca2+ responses of endothelial cells in arteries from WT and RPTPγ KO mice when they were investigated in the nominal absence of CO2/HCO3– ( Figure 5I , J , M , N ) . Reduced endothelial intracellular Ca2+ responses in arteries from RPTPγ KO mice in presence of CO2/HCO3– likely explain the observed mixed attenuation of NO-mediated and EDH-type vasorelaxation ( Figures 2 and 3 ) as these responses rely on the Ca2+-sensitive NO-synthase and on intermediate- ( IKCa ) and small-conductance ( SKCa ) Ca2+-activated K+-channels , respectively ( Figure 5O ) . Intracellular H+ and Ca2+ can compete for buffer binding ( Batlle et al . , 1993 ) , many Ca2+ handling proteins are pH-sensitive ( Boedtkjer and Aalkjaer , 2012 ) , and pH is an important regulator of NO-synthase activity ( Boedtkjer et al . , 2011 ) . Therefore , we next evaluated whether RPTPγ KO influences intracellular pH—measured using the pH-sensitive fluorophore BCECF—in endothelial cells of small arteries . In basilar ( Figure 5P ) as well as mesenteric ( Figure 5Q ) arteries , intracellular pH of the endothelial cells was similar for RPTPγ KO and WT mice , irrespective of whether experiments were performed with or without CO2/HCO3– . Omission of CO2/HCO3– from the bath solutions had no net effect on endothelial steady-state intracellular pH ( Figure 5P , Q ) , which is consistent with previous studies based on arteries from C57BL/6 mice ( Boedtkjer et al . , 2011; Boedtkjer et al . , 2012 ) . If the vasomotor impact of RPTPγ is sufficiently widespread to affect total peripheral resistance , we predicted that KO of RPTPγ would influence blood pressure regulation . We therefore measured systemic blood pressure and heart rate by radiotelemetry ( Figure 6A , B ) . Based on these recordings , we observed no difference in resting blood pressure or heart rate ( Figure 6A ) or in the blood pressure-elevating effect of L-NAME ( Figure 6B ) between RPTPγ KO and WT mice . As the lack of difference in resting blood pressure between WT and RPTPγ mice could be due to compensatory adaptations , we next studied acute blood pressure responses of endotracheally intubated , mechanically ventilated mice under capnographic control . When mice were ventilated to a normal expiratory end-tidal CO2 fraction ( FETCO2 ) of 3 . 5% , arterial blood gas parameters were very similar in RPTPγ KO and WT mice ( Table 1 ) consistent with previous findings ( Zhou et al . , 2016 ) . Hyperventilating mice until FETCO2 was lowered to 2% caused the expected decrease in PaCO2 and increase in pHa ( Table 1 ) . The shift of the chemical equilibrium CO2+H2O⇄HCO3−+H+ also resulted in the anticipated decrease in [HCO3–]a ( Table 1 ) . The mean arterial blood pressure of WT mice increased around 20% during hyperventilation and this blood pressure response was reduced by approximately 1/3 in RPTPγ KO mice ( Figure 6C ) . Although the observed hyperventilation-induced reduction in arterial [HCO3–] was of modest magnitude ( 1 . 4 mM , Table 1 ) , the associated increase in blood pressure is consistent with attenuated RPTPγ-dependent vasodilator influences when extracellular [HCO3–] decreases ( Boedtkjer et al . , 2016c ) . The remaining hyperventilation-induced blood pressure elevation likely results from a direct vasocontractile effect of alkalosis ( Boedtkjer et al . , 2016c ) . Metabolic waste products—that are generated at elevated rate and locally accumulate when neuronal activity increases—influence cerebral blood flow . We studied cerebral perfusion using laser speckle imaging ( Figure 6D–F ) in order to evaluate the involvement of RPTPγ in sensing of acid-base disturbances . Tissue perfusion in the somatosensory barrel cortex increased twice as much in RPTPγ KO mice compared to WT mice during whisker stimulation ( Figure 6E ) . This finding is consistent with metabolically produced H+ leading to a local decrease in [HCO3–] that is sensed via RPTPγ in arteries from WT mice ( Figure 5O ) . As RPTPγ-dependent vasorelaxation wanes at low [HCO3–] ( Boedtkjer et al . , 2016c ) , our data support that RPTPγ limits increases in cerebral perfusion induced by metabolic acidosis ( Figure 6E ) . Acid-base conditions during increased neuronal activity are complicated , as the increased metabolism will also induce some degree of CO2 accumulation that tends to increase [HCO3–] . In order to produce a simpler acid-base disturbance with predictable changes in extracellular [HCO3–] , we next hyperventilated mice , which reduced pCO2 , increased pH , and lowered [HCO3–] ( Table 1 ) . Under these conditions , cerebral vasoconstriction is expected in WT mice based on the combined effect of alkalosis and decreased [HCO3–] ( Boedtkjer et al . , 2016c ) . If RPTPγ is required for sensing of HCO3– , only the effect of the elevated pH is expected in the RPTPγ KO mice; and indeed , we saw a 30% smaller drop in cerebral perfusion in RPTPγ KO mice compared to WT mice during hyperventilation ( Figure 6F ) . These recordings may in fact underestimate the difference in cerebrovascular resistance between RPTPγ KO and WT mice since the cerebral perfusion pressure—based on the blood pressure changes shown in ( Figure 6C ) —increased less in hyperventilated RPTPγ KO than WT mice . When whisker stimulation was performed during hyperventilation , we saw an even more prominent difference in cerebral perfusion between WT and RPTPγ KO mice ( Figure 6F ) . Together , our findings demonstrate that RPTPγ plays a substantial role for control of blood pressure ( Figure 6C ) and cerebral perfusion ( Figure 6E , F ) during acid-base deviations where [HCO3–] decreases below the normal level . Based on human exome sequencing data from the UK Biobank , we next explored whether the functional role of RPTPγ in regulating endothelial function and responding to metabolic disturbances in murine cerebral , mesenteric , and coronary arteries is corroborated by and translates to an altered risk for human ischemic vascular disease amongst carriers of predicted loss-of-function variants within PTPRG . We identified 334 missense variants and 75 predicted loss-of-function variants ( categorized as low , moderate , and high impact ) in PTPRG among the ~50 , 000 UK Biobank participants with available exome sequencing data . The 72 carriers of predicted loss-of-function PTPRG variants with moderate or high impact showed substantially elevated risk ( Figure 7A ) of a combined vascular disease phenotype encompassing the three vascular beds functionally investigated in this study ( Figures 2–6 ) : cerebral infarct ( ICD-10 diagnosis code I63 ) , angina pectoris ( ICD-10 diagnosis code I20 ) , acute myocardial infarction ( ICD-10 diagnosis code I21 ) , and acute vascular disorders of the intestine ( ICD-10 diagnosis code K55 . 0 ) . In contrast , the 123 carriers of predicted loss-of-function PTPRG variants with low impact as well as the 29 , 975 carriers of one or more missense PTPRG variants had a risk similar to non-carriers for this aggregated phenotype of acute vascular disease ( Figure 7A ) . We followed-up by investigating the separate occurrences of diagnosed angina pectoris , acute myocardial infarction , and cerebral infarction amongst the individuals , who carried a predicted loss-of-function variant with moderate or high impact . The carriers of these most severe loss-of-function variants had dramatically increased risk ( ~7 fold ) of cerebral infarction whereas the risk of acute myocardial infarction showed a tendency toward elevation that did not reach statistical significance ( Figure 7B ) . The number of exome-sequenced UK Biobank participants with acute intestinal vascular disease was too low to provide statistical power for a separate analysis . To complement the evidence based on ICD-10 diagnosis codes , we further evaluated whether carriers of missense and loss-of-function variants in PTPRG differed in self-reported diseases compared to non-carriers ( Figure 7C ) . The risk of self-reported heart attacks was greatly elevated ( ~4 fold ) amongst individuals carrying loss-of-function PTPRG variants of moderate and high impact ( Figure 7C ) . The risk of self-reported stroke also showed a strong tendency toward elevation—with borderline statistical significance—amongst the carriers of moderate- and high-impact loss-of-function PTPRG variants ( Figure 7C ) . As cardiac pumping function often deteriorates following coronary ischemia , we next evaluated the cardiac contractile function in PTPRG loss-of-function carriers amongst the 12 , 851 exome-sequenced UK Biobank participants with available cardiac magnetic resonance imaging ( MRI ) data . We observed a significantly lower left ventricular ejection fraction in carriers of high-impact loss-of-function PTPRG variants compared to non-carriers ( Figure 7D ) consistent with the greater risk of ischemic heart disease ( Figure 7C ) . The statistical analyses linking PTPRG to risk of ischemic vascular disease ( Figure 7A–C ) and reduced left ventricular ejection fraction ( Figure 7D ) were all adjusted for sex , age , body mass index , genetic principal component , smoking status , dyslipidemia , diabetes , and hypertension between carriers and non-carriers of the evaluated PTPRG variants . Taken together , we demonstrate that RPTPγ ( a ) CO2/HCO3–-dependently enhances endothelial intracellular Ca2+ responses and endothelium-dependent vasorelaxation , ( b ) regulates microvascular perfusion and blood pressure during acid-base disturbances , and ( c ) is associated with human ischemic vascular disease of the brain and heart . Local acid-base-dependent mechanisms of arterial tone regulation match tissue perfusion to the oxidative metabolic demand and thereby dynamically control cardiovascular function during cycles of intermittent rest and activity . Re-establishing proper tissue perfusion is of obvious clinical importance when managing patients with latent or fulminant ischemia . Responses of coronary and cerebral arteries to acid-base disturbances have been recognized for almost 140 years ( Gaskell , 1880; Lassen , 1968 ) but this information has not yet been harnessed for therapeutic intervention for lack of identified molecular players and their interactions . In the current study , we identify PTPRG as a susceptibility locus for human ischemic vascular disease ( Figure 7 ) and provide mechanistic evidence that RPTPγ regulates endothelial intracellular Ca2+ responses ( Figure 5H , L ) , endothelium-dependent vasorelaxation ( Figures 2 and 3 ) , cerebral perfusion ( Figure 6E , F ) , and blood pressure ( Figure 6C ) during acid-base disturbances and increased metabolic demand . Variation in buffer composition accompanies changes in pH during acid-base disturbances . We show that the CO2/HCO3– buffer plays a hitherto unappreciated role for regulation of endothelial function and arterial tone through mechanisms that require RPTPγ ( Figures 2–4 ) . As schematically illustrated in Figure 5O , we demonstrate that RPTPγ enhances endothelial intracellular Ca2+ signals ( Figure 5G–N ) that in turn activate endothelial NO synthesis and EDH-type responses ( Figures 2 and 3 ) . These effects are in congruence with the Ca2+-sensitivity of eNOS and the SK and IK Ca2+-activated K+-channels . The vasorelaxant influence of RPTPγ wanes at low extracellular [HCO3–] ( Boedtkjer et al . , 2016c ) , which attenuates by half the elevation of cerebral perfusion during sensory input ( Figure 6E ) and amplifies by 30% the decrease in perfusion during hyperventilation ( Figure 6F ) . We also observe that RPTPγ is necessary for approximately 1/3 of the blood pressure increase during hyperventilation ( Table 1 and Figure 6C ) . Considering the prominent influence of RPTPγ on endothelial function in several vascular beds , somewhat surprisingly , we observe blood pressure consequences of RPTPγ KO only when we impose acute acid-base disturbances ( Figure 6A–C ) . The unaltered resting blood pressure in RPTPγ KO mice ( Figure 6A ) most likely reflects that numerous discrete mechanisms are involved in blood pressure control and that compensation—for example , through nervous , hormonal or renal influences ( Boedtkjer and Aalkjaer , 2013b ) —can maintain blood pressure in the sustained phase and mask hemodynamic consequences of RPTPγ except when acute acid-base disturbances are imposed . Previously recognized cellular acid-base sensors include G-protein coupled receptors ( e . g . OGR1 , GPR4 , and TDAG8 ) , ion channels ( e . g . ASIC , BKCa ) , and enzymes ( e . g . NO-synthase and rho-kinase ) sensitive to H+ ( Boedtkjer et al . , 2011; Boedtkjer et al . , 2012; Fleming et al . , 1994; Schubert et al . , 2001; Ludwig et al . , 2003; Wemmie et al . , 2006; Wenzel et al . , 2020 ) . In addition , we and others have described cellular functions responsive to changes in [HCO3–] and modified by RPTPγ ( Zhou et al . , 2016; Boedtkjer et al . , 2016c ) or the soluble adenylyl cyclase ( Chen et al . , 2000 ) . Evidence connecting acid-base sensors to human disease has so far been scarce; but in the current study , we identify PTPRG as an ischemia susceptibility locus ( Figure 7 ) , which is supported by a recent meta-analysis of genome-wide association studies linking PTPRG to ischemic stroke in African Americans ( Carty et al . , 2015 ) . Using out-of-equilibrium technology—that permits separate control of pH and the individual CO2/HCO3– buffer components—we previously demonstrated that RPTPγ responds to changes in extracellular [HCO3–] independently of pCO2 and pH ( Boedtkjer et al . , 2016c ) . The homology between the extracellular domain of RPTPγ and the active site of the carbonic anhydrases supports the ability of RPTPγ to bind HCO3– even though it lacks histidine residues required for carbonic anhydrase activity ( Zhou et al . , 2016 ) . The intracellular aspect of RPTPγ contains phosphatase domains with suggested auto-inhibitory activity; and HCO3– possibly alters the phosphatase activity by influencing the degree of RPTPγ dimerization ( Barr et al . , 2009 ) . The separate signaling effects of H+ and HCO3– are further confirmed by the similar steady-state intracellular pH of endothelial cells in arteries from WT and RPTPγ KO mice ( Figure 5P , Q ) . The current study adds to our growing appreciation that acid-base equivalents , and HCO3– in particular , fulfill multifaceted functions ( Boedtkjer et al . , 2016a ) . In cerebral resistance arteries , HCO3– ( a ) contributes to buffering of acute acid loads ( Rasmussen and Boedtkjer , 2018 ) , ( b ) serves as substrate particularly for the Na+ , HCO3–-cotransporter NBCn1 that protects against intracellular acidification ( Thomsen et al . , 2014 ) , and ( c ) is sensed by RPTPγ to regulate cerebral perfusion ( Figure 6 ) . In migrating vascular smooth muscle cells from conduit arteries , the high spatial mobility of the CO2/HCO3– buffer system also contributes to dissipating local pH gradients in diffusion-restricted spaces of filopodia ( Boedtkjer et al . , 2016b ) . Our current ( Figures 1–4 and 6 ) and previous ( Boedtkjer et al . , 2016c ) findings demonstrate that the endothelium of resistance arteries senses the local acid-base composition and has capacity to modify vascular resistance and perfusion in response to disturbances in extracellular [HCO3–] . Earlier studies of mouse arteries provide additional evidence that intracellular acidification inhibits endothelial NO-synthesis ( Boedtkjer et al . , 2011; Boedtkjer et al . , 2012; Thomsen et al . , 2014 ) , intracellular alkalinisation interferes with myo-endothelial current transfer required for EDH-type responses ( Boedtkjer et al . , 2013a ) , and acid-base disturbances modify prostanoid-mediated endothelium-dependent vasocontraction ( Baretella et al . , 2014 ) . Together , these studies highlight the sophisticated nature of vasomotor control that integrate the requirement for local blood flow and the necessity to minimize deviations in capillary pressure and fluid filtration during local metabolic disturbances , perturbed nutrient delivery , and restricted waste product elimination . Acting as a brake on vasodilation in regions of unmet metabolic demand , RPTPγ could reduce the degree of edema and consequent tissue damage caused by unopposed H+-induced vasodilation ( Boedtkjer , 2018; Boedtkjer et al . , 2016a ) . As illustrated in Figure 1D and E , Ptprg mRNA is found in arteriolar , capillary , as well as venular endothelial cells ( Kalucka et al . , 2020; Vanlandewijck et al . , 2018 ) , and it remains a future task to evaluate physiological and pathophysiological roles of RPTPγ in these individual blood vessel segments . Transcript levels for Ptprg appear particularly high in a smaller subset of endothelial cells ( up to 20–40% , depending on the vascular bed; Figure 1C–E ) , which raises the intriguing possibility that a distinct population of endothelial cells act as the primary sensors of the metabolic environment . Hyperventilation-induced cerebral vasoconstriction can cause syncope but is also a therapeutic tool for rapidly reducing cerebral blood flow , capillary filtration , and intracranial pressure . Hypocapnia during hyperventilation leads to decreased [HCO3–] and alkalosis ( Table 1 ) ; and the current study demonstrates that RPTPγ is necessary for approximately 30% of the associated decrease in cerebral perfusion ( Figure 6F ) . In conclusion , the single-pass transmembrane protein RPTPγ enhances endothelium-dependent vasorelaxation of resistance arteries—by amplifying endothelial intracellular Ca2+ responses , NO synthesis , and EDH-type responses—and adjusts cerebral perfusion during increased neuronal activity and acid-base disturbances . Although RPTPγ shows no major impact on resting blood pressure , it contributes markedly to the blood pressure increase observed during hyperventilation . Supporting the translational value of our findings , predicted loss-of-function variants in PTPRG are associated with human ischemic vascular disease in the brain and heart . The influence of RPTPγ on vascular resistance specifically in tissues with disturbed acid-base composition makes RPTPγ a promising focus for targeted therapeutic intervention against ischemia . Male RPTPγ KO and matched WT mice ( >8 weeks old ) deeply anesthetized by intraperitoneal injection of pentobarbital were sacrificed by exsanguination . Approximately 2 mm long segments of basilar , mesenteric , and coronary septal arteries were dissected under a stereomicroscope , mounted in four-channel wire myographs ( 610M; DMT , Denmark ) for isometric investigation , heated to 37°C , and normalized to 90% of the internal diameter corresponding to a transmural pressure of 100 mmHg ( Mulvany and Halpern , 1977 ) . The myograph chambers were aerated with 5% CO2/balance air ( for CO2/HCO3–-containing physiological saline solutions ) or nominally CO2-free air ( for CO2/HCO3–-free solutions ) . The CO2/HCO3–-containing solution consisted of ( in mM ) : 119 NaCl , 22 NaHCO3 , 10 HEPES , 1 . 2 MgSO4 , 2 . 82 KCl , 5 . 5 glucose , 1 . 18 KH2PO4 , 0 . 03 EDTA , 1 . 6 CaCl2 . The CO2/HCO3–-free solution was produced by substituting NaHCO3 with NaCl . HEPES is necessary for pH control in the CO2/HCO3–-free solution; and we included HEPES also in the CO2/HCO3–-containing solution in order to separate vascular effects of CO2/HCO3– omission from potential effects of adding HEPES ( Altura et al . , 1980 ) . For the control experiments in Figures 2C and 3C , we used a CO2/HCO3–-containing solution without HEPES , i . e . , where the 10 mM HEPES was substituted with 5 mM NaCl . After all contents were added , the solutions were vigorously bobbled with the appropriate CO2-containing or nominally CO2-free gas mixture at 37°C before pH was adjusted to 7 . 40 while allowing ample time for buffer equilibration . Arterial force was acquired using a PowerLab 4/25 recorder and LabChart 7 Pro software ( RRID:SCR_001620; ADInstruments , New Zealand ) . As standard warm-up protocol , arteries were exposed twice to 3 µM of the thromboxane A2 analog U46619 for 2 min , and arteries that produced less than 1 mN force were excluded from further analysis . Vasorelaxation in response to acetylcholine , the PAR2 agonist SLIGRL-amide , and the NO-donors SNAP , and spermine NONOate was tested in U46619-contracted arteries that developed stable tension equivalent to ~70% of the maximal response during the standard warm-up procedure . Vasorelaxation , from the pre-contraction level , was quantified as the relative decrease in active tone during the last 30 s of each 2 min agonist application . We used two NO donors that differ , for instance , in their rate and location of NO release ( Bradley and Steinert , 2015 ) . Because acetylcholine-induced vasorelaxation shows substantial tachyphylaxis in arteries from C57BL/6 mice , we tested single doses of acetylcholine applied on separate U46619 pre-contractions . In contrast , SLIGRL-amide does not show this degree of tachyphylaxis , and we therefore tested its vasorelaxant properties through cumulative additions . Ca2+-sensitive fluorophores were loaded preferentially into endothelial cells by slow ( ~1 mL/hr for 45 min ) perfusion of physiological saline solution containing 2 . 5 µM Fluo-4 and 3 . 5 µM Fura Red ( Invitrogen ) through the lumen of basilar and mesenteric arteries mounted in a pressure myograph ( 120CP; DMT ) ( Boedtkjer et al . , 2011 ) . The cell permeant acetoxymethyl ester ( AM ) forms of the fluorophores were first solubilized in a load mix containing dimethyl sulfoxide , Pluronic F127 , and Cremophor EL . The loaded arteries were next mounted in a confocal wire myograph ( 360CW; DMT ) and studied using a Zeiss Axiovert 200M confocal microscope equipped with an LSM Pascal exciter and a 40 × objective ( LD C-Apochromat; N . A . 1 . 10; Zeiss , Germany ) . The arteries were excited at 488 nm and emission light collected at wavelengths in the range of 505–530 nm ( F505-530; representing Fluo-4 signals that increase at elevated [Ca2+] ) and longer than 600 nm ( F>600; representing Fura Red signals that decrease at elevated [Ca2+] ) . Arteries with no or only solitary loaded endothelial cells were excluded from analysis . The F505-530/F>600 ratio normalized to the baseline ratio ( the first 2 min of the recording ) was used to evaluate relative intracellular Ca2+ dynamics . Inverted basilar and mesenteric arteries were mounted on a 40 µm wire in a confocal wire myograph ( 360CW , DMT ) and loaded with the pH-sensitive fluorophore 2' , 7'-bis- ( 2-carboxyethyl ) −5- ( and-6 ) -carboxyfluorescein ( BCECF ) in physiological saline solution ( 1 µM BCECF-AM in 0 . 05‰ DMSO ) to preferentially load endothelial cells , essentially as previously described ( Boedtkjer et al . , 2011; Boedtkjer and Aalkjaer , 2009 ) . The loaded arteries were studied using an Olympus IX83 microscope equipped with a 20 × objective ( LUCPlanFLN; N . A . 0 . 45; Olympus , Japan ) and an ORCA-Flash 4 . 0 camera ( Hamamatsu , Japan ) . The arteries were alternatingly excited at 490 and 436 nm and emission light collected at 530 nm . The F490/F436 fluorescence ratio was calibrated to intracellular pH using the high-[K+] nigericin method ( Boedtkjer and Aalkjaer , 2009; Aalkjaer and Cragoe , 1988 ) , which in arteries agrees well with the null-point technique ( Danielsen et al . , 2013 ) . Steady-state intracellular pH was recorded in the presence and absence of CO2/HCO3– . Membrane potentials were measured in vascular smooth muscle cells of isolated basilar arteries using sharp electrodes as previously described ( Boedtkjer et al . , 2013a ) . The arteries were mounted in a wire myograph ( 420A; DMT ) and microelectrodes that had resistances of 40–120 MΩ when backfilled with 3 M KCl were inserted into the vascular wall from the adventitial side . Cell impalement was observed as a sudden drop in voltage followed by sharp return to baseline upon retraction . Measurements with more than 10 mV difference between the baseline recording before and after impalement were excluded from analysis . The genetic insert that disrupts RPTPγ expression in the employed KO mice contains a promotorless LacZ sequence allowing for β-galactosidase expression under control of the Ptprg promotor ( Lamprianou et al . , 2006 ) . Four homozygous KO mice ( PtprgLacZ/LacZ ) and two WT mice were perfusion fixed with 4% ( weight/volume ) paraformaldehyde in phosphate-buffered saline ( PBS , in mM: 137 NaCl , 2 . 5 KCl , 4 . 3 Na2HPO4 , and 1 KH2PO4 ) and investigated for promoter activity essentially as previously described ( Boedtkjer et al . , 2008 ) . Segments of basilar , middle cerebral , gracilis , coronary , mesenteric , and pulmonary arteries as well as thoracic aorta were dissected free from surrounding tissue and washed in PBS overnight at 4°C . The arteries were then placed in staining solution ( in mM: 5 K4Fe ( CN ) 6 , 5 K3Fe ( CN ) 6 , 2 MgCl2 , 0 . 1% ( weight/volume ) sodium dodecyl sulfate ( SDS ) , 0 . 1% ( volume/volume ) TWEEN−20 , and 0 . 1% ( weight/volume ) 5-bromo-4-chloro-indolyl-13-D-galactoside ( X-Gal ) ) for 24 hr at room temperature ( ~21°C ) . Finally , samples were transferred to PBS containing 1% ( weight/volume ) ethylenediaminetetraacetic acid ( EDTA ) and 4% paraformaldehyde in order to stop the staining reaction . Whole mount micrographs of the arteries were captured using a Leica M165 C stereomicroscope equipped with a Leica M170 HD camera ( Germany ) . In addition , we paraffin-embedded and cut basilar and middle cerebral arteries—that showed the strongest staining—to 4-µm-thick histological sections that were visualized on an upright Leica DM light microscope equipped with a Leica DM300 digital camera in order to identify the cellular expression pattern . Arteries dissected free from surrounding tissue in cold physiological saline solution were stored in RNALater ( Qiagen , Denmark ) at 4°C . The isolated arteries were homogenized in RLT lysis buffer ( Qiagen ) using a TissueLyser II ( Qiagen ) at 30 Hz for 2 min . RNA was isolated using the RNeasy Micro Qiacube kit including carrier RNA ( Qiagen ) . Samples were reverse transcribed using random decamer primers and Superscript III Reverse Transcriptase ( Invitrogen , Fisher Scientific , Denmark ) . Reactions without reverse transcriptase were performed in order to test for genomic amplification . Quantitative PCR was performed on an MX3000P system ( Agilent , USA ) based on Maxima Hot Start Taq DNA polymerase ( ThermoFisher , Denmark ) . Ptprg mRNA levels relative to the reference genes Rn18s ( 18S ribosomal subunit ) and Actb ( β-actin ) were evaluated in the different vascular beds based on the 2–ΔΔCT method ( Livak and Schmittgen , 2001 ) . We used the following forward ( F ) and reverse ( R ) primers and probes ( P ) purchased from Eurofins Genomics ( Germany ) : Ptprg ( F: 5’ TGG TTA CAA CAA AGC GAA AGC CT 3’ , R: 5’ ATA CTG ATC ACA CTT TCT CCT TCC 3’ , P: 5’ ATC TGG GAA CAA AAC ACG GGA ATC ATC AT 3’ ) , Rn18s ( F: 5’ AAT AGC CTT CGC CAT CAC TGC 3' , R: 5’ GTG AGG TCG ATG TCT GCT TTC C 3’ , P: 5’ TGG GGC GGA GAT ATG CTC ATG TGG TGT T 3’ ) , and Actb ( F: 5’ TGA CGT TGA CAT CCG TAA AG 3’ , R: 5’ CTG GAA GGT GGA CAG TGA GG 3’ and P: 5’ AGT GCT GTC TGG TGG TAC CAC CAT GTA CC 3’ ) . Probes were modified with 5’ 6-FAM and 3’ TAMRA . Each reaction consisted of 10 min at 95°C followed by 50 cycles of 30 s at 95°C , 60 s at 55°C , and 60 s at 72°C . We explored levels of Ptprg transcripts in individual endothelial cells based on the EC Atlas database from VIB-KU Leuven , which is accessible at https://endotheliomics . shinyapps . io/ec_atlas/ ( downloaded on 11 June 2020 ) . This online database includes data from a recent single-cell RNA sequencing study on endothelial cells from healthy mice ( Kalucka et al . , 2020 ) . We determined protein expression levels of eNOS in basilar and mesenteric arteries by immunoblotting using previously described antibodies ( Boedtkjer et al . , 2011; Voss et al . , 2019 ) . Arteries were snap frozen in liquid nitrogen and then homogenized using pellet pestles ( Sigma-Aldrich , Denmark ) in a lysis buffer at pH 7 . 5 containing ( in mM ) 20 Tris-HCl , 150 NaCl , 5 ethylene glycol tetraacetic acid ( EGTA ) , 10 NaF , 20 β-glycerophosphate sodium salt , and HALT protease and phosphatase inhibitor cocktail ( Thermo Scientific , Denmark ) . Samples were sonicated for 45 s and centrifuged at ~16 , 000 g for 10 min . Total protein concentrations in the supernatants were measured using a bicinchoninic acid ( BCA ) protein assay kit ( Thermo Scientific ) ; and 10 µg total protein diluted in Laemmlie sample buffer ( Biorad , Denmark ) was loaded in each lane of an SDS polyacrylamide gel ( Biorad ) . Membranes were first probed with anti-eNOS ( RRID:AB_304967; 0 . 2 µg/mL ab5589; Abcam , UK ) or anti-pan-actin ( RRID:AB_2313904; 40 ng/mL #4968; Cell Signaling Technology , USA ) primary antibody and then with secondary goat anti-rabbit antibody ( RRID:AB_2099233; 30 ng/mL #7074; Cell Signaling Technology ) conjugated to horseradish peroxidase . Bound antibody was detected by enhanced chemiluminescence ( ECL Plus; GE Healthcare , Denmark ) using an ImageQuant LAS 4000 luminescent image analyzer ( GE Healthcare ) . Densitometric analyses were performed using ImageJ software ( RRID:SCR_003070; Rasband; NIH , USA ) . Band densities of eNOS relative to pan-actin were reported after normalization to the average WT level . Mice were anesthetized by subcutaneous injection of ketamine and xylazine ( 80 mg/kg Ketaminol vet and 8 mg/kg Narcoxyl vet; Intervet International , The Netherlands ) and placed on a thermostatically controlled heating platform . For telemetry-based measurements of resting blood pressure and effects of L-NAME ingestion , the catheter of a telemetry transmitter ( HD-X11; Data Sciences International , USA ) was inserted in the common carotid artery through a midline incision in the neck , and the transmitter body placed in a subcutaneous pocket during stereomicroscopy . Pain relief was achieved through subcutaneous injection of buprenorphine ( 0 . 2 mL/kg , Temgesic , Schering-Plough , Europe ) . Telemetry measurements started one week after the operation . One 48 hr long registration at baseline was followed by two 72 hr long registrations during which first 0 . 5 mg/mL and then 3 mg/mL L-NAME was added to the drinking water . Telemetry signals were recorded for 10 s every minute using Dataquest A . R . T . 4 . 3 and analyzed with Ponemah 5 . 0 software ( RRID:SCR_017107; Data Sciences International ) . We evaluated daytime and nighttime blood pressure from 11 AM to 1 PM and 11 PM to 1 AM , respectively . For measurements of acute blood pressure responses to hyperventilation , a catheter was inserted in the common carotid artery through a midline incision and connected to a pressure transducer ( MLT0699 , ADInstruments ) . The mice were intubated and ventilated on a Minivent type 845 ventilator ( Harvard Apparatus , USA ) at a frequency of 125 min−1 and with tidal volume adjusted until capnography readings ( Capnograph Type 240 , Hugo-Sachs Electronics , Germany ) showed an FETCO2 of 3 . 5% . During experiments , hypocapnia was induced by elevating the ventilation until FETCO2 decreased to 2% . Mean arterial blood pressure was derived from the pressure traces using the blood pressure add-on for LabChart 8 Pro ( ADInstruments ) . A few mice with an initial systolic blood pressure below 70 mmHg were excluded from the analysis . Mice were initially anesthetized by subcutaneous injection of ketamine ( 80 mg/kg ) and xylazine ( 8 mg/kg ) followed by injection with 1/4 the initial dose every 45 min . After endotracheal intubation , mice were ventilated under capnography control as described above . The head of the anesthetized mouse was fixed in an adaptor for a stereotaxic frame ( World Precision Instruments , UK ) while the rest of the body was kept warm on a heating pad ( Fine Science Tools Inc , Canada ) . The skin covering the top of the skull was removed , the surface of the bone cleaned , and a coverslip mounted in agarose in order to minimize optical reflections . Whiskers were fixed to a metal pole mounted on a cylindrical solenoid controlled by an Arduino circuit ( Funduino Kit , Germany ) . The whiskers on one side of the head were first stimulated by moving whiskers vertically with an amplitude of 8 mm and frequency of 4 Hz . Then , the stimulation protocol was repeated on the opposite side of the head , and the two responses were averaged for each mouse . Whisker stimulation was performed under control conditions ( FETCO2 = 3 . 5% ) and after FETCO2 was reduced to 2% by mechanical hyperventilation . Speckle images of 1088 × 1088 pixels were captured with a Basler acA2000-165uc camera mounted on a VZM 200i Zoom Imaging Lens ( Edmund Optics , USA ) during transcranial illumination with near-infrared laser light ( CLD 1011LP , Thorlabs Inc , USA ) . Speckle data were analyzed with MATLAB software ( RRID:SCR_001622; MathWorks , USA ) . In each experiment , values were calculated for a region of interest ( 100 × 100 pixels ) in the second posterior bifurcation of the middle cerebral artery corresponding to the somatosensory barrel cortex contralateral to the whisker stimulation ( Aronoff and Petersen , 2008 ) . Arterial blood samples were collected through a catheter implanted in the common carotid artery of anesthetized , intubated mice ventilated to normocapnia ( FETCO2 = 3 . 5% ) or experimental hypocapnia ( FETCO2 = 2% ) on a Minivent type 845 ventilator ( Harvard Apparatus ) . The blood was immediately analyzed using an ABL80 Flex blood gas analyzer ( Radiometer , Denmark ) . Data are expressed as mean ± SEM and n equals number of mice ( i . e . biological replicates ) . Probability ( p ) values less than 0 . 05 were considered statistically significant . Sample sizes were selected based on previous experience ( Boedtkjer et al . , 2006; Boedtkjer et al . , 2016c; Boedtkjer et al . , 2011; Thomsen et al . , 2014 ) to allow detection of biologically relevant differences . If distributions were approximately Gaussian and variances equal between groups , we compared ( a ) one variable between two groups using unpaired two-tailed Student’s t-test , ( b ) one variable between three or more groups using one-way ANOVA followed by Dunnett’s post-tests , and ( c ) effects of two variables on a third variable using two-way ANOVA followed by Sidak’s post-tests . If the distributions showed unequal variance ( i . e . p<0 . 05 by F test or by Brown-Forsythe and Bartlett’s tests ) or significant difference from normality ( i . e . p<0 . 05 by D'Agostino and Pearson or Shapiro-Wilk normality tests ) due to right-skewness , we performed square root- or log-transformation . If variances were still unequal , we compared one variable between two groups based on unpaired two-tailed t-tests with Welch’s correction . If data distributions still did not pass normality tests , we used the non-parametric Mann-Whitney test to compare one variable between two groups . Concentration-response relationships were fitted to sigmoidal functions using least-square regression analyses , and the derived log ( EC50 ) , Hill Slope , and bottom values were compared using extra sum-of-squares F-tests . Investigators were not blinded for genotype during experiments . Data processing and statistical analyses were performed using Microsoft Office Excel 2016 ( RRID:SCR_016137 ) and GraphPad Prism 7 . 05 ( RRID:SCR_002798 ) software . We studied association between rare variation in PTPRG and human ischemic vascular disease based on exome sequencing data covering ~50 , 000 participants from the UK Biobank ( RRID:SCR_012815 ) , which is a cohort with deep genetic and phenotypic data collected from the United Kingdom ( Bycroft et al . , 2018; Van Hout et al . , 2019 ) . Population-level variants ( UK Biobank data field 23170 ) generated with the Functionally Equivalent pipeline ( Regier et al . , 2018 ) were used after ensuring that the genomic region containing PTPRG was unaffected by the current known error in the exome analysis protocol ( UK Biobank Resource 3802 ) . We predicted the consequences of alternative alleles within the full-length transcript for PTPRG ( ENST00000474889 . 6 ) using the Ensembl Variant Effect Predictor ( VEP ) ( McLaren et al . , 2016 ) ; and included in our analyses , the 334 missense variants localized outside exon-intron boundary regions and 75 variants with predicted loss-of-function . Carriers of the identified variants were extracted using the R-package qgg ( Rohde et al . , 2020 ) . The identified loss-of-function variants included 36 with low ( intronic splice region variants , synonymous splice region variants ) , 17 with moderate ( in-frame insertions , in-frame deletions , missense splice region variants ) , and 22 with high ( splice acceptor variants , splice donor variants , stop-gain variants , frameshift variants ) predicted impact . In total , we identified 29 , 975 individuals who carried at least one missense variant and 195 individuals , who carried a variant with predicted loss-of-function . Of the 195 loss-of-function carriers , 123 , 32 , and 40 individuals carried variants with low , moderate , and high predicted impact , respectively . Known disease diagnoses for UK Biobank participants ( data field 41270 ) were extracted from hospital inpatient records coded according to the International Classification of Disease , version 10 ( ICD-10; RRID:SCR_010349 ) . Amongst the 49 , 953 participants with exome data , 2529 individuals were diagnosed with angina pectoris ( I20 ) , 972 with acute myocardial infarction ( I21 ) , 392 with cerebral ischemia ( I63 ) , and 153 with vascular intestinal disorders ( K55 . 0 ) . Self-reported health status was extracted for angina ( data field 1074; reported by 1621 individuals ) , heart attack ( data field 1075; reported by 1158 individuals ) , and stroke ( data field 1081; reported by 686 individuals ) . We further extracted MRI-based values for 12 , 851 exome-sequenced individuals for whom left ventricular ejection fraction ( data field 22420 ) was available from a fully automated analysis approach that has shown good correlation with manual analysis results performed by trained readers ( Suinesiaputra et al . , 2018 ) . The cardiac MRI data set included 6482 carriers of missense PTPRG variants , 32 carriers of low impact loss-of function variants , 13 carriers of moderate impact loss-of-function variants , and 13 carriers of high impact loss-of-function variants . We performed logistic regression analyses to calculate odds ratios ( ± SEM ) for carrying missense or predicted loss-of-function PTPRG variants when diagnosed with ischemic vascular disease . We fist evaluated acute ischemic vascular disease as a whole for the brain , heart , and intestine ( aggregate of ICD-10 diagnosis codes I20 , I21 , I63 , and K55 . 0 ) . For the group of individuals carrying a predicted loss-of-function variant of moderate or high impact , we then calculated separate odds ratios for I20 , I21 , and I63 . The number of exome-sequenced participants in the UK Biobank diagnosed with K55 . 0 was too low for a meaningful separate statistical analysis . We next evaluated association of the PTPRG loss-of-function variants with self-reported angina , heart attack , and stroke . Finally , we evaluated the relation between carrier status for PTPRG variants and left ventricular ejection fraction by multiple linear regression analysis . The logistic as well as multiple linear regression analyses were corrected for sex ( data field 31 ) , age ( data field 21022 ) , body mass index ( data field 21001 ) , the first four genetic principal components ( data field 22009 ) , smoking status ( data field 20116 ) , dyslipidemia ( ICD-10 diagnosis code E78 . 5 ) , diabetes mellitus ( ICD-10 diagnosis code E11 ) , and hypertension ( ICD-10 diagnosis code I10 ) .

Restricted blood flow in the heart or brain can deprive these vital organs of oxygen , thereby causing a heart attack or stroke . However , the body has compensatory mechanisms to mitigate damage: if the blood flow is reduced in one blood vessel , acidic waste accumulates locally . This causes nearby blood vessels to widen and increase the oxygen supply . Although scientists first observed this process 140 years ago , they have not yet devised a way to use it for treatment of heart attack or stroke . Now , Hansen et al . discovered that a protein called RPTPγ , which is found on the lining of blood vessels , could be a good target for drugs intended to reduce the consequences of heart attacks and strokes . The protein RPTPγ has a similar structure to other proteins that bind bicarbonate , an important ion that buffers acids in the body . RPTPγ can also trigger signals to nearby cells , which suggests that the protein can monitor bicarbonate levels in the blood and tissue and alert blood vessels of the need to widen . Hansen et al . found that the blood vessels of mice that lacked RPTPγ were unable to widen when needed . Moreover , mice without RPTPγ experienced abnormal changes in blood pressure and blood flow to the brain when oxygen consumption was elevated or pH was disrupted . Hansen et al . further analyzed genetic and health data from nearly 50 , 000 individuals in the UK Biobank . These analyses revealed that people with genetic changes that render RPTPγ ineffective are at higher risk of having a heart attack or stroke . People with these genetic variants also have reduced heart pumping ability . The experiments suggest that a lack of functional RPTPγ affects an individual’s ability to adjust local blood flow in response to acid-base disturbances and oxygen deficits , increasing the risk of a heart attack or stroke . This information may help scientists develop new ways to prevent or treat heart attacks and strokes , or to treat other conditions like cancer , where pH is disturbed .

2020

PTPRG is an ischemia risk locus essential for HCO3–-dependent regulation of endothelial function and tissue perfusion

Vascular network density determines the amount of oxygen and nutrients delivered to host tissues , but how the vast diversity of densities is generated is unknown . Reiterations of endothelial-tip-cell selection , sprout extension and anastomosis are the basis for vascular network generation , a process governed by the VEGF/Notch feedback loop . Here , we find that temporal regulation of this feedback loop , a previously unexplored dimension , is the key mechanism to determine vascular density . Iterating between computational modeling and in vivo live imaging , we demonstrate that the rate of tip-cell selection determines the length of linear sprout extension at the expense of branching , dictating network density . We provide the first example of a host tissue-derived signal ( Semaphorin3E-Plexin-D1 ) that accelerates tip cell selection rate , yielding a dense network . We propose that temporal regulation of this critical , iterative aspect of network formation could be a general mechanism , and additional temporal regulators may exist to sculpt vascular topology . The unique vascular topology of different organs exists for organ-specific functions . Different vascular network densities determine the specific amount of oxygen and nutrients to be delivered to each host tissue . Development of new vascular networks depends upon two types of specialized endothelial cells that work together: ( 1 ) The endothelial 'tip cell' , which is located at the front of a growing vessel and guides its extension by sensing and responding to environmental cues , analogous to the axonal growth cone ( Gerhardt et al . , 2003; Kurz et al . , 1996 ) . ( 2 ) 'Stalk cells' , which trail behind the tip cell and elongate the sprout . Tip and stalk cell identities are primarily controlled by the Dll4-Notch lateral inhibition pathway , which is activated in endothelial cells in response to VEGF from the local environment . VEGF-induced Dll4 activates Notch1 on the neighboring cell , leading to the down-regulation of VEGF receptor levels ( Figure 1B ) ( Hellström et al . , 2007; Leslie et al . , 2007; Lobov et al . , 2007; Suchting et al . , 2007 ) , ( Benedito et al . , 2009; Phng et al . , 2009; Roca and Adams , 2007 ) . Thus , lateral inhibition between endothelial cells generates an alternating pattern of active tip cells ( Dll4 high ) and inhibited stalk cells ( Dll4 low ) . Moreover , tip cell selection is a dynamic process , and tip cell identity is transient ( Arima et al . , 2011; Jakobsson et al . , 2010 ) . Reiteration of tip cell selection , sprout extension , and connection of neighboring sprouts ( anastomosis ) is the basis for building a sophisticated vascular network ( Adams and Eichmann , 2010; Carmeliet , 2000 ) . Although this VEGF/Notch signaling pathway has been well studied and found to be conserved among different vascular beds and species , how this central pattern generator is modified by various target tissue-specific signals to yield diverse network topologies is not known . 10 . 7554/eLife . 13212 . 003Figure 1 . Calibrated computational model predicts delayed tip cell selection in the absence of Sema3E-Plexin-D1 signaling . ( A ) Whole-mount vascular staining ( Isolectin B4 ) of retinas from Sema3e-/- and wildtype littermates at P4 . The mutant vasculature exhibits a reduced number of tip cells and branching points ( asterisks ) and an uneven growth front ( arrows and arrowheads ) . Scale bar: 500 μm . ( B ) Feedback between the VEGF/Notch and Sema3E-Plexin-D1 signaling pathways included in the extended agent-based computational model of tip cell selection . D1-D4: transcriptional delays . r1-r3: recovery delays representing degradation . δ , s , σ: change in expression levels in response to receptor activation . ( C ) Simulated tip cell selection . Colors represent Dll4 levels on a continuum from purple ( low ) to green ( high ) . The red boxes highlight a time frame in which a salt and pepper pattern has formed in the control vessel , while in the absence of Sema3E-Plexin-D1 signaling , only few early tip cells have been selected . ( D ) Average number of selected tip cells in simulated vessels . At a timepoint where the simulated control vessel ( black line ) already exhibits an alternating pattern of tip and stalk cells , the simulated vessel lacking Sema3E-Plexin-D1 signaling ( blue line , for a given set of parameter values: δ =5 , s=3 ) shows a 50% reduction in tip cells . Thin lines: standard deviation . n=50 . ( E ) In silico Dll4 levels in single endothelial cells during simulated tip cell selection . In the control situation ( top ) , Dll4 levels quickly stabilize . In the absence of Sema3E-Plexin-D1 signaling ( bottom ) Dll4 levels fluctuate in near synchrony before they finally stabilize . DOI: http://dx . doi . org/10 . 7554/eLife . 13212 . 003 Here , we propose a general principle of how collective cell behavior determines the diverse densities of different networks: the generation of vascular topologies depends heavily on the temporal regulation of tip cell selection . Integrated simulations predict that as cell neighborhoods change , due to anastomosis or cell rearrangement events , lateral inhibition patterns will necessarily be disrupted , requiring continual re-selection of new tip cells ( Bentley et al . , 2009; 2014a ) . In fact , mouse genetics experiments demonstrated that tip cell numbers are positively correlated with the branching points of the network ( Hellström et al . , 2007; Kim et al . , 2011 ) . Therefore , the length of time it takes to establish ( and re-establish ) the alternating pattern of tip and stalk cells may be a missing , critical determinant of vascular topology ( Bentley et al . , 2014b; 2014c ) . Here , we took an integrated approach combining computational modeling , mouse genetics , and in vivo endothelial cell tracking to determine whether tip/stalk patterning can be temporally modulated to generate different topologies . We hypothesize that the frequency of tip cell selection determines the length of linear extension vs . branching , thus dictating the density of the network . To begin to test this hypothesis , it is crucial to analyze dynamic single cell behavior and collective movement in the context of network formation ( Arima et al . , 2011; Jakobsson et al . , 2010 ) . Previously , we used static analyses of the postnatal mouse retina as a model to understand how neural signals shape vascular topology ( Kim et al . , 2011 ) . We discovered that retina ganglion cell-derived Semaphorin3E ( Sema3E ) and its receptor Plexin-D1 , which is expressed in endothelial cells at the front of actively sprouting blood vessels , control the VEGF/Notch pathway via a feedback mechanism . Mice lacking either Sema3E or Plexin-D1 exhibited an uneven vascular growth front and a reduction of tip cells that resulted in a less branched network compared to their wildtype littermate controls ( Kim et al . , 2011 ) ( Figure 1A ) . However , it is not clear how this phenotype is generated: specifically , how the Sema3E-Plexin-D1 feedback mechanism regulates VEGF/Notch signaling at a dynamic cellular level , and whether changes in temporal modulation of this pathway lead to the overall vascular topology phenotype . To begin to understand how Sema3E-Plexin-D1 signaling modifies vascular topology formation in a dynamic , spatiotemporal manner , we took advantage of an existing agent-based computational model ( the 'MemAgent-Spring Model' or MSM ) that simulates the cellular processes during tip cell selection making explicit the time it takes for gene expression ( e . g . transcription/translation ) changes to occur ( Figure 1B , C – note time delay parameters D1 and D2 ) ( Bentley et al . , 2008; 2009 ) . The MSM has been tested against numerous independent experimental data sets and validated as predictive of new mechanisms in vivo/in vitro ( Bentley et al . , 2014a; Guarani et al . , 2011; Jakobsson et al . , 2010 ) . To now simulate tip cell selection in the context of Sema3E-Plexin-D1 crosstalk signaling with VEGF/Notch signaling ( Fukushima et al . , 2011; Kim et al . , 2011 ) the MSM model was extended by adding four new parameters ( Figure 1B , Video 1–5 ) , with sensitivity analyses and calibration simulations performed , which include modulation of the existing parameter ( δ ) representing the induction level of Dll4 by VEGFR-2 activation ( See methods section ) . These four new parameters represent the time delay for induction of Plexin-D1 by VEGF ( D3 ) , the time delay ( D4 ) and strength ( s ) of the reduction of Dll4 levels in response to Sema3E-Plexin-D1 signaling , based on the experimental data previously shown ( Kim et al . , 2011 ) , as well as the degradation rate of Plexin-D1 ( r3 ) . Loss of Sema3E-PlexinD1 signaling was simulated by setting all Plexin-D1 levels to zero . Loss of function simulations recapitulated the two prominent features of Plxnd1-/- and Sema3e-/- mutant retinal vasculature remarkably closely: In vivo , mutant vascular networks exhibit fewer tip cells and 1 . 5–2 fold reduction in branching points , as well as an uneven growth front ( compare Figure 1A with 1C [red boxes] ) ( Kim et al . , 2011 ) . Furthermore , the dynamic nature of the simulations provided a novel insight to explain how this phenotype is generated . Simulations predict that higher Dll4 levels in the cells generates a shift towards synchronized Dll4 fluctuations overtime in contiguous cells , as they collectively battle more strongly via lateral inhibition negative feedback , causing an overall delay in amplification of differences needed to select the alternating pattern of tip and stalk cells ( Figure 1C–E , Video 6 , 7 ) . Therefore , computational simulations suggest that Sema3E-Plexin-D1 signaling enhances the speed and frequency of tip cell selection and thus increases the density of retinal vascular networks . 10 . 7554/eLife . 13212 . 004Video 1 . Calibration of time delay parameters ( induction of Plexin-D1 , reduction of Dll4 levels ) . Simulation of tip cell selection with Sema3E-Plexin-D1 related regulatory time delays equal to the time it takes for pVEGFR-2 to up-regulate Dll4 ( D3+D4=D1 ) . Color indicates cell Dll4 levels ( green = high , purple = low ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13212 . 00410 . 7554/eLife . 13212 . 005Video 2 . Calibration of time delay parameters ( induction of Plexin-D1 , reduction of Dll4 levels ) . Simulation of tip cell selection with Sema3E-Plexin-D1 related regulatory time delays slightly slower than the time it takes for pVEGFR-2 to up-regulate Dll4 ( D3+D4=D1+1 ) . Dll4 levels flash irregularly between very high and very low with no stable selection possible . Color indicates cell Dll4 levels ( green = high , purple = low ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13212 . 00510 . 7554/eLife . 13212 . 006Video 3 . Calibration of time delay parameters ( induction of Plexin-D1 , reduction of Dll4 levels ) . Simulation of tip cell selection with Sema3E-Plexin-D1 related regulatory time delays slightly faster than the time it takes for pVEGFR-2 to up-regulate Dll4 ( D3+D4=D1-1 ) . Dll4 levels flash irregularly between very high and very low with no stable selection possible . Color indicates cell Dll4 levels ( green = high , purple = low ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13212 . 00610 . 7554/eLife . 13212 . 007Video 4 . Calibration of Plexin-D1 degradation rate . Simulation of tip cell selection with slowed degradation of Plexin-D1 allowing it to affect transcription of Dll4 for one timestep ( 12 s ) longer . Dll4 levels flash irregularly between very high and very low with no stable selection possible . Color indicates cell Dll4 levels ( green = high , purple = low ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13212 . 00710 . 7554/eLife . 13212 . 008Video 5 . Calibration of Plexin-D1 degradation rate . Simulation of tip cell selection with slowed degradation of Plexin-D1 allowing it to affect transcription of Dll4 for two timesteps ( 30 s ) longer . Dll4 levels flash irregularly between very high and very low with no stable selection possible . Color indicates cell Dll4 levels ( green = high , purple = low ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13212 . 00810 . 7554/eLife . 13212 . 009Video 6 . Simulation of a vessel with 20 endothelial cells in the absence of Sema3E-Plexin-D1 signaling . Extended regions occur with no sprouting as cells battle for longer undergoing fluctuations as the Dll4 up-regulation is higher . These eventually resolve and tip cells are selected across the whole region . DOI: http://dx . doi . org/10 . 7554/eLife . 13212 . 00910 . 7554/eLife . 13212 . 010Video 7 . Simulation of a wildtype vessel for comparison with Video 6 . The selection occurs much faster and more regularly than in the vessel without Sema3E-Plexin-D1 signaling . DOI: http://dx . doi . org/10 . 7554/eLife . 13212 . 010 We next examined single cell behavior in a mosaic environment , because lateral inhibition requires collective coordination , and the relative , not absolute , levels of Dll4 expression among neighboring cells determine the outcome of the tip cell selection process ( Jakobsson et al . , 2010 ) . A Plxnd1-/- cell will behave differently in competition with a second Plxnd1-/- cell than in competition with a wildtype cell . Thus , we investigated whether the cell autonomous function of Sema3E-Plexin-D1 signaling in the direct competition process between two neighboring endothelial cells can drive changes to the tip cell selection dynamics . In a simulated mosaic vessel , where cells lacking Sema3E-Plexin-D1 signaling are intermingled with normal cells , the contribution of cells lacking Sema3E-Plexin-D1 signaling to the tip cell population was predicted to increase to 68–74% ( robustly across a range of parameter values: δ =4–6 , s=2–4 ) in comparison to control cells ( Figure 2A , B , Video 8 ) , with an increased speed of patterning compared to a vessel entirely lacking Sema3E-Plexin-D1 signaling . This result suggests that normally , Plexin-D1 cell-autonomously suppresses the tip cell phenotype . 10 . 7554/eLife . 13212 . 011Figure 2 . Sema3E-PlexinD1 signaling is cell-autonomously required to suppress tip cell identity . ( A ) Simulated tip cell selection in a mosaic vessel with 50% mutant cells placed randomly . Cells without Sema3E-Plexin-D1 signaling are indicated by bright pink color , which turns to yellow if Dll4 levels increase . For wildtype cells: purple = low Dll4 , pink = high Dll4 . ( B ) Comparison of simulated and in vivo contribution to the tip cell population by cells lacking Sema3E-Plexin-D1 signaling at 45% mosaicism . A range of δ values simulates different strengths of loss of Sema3E-Plexin-D1 signaling . Simulations , n=50 , in vivo , n=6 . Data is represented as mean +/- SEM . ( C ) Analysis of the occupation of tip or stalk cell position by Plxnd1 expressing wildtype cells in control retina ( left ) , and Plxnd1-/- cells ( middle ) or GFP+ cells ( right ) in mosaic retinas at P5 . Red: vascular membrane staining ( Isolectin B4 ) , blue: vascular nuclear staining ( α-ERG ) , green: in situhybridization ( left , middle ) , α-GFP staining ( right ) . ( D ) Quantification of c . n . s . = not significant , **p=0 . 0033 . WT retinas , n=6; Plxnd1-/- mosaic retinas , n=6; GFP+ mosaic retinas , n=4 . Scale bar: 50 μm . Data is represented as mean +/- SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 13212 . 01110 . 7554/eLife . 13212 . 012Video 8 . Simulated mosaic vessel with 50% mutant cells placed randomly . Cells without Sema3E-Plexin-D1 signaling are indicated by bright pink color , which turns to yellow if Dll4 levels increase . For wildtype cells: purple = low Dll4 , pink = high Dll4 as before . This chimera achieves a 75% contribution of mutant cells to the tip . DOI: http://dx . doi . org/10 . 7554/eLife . 13212 . 012 To test the direct competition of Plxnd1-/- cells and wildtype cells in vivo , we performed mosaic analysis using mice with tamoxifen inducible loss of Plxnd1 expression in the vasculature ( Cdh5-Cre-ERT2; Plxnd1flox/flox ) . In the sprouting front of the wildtype retina , cells expressing Plexin-D1 were equally distributed at both tip cell and the adjacent stalk cell positions ( Figure 2C , left panel , Figure 2D ) , showing they have no preference for either position . However , at approximately 45% of Plxnd1-/- mosaicism , 75% of the Plxnd1-/- cells became tip cells ( Figure 2C , middle panel , Figure 2D ) indicating that these cells do have a competitive advantage over Plexin-D1 expressing wildtype cells during tip cell selection . In contrast , in control experiments using mice with tamoxifen inducible expression of GFP ( Cdh5-Cre-ERT2; Z/EG+ ) , GFP positive cells showed no preference for either position ( Figure 2C , right panel , Figure 2D ) , demonstrating that preferential tip cell occupancy of mutant cells is due to lack of Sema3E-Plexin-D1 signaling . Taken together , in silico prediction and in vivo mosaic retinal analyses demonstrate that Sema3E-Plexin-D1 signaling suppresses the tip cell phenotype in a cell-autonomous manner . Interestingly , this cell-autonomous effect at single cell level contrasts with the effect of Sema3E-Plexin-D1 signaling at the collective level , where more new tip cells are selected in the presence than in the absence of Sema3E-Plexin-D1 signaling ( Figure 1 ) . This finding highlights the differences between collective and cell autonomous behaviors , and the challenge of intuiting one from the other . Having confirmed that the calibrated set of parameters is valid to model contributions of the Sema3E-Plexin-D1 pathway to VEGF/Notch patterning and to predict new in vivo data of a static nature , we next employed the model to predict the effect of Sema3E-Plexin-D1 signaling on the rate of tip cell selection during the fuller dynamic network formation processes in which tip and stalk cell identities are constantly re-defined during cell rearrangement/position switching . Here , we used a new 'MSM-CPM' model that has been previously extended , parameterized and experimentally validated to simulate tip/stalk patterning together with cellular rearrangements within a vascular sprout ( Bentley et al . , 2014a ) . In this model , a cell can move within the adhered collective of the sprout powered by multiple local junctional movements . By incorporating the newly calibrated Sema3E-PlexinD1 signaling extension the MSM-CPM model we simulated cell rearrangements in vascular sprouts in the presence or absence of Sema3E-Plexin-D1 signaling ( Figure 3A , B , Video 9 , 10 ) . Simulations predicted that in the presence of Sema3E-Plexin-D1 signaling , the tip cell is repeatedly overtaken by another cell . In contrast , in the absence of Sema3E-Plexin-D1 signaling , tip cell overtaking frequency is reduced by factor of 1 . 26; a given tip cell occupies the tip cell position for a longer time ( Figure 3C ) . Given the MSM is a qualitative not quantitative model we also performed simulations over a range of parameters and found that if the strength of Dll4 upregulation by VEGF ( δ ) is increased then this delay in overtaking is further exaggerated; a given tip cell occupies the tip cell position longer ( Figure 3C ) . 10 . 7554/eLife . 13212 . 013Figure 3 . Computational simulation predicts that lack of Sema3E-Plexin-D1 signaling leads to prolonged tip cell occupancy and reduced tip cell overtaking frequency . ( A ) Single frames of cell rearrangements in simulated sprouts of 10 cells . VEGF gradient extends in direction of vessel . ( B ) Kymograph plots of cell rearrangements in simulated sprouts . Each line represents one endothelial cell . Arrows indicate overtaking events at the tip cell position in a and b . ( C ) Quantification of overtaking events at the tip cell position in the presence and absence of Sema3E-Plexin-D1 signaling . A range of δ values simulates different strengths of loss of Sema3E-Plexin-D1 signaling . The setting δ=5 , which matched loss of Sema3E-Plexin-D1 signaling in other conditions in the paper exhibits a 1 . 26 slower tip cell overtaking frequency ( events/hour . ) As δ increases there is a clear trend towards slower tip cell overtaking across δ values . Data is presented as mean +/- SD , n=50 . DOI: http://dx . doi . org/10 . 7554/eLife . 13212 . 01310 . 7554/eLife . 13212 . 014Video 9 . Simulation of normal cell rearrangement and tip cell overtaking in a sprout consisting of ten cells , two per vessel cross section . VEGF gradient extends in the direction of the sprout . Each cell indicated by a different color . DOI: http://dx . doi . org/10 . 7554/eLife . 13212 . 01410 . 7554/eLife . 13212 . 015Video 10 . Simulation of cell rearrangement in a sprout , in the absence of Sema3E-Plexin-D1 , consisting of ten cells , two per vessel cross-section . VEGF gradient extends in the direction of the sprout . Each cell indicated by a different color . DOI: http://dx . doi . org/10 . 7554/eLife . 13212 . 015 To test experimentally if tip cell selection and overtaking rates are slowed down in the absence of Sema3E-Plexin-D1 signaling , we next performed live endothelial cell tracking to follow the behavior of individual cells in an actively forming vascular network . Previously , live imaging with single cell resolution has mainly been described in sprouting assays using aortic rings ( Arima et al . , 2011 ) or embryoid bodies ( EBs ) generated from ES cells ( Jakobsson et al . , 2010 ) . However , those assays primarily give rise to simple linear sprouts with less frequent branching than observed during retinal angiogenesis . To better mimic in vivo temporal and spatial events within a highly branched network , we developed a new ex vivo system , using explants from embryonic lungs . Whole embryonic lungs ( E12 . 0 ) were embedded within a collagen matrix in a tissue culture dish containing medium supplemented with recombinant human VEGF . After one day , endothelial sprouts grow out of the explant ( Figure 4—figure supplement 1A ) with tip cells extending filopodia into the collagen matrix ( Figure 4—figure supplement 1B ) . We observed branching events at various positions of the sprouts ( Figure 4B ) , as well as anastomosis ( Figure 4—figure supplement 1C ) , showing that all steps of the in vivo angiogenic sprouting process are recapitulated in our system even though there is no gradient of VEGF signaling . Lung explants endogenously express Sema3e and Plxnd1 ( Figure 4—figure supplement 1D ) . We could also detect Plexin-D1 and Dll4 in sprouting endothelial cells ( Figure 4—figure supplement 1E , F ) . We next performed long-term live imaging of wildtype and Plxnd1-/- explants and analyzed tip cell selection frequency . In contrast to the computational model where a tip cell selection event can only occur via a switch between two cells at the tip of the sprout ( Figure 3A ) , in the live imaging assay , tip cell selection events occur in the following distinct categories ( Figure 4A , B ) : a positional switch between a tip and a stalk cell in a linear sprout ( switch ) , the selection of a new tip cell at the front of the sprout ( branch , type I ) or at more proximal sites ( branch , type II ) . Single cells were tracked manually using a nuclear live stain . Consistent with the computational model prediction , in vivo live imaging data indeed showed a significant delay in the selection of new tip cells in Plxnd1-/- vascular sprouts compared to wildtype sprouts . The overall appearance of new tip cells , i . e . the tip cell selection frequency , was slowed down by factor 1 . 5 in the mutants ( Figure 4C , D , Video 11 , 12 ) . When analyzing the different categories separately , the number of events per total imaging time was significantly reduced in the categories 'switch' and 'branch , type II' in the absence of Sema3E-Plexin-D1 signaling ( Figure 4E ) . 10 . 7554/eLife . 13212 . 016Figure 4 . Endothelial cell live tracking in ex vivo lung explants reveals a reduction in tip cell selection frequency and a less branched lung vascular network in the absence of Sema3E-Plexin-D1 signaling . ( A ) Different types of tip cell selection events observed during live imaging . ( B ) Single frames from live imaging experiments illustrating the different types of tip cell selection events . Arrowheads point out the newly selected tip cells . Nuclei: blue . Scale bar: 50 μm . ( C ) Long-term live imaging experiments of vascular sprouts from wildtype and Plxnd1-/- lung explants . Single planes from z-stacks are shown . Arrowheads indicate a tip cell selection event . Nuclei: blue ( top and middle ) , white ( bottom ) . Individual nuclei are outlined by different colors in the middle panel . ( D ) Quantification of tip cell selection frequency calculated as events per hour . Tip cell selection frequency is reduced by factor 1 . 5 in sprouts from Plxnd1-/- explants . WT , n=24 sprouts from 12 explants . Plxnd1-/- , n=30 sprouts from 11 explants . Data is represented as mean +/- SEM . ( E ) Quantification of tip cell selection frequency calculated as incidence of events in each category as illustrated in ( A ) during total imaging time . **p=0 . 013 ( D ) , *p=0 . 029 ( e ‘switch’ ) , 0 . 041 ( e ‘branch , type II’ ) , permutation test with shuffled genotypes . ( F ) Vascular sprouts originating from Plxnd1+/- and Plxnd1-/- lung explants on day 3 . Left: whole-mount vascular staining ( green , PECAM ) , right: reconstructed/skeletonized network , Scale bar: 250 μm . ( G ) Quantification of branching points per area . The number of branching points is significantly reduced in Plxnd1-/- lung explants . Data is represented as mean +/- SEM . Plxnd1+/- , n=7 explants; Plxnd1-/- , n=6 explants . ***p=0 . 0005 , permutation test with shuffled genotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 13212 . 01610 . 7554/eLife . 13212 . 017Figure 4—figure supplement 1 . The ex vivo sprouting assay . ( A ) Sprouts originating from the lung explant express endothelial specific marker PECAM . Explants at day 1 ( left ) and day 3 ( right ) are shown . Scale bar: 300 μm . ( B ) Tip cells extend filopodia ( arrows ) into the collagen matrix . Scale bar: 50 μm . ( C ) Anastomosis: a new connection ( arrow ) is formed between neighboring sprouts . Scale bar: 50 μm ( D ) Sema3e and Plxnd1 are expressed by embryonic lung explants as shown by RT PCR . ( E ) Sprouting endothelial cells express Plexin-D1 ( green ) . PECAM staining shown in red . Scale bar: 10 µm . ( F ) In wildtype endothelial sprout Dll4 ( green ) localized in the tip cells ( arrow ) . Dll4-positive area is increased in Plxnd1-/- explants . Scale bar: 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 13212 . 01710 . 7554/eLife . 13212 . 018Figure 4—figure supplement 2 . Computational method used for vascular network analysis . The region of interest ( ROI ) was defined as the vascular network originating from the lung explants . Vasculature inside the lung tissue was excluded from the analysis . Stacks of confocal images were binarized and then skeletonized by thinning . Branching points were identified as skeleton pixels with at least three neighbors and quantified to characterize the vascular network . The 2D images shown are the maximum intensity z-projections of the original 3D stacks , except for the binary image , which is an intensity sum z-projection . DOI: http://dx . doi . org/10 . 7554/eLife . 13212 . 01810 . 7554/eLife . 13212 . 019Figure 4—figure supplement 3 . Model: Modifications of the central pattern generator lead to the formation of diverse vascular topologies . The unique topologies of organ-specific vascular networks are dependent on the tight temporal control of the tip cell selection process . The VEGF-Notch lateral inhibition pathway is the central pattern generator of tip cell selection . We suggest that in different tissues , specific target-derived signals functioning as 'molecular metronomes' regulate the pace of the pattern generator to ensure the formation of networks that cater each tissue’s specific need . A 'fast molecular metronome' ( e . g . Sema3E-Plexin-D1 signaling ) will speed up the Dll4/Notch feedback loop and increase the frequency of tip cell selection , leading to the formation of a dense network with a small pore size , while a 'slow molecular metronome' will slow down the Dll4/Notch feedback loop and decrease the tip cell selection frequency , resulting in a network with larger pore sizes . DOI: http://dx . doi . org/10 . 7554/eLife . 13212 . 01910 . 7554/eLife . 13212 . 020Video 11 . Wildtype sprout in ex vivo endothelial cell tracking assay . The movie represents 9 hr of live imaging and corresponds to Figure 4C , left panel . Merge of fluorescent channel ( nuclear live stain ) and brightfield channel . Arrows indicate newly selected tip cells that give rise to a new sprout . Arrowhead indicates a switching event . Single planes from z-stacks are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 13212 . 02010 . 7554/eLife . 13212 . 021Video 12 . Plxnd1-/- sprout in ex vivo endothelial cell tracking assay . The movie represents 9 hr of live imaging and corresponds to Figure 4C , right panel . Merge of fluorescent channel ( nuclear live stain ) and brightfield channel . Arrows indicate newly selected tip cells that give rise to a new sprout . Single planes from z-stacks are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 13212 . 021 Finally , to directly test experimentally whether the reduced tip cell selection rate observed by live imaging in our Plxnd1-/- lung explant indeed leads to a less branched network over time , we analyzed the topology of the lung explant under the same culture conditions as the live imaging paradigm . Using a computational method ( Figure 4—figure supplement 2 and detailed description in the material and methods section ) to analyze the number of branching points in an unbiased way we found a significant reduction in branching points of the network ( Figure 4F , G ) . These data further demonstrate that a slowed tip cell selection rate results in a less branched vascular network . Together , the ex vivo and in silico results demonstrate that Sema3E-Plexin-D1 signaling modulate the pace of tip cell selection . In the absence of Sema3E-Plexin-D1 signaling , the rate of tip cell selection is reduced , which overall leads to longer linear sprout extension with less frequent branching substantially influencing the architecture of the growing vasculature and resulting in a less dense network , as seen in the Plxnd1 and Sema3e mutant lung explant as well as in retina ( Kim et al . , 2011 ) . During angiogenesis , the topology of the network is shaped essentially through the dynamic process of stalk cells turning into a new tip cell ( tip cell selection ) , which is dependent on the Delta-Notch lateral inhibition pathway , a widely used machinery to regulate aspects of development that require temporal control by a 'molecular clock' . For example , during somitogenesis , Delta-Notch oscillations determine the frequency of new somite formation ( 'the somite clock' ) ( Aulehla and Pourquié , 2008 ) . In endothelial cells , we propose that modulation of any of the components of the central pattern generator will result in an altered pace of Delta-notch oscillations and an altered vascular patterning . Deceleration of the Delta-notch feedback loop ( by a 'slow molecular metronome' ) will lead to the selection of fewer tip cells within a certain time frame and thus to the formation of a less dense network with bigger pore sizes . Acceleration of the selection process ( by a 'fast molecular metronome' ) would result in an overly dense network ( Figure 4—figure supplement 3 ) . However , complex nonlinear feedback dynamics are often hard to intuit , and further careful simulation integrated with experimentation will be required to fully elucidate the temporal modulations and topological outcomes possible . In this work , we describe how Sema3E-Plexin-D1 signaling can modify vascular density by impinging on the central pattern generator VEGF/Notch signaling . Our computational modeling predictions , mouse genetics mosaic analysis , and live imaging of individual cell dynamics in actively forming blood vessel networks and computational quantification of branching points show that the lack of Sema3E-Plexin-D1 signaling slows down the rate of tip cell selection and rearrangement , resulting in a less branched vascular network . Therefore , the Sema3E-Plexin-D1 pathway represents a 'faster molecular metronome' that results in a relatively dense network . These data suggest that temporal regulation of this critical , iterative aspect of network formation could be a general mechanism , and additional temporal regulators with varying pace ( fast vs . slow ) may exist to sculpt vascular topology in different tissues . Furthermore , our findings may provide insights into our understanding of morphogenesis in general , and aid in efforts to develop therapeutic approaches for tissue engineering and control of tumor progression and vascular diseases . Plxnd1flox/floxmice ( Zhang et al . , 2009 ) , Plxnd1+/- mice ( Gu et al . , 2005 ) and Cdh5-Cre-ERT2 mice ( Monvoisin et al . , 2006 ) were maintained on a C57Bl/6 background . Z/EG+ reporter ( Novak et al . , 2000 ) mice were maintained on a 129P3J;C57Bl/6 mixed background . Pregnant mice were obtained following overnight mating ( day of vaginal plug was defined as embryonic day 0 . 5 ) . All animals were treated according to institutional and US National Institutes of Health ( NIH ) guidelines approved by the Institutional Animal Care and Use Committee ( IACUC ) at Harvard Medical School . Cre-mediated recombination was induced by intraperitoneal injection of tamoxifen ( T5648 , Sigma-Aldrich ) dissolved in safflower oil on postnatal day ( P ) 4 . Due to the different Cre sensitivities at Plxnd1 and Z/EG locus , we experimentally determined the dosage of tamoxifen necessary for 45% of mosaicism . 25 pg of tamoxifen was injected in Cdh5-Cre-ERT2; Plxnd1flox/flox pups , and 10 ng of tamoxifen was injected in Cdh5-Cre-ERT2; Z/EG+ pups . Mice were sacrificed at P5 and retinas were isolated for analysis . In situ hybridization , Isolectin B4 staining and ERG immunohistochemistry ( 1:200; SC353 , Santa Cruz ) were performed as described previously . Images of flat mounted retinas were taken at 40x magnification using Zeiss LSM 510 META confocal microscope . Images were processed using Adobe Photoshop and Image J ( National Institutes of Health ) . Mosaic recombination in Cdh5-Cre-ERT2;Plxnd1flox/flox or Z/EG; Cdh5-Cre-ERT2 was analyzed by in situ signal or GFP positivity , respectively , in combination with Isolectin B4 and endothelial nuclear ERG staining . The tip cells were determined as blind-ended endothelial cells that are associated with filopodia protrusions at the sprouting front . Endothelial cells ( either tip or stalk cells ) were counted by combination of Erg positive staining and morphological definition . 84 endothelial cells from 3 Cdh5-Cre-ERT2;Plxnd1flox/flox animals and 64 endothelial cells from 3 wildtype animals were counted for tip cells and stalk cells that were either Plxnd1 positive or negative . 37 endothelial cells from 2 Z/EG;Cdh5-Cre-ERT2 animals were counted for tip cells and stalk cells that were either GFP positive or negative . Statistical significance was tested using two-way Anova . Total RNA was extracted from collagen embedded lung explants , or whole retina using the RNeasy Micro Plus RNA extraction kit ( Qiagen ) according to manufacturer’s instructions . cDNA was generated from 100 ng total RNA using the Superscript III reverse transcription kit ( Invitrogen ) . The following primers were used to detect Plxnd1 and Sema3e transcripts: Sema3e forward 5’-aggctacgcctgtcacataaa-3’ , Sema3e reverse 5’-ccgttcttgatactcatccagc-3’; Plxnd1 forward 5’-gctgactgtagcctatgggga-3’ , Plxnd1 reverse 5’- gccatctggtgggatgtcat-3’ . Lungs were dissected from E12 . 0 embryos in ice-cold dissecting medium ( DMEM + Penicillin/Streptomycin ) , divided into single lobes , washed in dissecting medium ( 30 min , 4°C ) and embedded in a glass-bottom dish ( Mattek ) between two layers of polymerized collagen IA gel ( 1 . 5 mg/ml , Cellmatrix ) prepared as previously described ( Jakobsson et al . , 2006 ) . After solidification of collagen ( 30 min , 37°C ) , imaging medium was added ( DMEM without phenol red , 15% ES cell grade FBS , 30 ng/ml rhVEGF ( R&D systems ) , antibiotics ) . Explants were grown overnight at 37°C , 5% CO2 . Next day , nuclear live stain solution ( NucBlue , Life Technologies , Carlsbad , CA ) was added ( 2 drops/ml for 45 min ) . Imaging medium was replaced and explants were imaged immediately using a Leica SP8 confocal microscope equipped with a Tokai Hit chamber ( 37°C , 5% CO2 ) . 70 μm stacks were acquired at 20x magnification every 10 min . Single cells were tracked manually using a combination of nuclear staining and bright-field images of the sprouts . Per explant , 1–4 sprouts were analyzed for 9–20 hr . Tip cell selection frequency was calculated as selection events observed per hour live imaging . All 3 types of tip cell selection events were considered for quantification of selection frequency . Statistical significance was tested using a permutation test with shuffled genotypes . Explants were dissected at E12 . 0 or E13 . 5 , prepared as described above , and fixed on day 2 or 3 in 4% PFA for 1 hr at RT . Then they were washed in PBS ( 3 x 10 min ) , permeabilized in blocking solution ( PBS + 2% BSA + 0 . 2% Triton ) for 1 hr at RT ( 2 x 30 min ) and incubated with PECAM ( 1:300 , BD ) , Plexin-D1 ( abcam ) or Dll4 ( R&D ) primary antibody in PBS + 0 . 2% Triton overnight at 4°C and additional 6 hrs at RT . Explants were then washed in PBS + 0 . 1% Triton ( 6x 30 min ) , incubated with secondary antibody in PBS + 0 . 1% Triton overnight at 4°C and washed in PBS + 0 . 1% Triton ( 2 x 30 min ) . If HRP-conjugated secondary antibody was used , explants were washed longer ( 3 x 30 min in PBS ) and stained with DAB staining solution ( 1 mg/ml Diaminobenzindine tetrahydrochloride in PBS , Sigma ) for 3 min . For network topology analysis , Alexa488 conjugated secondary antibody was used and explants were imaged at a Leica SP8 confocal microscope at 10x magnification . In the previously established 'MemAgent-Spring Model' ( MSM ) each endothelial cell is comprised of multiple smaller computational elements ( 'agents' ) that represent local sections of cell membrane and actin tension beneath . The memAgents have dynamic internal levels of proteins , which enable them to sense protein levels in the local extracellular environment ( primarily VEGF ligands ) . The endothelial cell integrates this local spatial information to determine its behavior ( e . g . extension of filopodia ) and perform genetic regulatory processes ( e . g . Dll4-Notch ) after a time delay representing the processes of transcription and translation ( parameters D1 and D2 in Figure 1B ) . As the time delay for Notch/VEGF gene expression is not currently known in the mouse , the model was calibrated to match the known Notch periodicity ( 30 min ) of the zebrafish somite clock ( Guidicelli and Lewis , 2004 ) so D1=D2=28 time steps ( representing 7 min of real time ) as it was previously shown that , consistent with period/delay relations in the Notch somite clock models ( Guidicelli and Lewis , 2004 ) , the periodicity of Notch/VEGF signaling in the MSM model = 2 x ( D1+D2+R1+R2 ) , where R1 and R2 are the recovery delays , representing degradation rates , which were both set to 1 ( Bentley et . al . , 2008 ) . To include Sema3E-Plexin-D1 interactions to the existing MSM model of VEGF/Notch/Dll4 signaling in endothelial cells during tip/stalk selection ( full model described in [Bentley et al . , 2008; 2009] ) , four new parameters were included . Two new time delay parameters: D3 controls how long it takes for VEGFR-2 to increase Plexin-D1 protein levels at the membrane and D4 determines the time it takes for an active Plexin-D1 receptor to lower Dll4 expression . Additionally s was added to determine the strength of Plexin-D1 down-regulation of Dll4 ( specifically how many fewer Dll4 are produced for one active Plexin-D1 receptor ) and r3 , which controls how long the down regulation effect lasts for , encompassing the factors such as Plexin-D1 degradation rate , see Figure 1B for schematic . As Sema3E is assumed to be uniformly present around the cell based on experimental data in the mouse retina ( Kim et al . , 2011 ) , activation of Plexin-D1 by Sema3E is not directly modeled , but simply assumed to occur at a constant level . As the exact number of Plexin-D1 receptors on the cell surface is also not known we assume Plexin-D1 receptors vary within the same range as VEGFR-2 receptors ( see ( Bentley et al . , 2008 ) for details ) , and are instantly activated by Sema3E when present . These assumptions produce the most parsimonious model possible , permitting Plexin-D1 levels to be controlled by just the D3 time delay parameter , and Sema3E-Plexin-D1 signaling strength to be determined by the modulation of a single parameter s , which varies the strength of effect of the signaling on Dll4 up-regulation specifically . Dll4 levels were then determined as follows: Dll4t+1=Dll4t+V''δ−P''s where V'' is the number of active VEGFR-2 receptors by VEGF after time delay D1 has been applied , representing the current active VEGFR-2 level affecting gene expression in the nucleus . Likewise P'' is the number of Plexin-D1 receptors ( assumed activated by Sema3E ) able to affect gene expression after time delay D4 . Previously δ , which represents the up-regulation strength of Dll4 by VEGF-VEGFR-2 signaling , was calibrated to 2 to generate matching tip/stalk pattern selection and sprouting behavior in vivo under different conditions ( Bentley et al . , 2008; 2009 ) . So now with a balancing inhibition term s representing reduction in Dll4 via Plexin-D1 , we know that δ - s = 2 is required for normal sprouting . Any combination of δ and s values such that this relation held true would give normal sprouting . To simulate loss of Sema3E-Plexin-D1 signaling s was set to zero . Thus the value for δ chosen ultimately determines the strength of the simulated Sema3E-Plexin-D1 mutant phenotype , hence results are shown throughout across a range of δ values when s = 0 . For control simulations s = δ-2 . Experiments indicate that the rate of Plexin-D1 up-regulation by pVEGFR-2 is fast compared to pVEGFR-2 up-regulation of Dll4 , indicating that together the delays D3+D4 >=D1 . To investigate the effects of varying the temporal regulation of Plexin-D1 on tip cell selection a sensitivity analysis was performed simulating with different delay settings for the new parameters D3 and D4 ( a full analysis of varying delays D1 and D2 is given in [Bentley et al . , 2008] ) . It was found that the lateral inhibition mechanism is strictly sensitive to the values of these new delay parameters relative to the existing delays D1 and D1 . In the model , only a setting of D3+D4 =D1 would allow for normal tip cell selection in control conditions ( Video 1 ) . Even a delay with D3 or D4 = ± 1 timestep in the model ( representing 15 s ) would disrupt the process and tip cells could not be selected and the system falls into unrealistic 'flashing' oscillations as the cells instantly raise and then lower dll4 each time step through the Notch/VEGF negative feedback loop resulting in a counter intuitive hypersprouting rather than inhibited phenotype as no cell is under the inhibition long enough to become a stalk cell ( Video 2 , 3 ) . Interestingly if all delay parameters are set to zero , representing the null hypothesis that no time delays are required to explain the phenotype , the same system behavior occurs as in Video 2 and 3 illustrating the importance of explicitly representing the amount of time that gene expression takes in computational models . Disrupting the degradation rate r3 of Plexin-D1 was also found to have drastic affects on the ability of the system to select tip cells . The r3 parameter was required to satisfy: r3 = r2 = r1 = 1 timestep ( representing 15 s ) . Any increase led to similar irregular flashing oscillations and abrogated tip cell selection as seen with increases to the delays D3 or D4 ( Video 4 and 5 ) . Thus for all simulations D3+D4 =D1 , where D3= 1 and D4 = 27 timesteps . Mosaic vessels follow the same simulation method as simulations of a fully wildtype or mutant vessel , except that at the start of the simulation each cell is randomly assigned a wildtype or mutant setting of the δ and s parameters ( Video 8 ) . 45% mosaicism was calculated as average of simulations with 40% and 50% mosaicism . Results were averaged over 50 runs . In this model a cell can move within the adhered collective of the sprout powered by multiple local junctional adhesion movements ( based on the Cellular Potts Model 'CPM' of differential adhesion [Graner and Glazier , 1992] ) , which are regulated by VEGF/Notch signaling . The stacks were processed and analyzed in 3D using the following Python 2 . 7 modules: Numpy , Scipy , Matplotlib , Opencv2 , Igraph and Networkx . The first step of the algorithm was to apply a Gaussian smoothing filter to the stacks . A standard deviation of 5 µm was used for the Gaussian kernel . Next , the moment-preserving threshold technique ( Tsai , 1985 ) was used in order to find a proper threshold for stack binarization . Pixels having intensity values larger than the calculated threshold were classified as belonging to a vessel . Remaining image components ( Stockman , and Shapiro , 2001 ) smaller than 20000 µm3 were considered background noise and removed from the binary image . A thinning procedure ( Palagyi and Kuba , 1998 ) was then applied to the binary image , resulting in what we call the skeleton ( Costa and Cesar , 2009 ) of the blood vessels . The skeleton tends to present some sets of connected pixels having more than two neighbors each . Such sets were erased from the skeleton and represented as a single pixel at the center of mass of the set . The remaining skeleton pixels having at least three neighbors were classified as branching points , while pixels having one neighbor were considered a termination point . Spurious skeleton segments were removed by an iterative algorithm . First , termination segments smaller than 20 µm were erased , where a termination segment is defined as a segment having one termination point . After erasing such segments , new small termination segments might appear . They were iteratively erased until no new termination segments smaller than 20 µm remained . The sample was then characterized by quantification of the remaining branching points . Finally , in order to validate the analysis , we created images containing both the original image and the final skeletons and verified that the obtained skeletons were accurately representing the original blood vessel structure . Statistical significance was tested using a permutation test with shuffled genotypes .

Many animals have a network of blood vessels that supplies oxygen and nutrients to every part of the body . Each organ contains a unique pattern of blood vessels; some have lots of densely packed vessels , while others have fewer vessels that are more widely spaced . New blood vessels typically form by sprouting from the side of pre-existing vessels . This involves the endothelial cells that line the inner wall of blood vessels moving outwards to create a sprout that is made up of ‘tip cells’ and ‘stalk cells’ . Tip cells are found at the front of the growing vessels and encourage the formation of new sprouts , while the stalk cells trail behind and elongate the sprout . Two signaling pathways that involve two proteins called VEGF and Notch interact with each other to control which cells become tip cells and which become stalk cells . Cells with higher levels of VEGF signaling will become tip cells . These cells also activate Notch signaling , which in turn blocks VEGF signaling in their neighboring cells . This feedback mechanism enables a new sprout to form by forcing cells present around a newly formed tip cell to become stalk cells . However , it was still not understood how the different organs develop blood vessel networks with different densities . In 2011 , researchers revealed that two other proteins , Semaphorin3E and its receptor Plexin-D1 , are expressed in tip cells in the back of the eye in mice and control the VEGF/Notch signaling pathway . Now Kur et al . – including some of the researchers involved in the 2011 work – have used a combination of predictive computer simulations and experimental approaches to understand this interaction in more detail . The analysis showed that Semaphorin3E and Plexin-D1 speed up VEGF/Notch signaling , which causes new tip cells to form at a faster rate , and results in a more densely packed network of blood vessels . For example , in mice that lack Semaphorin3E and Plexin-D1 , VEGF/Notch signaling was slower and new tip cells formed more slowly , which resulted in the blood vessel network at the back of the mice’s eyes being less dense . Kur et al . propose that different organs have different ‘molecular metronomes’ that control the pace of VEGF/Notch signaling . A fast acting metronome would yield a dense network , while a slower one would form a less dense network . This helps to explain how diverse densities of blood vessel networks are formed in different organs . This work may aid efforts to develop therapeutic approaches for controlling the development of new blood vessels in cancers and other diseases .

2016

Temporal modulation of collective cell behavior controls vascular network topology

Familial Advanced Sleep Phase ( FASP ) is a heritable human sleep phenotype characterized by very early sleep and wake times . We identified a missense mutation in the human Cryptochrome 2 ( CRY2 ) gene that co-segregates with FASP in one family . The mutation leads to replacement of an alanine residue at position 260 with a threonine ( A260T ) . In mice , the CRY2 mutation causes a shortened circadian period and reduced phase-shift to early-night light pulse associated with phase-advanced behavioral rhythms in the light-dark cycle . The A260T mutation is located in the phosphate loop of the flavin adenine dinucleotide ( FAD ) binding domain of CRY2 . The mutation alters the conformation of CRY2 , increasing its accessibility and affinity for FBXL3 ( an E3 ubiquitin ligase ) , thus promoting its degradation . These results demonstrate that CRY2 stability controlled by FBXL3 plays a key role in the regulation of human sleep wake behavior . Sleep is vital for all animals . Sleep-wake timing is regulated by the internal biological clock driving physiological rhythms with a period of approximately 24 hr ( Takahashi , 1995 ) . The circadian clock is composed of interlocked transcriptional and translational negative feedback loops ( Lowrey and Takahashi , 2004; Reppert and Weaver , 2001 ) . In mammals , a CLOCK-BMAL1 heterodimer binds to E-boxes and activates gene expression of the Period ( Per ) and Cryptochrome ( Cry ) genes . Translated PERs and CRYs proteins form a complex that enters the nucleus to inhibit their own transcription through direct interaction with CLOCK-BMAL1 heterodimers . PER and CRY proteins accumulating in the nucleus are then degraded over time . As protein levels fall ( depending on rate of degradation ) , the transcription-translation feedback loop begins anew . CRY2 is a principal component in mammalian circadian clocks ( Shearman et al . , 2000; van der Horst et al . , 1999; Vitaterna et al . , 1999 ) . While Drosophila and plant CRY proteins act as photoreceptors contributing to photoentrainment of the circadian clock and other biological processes by binding to flavin adenine dinucleotide ( FAD ) ( Partch and Sancar , 2005 ) , mammalian CRY2 has light-independent transcriptional repressor activity and strongly inhibits E-box-regulated gene expression ( Griffin et al . , 1999; Kume et al . , 1999; Shearman et al . , 2000 ) . The protein stability of CRY2 is fine-tuned by post-translational modification including phosphorylation and ubiquitylation . In addition , various enzyme modifications play a role in CRY2 regulation ( Reischl and Kramer , 2011; Stojkovic et al . , 2014 ) . Among them , FBXL3 is an F-box type E3 ubiquitin ligase which promotes CRY1 and CRY2 ubiquitylation thus leading to proteasome-mediated degradation ( Busino et al . , 2007 ) . Mutations in mouse Fbxl3 or knockout of the Fbxl3 gene dramatically lengthens the period of mouse behavioral rhythms in constant darkness ( Godinho et al . , 2007; Hirano et al . , 2013; Shi et al . , 2013; Siepka et al . , 2007 ) , indicating that the protein stability of CRY1 and CRY2 is a critical determinant of circadian period in mice . However , direct evidence supporting the significance of CRY2 and the post-translational regulation of CRY2 protein in the human circadian clock regulating the sleep-wake cycle has been lacking . Familial Advanced Sleep Phase ( FASP ) is a heritable sleep phenotype characterized by stable early sleep and wake times ( Jones et al . , 1999; Reid and Burgess , 2005; Reid et al . , 2001 ) . The FASP phenotype can segregate as a highly penetrant , autosomal dominant trait in human kindreds . Previously , we have identified mutations in clock genes , including Period2 , Period3 , casein kinase Iδ , and Dec2 causing circadian and sleep homeostasis phenotypes in humans ( He et al . , 2009; Toh et al . , 2001; Xu et al . , 2005; Zhang et al . , 2016 ) . A mutation at the phosphorylation priming site of PER2 attenuates sequential phosphorylation and consequently destabilizes PER2 proteins . The mouse model expressing mutant PER2 exhibits a shortened circadian period accompanied with large phase-advance in sleep-wake rhythms ( Xu et al . , 2007 ) . This sequential phosphorylation region of PER2 was later found to be modulated by another post-translational regulation , O-GlcNAcylation , demonstrating an interplay and competition between phosphorylation and O-GlcNAcylation of serine residues in this region ( Kaasik et al . , 2013 ) . These studies highlighted the important role of post-translational regulation of clock proteins in vivo in humans and also revealed mechanistic insight into the regulation of PER2 . Thus , human genetic studies have provided valuable and unique opportunities to elucidate novel molecular mechanisms of circadian/sleep regulation . Here we report the identification of a novel variant in the human hCRY2 gene that leads to FASP . Generation of a mouse model carrying the mutation revealed that the mutation causes a FASP-like phenotype in mice with altered circadian period and photic entrainment . We found that the mutation in the CRY2 FAD-binding-domain enhances its affinity for FBXL3 , thus destabilizing CRY2 via increased ubiquitylation and targeting for degradation by the proteasome . We conclude that regulation of CRY2 stability by a proper balance of FAD and FBXL3 is essential for the sleep-wake cycle in humans . Through candidate gene screening in FASP families , we identified a missense mutation in the human CRY2 gene , which causes an amino acid conversion from Ala→Thr at position 260 ( A260T ) ( Figure 1A ) . No other novel mutations were found in ~25 candidate circadian genes that were sequenced . The A260T mutation is associated with the circadian phenotype in this FASP family ( Figure 1A ) ( Jones et al . , 1999; Toh et al . , 2001; Xu et al . , 2005; Zhang et al . , 2016 ) . The fraternal twin sisters inherited the mutation from their mother and both reported a strong morning preference ( Horne-Ostberg scores of 84 and 72 ) ( Figure 1—source data 1 ) . The proband also had a very early melatonin onset ( 4:41 P . M . ) , while the averaged melatonin onset of normative samples is 8:50 P . M . ( Burgess and Fogg , 2008 ) . Her melatonin onset time is 3 . 35 standard deviations earlier than expected and among the earliest 0 . 05% of normative samples ( Burgess and Fogg , 2008 ) . Ala260 is located in the FAD binding domain of CRY2 and it is highly conserved in CRY1 and CRY2 proteins of various species ( Figure 1B ) . 10 . 7554/eLife . 16695 . 003Figure 1 . A CRY2 mutation in FASP kindred 50035 . ( A ) Pedigree of the family ( kindred 50035 ) segregating the CRY2 mutation ( A260T ) . Circles and squares represent women and men , respectively . An asterisk marks the proband . A missense mutation from G to A causes an amino acid conversion from Alanine to Threonine at position 260 . ( B ) Amino acid alignment around the mutation site . The A260T mutation is located in the N-terminal portion of the FAD binding domain in CRY2 . This residue is highly conserved among vertebrate species . CC denotes a Coiled-Coil sequence . DOI: http://dx . doi . org/10 . 7554/eLife . 16695 . 00310 . 7554/eLife . 16695 . 004Figure 1—source data 1 . Summary of sleep phenotype of human subjects . Subject IDs correspondence to numbers in Figure 1A . DOI: http://dx . doi . org/10 . 7554/eLife . 16695 . 004 To test whether the A260T mutation causes FASP and has a dominant effect on the circadian sleep-wake cycle , we generated wild-type hCRY2 ( hCRY2-WT ) and mutant hCRY2 ( hCRY2-A260T ) human BAC transgenic ( Tg ) mice ( Figure 2—figure supplement 1A ) . Transgenic mice were subjected to locomotor behavioral analysis using a video camera tracking system . Under conditions of 12 hr light and 12 hr dark ( LD 12:12 ) , both hCRY2-WT and hCRY2-A260T mice entrained stably to the LD cycle ( Figure 2A ) . However , the peak time of resting behavior , as determined by quadratic-function fitting , was significantly advanced in hCRY2-A260T mutant mice ( Figure 2B ) . The activity offset and onset times were also advanced in hCRY2-A260T mutant mice vs . hCRY2-WT mice ( Figure 2C and Figure 2—figure supplement 1B ) . Similarly , hCRY2-A260T mutant mice on a Cry2 null background demonstrated advanced activity onset and offset , especially around the LD transition ( ZT12-13 ) ( Figure 2-—figure supplement 1C ) . These results demonstrate that hCRY2-A260T mice recapitulate the advanced sleep phase seen in the human FASP subjects harboring the CRY2 mutation . 10 . 7554/eLife . 16695 . 005Figure 2 . hCRY2-A260T mice have advanced phase of sleep-wake behavior in a light-dark cycle and a shortened circadian period in constant darkness . ( A ) Mouse movement was tracked by an infrared video camera in LD . The ratio of immobilization time to total daily immobilization time ( upper panel ) and the ratio of walking distance to total daily distance ( bottom panel ) were plotted every 10 min . Data are shown as means with SEM ( n = 8 for hCRY2-WT and hCRY2-A260T ) . ( B ) Peak time of immobility was measured by fitting a quadratic function to data from ZT0 to 13 . Representative examples of curve fitting for hCRY2-WT and hCRY2-A260T are shown here . Data are shown as means with SEM ( n = 8 for hCRY2-WT and hCRY2-A260T , *p<0 . 05 by Student’s t-test ) . ( C ) Onset and offset of locomotor activity . Data are shown as means with SEM ( n = 8 for hCRY2-WT and hCRY2-A260T ) . ( D ) Actograms of wheel-running activity for hCRY2-WT , hCRY2-A260T , and littermate transgene-negative mice . The blue shadows indicate periods when the lights were on . Red lines were fitted to activity onset using ClockLab analysis software . ( E ) Phase-shifts in response to a 30-min light exposure at ZT14 indicated by red arrows in ( D ) . *p<0 . 05 by Tukey’s test ( n = 7 for hCRY2-WT , n = 11 for hCRY2-A260T , n = 10 for WT ) . ( F ) The distribution of period measurements for BAC transgenic mice and transgene negative controls . Period was determined by line fitting of activity onset and chi-square periodogram from day 7 to day 19 in DD . *p<0 . 05 ( n = 15 for hCRY2-WT , n = 14 for hCRY2-A260T and n = 7 for WT ) DOI: http://dx . doi . org/10 . 7554/eLife . 16695 . 00510 . 7554/eLife . 16695 . 006Figure 2—figure supplement 1 . Locomoter activity of h CRY2 BAC Tg mice . ( A ) CRY2 expression in total liver lysate at ZT18 . CRY2 protein was detected by human CRY2 antibody ( Santa Cruz ) and mouse CRY2 antibody ( Hirano et al . , 2013 ) . Copy number was determined by real-time PCR using common sequences in mouse and human Cry2 genes . ( B ) Representative profiles of locomotor activity measured by video recording . Activity onset ( black ) and offset ( red ) are indicated by arrows in the figures . The time of onset and offset for all animals of respective genotypes are averaged and plotted in Figure 2C . ( C ) Locomotor activity and resting behavior of transgenic mice on a mCry2 knockout background . Mouse movement was tracked by an infrared video camera in LD . The ratio of immobilization time to total daily immobilization time ( upper panel ) and the ratio of walking distance to total daily distance ( bottom panel ) were plotted every 10 min . Data are shown as means with SEM ( n = 6 for hCRY2-WT/Cry2 KO , n = 6 for hCRY2-A260T/Cry2 KO ) . The average of activity offset and onset in LD 12:12 are shown ( n = 6 for hCRY2-WT/Cry2 KO , n = 6 for hCRY2-A260T/Cry2 KO ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16695 . 00610 . 7554/eLife . 16695 . 007Figure 2—figure supplement 2 . Wheel-running activity of hCRY2 BAC Tg mice . ( A ) Activity profiles of transgenic mice in LD 12:12 . Ratio of wheel-running counts to total daily counts are plotted every 20 min ( n = 18 for hCRY2-WT , n = 14 for hCRY2-A260T ) . The average of activity offset in LD 12:12 are shown in the right graph . *p<0 . 05 by Tukey’s test ( n = 19 for hCRY2-WT , n = 19 for hCRY2-A260T and n = 7 for WT ) . ( B ) Actograms and activity profiles of transgenic mice on a mCry2 knockout background in LD 12:12 . Ratio of wheel-running counts to total daily counts are plotted every 5 min ( n = 5 for hCRY2-WT , n = 5 for hCRY2-A260T ) . The average of activity offset in LD 12:12 are shown in the right graph . *p<0 . 05 by Tukey’s test ( n = 5 for hCRY2-WT/mCry2 KO , n = 5 for hCRY2-A260T/mCry2 KO and n = 6 for mCry2 KO ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16695 . 00710 . 7554/eLife . 16695 . 008Figure 2—figure supplement 3 . Light phase-shift of hCRY2 BAC Tg mice at ZT22 . ( A ) Phase-shift in response to 30-min light exposure at ZT22 . Mice were exposed to a 30-min light pulse at ZT22 ( indicated by red arrows ) and released into DD after the light pulse . The blue shadows indicate periods when the lights were on . Red lines were fitted to activity onset using ClockLab analysis software . Phase-shift was determined by line-fitting to activity onset ( n = 6 for hCRY2-WT , n = 7 for hCRY2-A260T and WT ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16695 . 00810 . 7554/eLife . 16695 . 009Figure 2—figure supplement 4 . Wheel-running activity of hCRY2Cry2 BAC Tg mice on a m KO background and another BAC Tg mice line . ( A ) Circadian period of the transgenic mice on a mCry2 knockout background . Period was determined by chi-square periodgram from day 7 to day 14 in DD . *p<0 . 05 by Welch’s test ( n = 6 for hCRY2-WT/mCry2KO and n = 5 for hCRY2-A260T/mCry2KO ) . ( B ) Circadian period and phase-shift of hCRY2-A260T mice line#2 . Representative actograms of wheel-running activity of hCRY2-A260T line#2 and littermate controls ( WT ) are shown . The blue shadows indicate periods when the lights were on . Red lines were fitted to activity onset using ClockLab analysis software . Phase-shifts shown here are by 30-min light exposure at ZT14 ( left panel , n = 7 for hCRY2-A260T line#2 , n = 3 for WT ) . Period was determined by line fitting of activity onset from day 7 to day 19 in DD . *p<0 . 05 by Student’s t-test ( right panel , n = 7 for hCRY2-A260T line#2 , n = 3 for WT ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16695 . 009 We next analyzed voluntary wheel-running activity to evaluate phase-shift and free-running period of the circadian clock for the Tg mouse models . Similar to locomotor activity measured by video tracking ( Figure 2A ) , wheel-running activity offset times were advanced in hCRY2-A260T vs . hCRY2-WT in LD on both mCry2 WT and null backgrounds ( Figure 2D; Figure 2—figure supplement 2 ) , while there are no significant differences in activity onset time and acrophase . Interestingly , hCRY2-A260T showed reduced phase-delay when mice were subjected to a 30-min light pulse at ZT14 ( Figure 2D , E ) , whereas phase-advance was normal in response to a light pulse at ZT22 ( Figure 2—figure supplement 3 ) . Thus , mutant mice have reduced sensitivity to entrainment by light at early night compared to control mice . The mice were subsequently released into constant darkness ( DD ) to determine circadian period . The free-running period of hCRY2-A260T ( 23 . 52±0 . 04 hr ) was significantly shorter than that of hCRY2-WT ( 23 . 70±0 . 03 hr ) and WT mice ( 23 . 74±0 . 02 hr ) ( Figure 2F ) . The period shortening phenotype was further enhanced by crossing Tg mice onto the mCry2 null background ( Figure 2—figure supplement 4A ) . The shorter circadian period and reduced phase-delay were also observed in another mutant line ( 23 . 51±0 . 06 hr ) with a higher mutant transgene copy number ( Figure 2—figure supplements 1A , 4B ) , thus excluding the possibility that the phenotype was due to positional effects of the transgene insertion site in the genome . Of note , there is no significant difference in the periods and phase-shifting of hCRY2-WT transgenic vs . transgene negative mice ( Figure 2E , F ) , indicating that the shortening of circadian period and abnormal phase-delay are not simply due to overexpression of hCRY2 . Taken together , these data demonstrate that the phase-advances in mice and humans results from the CRY2 mutation . Data from the transgenic mice suggests that the phase advance may be due to a combination of shortened period and altered sensitivity to photic entrainment . 10 . 7554/eLife . 16695 . 010Figure 3 . Bioluminescence rhythms in tissue cultures from hCRY2-A260T mice and CRY2-A260T stable cell lines . ( A ) Representative rhythms of PER2::LUC bioluminescence in the liver and lung . Data were detrended by subtracting the 24 hr average of bioluminescence . ( B ) Period measurements of the bioluminescence rhythms in liver and lung tissues . Data are shown as means ± SEM ( n = 4 for liver and hCRY2-WT lung , n = 3 for hCRY2-A260T lung , *p<0 . 05 by Student’s t-test ) . ( C ) Peak and trough time of PER2::LUC bioluminescence rhythms of mouse liver and lung tissues . Data are shown as means ± SEM ( n = 4 for liver ) . Data are shown as means ± SD ( n = 2 to 4 for lung , *p<0 . 05 by Student’s t-test ) . ( D ) Representative rhythms of PER2::LUC bioluminescence in MEFs from hCRY2 transgenic mice . Data were detrended by subtracting the 24 hr average of bioluminescence . Period lengths of the bioluminescence rhythms are shown as means ± SEM ( n = 4 , *p<0 . 05 by Tukey’s test ) . ( E ) Representative examples of bioluminescence rhythms of mBmal1-luc in MEFs from transgenic mice on a mCry2 knockout background . Cells were transfected with mBmal1-luc vector 24 hr before the recording . Data were detrended by subtracting the 24 hr average of bioluminescence . Period lengths of the bioluminescence rhythms in the stable cell lines are shown as means ± SEM ( n = 4 , *p<0 . 05 by Games-Howell test ) . ( F ) Representative examples of bioluminescence rhythms of Bmal1-luc in NIH3T3 cells stably expressing FLAG-CRY2-WT or FLAG-CRY2-A260T . Data were detrended by subtracting the 24 hr average of bioluminescence . Period lengths of the bioluminescence rhythms in the stable cell lines are shown as means ± SEM ( n = 3 , *p<0 . 05 by Student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16695 . 01010 . 7554/eLife . 16695 . 011Figure 4 . CRY2-A260T is less stable , particularly in nuclei . ( A ) Luciferase activity driven by mouse Per1 E-box in HEK293 cells . 2 , 5 , 10 and 20 ng of hCRY2 expression vector ( WT or A260T ) was transfected into cells cultured in a 24-well plate . Luciferase activity was normalized to Renilla luciferase activity . Data are shown as means with SEM ( n = 4 , *p<0 . 05 by Student’s t-test ) . ( B ) Protein levels of CRY2-WT and CRY2-A260T in the nuclear and cytosolic fractions of HEK293 cells . Data are shown as means with SEM ( n = 3 , *p<0 . 05 by Student’s t-test ) . ( C ) Degradation assay of CRY2 protein in HEK293 cells . Forty-eight hours after transfection , cells were treated with 100 μg/ml CHX and fractionated into the nuclear and cytosolic fractions . CRY2 protein levels at the starting point ( t = 0 hr ) were normalized to 1 . Data are shown as means with SEM ( n = 3* , p<0 . 05 by Student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16695 . 01110 . 7554/eLife . 16695 . 012Figure 4—figure supplement 1 . The effects of mutations at A260 on protein stability , repressor activity . ( A ) Twenty-four hours after transfection , the culture media was changed to recording media containing 100 µg/ml CHX , and bioluminescence of CRY2-LUC ( WT , A260T or A260D ) was recorded continuously . Bioluminescence normalized to the start point was fitted to an exponential curve to determine half-life of CRY2-LUC ( right panel ) . Data are shown as means ± SEM ( n = 4 , *p<0 . 05 by Tukey’s test ) . Left panel shows representative bioluminescence decay of CRY2-LUC . ( B ) Luciferase activity driven by mouse Per1 E-box . 10 or 25 ng of CRY2 construct was transfected to HEK293 cells cultured in 24-well plate . Luciferase activity was normalized to Renilla luciferase activity . Data are shown as means ± SEM ( n = 3 , *p<0 . 05 by Tukey’s test ) . ( C ) Degradation assay of CRY1 protein in HEK293 cells . Sixty hours after transfection , cells were treated with 100 μg/ml CHX and fractionated into the nuclear and cytosolic fractions . CRY1 protein levels at the starting point ( t = 0 hr ) were normalized to 1 . Data are shown as means with SEM ( n = 3* , p<0 . 05 by Student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16695 . 012 The effect of the mutation on the circadian period and phase angle in the peripheral clock was examined by crossing BAC transgenic mice with mPer2LUCknock-in mice ( Yoo et al . , 2004 ) . Consistent with the behavioral rhythms , shortened clock period was observed in PER2::LUC bioluminescence rhythms of liver and lung cultures from hCRY2-A260T vs . WT mice ( Figure 3A , B ) . The peak and trough time of the bioluminescence rhythms were advanced in both tissues of hCRY2-A260T mice , suggesting that phase of the peripheral clock is also advanced by the mutation in vivo ( Figure 3C ) . Circadian period shortening by the A260T mutation was also found using mouse embryonic fibroblasts ( MEFs ) derived from mice with a mutant vs . WT transgene on both WT and mCry2 knockout backgrounds ( Figure 3D , E ) . In addition , NIH3T3 cells stably expressing CRY2-A260T also displayed a shorter circadian period than CRY2-WT expressing cells ( Figure 3F ) , emphasizing the dominant effect of CRY2-A260T on the circadian period . Our results indicate that CRY2-A260T shortens the circadian period in both central and peripheral clocks , consistent with current understanding that core clock genes such as Cry2 influence physiologies in multiple mammalian organ systems . The CRY2 Ala260 residue resides in the ‘phosphate loop’ responsible for binding to the phosphate of FAD ( Hitomi et al . , 2009 ) ( Figure 1B ) . Mutations at amino acid residues critical for FAD binding affect CRY2 repressor activity of E-box-mediated transcriptional activation ( Czarna et al . , 2013; Hitomi et al . , 2009; Sanada et al . , 2004 ) . Furthermore , Ser265 of mouse CRY2 ( homologous residue of Ser266 in human CRY2 ) is a phosphorylation site , and the S265D mutation , mimicking a phosphorylated serine 265 , reduces CRY2 repressor activity ( Sanada et al . , 2004 ) . Using a Luciferase assay , we found that the A260T mutation weakened CRY2 repressor activity on Per1 E-box-mediated transcriptional activation by CLOCK-BMAL1 ( Figure 4A ) . However , the nuclear CRY2 protein levels in culture cells were decreased by the mutation ( Figure 4B ) , which could potentially account for the reduction of CRY2 repressor activity ( Figure 4A ) . We then examined the degradation of CRY2 proteins by cycloheximide ( CHX ) chase experiments . Consistent with the cellular distribution of CRY2 ( Figure 4B ) , CRY2-A260T was less stable than CRY2-WT in HEK293 cells , especially in the nucleus ( Figure 4C ) . Destabilization of CRY2 by the mutation was further verified in a CRY2-LUC based bioluminescence degradation assay , where the protein decay rate can be determined by recording CRY2-LUC bioluminescence in culture ( Hirano et al . , 2013; Hirota et al . , 2012 ) ( Figure 4—figure supplement 1A ) . Although Ala260 does not directly bind to FAD ( Hitomi et al . , 2009 ) , amino acid conversion from the hydrophobic and small amino acid , alanine , to threonine could alter the conformation of the phosphate loop . This idea is supported by an observation that a mutation from alanine to aspartic acid ( A260D ) caused a more severe effect on CRY2 repressor activity and protein stability than the A260T mutation ( Figure 4—figure supplement 1A , B ) . We found that human CRY1 harboring the corresponding mutation at position 241 is less stable than CRY1-WT , suggesting a common regulatory mechanism for CRY1 and CRY2 by FAD binding ( Figure 4—figure supplement 1C ) . These results indicate that the conformation of the phosphate loop may play critical roles in regulating CRY2 stability and repressor activity . FBXL3 primarily localizes in the nucleus and promotes proteasomal degradation of CRY2 , consequently having a strong impact on the circadian period of mice ( Busino et al . , 2007; Godinho et al . , 2007; Siepka et al . , 2007; Stojkovic et al . , 2014 ) . A previous structural study demonstrated that C-terminal region of FBXL3 interacts with CRY2 through the FAD binding pocket and mutations in the FAD binding domain alter CRY2-FBXL3 interaction ( Xing et al . , 2013 ) . We therefore speculated that the A260T mutation affects the FBXL3-CRY2 interaction , thus altering CRY2 protein stability . We first examined WT and mutant CRY2 stability in the absence of FBXL3 . As anticipated , hFBXL3 knockdown in HEK293 cells increased the stability of both CRY2-WT and CRY2-A260T . Interestingly , the destabilizing effect of the A260T mutation was abrogated by hFBXL3 knockdown ( Figure 5A ) , suggesting the effect of the mutation requires FBXL3 . 10 . 7554/eLife . 16695 . 013Figure 5 . CRY2-A260T binds more strongly to FBXL3 leading to faster degradation of mutant CRY2 . ( A ) Effect of human FBXL3 knockdown on CRY2 protein stability in HEK293 cells . Forty-eight hours after transfection , the culture medium was changed to the recording medium containing 100 μg/ml CHX , and bioluminescence of CRY2-LUC was recorded continuously . Bioluminescence normalized to the value at time 0 was fitted to an exponential curve to determine half-life of CRY2-LUC . Data are shown as means with SEM ( n = 4 , *p<0 . 05 by Welch’s t-test ) . ( B ) Effect of FAD on CRY2 protein levels . Forty-two hours after transfection , HEK293 cells were treated with 100 μM FAD for 6 hr . Data are shown as means with SEM ( n = 3 , *p<0 . 05 by Student’s t-test ) . ( C ) FAD and FBXL3 competition assay . CRY2-FBXL3 complex expressed in HEK293 cells was purified using FLAG antibody . FAD was added to CRY2-FBXL3 complex and incubated at 4°C for 2 hr in vitro . ( D ) CRY2 protein stability in the cells treated with KL001 . Twenty-four hours after transfection , the culture medium was changed to the recording medium containing 100 μg/ml CHX and KL001 . Recording of bioluminescence and calculation of half-life was performed as described above . *p<0 . 05 by Student’s t-test ( n = 3 ) . ( E ) Interaction of CRY2 with FBXL3 in HEK293 cells treated with MG132 for 6 hr prior to harvesting . ( F ) Ubiquitylation of CRY2 . HEK293 cells expressing FLAG-CRY2 were treated with MG132 for 6 hr before harvesting . FLAG-CRY2 was then purified and blotted with anti-Ubiquitin antibody . Quantitative data are shown as means with SEM ( n = 3 , *p<0 . 05 by Student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16695 . 01310 . 7554/eLife . 16695 . 014Figure 5—figure supplement 1 . Structural modeling of A260T CRY2 . ( A ) CRY2 protein stability in the cells treated with KL001 . Twenty-four hours after transfection , the culture medium was changed to the recording medium containing 100 μg/ml CHX and KL001 at the indicated concentration . *p<0 . 05 by Student’s t-test ( n = 3 ) . ( B ) Structure of mouse CRY2 bound to FAD ( PDB code 4I6G ) ( orange ) or FBXL3 ( 4I6J ) ( red ) . Green represents FBXL3 . Short dashed lines indicate the side chain of A259 ( A260 of human CRY2 ) . CRY2 bound to FBXL3 is in a more open conformation compared to the FAD-bound CRY2 . ( C ) Structure of mouse CRY2-WT ( top ) or mouse CRY2-A259T ( corresponded to human CRY2-A260T , bottom ) bound to FBXL3 . The Ala to Thr mutation increases the molecular density and likely changes the conformation of CRY2 from the FAD-bound form to the FBXL3-bound form , in which main chain around A259 moved to left side in this image . DOI: http://dx . doi . org/10 . 7554/eLife . 16695 . 014 FAD stabilizes CRY2 by structurally interfering with the interaction between FBXL3 and CRY2 ( Xing et al . , 2013 ) . We found that treatment of HEK293 cells with FAD increased CRY2-WT protein levels much more than CRY2-A260T ( Figure 5B ) , suggesting that stabilization of CRY2 by FAD was reduced by the mutation . This result supports the hypothesis that the A260T mutation alters the FBXL3-CRY2 interaction . We thus carried out a competitive assay by adding FAD to complexed CRY2-FBXL3 to examine the effect of the mutation on the release of CRY2 from purified CRY2-FBXL3 complexes in vitro . Free CRY2-WT protein levels increased with addition of FAD in a dose-dependent manner ( Figure 5C ) . In contrast , mutant CRY2 was released less readily than CRY2-WT from the complexes even though both forms of CRY2 were bound to FBXL3 at the same level ( Figure 5C ) . In addition , we tested FBXL3-mediated CRY2 degradation using KL001 , a small synthetic molecule known to stabilize CRY1 and CRY2 ( Hirota et al . , 2012 ) due to its structural similarity to FAD ( Nangle et al . , 2013 ) . KL001 stabilizes CRY2-WT in a dose-dependent manner as was previously reported ( Hirota et al . , 2012 ) . However , KL001 failed to stabilize CRY2-A260T to the same extent it did for CRY2-WT ( Figure 5D , Figure 5—figure supplement 1A ) . These results indicate that the A260T mutation weakened the function of FAD and KL001 as inhibitors of FBXL3-mediated degradation of CRY2 and that CRY2-A260T is less stable than CRY-WT , likely due to strengthened interaction with FBXL3 . To determine whether the A260T mutation indeed modifies the interaction of CRY2 and FBXL3 , we performed co-immunoprecipitation analysis . As expected , CRY2-A260T binds more strongly to FBXL3 than CRY2-WT in HEK293 cells ( Figure 5E ) , consequently leading to more ubiquitylation of the mutant protein ( Figure 5F ) . The effect of the mutation on the binding affinity under in vivo conditions will need to be further evaluated when better human CRY2 antibodies ( for immunoprecipitation ) become available . We next performed structural modeling of mutant CRY2 to address how the A260T mutation modulates conformation of the phosphate loop . For modeling , we used the mouse CRY2 structure as the crystal structure of mouse CRY2-FBXL3 complex is available ( Xing et al . , 2013 ) and amino acid sequence in the phosphate loop perfectly conserved ( Figure 1B ) . The published CRY2-FBXL3 structure ( Xing et al . , 2013 ) revealed that space between Ala260 ( corresponding to A259T in mouse ) and Asp442 ( mouse Asp441 ) in FBXL3-binding form ( red ) is more opened vs . the FAD-binding form ( orange , Figure 5—figure supplement 1B ) . The amino acid change from Ala260 to Thr increases molecular density in this space and may alter electrostatic interactions between Ala260Thr and Asp442 . As a result , the mutation likely renders the CRY2-A260T more accessible to FBXL3 binding ( Figure 5—figure supplement 1C ) . This model is consistent with the results from CRY2-A260D mutation ( Figure 4—figure supplement 1 ) . Taken together , we demonstrated that the A260T mutation in the FAD binding pocket endows mutant CRY2 with a higher accessibility and affinity for FBXL3 , therefore leading to faster proteasomal degradation of mutant vs . wild-type CRY2 . Our in vitro studies demonstrate that CRY2-A260T is less stable than CRY2-WT ( Figure 4 ) . To examine the protein levels of mutant vs . wild-type CRY2 under physiological conditions , we used MEFs prepared from transgenic mice . Total protein levels of endogenous CRY2 were significantly lower in synchronized cells derived from hCRY2-A260T vs . hCRY2-WT mice at two different time points ( Figure 6A ) , even though Cry2 transcript levels in hCRY2-A260T mice was higher than in hCRY2-WT mice ( Figure 6B ) . Similarly , CRY2 protein levels from liver nuclear extracts were lower in hCRY2-A260T vs . hCRY2-WT mice ( Figure 6C , D ) , while hCRY2 mRNA levels were higher in mutant mice ( Figure 6E ) . CRY1 protein levels were also decreased by the mutation in nuclear extracts from liver ( Figure 6C , D ) , suggesting that CRY1 is destabilized in hCRY2-A260T mice . At the same time , we found that nuclear expression of PER1 and PER2 were up-regulated in liver extracts from mutant mice , particularly at ZT14 ( Figure 6C , D ) , suggesting that the timing of nuclear accumulation of PER proteins is advanced in hCRY2-A260T mice . Although PER1 and PER2 protein levels were significantly altered , their mRNA levels were not different ( Figure 6E ) . A similar alteration of PER1 and PER2 protein levels in the absence of noticeable changes in mRNA levels was previously reported for Psttm mutant mouse liver ( Yoo et al . , 2013 ) . Psttm mice have a mutation in the Fbxl21 gene and this mutation decreases the protein level of FBXL21 , which functionally competes with FBXL3 . The Psttm mutation resulted in CRY1 and CRY2 protein destabilization and a shorter circadian period in mice ( Yoo et al . , 2013 ) , which parallels the phenotype of our hCRY2-A260T mice ( Figure 2E ) . We found that the expression of clock genes in liver was not significantly altered by the mutation , while the effect was obvious in MEFs ( Figure 6B , E , and Figure 6—figure supplement 1 ) . This is congruent with a previous suggestion that the effect of the CRY2 destabilization has diverged among different tissues ( Hirano et al . , 2013; Yoo et al . , 2013 ) . Collectively , we demonstrated that the A260T mutation destabilizes CRY2 proteins in vivo likely through alteration of FBXL3-mediated CRY2 degradation , leading to perturbation of the circadian clock . 10 . 7554/eLife . 16695 . 015Figure 6 . CRY2-A260T expression is down-regulated in hCRY2-A260T mice . ( A ) CRY2 protein levels in synchronized MEFs . Cells were treated with 100 nM Dex for 2 hr to synchronize the cellular rhythms . Media was change and MEFs were cultured for 24 or 36 hr before harvesting . Quantified band intensities of CRY2 ( mouse CRY2 and human CRY2 ) are shown as means ± SEM ( n = 3 , *p<0 . 05 by Student’s t-test ) . ( B ) mRNA levels of clock genes in synchronized MEFs . Cellular rhythms of MEFs were synchronized with 100 nM Dex for 2 hr . mRNA levels were quantified by real-time PCR . Data are shown as means ± SEM ( n = 3 , *p<0 . 05 by Student’s t-test ) . ( C ) Temporal expression profiles of PER1 , PER2 , CRY1 and CRY2 in mouse liver . Mice were sacrificed every 4 hr on the second day in DD . Asterisks mark non-specific bands . ( D ) Quantification of protein levels in ( C ) . Data are shown as means ± SEM ( n = 3 ) . ( E ) mRNA levels of indicated clock genes in mouse liver . Mice were sacrificed every 4 hr on the second day in DD . mRNA levels of indicated genes were quantified by real-time PCR using gene specific primers . Data are shown as means ± SEM ( n = 3 ) . ( F ) Model of CRY2 protein regulation . In wild-type , FAD binding to CRY2 acts to stabilize by competing with FBXL3 . In hCRY2-A260T transgenic mice or FASP human subjects with CRY2 mutations , FAD does not protect CRY2 from FBXL3-mediated degradation . Destabilization of CRY2 results in shortened period , leading to advanced sleep phase . In Fbxl3 knockout mice or mutant mice ( Overtimeand After-hours ) ( Godinho et al . , 2007; Siepka et al . , 2007 ) , CRY2 is stabilized in the nucleus , thus lengthening the circadian period . DOI: http://dx . doi . org/10 . 7554/eLife . 16695 . 01510 . 7554/eLife . 16695 . 016Figure 6—figure supplement 1 . mRNA levels of clock genes in MEF and liver of hCRY2 BAC Tg mice . ( A ) mRNA levels of indicated genes in synchronized MEFs . Cellular rhythms of MEFs were synchronized with 100 nM Dex for 2 hr . mRNA levels were quantified by real-time PCR . Data are shown as means ± SEM ( n = 3 , *p<0 . 05 by Student’s t-test ) . ( B ) Expression patterns of indicated clock genes in mouse liver . Mice were sacrificed every 4 hr on the second day in DD . mRNA levels of indicated genes were quantified by real-time PCR using gene specific primers . Data are shown as means ± SEM ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16695 . 016 We report here a mutation in hCRY2 that causes FASP in humans . We initially identified this as a novel variant . Since that time , it has been recognized as a rare variant in the SNP database ( rs201220841 ) . The frequency of the A260T allele ( 0 . 00008 in the ExAc database ) is much lower than that of FASP ( 0 . 5% , our unpublished data ) in the general population . This is consistent with the A260T variant found in one of our FASP families being responsible for a small portion of FASP in the general population . Among the mutation carriers of this family ( Figure 1A ) , the proband , her twin sister , and her mother have clear advanced sleep phase . In contrast , the nephew of the proband ( 101374 ) did not have early sleep onset and offset although his genotype is A/G ( Figure 1—source data 1 ) . Considering his age , we classified him as ‘unknown’ , since adolescents and young adults are typically more difficult to categorize as having a definite circadian phenotype due to normal phase delays seen in many people beginning in adolescence and persisting into young adult life ( boys/men > than girls/women ) ( Roenneberg et al . , 2004 ) . When phenotyped at age 21 , subject 101374 was at the statistical peak age of maximum phase delay due to these developmental effects and was prone to be even more phase-delayed by his male sex . Therefore , he may become progressively more phase advanced as he grows older as a result of the CRY2 FASP allele as it is unmasked by these developmental changes . It is also possible that the mutation may not have 100% penetrance and therefore the nephew will never manifest the FASP trait . In order to confirm that the A260T mutation is causative of FASP , we generated mice carrying the mutation and subjected them to detailed behavioral analysis ( Figure 2 ) . Consistent with the other FASP mutations previously reported in hCK1δ and hPER3 , the effects of the human mutation observed in mouse models and in vitro are subtle compared to those found in forward mutagenesis screens . This is expected since the mutations that we identified are found in extant humans in the ‘real world’ . The circadian body clock plays crucial roles in maintaining normal physiological functions . Thus , any mutation manifesting the strong phenotypes seen in mutagenesis screens would almost certainly have been a selective disadvantage if they arose spontaneously in humans . Furthermore , while the phase advance in mice carrying the human PER2 mutation appears to be due largely to a shortening of circadian period , the published PER3 mice and the CRY2 mice reported here both have altered entrainment . We speculate that some of FASP in humans is caused by altered entrainment properties that would not have been detected in forward mutagenesis screens because they have focused almost entirely on measuring period ( not phase ) as the target phenotype . Previously , we have used wheel-running behavior analysis to characterize mouse models carrying human FASP mutations ( Xu et al . , 2007 , 2005 ) . Here , we investigated an additional behavior analysis method . For the hCRY2-A260T mouse model , we also employed continuous video recording ( Figure 2A and Figure 2—figure supplement 1 ) . We found that video recording was quite sensitive to detect advanced sleep phase ( Figure 2A–C and Figure 2—figure supplement 1 ) . Although wheel-running also displays the phase advance of activity offset in the hCRY2-A260T mice ( Figure 2—figure supplement 2 ) , the data from wheel-running is less robust for detecting a phase advance of activity onset . This is likely due to a strong light-masking effect for mouse at the light-to-dark transition for activity onset . Thus , when the lights are on , mice are less likely to run on a wheel . However , smaller amplitude movements like grooming behavior and moving in the cage were detected by video recording . Wheel running was quite sensitive to detect phase advance of activity offset as this is seen during the dark phase of LD 12:12 . In this study , we found that the circadian period was significantly shortened by the A260T mutation in the central and peripheral clocks ( Figure 2F , 3A–E ) . Growing evidence indicates that stability of CRY proteins dominantly determines the circadian period length ( Godinho et al . , 2007; Hirano et al . , 2014; Shi et al . , 2013; Siepka et al . , 2007; St John et al . , 2014 ) : stabilization of CRY lengthens circadian period in mice whereas destabilization of CRY shortens the period . Consistent with this model , our human mutation destabilizes CRY2 and leads to a short circadian period of mouse behavioral rhythms ( Figure 6F ) . Shortened circadian periods have been measured in one FASP human subject ( Jones et al . , 1999 ) , and in mouse models of human FASP mutations ( Xu et al . , 2007 , 2005 ) . Mutant animals having a shorter free-running period such as tau mutant hamster also tend to exhibit advanced phase of behavioral rhythms in LD ( Lowrey et al . , 2000 ) . Thus , it is likely that the CRY2 mutation results in FASP , at least in part , through shortening of the free-running period . Interestingly , our data also indicates that another circadian clock feature , phase resetting by light , is dysregulated in the hCRY2 mutant mouse model . hCRY2-A260T mice have a smaller phase-shift in response to a light pulse in early subjective night compared to hCRY2-WT mice ( Figure 2E; Figure 2—figure supplement 3 ) . In order to live on a 24 hr day in LD 12:12 , wild-type mice need to phase delay a small amount each day because the endogenous circadian period is slightly shorter than 24 hr . A reduced ability to phase-delay observed in the hCRY2-A260T mouse model could contribute to advanced sleep phase , though the mechanism for the altered phase-shift remains to be elucidated . Although the difference in the free-running period of the transgenic mice was subtle as compared to the degree of phase-advance manifested in human mutation carriers or mice carrying the mutant hPER2 transgene ( Figure 2B , F ) , it is reasonable to expect that the alteration of both circadian period and light-induced phase-shifting together will strongly influence the phase angle of entrainment . The after-hours mutation in the mouse Fbxl3 gene causes stabilization of CRY leading to an extremely long circadian period ( Godinho et al . , 2007 ) . These mutant mice also exhibit large phase-shifts in response to light ( Guilding et al . , 2013 ) . The authors speculated that reduced amplitude of the circadian clock in after-hours mutant mice leads to the abnormal enhancement of phase-resetting . In this study , the A260T mutation elevated PER1 and PER2 levels , especially at ZT14 , and the amplitude of PER1 protein rhythm was greater in hCRY2-A260T vs . hCRY2-WT mice ( Figure 6D ) . Thus , the effect of light pulses in the early night may be decreased by the perturbed protein profiles ( increased amplitude ) of PER1 and PER2 in hCRY2-A260T mice . FAD is a chromophore binding to flavo proteins regulating various biological processes and it is required for the light-sensing activity of CRY in various species ( Lin and Todo , 2005; Partch and Sancar , 2005 ) . Drosophila CRY is degraded by the proteasome in response to light signals , which is a trigger for phase-resetting of the circadian clock in flies . However , the ability of mammalian CRYs as a photoreceptor remains controversial as double knockout mice of Cry1 and Cry2 are still able to entrain to light and show Per1 gene induction in SCN in response to light pulses ( Okamura et al . , 1999 ) . These double knockout mice completely lack behavioral rhythms in constant darkness ( van der Horst et al . , 1999 ) . Furthermore , the repressor activity of CRY on E-boxes and its interaction with other clock proteins are independent of light ( Griffin et al . , 1999 ) . These findings emphasize the light-independent role of CRY proteins in mammals . Thus , the physiological role of FAD binding in the mammalian clock has been totally unknown , while a previous study implied that FAD can structurally compete with FBXL3 . Our findings provide the first evidence that FAD functions as a stabilizer of CRY2 protein by modulating FBXL3-CRY2 interaction ( Figure 5 ) . The results presented here demonstrate that the protein stability of CRY2 regulated by the balance of FBXL3 and FAD controls clock speed and sleep/wake timing in mice and humans . Several genetic studies reported that the human CRY2 gene is associated with mood regulation , cancer and glucose homeostasis ( Dupuis et al . , 2010; Hoffman et al . , 2010; Kovanen et al . , 2013; Lavebratt et al . , 2010; Sjöholm et al . , 2010; Zhang et al . , 2013 ) . Although psychiatric disorders , cancer , and metabolic disorders are tightly connected with dysfunction of the biological clock , associations of CRY2 polymorphisms with morning/evening preference or other circadian phenotypes have not been described . One polymorphism in the FBXL3 gene was reported to be associated with diurnal preference ( Parsons et al . , 2014 ) , implying the conserved role of FBXL3 in the human circadian clock . However , it remains to be elucidated whether that variant in FBXL3 is causative ( vs . merely be associated with ) the human circadian phenotype and whether it acts through CRY regulation . Here , we demonstrate that regulatory mechanisms for CRY2 protein are well conserved between mice and humans and that control of CRY2 stability is critical for appropriate phase angle and period of the circadian clock in humans . All human subjects signed a consent form approved by the Institutional Review Boards at the University of Utah and the University of California , San Francisco ( IRB# 10–03952 ) . The consent form includes all confidentiality and ethic guidelines and also indicates not revealing subject information in the publication . All experimental protocols ( Protocol no . AN111686-02 ) were conducted according to US National Institutes of Health guidelines for animal research and were approved by the Institutional Animal Care and Use Committee at the University of California , San Francisco . Subjects were characterized by a previously published procedure established by one of the authors ( CRJ , Jones et al . , 1999 ) . The data were interpreted by one of the authors ( CRJ ) as possible , probable , definite , or severe advanced sleep phase syndrome by at least age 30 . Though ancillary features of ASP ( earlier spontaneous wake time if an earlier bed time is selected ) and potential confounding or masking influences were considered , most participants categorized as ‘definite ASP’ reported spontaneous vacation sleep onset and offset time no later than 21:30 and 05:30 , respectively and had H-O score ( or numerically equivalent childhood MEQ score ) of at least 72 . We considered children and adolescents more difficult to categorize as having a definite , life-long circadian phenotype unless it was severe by all measures including DLMO phase . DNAs purified from blood samples were used to screen for mutations . The salivary dim light melatonin onset ( DLMO ) of the proband was obtained on the last night of the home recordings . DLMO phase was assessed from serial saliva samples ( ~1 mL ) collected at 30 min intervals using ‘Salivette’ saliva collection tubes ( Sarstedt , Inc . , Newton , NC ) in dim light ( </=10 lux ) confirmed by recording the ambient light level before each sample using a luxmeter ( Sinometer , ShenZhen , China ) . Samples were collected , beginning 6 hr before the subject’s typical bedtime . Saliva samples were frozen overnight and then shipped the next day in an insulated box with frozen coolant to another laboratory ( SolidPhase Inc . , Portland , ME ) for radioimmunoassay by test kit ( ALPCO Diagnostics , Windham , NH ) . The lower limit of detection of this assay is 0 . 2 pg/mL . The salivary dim light melatonin onset ( DLMO ) in adults was calculated and compared with a population sample not purposely selected for morning or evening preference by the method and data of Burgess and Fogg ( Burgess and Fogg , 2008 ) . Concurrent sleep logs and Zeo ( Zeo Incorporated , Boston , MA ) EEG recordings were obtained for ten consecutive nights of sleep at home . DNAs purified from blood samples were used to screen for mutations . For this particular family , a list of candidate genes including CLOCK , BMAL1 , PER1-3 , CRY1-2 , DEC1-2 , CSNK1D , CSNK1E , PRKAA2 , NPAS2 , CSNK2A2 , CSNK2B , FBXL3 , GSK3B , PKCA , PRKAA1 , PRKAA2 , RAB3A , RORA , TIMELESS , NR1D1 , and PRKCG were screened . CRY2 ( Accession number; EAW68030 ) A260T was identified as a novel variant specific for mutation carriers of this family ( at the time of identification in 2008 ) . The prevalence of the A260T allele ( rs201220841 ) is 0 . 008% and 0 . 1% in the two sets of public genome databases , of which sample sizes are 121 , 412 and 1323 , respectively . A human BAC RP11-1084E2 containing the entire CRY2 gene on a 189 kb genomic insert was obtained from CHORI ( Children’s Hospital Oakland Research Institute ) . The BAC clone was modified by homologous recombination using the Counter-Selection BAC Modification Kit ( Gene Bridges GmbH , Heidelberg , Germany ) as previously described ( Lee et al . , 2012 ) . Briefly , a linear PCR fragment containing a streptomycin/kanamycin counter selection gene was amplified . The primers for this reaction were designed so that 20 nucleotides would anneal to the streptomycin/kanamycin gene and an additional 40 nucleotides homologous to sequences flanking the mutation site . This PCR product was transferred into the RP11-1084E2 BAC to initiate homologous recombination in the DH10B Escherichia coli strain that already contained the plasmid pSC101- BAD-gbaAtet . The counter selection gene was then removed by a second recombination event using an oligonucleotide carrying the mutation ( G-to-A ) in the center . All relevant segments generated by PCR and recombination were sequence confirmed . Detailed mapping was carried out for the modified BACs to ensure that correct constructs were obtained . Transgenic mice were generated using standard microinjection procedures . The transgenic founders were on a C57BL/6 × SJL F1 background and were backcrossed to C57BL/6 mice in successive generations . The copy number for each transgenic line was calculated by quantitative real-time PCR using common sequences for mouse Cry2 ( reference ) and human CRY2 genes . mPer2Luc knockin mice ( RRID IMSR_JAX006852 ) and mCry2 knockout mouse ( RRID IMSR_JAX016185 ) were purchased from The Jackson Laboratory and crossed with hCRY2 transgenic mice . All mice tested were ~8 week-old males maintained on a C57BL/6J background . Mice were kept in individual wheel running cages with free access to food and water . First , mice were entrained to LD 12:12 . Activity profiles , offset time and acrophase were analyzed using data from day 10 to day 14 in LD . After entrainment to LD for approximately 3 weeks , mice were released into constant darkness ( DD ) for measurement of free-running period . Circadian periods were calculated by line fitting of activity onsets from day 7 to day 19 in DD . To analyze for phase-shifts , mice were given a 30 min-light pulse ( 200 lux ) beginning at ZT14 ( 2 hr after lights-off ) or at ZT22 ( 2 hr before lights-on ) , and then released into DD . Phase-shifts were determined by line fitting of activity onsets from day1 to day7 in DD . All data collection and analysis was done using ClockLab software ( Actimetrics , Wilmette , IL; RRID SCR_014309 ) . Activity onset and offset were defined using the ClockLab software algorithm . The default template is 6 hr of inactivity followed by 6 hr activity for onset ( or vice versa for offset ) . All mice tested were ~16 week-old males maintained on a C57BL/6J background . Mice were kept in individual cages with free access to food and water . Mice were monitored by infrared camera and tracked by an automatic video tracking system ( Stoelting Co . , Wood Dale , IL; RRID SCR_014289 ) . Mice were entrained to LD 12:12 for 1 week and then locomotor activity was recorded for 3 or 4 days . Walking distance and immobility times were calculated using ANY-maze software and data were averaged . Samples with over 500-meter walking distance or below 10 , 000-s immobility time each day were excluded from the statistical analysis due to the failure of automatic tracking . HEK293 cells ( ATCC CRL-1573; RRID CVCL_0045 ) and NIH 3T3 cells ( ATCC CRL-1658; RRID CVCL_0594 ) were purchased from ATTC . Authentication of the cell lines was performed using STR profiling by ATCC . Mycoplasma contamination was checked every 6 months and mycoplasma-free cell lines were used for all experiments in this study . Cells were cultured in DMEM ( Sigma Aldrich , St . Louis , MO ) containing 10% FBS and 100 U/ml Penicillin-Streptomycin ( Thermo Fisher Scientific , South San Francisco , CA ) and maintained by standard methods . Mouse embryonic fibroblasts ( MEFs ) were prepared from E12 . 5 embryos of hCRY2-WT and hCRY2-A260T transgenic mice . After removing the head , paws and internal organs , embryos were chopped and incubated in 0 . 25% trypsin in PBS for 24 hr at 4°C . After incubation for 20 min at 37°C in 0 . 25% trypsin in PBS , cells were dissociated by pipetting in DMEM . Supernatant was cultured in a cell culture dish with DMEM and maintained by standard methods . Cells were transfected with Lipofectamine 3000 transfection reagent ( Thermo Fisher Scientific ) according to manufacturer’s protocol . DNA constructs used for transfections are as follows: hCRY2-WT-Myc-His/pcDNA3 . 1 , hCRY2-A260T-Myc-His/pcDNA3 . 1 , hCRY2-A260D-Myc-His/pcDNA3 . 1 , hCRY2-WT-HA/pCMV-tag2B , hCRY2-A260T-HA/pCMV-tag2B , FLAG-hCRY2-WT/p3×FLAG-CMV-10 , FLAG-hCRY2-A260T/p3×FLAG-CMV-10 , FLAG-hCRY1-WT/p3×FLAG-CMV-10 , FLAG-hCRY1-A241T/p3×FLAG-CMV-10 , FLAG-hFBXL3/p3×FLAG-CMV-10 , FLAG-hBMAL1/p3×FLAG-CMV-10 , FLAG-hCLOCK/p3×FLAG-CMV-10 , hCRY2-WT-LUC/p3×FLAG-CMV-10 , hCRY2-A260T-LUC/p3×FLAG-CMV-10 , hCRY2-A260D-LUC/p3×FLAG-CMV-10 , mPer1-luc/pGL3 , pRL-TK ( Renilla luc expression for internal control in luciferase assay , Promega , Fitchburg , WI ) . 0 . 3kb-mBmal1-luc/pGL3 is a gift by Dr . Yoshitaka Fukada ( University of Tokyo ) . Mutant hCRY2 and hCRY1 expression vectors were generated by PCR-based site-directed mutagenesis , and the mutation was verified by sequencing . For knockdown of human FBXL3 , Hs_FBXL3_1 , Hs_FBXL3_2 FlexiTube siRNA ( QIAGEN , Hilden , Germany ) and control siRNA ( QIAGEN ) were purchased . hCRY2 transgenic mice were crossed with mPer2Luc knock-in mice ( Yoo et al . , 2004; RRID IMSR_JAX006852 ) . Mice were sacrificed between ZT11 and ZT12 . Dissected liver tissues were cultured on Millcell culture membrane ( PICMORG50 , EMD Millpore , Billerica , MA ) in 35 mm dishes . For recording of lung rhythms , dissected lung tissue was placed in 35 mm dishes without Millcell culture membrane . Recording medium was phenol-red free DMEM ( Sigma Aldrich ) containing 10 mM HEPES-pH7 . 0 , 3 . 5 g/L D-glucose , 0 . 2 mM luciferin potassium salt , 0 . 35 g/L sodium bicarbonate , 2% B-27 supplement ( Thermo Fisher Scientific ) , 50 U/ml penicillin-streptomycin ( Thermo Fisher Scientific ) . Bioluminescence was continuously recorded in a LumiCycle 32 instrument ( Actimetrics , Wilmette , IL ) . Bioluminescence was detrended by subtracting 24 hr average of bioluminescence using the LumiCycle analysis software . The periods were determined by dampened sine-curve fitting using LumiCycle analysis . For hCRY2 transgenic/mCry2 knockout MEFs and NIH3T3 stable cells , cells were transfected with 500 ng 0 . 3kbp-mBmal1-luc/pGL3 by Lipofectamine3000 before recordings . Cellular rhythms were synchronized by treatment with 100 nM dexamethasone ( DEX ) for 2 hr . Medium was changed to the recording medium: phenol-red free DMEM ( Sigma Aldrich ) containing 10 mM HEPES-pH7 . 0 , 3 . 5 g/L D-glucose , 0 . 1 mM luciferin potassium salt and 50 U/ml penicillin-streptomycin ( Thermo Fisher Scientific ) . Bioluminescence recording and data analysis were as described in the methods for “Bioluminescence rhythms in tissue culture” . HEK293 cells were transfected with 50 ng Per1-luc expression vector , 25 ng Renilla luc control vector and 2 , 5 , 10 or 20 ng hCRY2 expression vectors . The luciferase assay was performed with Dual-Luciferase Reporter Assay System ( Promega , Fitchburg , WI ) according to the manufacturer’s protocol . Bioluminescence was detected by Synergy H4 Hybrid Multi-Mode Microplate Reader ( BioTek , Winooski , VT ) . Bioluminescence of Firefly LUC was normalized to bioluminescence of Renilla LUC . The hCRY2-LUC fusion protein expressing vector was created by inserting a CRY2-Luc cDNA between EcoRI and BamHI sites in the p3×FLAG-CMV-10 vector . HEK293 cells were transfected with 50 ng hCRY2-LUC vectors and cultured for 24 hr . The culture medium was replaced with recording medium [phenol-red free DMEM ( Sigma Aldrich ) supplemented with 10% fetal bovine serum , 3 . 5 mg/ml glucose , 50 U/ml penicillin-streptomycin ( Thermo Fisher Scientific ) , 0 . 05 mM luciferin , and 10 mM HEPES-NaOH; pH 7 . 0] containing 100 μg/ml cycloheximide ( CHX; Santa Cruz Biotechnology Inc . , Santa Cruz , CA ) . Luciferase activity of hCRY2-LUC was recorded at 10-min intervals at 37°C with a LumiCycle 32 instrument ( Actimetrics ) . The luminescence signals were fitted to an exponential function to quantify the half-life of CRY2-LUC . KL001 ( Cayman Chemical , Ann Arbor , MI ) was diluted in DMSO to a final concentration of 20 mM . Mice were entrained to LD 12:12 for at least 10 days . Mice were transferred to DD , and mice were sacrificed in dim red light on the 2nd day of DD . Liver tissues were collected , followed by nuclear extraction ( Yoshitane et al . , 2009 ) and mRNA extraction . Protein levels and mRNA levels were normalized to LaminB levels and Gapdh levels , respectively . For whole-cell extracts , HEK293 cells were lysed in SDS sample buffer [62 . 5 mM Tris-HCl ( pH 6 . 8 ) , 50 mM DTT , 2% SDS , 10% glycerol] . Preparation of the cytosolic and the nuclear fractions of mouse liver was performed as previously described ( Yoshitane et al . , 2009 ) . Protein samples were separated by SDS-PAGE . Tissues were transfered to PVDF membranes ( EMD Millipore ) with blocking in T-TBS [50 mM Tris-HCl ( pH 7 . 4 ) , 137 mM NaCl , 0 . 1% Tween 20] containing 1% Skim milk . Primary antibodies were reacted in the blocking solution at 4°C overnight . Then , secondary antibodies were reacted in the blocking solution at RT for 2 hr . Proteins were detected with the Western Lightning Plus ECL ( PerkinElmer , Waltham , MA ) . Band intensities were determined using Image J software . β-actin and Vinculin were used as loading controls for total cell lysates , and LaminB and TBP were used as nuclear markers . Proteins were detected with the following antibodies: anti-cMyc 9E10 ( Santa Cruz , sc-40 ) , anti-FLAG M2 ( Sigma Aldrich , F1804 ) , anti-HA Y11 ( Santa Cruz , sc-805-G ) , anti-β-actin ( Abcam , AC-15 , Cambridge , UK ) , anti-Vinculin ( Abcam , ab18058 ) , anti-TBP ( Santa Cruz , sc-273 ) , anti-Ub ( Santa Cruz , sc-8017 ) , anti-hPER1 ( Thermo Fisher Scientific , PA1-524 ) , anti-LaminB1 ( Abcam , ab16048 and Santa Cruz , C20 ) , anti-mPER2 ( Alpha Diagnostic International , PER-21A , San Antonio , TX ) , anti-hCRY2 ( Santa Cruz , sc-130731 ) and anti-mCRY1 ( MBL , PM081 , Woburn , MA ) . Rabbit polyclonal anti-mCRY2 antibody was provide by Dr . Yoshitaka Fukada ( University of Tokyo ) ( Hirano et al . , 2013 ) . Secondary antibodies used were goat anti-mouse IgG-HRP ( Santa Cruz , sc-2005 ) , goat anti-rabbit IgG-HRP ( Santa Cruz , sc-2006 ) and goat anti-guinea pig IgG-HRP ( Santa Cruz , sc2438 ) . Flavin adenine dinucleotide disodium salt hydrate ( FAD , Sigma Aldrich ) was diluted in PBS to a final concentration of 100 mM . HEK293 cells were transfected with plasmid vectors for 10 μg hCRY2-His-Myc ( WT or A260T ) and 10 μg FLAG-hFBXL3 . Forty-two hours after transfection , the cells were treated with 10 μM MG132 ( EMD Millipore ) for 6 hr . CRY2-FBXL3 complex was purified with anti-FLAG M2 affinity gel ( Sigma Aldrich ) . FAD was incubated with CRY2-FBXL3 complex binding to anti-FLAG M2 affinity gel in 40 μl PBS for 2 hr on ice . After centrifugation , the supernatant was collected as the ‘released CRY2’ sample . CRY2 still binding to FLAG-FBXL3 was eluted by adding SDS sample buffer to FLAG-M2 affinity gel . HEK293 cells were transfected with plasmid vectors for hCRY2-His-Myc ( WT or A260T ) and FLAG-hFBXL3 . Forty-two hours after transfection , the cells were treated with 10 μM MG132 ( EMD Millipore ) for 6 hr . CRY2-FBXL3 complex was purified with anti-FLAG M2 affinity gel ( Sigma Aldrich ) and eluted by 300 μg/ml 3×FLAG peptide ( Sigma Aldrich ) . Total RNA was extracted by TRIzol reagent ( Thermo Fisher Scientific ) from MEFs or liver samples of transgenic animals . cDNA was synthesized by Superscript III ( Thermo Fisher Scientific ) for MEFs or GoScript ( Promega ) for liver samples . Quantification of mRNA was performed with GoTaq Real-Time qPCR Kits ( Promega ) using gene specific primers . mRNA levels were normalized to mouse Gapdh levels . Primers: mouse Per1-fw; CAGGCTAACCAGGAATATTACCAGC , mouse Per1-rv; CACAGCCACAGAGAAGGTGTCCTGG , mouse Per2-fw; ATGCTCGCCATCCACAAGA , mouse Per2-rv; GCGGAATCGAATGGGAGAAT , mouse Gapdh-fw; ACGGGAAGCTCACTGGCATGGCCTT , mouse Gapdh-rv; CATGAGGTCCACCACCCTGTTGCTG , mouse Cry2-fw; GGGACTCTGTCTATTGGCATCTG , mouse Cry2-rv; GTCACTCTAGCCCGCTTGGT , mouse Cry1-fw:CCCAGGCTTTTCAAGGAATGGAACA mouse Cry1-rv:TCTCATCATGGTCATCAGACAGAGG human CRY2-fw; CCAAGAGGGAAGGGCAGGGTAGAG , human CRY2-rv; AGGATTTGAGGCACTGTTCCGAGG mouse Dbp FW , AATGACCTTTGAACCTGATCCCGCT mouse Dbp RV , GCTCCAGTACTTCTCATCCTTCTGT mouse Bmal1 FW , GCAGTGCCACTGACTACCAAGA mouse Bmal1 RV , TCCTGGACATTGCATTGCAT mouse Rev-erbα FW , GGGCACAAGCAACATTACCA mouse Rev-erbα RV , CACGTCCCCACACACCTTAC mouse REV-erbβ FW , TGGGACTTTTGAGGTTTTAATGG mouse REV-erbβ RV , GTGACAGTCCGTTCCTTTGC mouse Dec1 FW , ATCAGCCTCCTTTTTGCCTTC mouse Dec1 RV , AGCATTTCTCCAGCATAGGCAG mouse Dec2 FW , ATTGCTTTACAGAATGGGGAGCG mouse Dec2 RV , AAAGCGCGCGAGGTATTGCAAGAC Structural modeling was based on the structure of mouse CRY2 bound to FAD ( PDB code 4I6G ) and mouse FBXL3 ( PDB code 4I6J ) ( Xing et al . , 2013 ) . Modeling of mutant CRY2 was performed using Molecular graphics and analyses were performed with the UCSF Chimera package ( RRID SCR_004097 ) . Chimera was developed by the Resource for Biocomputing , Visualization , and Informatics at the University of California , San Francisco ( supported by NIGMS P41-GM103311 ) ( Pettersen et al . , 2004 ) . All error bars in the figures represent SEM except for Figure 3C . In Figure 3C , the error bars represent SD . No statistical analysis was used to predetermine the sample sizes . Experiments were not randomized and not analyzed blindly . In Figure 2A and Figure 2—figure supplement 1 , the sample with extremely abnormal walking distance ( >500 meter walking distance , due to a failure of automatic video tracking ) was excluded as an outlier according to the Smirnov-Grubbs test . Data was statistically analyzed using R software ( RRID SCR_001905 ) . To assess statistical significance , data was obtained from three or more independent experiments . All data sets were assumed to follow normal distributions by the Kolmogorov-Smirnov test , and homogeneity of variance between compared groups was tested by F-test ( comparison of 2 groups ) or Bartlett test ( comparison of multiple groups ) . Two-tailed paired Student’s t-test or Welch’s t-test was used for the comparison of 2 groups with or without homogeneity of variance . Tukey’s test or Games-Howell test were used for multiple comparisons with or without of homogeneity of variance . Differences with a p value <0 . 05 were considered statistically significant .

Sleep is an essential process in animals . In humans , the disturbance of normal sleep-wake cycles through shift-work or long-term sleep disorders increases the risk of developing conditions including mental illness , cancer and metabolic syndromes . Understanding how sleep-wake behavior is controlled within cells may help researchers to develop effective therapies to reduce the ill effects of disturbed sleep-wakLouise cycles on health . To understand how our sleep-wake cycles are regulated in cells , researchers have been looking for genetic mutations that affect human sleep schedules . For example , some people have a ‘morning lark’ schedule that makes them prone to go to sleep early and rise early the next day . Others are prone to be ‘night owls’ , staying up later at night and waking up later in the morning . By studying the mutations that underlie these behaviors , researchers hope to understand precisely how these genes regulate sleep schedules . Now , Hirano et al . have identified a particular mutation in a gene called Cryptochrome 2 ( CRY2 ) that causes people to have shorter sleep-wake cycles so that they wake up very early in the morning and struggle to stay awake in the evening . For the experiments , mice were genetically engineered to carry the mutant human CRY2 gene , which shortened the sleep-wake cycles of the mice and their responses to light so that they both woke up earlier and went to sleep earlier . Further experiments examined what effect the mutation has on the protein that is produced by CRY2 . The mutation changes the shape of the protein , which allows an enzyme called FBXL3 to bind to the mutant protein more easily and rapidly break it down . The length of sleep cycles may be determined by how long it takes FBXL3 to break down the protein produced by CRY2 . The findings of Hirano et al . may help researchers to develop treatments for people with sleep problems .

2016

A Cryptochrome 2 mutation yields advanced sleep phase in humans

The amyloid β peptide ( Aβ ) is a key player in the etiology of Alzheimer disease ( AD ) , yet a systematic investigation of its molecular interactions has not been reported . Here we identified by quantitative mass spectrometry proteins in human brain extract that bind to oligomeric Aβ1-42 ( oAβ1-42 ) and/or monomeric Aβ1-42 ( mAβ1-42 ) baits . Remarkably , the cyclic neuroendocrine peptide somatostatin-14 ( SST14 ) was observed to be the most selectively enriched oAβ1-42 binder . The binding interface comprises a central tryptophan within SST14 and the N-terminus of Aβ1-42 . The presence of SST14 inhibited Aβ aggregation and masked the ability of several antibodies to detect Aβ . Notably , Aβ1-42 , but not Aβ1-40 , formed in the presence of SST14 oligomeric assemblies of 50 to 60 kDa that were visualized by gel electrophoresis , nanoparticle tracking analysis and electron microscopy . These findings may be relevant for Aβ-directed diagnostics and may signify a role of SST14 in the etiology of AD . The amyloid beta ( Aβ ) peptide represents a 37 to 49 amino acids endoproteolytic fragment of the amyloid precursor protein ( APP ) , a single-span transmembrane protein , which in humans is coded on the long arm of Chromosome 21 within cytogenetic band q21 . 3 ( Kang et al . , 1987 ) . The cellular biology that governs the formation and clearance of Aβ has been understood to play a critical role in Alzheimer disease ( AD ) since it was shown that fibrillary aggregates of Aβ represent the main constituents of extracellular amyloid plaques that accumulate in the brains of individuals afflicted with the disease ( Glenner and Wong , 1984 ) . The endoproteinases responsible for the release of Aβ are broadly referred to as secretases , with β- and γ-secretases being responsible for the critical cleavage reactions that cause the N- and C-terminal release of the Aβ peptide from its APP precursor , respectively . Mutations in the human APP gene ( Goate et al . , 1991 ) or the genes coding for presenilin 1 or 2 , catalytic subunits of γ-secretases ( Rogaev et al . , 1995; Sherrington et al . , 1995 ) , which cause an increase in Aβ levels or shift the balance of Aβ peptides of different lengths in favor of the production of Aβ1-42 , remain the only known causes for early-onset familial manifestations of AD . Although the extracellular amyloid deposits themselves are increasingly being viewed as innocuous sinks for misfolded Aβ , they may continue to play a role as reservoirs from which monomeric or oligomeric Aβ , hereafter referred to as mAβ or oAβ ( see also ( Yang et al . , 2017 ) ) , can emanate ( Selkoe , 2001 ) . These soluble forms of Aβ may interact with other molecules they meet in the extracellular space ( Narayan et al . , 2011 ) , or may bind to molecules embedded in the plasma membrane ( Laurén et al . , 2009 ) , which may be a precursor to their endocytosis ( Jin et al . , 2016 ) . Although it is not known how precisely Aβ , which has been taken up into the cell by endocytosis , can overcome the bilayer that surrounds the endolysosomal vesiculo-tubular network , it is well established that the peptide can also reach the cytoplasm and associates with other intracellular compartments , including mitochondria ( Lustbader et al . , 2004 ) . The ability of Aβ to overcome compartmental boundaries allows the peptide to encounter a wide variety of proteins that may essentially exist anywhere in the brain . Within the extracellular space , Aβ has , for instance , been shown to interact with apolipoproteins E ( Strittmatter et al . , 1993 ) and J ( the latter is also known as clusterin ) ( Narayan et al . , 2011; Ghiso et al . , 1993 ) . These interactions with apolipoproteins are relevant in the AD context , as certain polymorphisms in the genes coding for them have been shown to bestow an increased risk to develop late-onset AD ( Lambert et al . , 2009; Harold et al . , 2009; Corder et al . , 1993 ) . When neurons are exposed to soluble oAβ , a cascade of events is thought to unfold , which leads to the intracellular deposition of hyperphosphorylated tau in the form of so-called neurofibrillary tangles ( NFT ) ( Grundke-Iqbal et al . , 1986a , 1986b ) . The relationship between these two pathobiological features of the disease has remained enigmatic and , although there is broad agreement that signaling downstream of oAβ is toxic for cells , a bewildering number of hypotheses exist regarding the mechanism by which oAβ-dependent toxicity manifests . Receptor candidates proposed to mediate oAβ toxicity include RAGE receptors ( Origlia et al . , 2008 ) , insulin receptor-sensitive Aβ-binding protein ( De Felice et al . , 2009 ) , P/Q calcium channels ( Nimmrich et al . , 2008 ) , sphingomyelinase ( Grimm et al . , 2005 ) , the Aβ parent molecule APP itself ( Lorenzo et al . , 2000; Shaked et al . , 2006; Sola Vigo et al . , 2009 ) , amylin receptors ( Fu et al . , 2012 ) , a subset of integrins ( Wright et al . , 2007 ) , the prion protein ( PrP ) ( Laurén et al . , 2009 ) and the metabotropic glutamate receptor 5 ( which may act as a co-receptor for PrP ( Um et al . , 2013; Hu et al . , 2014 ) ) . Within cells , Aβ has been reported to associate with more than a dozen additional proteins , including glyceraldehyde dehydrogenase ( GAPDH ) ( Verdier et al . , 2005 ) , the mitochondrial ATP synthase complex ( Schmidt et al . , 2008 ) , and a 17-β-hydroxysteroid dehydrogenase X ( HSD10 ) ( Yan et al . , 1997 ) , also known as 3-hydroxyacyl-CoA dehydrogenase type-2 or Aβ-binding alcohol dehydrogenase ( ABAD ) . Countless isolated experimental paradigms , differences in Aβ preparations and a broad spectrum of methods underlie the discoveries of the aforementioned Aβ binding candidates , a dissatisfactory status quo , also lamented by others ( Mucke and Selkoe , 2012 ) . We therefore set out to locate pertinent literature studies that made use of a more systematic mass spectrometry ( MS ) -based discovery approach for the identification of Aβ interactors . Surprisingly , whereas several studies are available that interrogated the APP interactome by affinity purification MS ( Kohli et al . , 2012; Bai et al . , 2008 ) , a similar investigation has , to our knowledge , not yet been reported for the Aβ peptide . To address this shortcoming , we have now undertaken an in-depth search for proteins that can bind to Aβ using biotinylated Aβ peptides as baits and human frontal lobe extract as the biological source for the capture of Aβ binding proteins . We made use of isobaric tagging for the relative quantitation of proteins and capitalized on recent advances in mass spectrometry instrumentation . In addition to confirming many of the previously known Aβ interactors and revealing more than a hundred novel Aβ binding candidates , the study uncovered a surprising and selective interaction between oligomeric Aβ and the small cyclic peptide somatostatin ( SST ) . We demonstrate that SST ( i ) preferentially binds to aggregated Aβ , ( ii ) influences Aβ aggregation , ( iii ) traps a proportion of Aβ in oligomeric assemblies of 50 to 60 kDa , ( iv ) masks the ability of several widely-used antibodies to detect Aβ , and ( v ) may form a complex with monomeric Aβ that can induce tau hyperphosphorylation in primary hippocampal neurons . The primary objective of this study was to generate an in-depth inventory of human brain proteins that oligomeric preparations of Aβ1-42 can bind to using an unbiased in vitro discovery approach . Synthetic Aβ1-42 peptides and brain extracts generated from adult human frontal lobe tissue served in these studies as baits and biological source materials , respectively . oAβ1-42 was prepared by aggregating the peptide at 4°C for 24 hr , using previously described procedures known to generate amyloid-β-derived diffusible ligands ( ADDLs ) ( De Felice et al . , 2009; Krafft and Klein , 2010 ) . ADDLs are understood to be composed of a heterogenous mixture of oligomeric and prefibrillar Aβ aggregates . This heterogeneity ensured that the analysis was not limited to particular oligomeric Aβ assemblies , which were observed to predominate with some alternative preparation protocols ( Barghorn et al . , 2005; Ahmed et al . , 2010 ) . Because the interaction with a given binding partner may involve a binding epitope that comprises N- or C-terminal residues of Aβ1-42 , initially two separate experiments ( I and II ) were conducted , which differed in the orientation designated for tethering the oAβ1-42 bait to the affinity matrix ( Figure 1A ) . To facilitate meaningful comparisons across experiments , the method of Aβ1-42 capture was not based on immunoaffinity reagents . Instead , alternative Aβ1-42 baits were equipped with biotin moieties attached to the N- or C-terminus by a 6-carbon linker chain , enabling their consistent affinity-capture on streptavidin agarose matrices . Large aggregates were removed prior to the bait capture step by centrifugal sedimentation . Biotin-saturated streptavidin agarose matrices served as negative controls and three biological replicates of samples and controls were generated for each interactome dataset by reproducing the affinity-capture step side-by-side on three separate streptavidin agarose affinity matrices that had been saturated with the biotinylated baits . To identify differences in protein-protein interactions of monomeric versus oligomeric Aβ1-42 , a third interactome experiment was conducted in which oAβ1-42-biotin or mAβ1-42-biotin served as baits . Digitonin-solubilized brain extracts , which are known to primarily comprise extracellular and cellular proteins ( except for nuclear proteins ) served as biological starting materials , consistent with the main subcellular areas previously reported to harbor Aβ . Following extensive washes of affinity matrices in their protein-bound state , binders to the bait peptides were eluted by rapid acidification , fully denatured in 9 M urea , and trypsinized . To avoid notorious confounders related to variances in the subsequent handling and analysis of samples , individual peptide mixtures were labeled with distinct isobaric tandem mass tags ( TMT ) in a six-plex format , then combined and concomitantly subjected to ZipTip-based pre-analysis clean-up by strong cation exchange ( SCX ) and reversed phase ( RP ) separation . Four-hour split-free reversed phase nanospray separations were online coupled to an Orbitrap Fusion Tribrid mass spectrometer , which was configured to run an MS3 analysis method . Tandem MS spectra were matched to peptide sequences by interrogating the human international protein index ( IPI ) using Sequest and Mascot algorithms . The relative levels of individual peptides in the six samples could be determined by comparing the intensity ratios of the corresponding TMT signature ions in the low mass range of MS3 fragment spectra . 10 . 7554/eLife . 28401 . 003Figure 1 . Summary of Aβ1-42 interactome analyses . ( A ) Workflow of interactome studies designed to capture binders to oligomeric Aβ1-42 tethered to the streptavidin matrix by N-terminal ( Experiment I ) or C-terminal ( Experiment II ) biotin groups , or comparing binders to oligomeric versus monomeric Aβ1-42 ( Experiment III ) . ( B ) Representative chart from interactome dataset generated in Experiment I , depicting the false discovery rates of peptide-to-spectrum matches and benchmarks of the analysis depth . ( C ) ‘Cellular Component’ Gene Ontology analysis of top 200 proteins that exhibited the most pronounced oAβ1-42 co-enrichment in Experiment I on the basis of their isobaric signature ion distribution . Asterisks indicate TMT labels that were omitted in a subset of quantitative mass spectrometry experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 28401 . 00310 . 7554/eLife . 28401 . 004Figure 1—source data 1 . Experiment I-III interactome data ( alphabetically sorted ) ( Excel file ) . DOI: http://dx . doi . org/10 . 7554/eLife . 28401 . 004 The three comparative Aβ1-42 interactome analyses ( Experiments I-III ) conducted in this study led to mass spectrometry datasets , which were characterized by similar benchmarks of data quality and enabled confident assignments of several thousand mass spectra to human peptides . For example , Experiment I generated 9661 spectra , which passed confidence criteria applied . These filtered spectra could be matched to 4352 unique peptides , which in turn formed the basis for the identification of 1074 protein groups . The designation ‘protein groups’ , as opposed to ‘proteins’ , reflects a reality of a subset of tryptic peptide sequences not being uniquely associated with a specific protein . Whenever encountered , only ambiguous assignments are possible . For this specific dataset , the 1074 protein groups were annotated to comprise 1619 unique proteins . No attempt was made to resolve this residual source of ambiguity at the individual peptide level . Instead , a majority of uncertain identifications were removed by requiring protein identifications to be based on the confident assignment of at least three unique peptides with a minimum length of six amino acids ( Figure 1B ) . To begin to characterize the Aβ1-42 interactome , the top-listed 200 proteins , whose levels were most pronouncedly co-enriched with biotin-oAβ1-42 , were subjected to a gene ontology ( GO ) analysis . This analysis revealed amongst the biotin-oAβ1-42 binding candidates a highly significant overrepresentation ( p<1 . 0 E-50 ) of ‘Cellular component’ GO classifiers that indicate an association with the ‘extracellular vesicular exosome’ , ‘cytosol’ , ‘vesicle’ and ‘extracellular region part’ ( Figure 1C ) . A more detailed analysis of the list of biotin-oAβ1-42 ( B-oAβ1-42 ) candidate interactors revealed that all top-listed 50 candidates were identified on the basis of more than ten peptide-to-spectrum matches ( PSMs ) and with sequence coverages exceeding 50% of their primary structures ( Table 1 ) . Moreover , indicative of high selectivity of the affinity capture procedures applied , the abundance levels of these 50 candidate interactors in the B-oAβ1-42-specific samples exceeded their levels in the negative control sample by more than four-fold . Not surprisingly , the Aβ peptide itself was amongst the proteins identified whose enrichment levels in specific versus unspecific affinity capture samples were most pronounced . In contrast , assignments of APP peptides that map to regions outside of Aβ were low scoring ( that is , did not pass a 95% significance threshold ) and therefore not credible . The highest levels of B-oAβ1-42 co-enrichment were exhibited by peroxiredoxin-5 , rab-1A , lactate dehydrogenase , hypoxanthine-guanine phosphoribosyltransferase , ATP synthase subunits beta and gamma , and acyl CoA thioester hydrolase . Several of the Aβ1-42-biotin candidate interactors revealed by this analysis were previously known to interact with Aβ peptides , including ATP synthase ( Verdier et al . , 2005; Schmidt et al . , 2008 ) and glyceraldehyde-3-phosphate dehydrogenase ( Verdier et al . , 2005 , 2008 ) , and/or had been shown to be present at altered abundance levels in cells exposed to Aβ ( Lovell et al . , 2005 ) ( Table 1 ) . Observations by others , which preceded this study , have provoked a hypothesis that a free N-terminus of Aβ1-42 might be critical for inducing neurofibrillary degeneration ( Jin et al . , 2011 ) and synapse loss ( Shankar et al . , 2008 ) . We therefore explored how the orientation of Aβ-tethering might affect its protein-protein interactions . When the C-terminally biotinylated oAβ1-42-B bait was employed for affinity capture , a lower number of proteins was consistently observed to co-enrich with the bait and their ratios of co-enrichment ( relative to biotin-only negative controls ) were generally lower than those observed with the N-terminally tagged B-oAβ1-42 bait ( see also Figure 1—source data 1 ) . 10 . 7554/eLife . 28401 . 005Table 1 . Top-listed 50 interactors of biotin-Aβ1-42 observed in Experiment I . Except for the Aβ1-42 bait , which is shown in first position , proteins are listed by their enrichment ( relative to the biotin-only saturated negative control matrix ) observed in Experiment I . Note the extensive amino acid sequence coverage exceeding 44% for all proteins listed . Whenever the same proteins were also observed in Experiments II and III , their corresponding enrichment ratios and counts of peptides quantified are shown in additional columns on the right . ( see also Figure 1—source data 1 and Supplementary file 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 28401 . 005PeptidesExperiment I B-oAβ1-42/BExperiment II oAβ1-42-B/B*Experiment III oAβ1-42-B/ mAβ1-42-BAccessionDescriptionCoverageUniqueTotalRatioCountRatioCountRatioCountAAs IPI00006608 . 1APP770 of Amyloid beta A4 protein ( Fragment ) 62 . 34%2488 . 33185 . 77132 . 8037770 IPI00024915 . 3Peroxiredoxin-5 , mitochondrial86 . 45%81812 . 2332 . 1535 . 9023214 IPI00005719 . 1Ras-related protein Rab-1A80 . 49%41810 . 5233 . 286205 IPI00219217 . 3L-lactate dehydrogenase B chain74 . 55%6269 . 8792 . 463334 IPI00218493 . 7Hypoxanthine-guanine phosphoribosyltransferase64 . 22%7229 . 00122 . 458218 IPI00303476 . 1ATP synthase subunit beta , mitochondrial83 . 55%14488 . 5932 . 6121529 IPI00395769 . 2ATP synthase subunit gamma , mitochondrial59 . 60%6258 . 5733 . 2919297 IPI00219452 . 1Cytosolic acyl CoA thioester hydrolase76 . 60%12318 . 38314 . 0331329 IPI00027223 . 2Isocitrate dehydrogenase [NADP] cytoplasmic62 . 56%5308 . 204414 IPI00743713 . 4Pyruvate kinase64 . 77%3417 . 9551 . 8974 . 508599 IPI00029469 . 1Centractin , beta76 . 06%3267 . 8773 . 2539376 IPI00003482 . 12 , 4-dienoyl-CoA reductase , mitochondrial88 . 06%12377 . 74214 . 3228335 IPI00000949 . 1Mu-crystallin homolog80 . 25%11227 . 65112 . 4117314 IPI00028520 . 2NADH dehydrogenase flavoprotein 177 . 59%12317 . 52323 . 4844 . 0820464 IPI00018352 . 1Ubiquitin carboxyl-terminal hydrolase L163 . 68%9177 . 49381 . 8721 . 3919223 IPI00478231 . 2Transforming protein RhoA51 . 30%5117 . 4875 . 748193 IPI00028091 . 3Actin-related protein 354 . 78%6247 . 4591 . 814418 IPI00029468 . 1Centractin , alpha72 . 34%7367 . 34153 . 9433 . 1138376 IPI00335509 . 3Dihydropyrimidinase-related protein 577 . 13%8427 . 0563 . 798564 IPI00479186 . 7Pyruvate kinase isozymes M1/M293 . 03%3707 . 041381 . 681004 . 33123531 IPI00219018 . 7Glyceraldehyde-3-phosphate dehydrogenase98 . 21%35506 . 961602 . 361212 . 62251335 IPI00873027 . 2Carbohydrate kinase domain-containing protein95 . 68%15316 . 95182 . 1731 . 7232347 IPI00179187 . 4DnaJ homolog subfamily A member 374 . 17%6316 . 9339 . 578453 IPI00216456 . 5Histone H2A type 1 C92 . 31%3216 . 83152 . 03100 . 5954130 IPI00016801 . 1Glutamate dehydrogenase 1 , mitochondrial85 . 13%13466 . 82162 . 9922558 IPI00218448 . 4Histone H2A . Z83 . 59%5146 . 7841 . 8850 . 5130128 IPI00220301 . 5Peroxiredoxin-680 . 80%12186 . 6971 . 4221 . 5422224 IPI00472047 . 1NAD-dependent deacetylase sirtuin-278 . 32%7286 . 6842 . 5815369 IPI00169383 . 3Phosphoglycerate kinase 192 . 09%18516 . 61481 . 38131 . 7321417 IPI00895801 . 1Medium-chain specific acyl-CoA dehydrogenase69 . 41%8396 . 3931 . 8212425 IPI00026268 . 3Guanine nucleotide-binding protein beta-183 . 82%7356 . 3842 . 5830340 IPI01015522 . 1Actin , cytoplasmic 196 . 83%8376 . 25332 . 07264 . 71129347 IPI00909207 . 1Peroxiredoxin-292 . 35%9236 . 00131 . 51142 . 4013183 IPI00970967 . 1GSTT2 protein51 . 64%5146 . 007244 IPI00298547 . 3DJ-195 . 24%11225 . 87171 . 4241 . 8430189 IPI00796462 . 1GTP-binding nuclear protein Ran90 . 60%13245 . 82264 . 82144 . 4246234 IPI00784154 . 1Heat shock protein , 60 kDa , mitochondrial94 . 76%35715 . 68692 . 27174 . 2173573 IPI00003362 . 378 kDa glucose-regulated protein58 . 72%6445 . 65201 . 3912 . 427654 IPI00011229 . 1Cathepsin D84 . 95%11355 . 52152 . 1026412 IPI00013683 . 2Tubulin beta-3 chain84 . 44%10435 . 3951 . 1812 . 0485450 IPI00032406 . 1DnaJ homolog subfamily A member 290 . 29%10385 . 2856 . 1019412 IPI00419802 . 43-hydroxyisobutyryl-CoA hydrolase64 . 25%5225 . 277386 IPI00219037 . 5Histone H2A . x92 . 31%5235 . 2331 . 6990 . 6748143 IPI00556376 . 2Dihydropyrimidinase-related protein 189 . 21%21605 . 21571 . 39273 . 7167686 IPI00000792 . 1Quinone oxidoreductase68 . 69%3185 . 184329 IPI00295400 . 1Tryptophanyl-tRNA synthetase , cytoplasmic66 . 67%6335 . 1793 . 972471 IPI00440493 . 2ATP synthase subunit alpha , mitochondrial99 . 46%38895 . 17622 . 16324 . 88131553 IPI00029111 . 3Dihydropyrimidinase-related protein 380 . 99%4495 . 13151 . 42103 . 4019684 IPI00019171 . 1Endophilin-A178 . 41%10375 . 0371 . 2811 . 0740352 IPI00005198 . 2Interleukin enhancer-binding factor 244 . 36%3134 . 9841 . 416390 IPI00030363 . 1Acetyl-CoA acetyltransferase , mitochondrial86 . 65%19504 . 81332 . 64102 . 4543427 IPI00413344 . 3Cofilin-290 . 96%6224 . 79251 . 81136 . 9311166 IPI00007068 . 1Actin-related protein 3B50 . 96%4214 . 7711418 IPI00012011 . 6Cofilin-193 . 98%15294 . 67561 . 75375 . 3456166*Note that SST14 is not included in this list because this protein only came to the fore as the most selective oAβ1-42-B interactor when the ≥6 amino acids and ≥3 peptides per protein requirements were waived . To determine the extent to which the binding of individual Aβ interactors was governed by Aβ aggregation , we next compared directly the interactome of proteins that bind to monomeric versus oligomeric preparations of the Aβ1-42-biotin bait . This experiment revealed a preference amongst the most highly enriched Aβ1-42 candidate interactors for binding to pre-aggregated bait peptides ( Table 1 ) . Exceptions represented the robust and preferred binding of histones H2 , H3 and H4 to monomeric Aβ1-42 bait matrices ( please see Supplementary Table 1 in Supplementary file 1 ) . Moreover , a majority of proteins , which exhibited lower levels of Aβ1-42 co-enrichment in Experiments I and II , were observed to bind preferentially to monomeric Aβ bait peptides ( Figure 1—source data 1 ) . When we waived the requirement that peptides had to be at least six amino acids long to be considered for protein identification , a tandem mass spectrum assigned to the five amino acid sequence ‘NFFWK’ came to the fore , which exhibited pronounced preference for binding to C-terminally tethered oAβ1-42-B . A query of human genome databases revealed that this peptide , owing to its unusual composition , and despite its short length , could only have originated from the well-known paralogs preprosomatostatin or preprocortistatin ( Figure 2A ) . This conclusion was strengthened by the fact that this peptide is naturally preceded by a tryptic cleavage site and the exquisite match between observed and in silico predicted fragment ions ( Figure 2B ) . In light of the high intensity ion counts of fragments observed for this peptide and its robust co-enrichment with oAβ1-42-B in three biological replicates ( Figure 2C ) , it first seemed puzzling that the identification of this protein group was not corroborated by other spectra that map to sequences outside of the ‘NFFWK’ sequence stretch . Searching for a plausible explanation , it became apparent that preprosomatostatin and preprocortistatin give rise to cyclic neuropeptide hormones through a series of posttranslational trimming steps , and the ‘NFFWK’ peptide is the only tryptic peptide derived from the mature hormone that is of sufficient length and distinct sequence to be readily identifiable by MS ( Figure 2A ) . To further explore if binding had occurred to the mature neuropeptide , as opposed to the precursor , and to characterize the Aβ binding requirements , streptavidin-based affinity matrices were next side-by-side pre-saturated with biotin , N- or C-terminally biotinylated Aβ1-42 or an N-terminally truncated Aβ17-42-biotin bait . The subsequent application of an identical interactome analysis workflow as outlined above ( but replacing TMT with iTRAQ isobaric labeling ) ( Figure 1A ) corroborated the propensity of ‘NFFWK’ to bind to oAβ1-42-B but also established that the free N-terminus is indeed essential for this peptide-peptide interaction ( Figure 2D ) . Although neither preprosomatostatin nor preprocortistatin had passed stringent filtering criteria required for inclusion in the Aβ interactome data tables ( because their identification could not be based on at least three unique peptides with a minimum length of six amino acids ) , close inspection of Experiment III data under omission of these filters suggested that mature somatostatin was not only present in the dataset but represented the protein , whose levels were most selectively enriched in oAβ1-42-B affinity capture eluates . This conclusion was further corroborated by two additional PSMs that mapped to regions outside the SST14 neuropeptide sequence and one additional five amino acid peptide of the sequence ‘TFTSC’ that could be assigned ( albeit not unambiguously ) to the mature SST14 neuropeptide sequence itself . More specifically , consistent with the notion that Aβ1-42 interacts separately with the SST14 neuropeptide and preprosomatostatin precursor , the distributions of TMT signature ions derived from tryptic ‘NFFWK’ and ‘TFTSC’ peptides derived from the mature SST14 neuropeptide exhibited identical TMT signature ion profiles patterns that differed fundamentally from the respective TMT profiles of preprosomatostatin peptides which mapped to sequences upstream of the SST14 sequence domain ( Figure 2E ) . In agreement with the interpretation that the ‘NFFWK’ had probably originated from SST14 , not CST17 , no peptides were observed in this or other experiments that could be uniquely assigned to preprocortistatin . Finally , these experiments repeatedly established that the ‘NFFWK’ peptide binds preferentially to oligomeric ( pre-aggregated ) but not monomeric Aβ1-42-biotin ( Figure 2E ) . 10 . 7554/eLife . 28401 . 006Figure 2 . Discovery of somatostatin as a candidate interactor of oligomeric Aβ1-42 . ( A ) Sequence alignment of preprocortistatin and preprosomatostatin . The signal sequence and the boundaries of the bioactive cortistatin and somatostatin peptides are indicated by horizontal bars . Identical residues are highlighted by black background shading , and peptide sequences observed by mass spectrometry are shown in colored fonts . ( B ) Example tandem MS spectrum supporting the identification of a peptide with the amino acid sequence ‘NFFWK’ . Fragment masses attributed to B- and Y- ion series are shown in red and blue colors , respectively . ( C ) Expanded view of MS3 spectrum derived from ‘NFFWK’ parent spectrum in interactome study based on oAβ1-42-biotin bait and biotin only negative control ( Experiment II ) . In this view , the relative intensities of signature ions reflect the relative abundances of the ‘NFFWK’ peptide in the six side-by-side generated affinity purification eluate fractions . ( D ) SST/CST in human frontal lobe extracts binds to oAβ1-42-biotin but not to N-terminal biotinylated or truncated Aβ baits . iTRAQ signature ion intensity distribution in experiment probing the relative ability of four different biotin baits to capture SST/CST from human brain extract . The exclusive presence of a high intensity 116 ion indicates that the ‘NFFWK’ fragment spectrum , which gave rise to this peak distribution , was dependent on SST/CST exclusively associating with oAβ1-42-biotin . ( E ) Preferential binding of SST to pre-aggregated oAβ1-42 . TMT signature ion intensity distributions of four MS3 spectra assigned to preprosomatostatin based on oligomeric or monomeric Aβ1-42-biotin baits ( Experiment III ) . PSMs derived from SST-14 ( ‘TFTSC’ and ‘NFFWK’ peptides ) had in common signature ion intensity distributions characterized by high intensity even-numbered TMT fragments . In contrast , signature ion intensity distributions of preprosomatostatin-derived tryptic peptides outside of the SST-14 coding region were relatively evenly distributed . DOI: http://dx . doi . org/10 . 7554/eLife . 28401 . 006 To further characterize the binding of SST14 to oAβ1-42 , we next undertook a series of biochemical binding experiments . In a first reductionist approach , we explored if binding of SST14 requires co-factors or occurs directly and , therefore , can be observed with synthetic peptides . For this experiment , oAβ1-42 preparations ( generated by incubating mAβ1-42 purified by gel filtration at 37°C for 2 hr with shaking at the speed of 700 rpm ) gave rise to a pronounced signal of high molecular mass ( HMM ) , not seen with mAβ1-42 , when analyzed by denaturing SDS-PAGE and immunoblotting with the 6E10 antibody directed against an N-terminal Aβ1-42 epitope ( Figure 3A ) . Next , streptavidin agarose matrices were pre-saturated with N-terminally biotinylated SST14 and exposed to unlabeled monomeric or oligomeric Aβ1-42 preparations . Finally , affinity capture matrices were boiled in Laemmli sample buffer and again immunoblotted with the 6E10 antibody ( Figure 3B ) . The analysis revealed in samples which had been exposed to oAβ1-42 robust 6E10 signals both in the HMM and low mass region of the blot , the latter migrating at the same level as mAβ1-42 . Because very little low mass signal was observed in samples , which had been incubated with mAβ1-42 , we concluded that the low mass signal in oAβ1-42 exposed samples most likely reflected a release of a fraction of Aβ1-42 molecules from SST14-bound oAβ1-42 aggregates under the harsh denaturing analysis conditions . Taken together , this experiment validated that SST14 binding only occurs when Aβ1-42 is available in oligomeric form and established a direct mode of interaction . 10 . 7554/eLife . 28401 . 007Figure 3 . Validation of SST binding to oAβ1-42 but not to mAβ1-42 . ( A ) Western blot analysis of synthetic mAβ1-42 and oAβ1-42 . The asterisk designates a signal caused by the partial release of mAβ1-42 from high molecular mass ( HMM ) oAβ1-42 in the presence of SDS . ( B ) Biotin-SST affinity capture of oAβ1-42 but not mAβ1-42 . ( C ) Evidence for fluorescence energy transfer between ( FRET ) between Edans-SST14 donor and Aβ1-42-TMR acceptor . Samples containing donor and acceptor peptides at 20 μM concentrations were incubated overnight at physiological pH . Note the profound quenching of the donor signal and increase in acceptor fluorescence relative to negative control preparations that contained only donor or acceptor peptides but were otherwise treated identically . ( D ) Competition FRET analysis based on configuration shown in panel ‘D’ but with unlabeled SST14 or the negative control AVP peptide being added at varying concentrations to the assay mix . Note the rescue of donor fluorescence in the presence of unlabeled SST14 but not AVP . DOI: http://dx . doi . org/10 . 7554/eLife . 28401 . 007 To further validate the SST14-oAβ1-42 interaction and derive a binding constant we employed fluorescence resonance energy transfer ( FRET ) methodology based on a pairing of an SST14 donor and Aβ1-42 acceptor labeled with Edans and TMR fluorophores , respectively . To enable the formation of Aβ1-42 oligomers and aggregates , fluorescence spectra were recorded following a 24 hr incubation . Following excitation at a wavelength of 335 nm , the recording of independent fluorescence spectra of donor or acceptor peptides over a window of wavelengths spanning 350 nm to 650 nm gave rise to the expected Edans and TMR fluorescence intensity maxima at 493 nm and 596 nm , respectively ( Figure 3C ) . Mixing of the two peptides caused profound quenching of the donor signal , concomitant with an increase in acceptor fluorescence , indicating energy transfer . The acceptor sensitization at 596 nm discouraged an interpretation whereby the donor quenching merely reflected an unspecific amorphous co-precipitation of SST14 with Aβ . Next , this experimental setup was exploited to derive a first indication of the binding constant that governs the interaction between SST14 donor and Aβ1-42 acceptor . To this end , various concentrations of unlabeled SST14 were added to the reaction mixture in a FRET competition experiment to determine the concentration ( EC50 ) for restoring donor fluorescence ( Figure 3D ) . A half-maximal effective concentration of 12 . 7 μM SST14 could be deduced from the recorded graph . In contrast , addition of [Arg8]-vasopressin ( AVP ) , a cyclic neuropeptide of similar size and biochemical characteristics as SST14 , did not restore donor quenching in a control experiment , further strengthening the conclusion that the energy transfer reflected a specific interaction between the SST14 donor and Aβ1-42 acceptor . To explore the influence of SST14 on Aβ1-42 more rigorously , a Thioflavin T ( ThT ) fluorescence assay was applied that incorporated previously reported methodology advancements ( Hellstrand et al . , 2010 ) , including the removal of residual Aβ1-42 aggregates by size-exclusion chromatography immediately prior to the recording of ThT fluorescence spectra ( Figure 4A ) . The application of this method revealed that the presence of 15 μM SST14 ( sourced from two independent vendors ) in the reaction mix reproducibly extended the lag phase of Aβ1-42 aggregation by more than two hours and lowered the signal amplitude during the subsequent stationary phase ( Figure 4B ) . The presence of an intact disulfide bridge within SST14 was not required for this delay to occur . No delay was observed when SST14 was replaced by the AVP negative control peptide . In further variations to this experimental setup it was established that pre-incubation of SST14 , at a concentration which causes this peptide to pre-aggregate ( see also Figure 4—figure supplement 1 ) , did not affect its ability to delay ThT incorporation into Aβ1-42 aggregates ( Figure 4C ) . Importantly , the SST14-dependent lag phase extension correlated directly with the concentration of SST14 in the reaction mix ( Figure 4D ) . Moreover , a similar lag phase extension and/or reduction in ThT fluorescence intensity was not observed when the Aβ1-42 peptide was replaced in the same assay with amylin , another well-known amyloidogenic peptide of similar size ( 37 amino acids ) . More specifically , the presence of SST had no influence on the ThT fluorescence curve collected with a reaction mix containing 10 μM amylin . This result was observed even when SST was added to the reaction well at a considerably higher concentration of 100 μM ( Figure 4—figure supplement 2 ) . Finally , an experiment in which SST14 was replaced by CST17 revealed that the latter peptide is even more potent in its ability to interfere with the Aβ1-42 aggregation-dependent ThT incorporation . Thus , whereas the addition of 15 μM SST14 delayed Aβ1-42 aggregation by a few hours , addition of CST at the same concentration completely abrogated ThT incorporation ( Figure 4E ) . 10 . 7554/eLife . 28401 . 008Figure 4 . SST14 and CST17 delay Aβ1-42 aggregation in ThT fluorescence assay . ( A ) Workflow of ThT-based aggregation assay . ( B ) Representative ThT fluorescence charts using synthetic Aβ1-42 alone or in combination with SST14 or negative control AVP peptides . Note that SST14 alone does not contribute to ThT fluorescence in this assay at these relatively low concentrations . ( C ) Pre-aggregation of SST14 did not alter its effect on Aβ1-42 aggregation . ( D ) Evidence that the SST14-dependent delay in Aβ1-42 aggregation is SST14 concentration dependent . ( E ) Like SST , CST17 causes a concentration-dependent inhibition of Aβ1-42-dependent ThT fluorescence . Note that at 15 μM concentrations , CST17 appears more potent in this regard than SST14 . Please see legend for experimental conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 28401 . 00810 . 7554/eLife . 28401 . 009Figure 4—figure supplement 1 . Oligomeric , but not monomeric , Aβ1-42 interferes with SST14-dependent ThT incorporation . ( A ) Workflow of ThT-based aggregation assay . Aβ1-42 was separated by gel filtration to isolate fractions that contained oligomeric or monomeric Aβ1-42 . ( B ) Representative ThT fluorescence charts monitoring SST14-dependent ThT incorporation . Note that SST14 required concentrations upward of 200 μM before it reliably assembled into aggregates that incorporated ThT . Interestingly , the presence of a 100-fold lower amount pre-aggregated Aβ1-42 , but not monomeric Aβ1-42 , reliably delayed SST14-dependent ThT incorporation by several hours . DOI: http://dx . doi . org/10 . 7554/eLife . 28401 . 00910 . 7554/eLife . 28401 . 010Figure 4—figure supplement 2 . The presence of SST14 does not affect the ThT fluorescence curve of amylin . ( A ) Workflow of amylin-dependent ThT fluorecence assay in the presence or absence of SST14 . ( B ) Representative ThT fluorescence charts monitoring amylin-dependent ThT incorporation . Even at concentrations of 100 μM SST14 the aggregation of amylin , measured on the basis of ThT fluorescence , remained unaffected . DOI: http://dx . doi . org/10 . 7554/eLife . 28401 . 010 At concentrations above 3% w/w ( ~20 mM ) the SST14 peptide had been observed to acquire amyloid characteristics , including Congo Red birefringence ( van Grondelle et al . , 2007 ) . In our hands , even considerably lower concentrations of SST14 exceeding 250 μM reproducibly caused SST14 to incorporate ThT following a 2 hr lag phase ( Figure 4—figure supplement 1 ) . We used this property to determine if the presence of monomeric or oligomeric 1 μM Aβ1-42 influences SST aggregation . Note that the presence of 1 μM Aβ1-42 in this experiment does not contribute to ThT incorporation in this reversed assay configuration . More specifically , whereas the presence of 15 μM merely causes a profound lag phase extension ( Figure 4D ) the addition of 250 μM SST14 abolishes the ability of 1 μM Aβ1-42 to form aggregates that can incorporate ThT ( not shown ) . Also note that in contrast to the first assay configuration , when low micromolar concentrations of SST14 were added to 1 μM Aβ1-42 , in this second configuration of the ThT assay , substoichiometric quantities of 1 μM Aβ1-42 were added to a large excess of 250 μM SST . Strikingly , this analysis revealed that the presence of 1 μM Aβ1-42 caused a profound delay in SST14 aggregation-dependent ThT incorporation but only when Aβ1-42 had been allowed to pre-aggregate ( Figure 4—figure supplement 1B ) . Taken together , these experiments corroborated the notion that the interaction of SST14 with Aβ1-42 relies on the presence of pre-aggregated oAβ1-42 . To determine if the presence of SST14 or CST17 simply prevented Aβ1-42 aggregation , or had caused the formation of alternative heteromeric complexes that failed to incorporate ThT , samples prepared as described in the previous section ( but under omission of ThT ) were analyzed by Western blotting using antibodies targeting N-terminal , central or C-terminal epitopes within Aβ1-42 ( Figure 5A ) . Initially , we were interested in observing the influence of incubation time on sample composition ( Figure 5B ) . As expected , the incubation of monomeric Aβ1-42 alone over a span of 18 hr caused its signals to gradually shift to higher molecular mass bands , reflecting the appearance of Aβ1-42 aggregates . In the presence of SST14 or CST17 , 6E10-reactive signals that migrated at the level of monomeric Aβ1-42 also disappeared over the course of 18 hr albeit at a slower rate than observed with Aβ1-42 alone , most likely reflecting the delayed aggregation ( Figure 4D and E ) . Strikingly , however , in the presence of SST14 or CST17 only very weak HMM signals of lower apparent molecular weight ( 50-70 kDa ) than those seen in preparations of Aβ1-42 alone ( 55-200 kDa ) could be seen following 18 hr incubation ( Figure 5B , lanes 8 and 10 ) . To further explore the apparent disappearance of 6E10-reactive bands during this time-course experiment , we next repeated the 18 hr incubation of Aβ1-42 in the presence of varying concentrations of SST14 or CST17 . This experiment documented again an inverse correlation of the signal intensity of SST14 and CST17 concentrations and 6E10-reactive blot signals ( Figure 5C ) . This effect was not limited to detection by the 6E10 antibody but was similarly observed with antibodies targeting a central Aβ1-42 epitope ( 4G8 ) or the very C-terminus of Aβ1-42 ( 12F4 ) ( not shown ) . The disappearance of signals could not be attributed to some unexpected disappearance of the Aβ1-42 peptide in these samples , because silver-stain analyses of the same fractions revealed relatively stronger Aβ1-42 signals in CST17 and SST14 containing samples . Rather , all available data suggest that the interaction of SST14 or CST17 masked the ability of these antibodies to bind to Aβ1-42 . 10 . 7554/eLife . 28401 . 011Figure 5 . Binding of SST14 or CST17 precludes detection of Aβ1-42 with commonly used antibodies . ( A ) Schematic highlighting Aβ1-42 binding epitopes of antibodies used to generate this and the subsequent figure . ( B ) Western blot-based time-course analysis of Aβ1-42 aggregation in the presence or absence of SST14 or CST17 . ( C ) SST14- and CST17-dependent masking of Aβ1-42 binding epitopes . Note the relatively more intense silver-stained bands of monomeric Aβ1-42 ( red arrowhead ) in samples containing the highest concentrations of SST14 or CST17 ( lanes 6 , 7 and 9 ) , yet virtual absence of Aβ-specific immunoblot signals in the corresponding lanes . Note also the faster migrating band ( green arrowhead ) representing monomeric CST or SST observed in samples , which contained the highest concentrations of CST or SST ( lanes 7 and 9 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 28401 . 01110 . 7554/eLife . 28401 . 012Figure 5—figure supplement 1 . Full view of silver-stained gels depicted in Figure 5 , panels B and C . Note the relatively more intense silver-stained bands of monomeric Aβ1-42 ( red arrowhead ) in samples containing the highest concentrations of SST14 or CST17 ( lanes 6 , 7 and 9 ) . Note also the faster migrating band ( green arrowhead ) representing monomeric CST or SST observed in samples , which contained the highest concentrations of CST or SST ( lanes 7 and 9 ) . The asterisk demarks non-specific silver-staining artefacts . Because these signals were not derived from somatostatin , cortistatin or Aβ ( as their intensities did not correlate at all with the differences in the amounts of these peptides in the samples ) they did not aid in interpreting the binding biology that was the focus of the experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 28401 . 012 We next sought to compare the effect of SST14 ( or CST17 ) on the aggregation of Aβ1-42 and Aβ1-40 . To learn more about how the presence of SST14 ( or CST17 ) affects Aβ aggregation , ThT assay fractions were in this experiment collected and further analyzed by Western blotting . To compensate for the abovementioned epitope masking effect of SST14 or CST17 ( Figure 5C ) , these experiments were undertaken with 5 to 10 times higher peptide concentrations than those described above ( Figures 4 and 5 ) . As expected , at these higher concentrations , Aβ1-42 and Aβ1-40 exhibited shorter lag phases in the ThT fluorescence assay of 30 min and 45 min , respectively ( Figure 6A ) . Consistent with the ThT fluorescence data collected at lower peptide concentrations ( Figure 4 ) , the presence of CST17 ( or SST14 , not shown ) , but not AVP , extended the lag phase of Aβ aggregation considerably ( by approximately one hour ) . However , except for slightly longer lag phases observed in fractions containing Aβ1-40 , there was no difference in Aβ1-42 and Aβ1-40 kinetics revealed by this assay . Next , Western blot analyses of ThT assay fractions were undertaken without or with prior boiling in SDS . Interestingly , Aβ1-42-containing samples subjected to the milder treatment , which had been incubated for 18 hr in the presence of SST14 or CST17 , gave rise to 6E10-reactive bands that migrated at 50-60 kDa , not seen in samples incubated either without these peptides or with AVP ( Figure 6B , lanes 2 and 3 ) . These intermediate-sized 6E10-reactive bands were even more pronounced when the same samples had been boiled in the presence of SDS ( Figure 6B , lanes 10 and 11 ) but escaped detection with an antibody directed against the C-terminus of Aβ1-42 ( Figure 6B , lanes 6 and 7 ) . Interestingly , similar bands were not observed when Aβ1-40 served as the substrate for aggregation ( Figure 6B , lanes 22 and 23 ) , suggesting that the two most C-terminal residues of Aβ1-42 confer properties that are essential for the formation of the 50-60 kDa oligomeric assemblies . Only when SDS was present in the sample buffer , low mass bands appeared in the Western blot that migrated at a level expected for the monomeric Aβ peptide , consistent with the interpretation that these peptides were released from larger oligomeric or fibrillar aggregates under denaturing conditions ( see also Figure 5 ) . Strikingly , in samples that contained SST14 ( or CST17 ) a fraction of these low mass bands , less than what would be needed for their detection by silver-staining , were occasionally ( in well-resolved gels ) observed to migrate at levels expected for heterodimers of Aβ1-42 and SST14 ( or CST17 ) ( Figure 6B , lanes 5 , 6 , 10 and 11 ) . Consistent with the 3-amino acid smaller size of SST14 , relative to CST17 , the respective heterodimers containing SST14 migrated slightly faster than those containing CST17 . A five-fold increase in Aβ1-42 levels in the reaction mix further emphasized the appearance of these low mass heterodimers but also revealed strong signals migrating at the expected size of Aβ1-42 dimers ( Figure 6B , lanes 13 to 16 ) . Notably , whereas evidence for heterotrimers composed of two Aβ1-42 molecules and one SST14 ( or CST17 ) molecule was never obtained , we occasionally observed signals that migrated at molecular masses expected for SDS-stable complexes consisting of three Aβ1-42 peptides linked to SST14 ( or CST17 ) ( Figure 6B , lanes 14 and 15 ) . Corroborating this interpretation was the fact that bands matching this size were never observed in fractions lacking SST14 ( or CST17 ) or containing AVP as a negative control peptide . 10 . 7554/eLife . 28401 . 013Figure 6 . Aβ1-42 forms SDS-stable oligomeric complexes of 50-55 kDa in the presence of SST14 ( or CST17 ) . ( A ) In the presence of CST17 ( or SST14 , not shown ) the ThT fluorescence curve or Aβ1-42 or Aβ1-40 is characterized by an extension of the lag phase and a reduction in ThT fluorescence . ( B ) CST17 ( or SST14 ) co-assemble with Aβ1-42 into oligomers of 50-55 kDa that withstand boiling ( lanes 2 and 3 ) but partially disintegrate in the presence of SDS . Immunoblot analyses with antibodies directed against the C-terminus ( x-42 ) ( lanes 6 and 7 ) or an N-terminal epitope ( 6E10 ) ( lanes 10 and 11 ) revealed bands of 5-6 kDa , consistent with the existence of SDS-resistant heterodimeric complexes of mAβ1-42 and SST14 ( or CST17 ) . Note the well-defined oligomeric bands of 50 and 55 kDa ( lanes 10 and 11 ) that were observed in samples derived from the co-incubation of SST14 ( or CST17 ) with Aβ1-42 ( lanes 10 , 11 , 18 , 19 ) . Note also that signals interpreted to represent trimeric Aβ1-42 ( t ) ( lanes 13-16 ) , but not dimeric Aβ1-42 ( d ) , can be seen to migrate slower in the presence of SST14 ( or CST17 ) ( lanes 14 and 15 ) . Finally , intensity levels of homodimeric Aβ1-42 bands are reduced in the presence of SST14 ( or CST17 ) ( compare lanes 17 and 20 with lanes 18 and 19 ) . Black arrowhead labeled with ‘m’ , ‘d’ , and ‘t’ designate bands interpreted to consist of monomeric , dimeric and trimeric Aβ1-42 . Green and red arrowheads were used to label bands interpreted to represent SDS-stable heteromeric building blocks consisting of SST14 ( or CST17 ) bound to monomeric and trimeric Aβ1-42 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 28401 . 013 These biochemical analyses indicated the formation of SDS-stable oligomeric complexes of Aβ1-42 and SST14 ( or CST17 ) made of building blocks comprising one or three Aβ1-42 peptides . To delineate key residues and minimal components required for SST to interact with Aβ1-42 , we next had a series of SST-derived peptides custom-synthesized that differed in one or two amino acids from the wild-type SST sequence or were truncated , thereby missing the internal disulfide bridge . Based on prior high-resolution NMR models available for the cyclic peptide and structure-activity relationship data derived from SST-receptor docking studies , we hypothesized that a striking hydrophobic ‘belt’ composed of three phenylalanines and a tryptophan , which is flanked by two lysine side-chains in the 3D rendering of SST ( Figure 7A , B and C ) , might also be critical for binding to Aβ1-42 . With regard to the juxtaposed phenylalanines , we were in particular interested to learn if the well-known ability of their side-chains to engage in pi stacking plays a role in the SST-Aβ1-42 interaction . Interestingly , the replacement of one or two of the phenylalanines to leucines did not rescue the lag phase extension phenotype in the ThT fluorescence assay , indicating that the aromatic nature of these residues is not essential for the interaction . However , when we replaced the single tryptophan present in position 8 both the lag phase extension and the reduction in total ThT fluorescence observed in the presence of wild-type SST were rescued ( Figure 7D ) . This outcome was documented when tryptophan was replaced with alanine , proline , histidine or tyrosine . The fact that even the replacement with tyrosine rescued the lag phase extension phenotype indicates that tryptophan does not merely act in this context by providing an aromatic side-chain but suggests that other structural features present in tryptophan provide the necessary fit for binding to Aβ1-42 in this PBS-based system . 10 . 7554/eLife . 28401 . 014Figure 7 . Tryptophan-8 in SST14 sequence is essential for lag phase extension of Aß1-42 in ThT incorporation assay . ( A–C ) NMR-structure of SST14 in 5% D-mannitol ( RCSF PDB structure ID: 2MI1 , Model 1 ) , adapted from Anoop et al . ( 2014 ) . Renderings were generated in NGL 3D viewer ( powered by MMTF ) . ( A ) Backbone of SST14 emphasizing secondary structure . ( B ) and ( C ) Stick and surface models of SST14 with coloring emphasizing relative hydrophobicity . ( D ) and ( E ) Thioflavin T absorbance assay data based on SST point mutants and deletion constructs , respectively . Please see legend for sample compositions . DOI: http://dx . doi . org/10 . 7554/eLife . 28401 . 01410 . 7554/eLife . 28401 . 015Figure 7—figure supplement 1 . In the presence of SST , Aß1-42 forms smaller quaternary assemblies . ( A–E ) Particle size distributions determined by nanoparticle tracking analysis . Traces represent averages of particle size distributions determined in five consecutive analyses . Note that the analysis buffer without peptides or preparations of SST alone did not give rise to measurable particles in this analysis ( not shown ) . ( E ) Nanotracking analysis of 100 nm polystyrene latex microsphere standard . ( F ) D10-D90 summary of particle size distributions . The centre line represents D50 , corresponding to data in panels ‘A’ to ‘D’ . DOI: http://dx . doi . org/10 . 7554/eLife . 28401 . 01510 . 7554/eLife . 28401 . 016Figure 7—figure supplement 2 . Negative stain electron microscopy of Aβ1-42 and Aβ1-42–SST14 complexes . ( A ) Aβ1-42 was fibrillized in PBS at a concentration of 50 μM . Individual Aβ1-42 amyloid fibrils and small clusters were visualized . ( B ) Incubation of equimolar concentrations ( 50 μM ) of Aβ1-42 and SST14 under identical conditions resulted in oligomeric assemblies only . No amyloid fibrils were observed . Magnification bars = 100 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 28401 . 016 Analogous experiments with truncated SST peptides revealed that a peptide encompassing SST residues 5-11 can replace wild-type SST with regard to its effect on Aβ1-42 ( Figure 7E ) , thereby upholding the conclusion that the ability of SST to form its internal disulfide bridge is not essential for this effect ( Figure 4B ) . However , further stripping of one amino acid from each end of this peptide resulted in a SST6-10 peptide , which lacked the ability to influence Aβ1-42 aggregation . We next compared Aβ aggregate size populations in the presence or absence of SST by nanoparticle tracking analysis ( based on Nanosight technology [Filipe et al . , 2010] ) . This orthogonal method documented that Aβ aggregates present in ThT aggregation assay fractions after 75 min of incubation were profoundly shifted toward smaller particle sizes when samples were co-incubated with SST ( Figure 7—figure supplement 1A and B ) . Importantly , the replacement of SST with an SST-W8P negative control peptide did not result in this particle size shift toward smaller aggregates ( Figure 7—figure supplement 1C ) , indicating that the altered particle size distribution is caused by the SST-Aβ interaction , rather than by some unspecific effect of the peptide . Negative stain electron microscopy corroborated the reduction in aggregate size when Aβ1-42 was incubated with SST versus Aβ1-42 alone ( Figure 7—figure supplement 2 ) . Incubating 50 μM of Aβ1-42 under aggregation conditions resulted in the appearance of characteristic amyloid fibrils ( Figure 7—figure supplement 2A ) , while in the presence of equimolar concentrations of SST only oligomeric structures were observed ( Figure 7—figure supplement 2B ) . Taken together , these results demonstrated exquisite specificity of the SST-Aβ1-42 interaction and corroborated the conclusion that Aβ assemblies in the presence of SST are smaller than those observed following Aβ-only aggregation . A 2009 milestone report revealed that signaling downstream of SST positively modulates transcription of the neprilysin gene ( Saito et al . , 2005 ) . Consistent with the known role of neprilysin as one of only a few proteases known to contribute to Aβ1-42 degradation , the authors reported an inverse relationship between SST and Aβ1-42 levels in mice deficient for the somatostatin gene . To begin to assess the physiological significance of the interaction between Aβ1-42 and SST14 observed in this study , an experimental paradigm was needed that is both amenable to manipulation and can be provoked to exhibit a phenotype considered relevant in the context of AD research . The primary hippocampal neuron assay , first described in detail by Jin et al . ( 2011 ) , fits this description ( Figure 8A ) and was favored over an analysis based on the technically challenging enrichment of the relatively small number of GABAergic interneurons known to release SST naturally . In agreement with prior data by others , robust tau hyperphosphorylation was observed when mouse hippocampal neurons , which had been cultured for 18 days , were exposed to oAβ1-42 but not when mAβ was added to the cell culture dish ( Figure 8B ) . For the purposes pursued here , we restricted the tau phospho-epitope characterization to monitoring occupancy of pThr231 ( AT180 ) and pSer202/pThr205 ( AT8 ) phospho-acceptor sites , two of several known sites that undergo robust hyperphosphorylation also in AD brains , because we had noted that these phospho-epitopes were most responsive to oAβ exposure in the original report ( Jin et al . , 2011 ) . We then modified the assay by the concomitant addition of synthetic SST14 , other cyclic control peptides , or bombesin , which does not carry an internal disulfide bridge but was observed to share aggregation characteristics with SST ( Maji et al . , 2009 ) . In a series of follow-up experiments , we added a pre-aggregation step for the peptide and/or Aβ1-42 , varied the pH or duration under which this pre-aggregation was conducted , or changed the duration of cellular exposure to the peptide mixture as described in detail in the figure legend ( Figure 8A , C , D and E ) . Because SST is naturally released from cells through the regulated secretory pathway when densely packed granules fuse with the plasma membrane ( Maji et al . , 2009 ) , their content does not dissolve instantly when it encounters the extracellular milieu but dissolves with a time-scale of hours ( Anoop et al . , 2014 ) . We therefore reasoned that low micromolar SST concentrations , including residual SST aggregates of varying size , might also be locally encountered by neurons in the brain in proximity to SST release sites . 10 . 7554/eLife . 28401 . 017Figure 8 . Exposure of primary hippocampal neurons to SST14 and Aβ1-42 can potentiate phosphorylation of tau at sites known to undergo hyperphosphorylation in AD . ( A ) Workflow of primary hippocampal neuron assay . Please see methods section for details on peptide preparation . ( B ) Addition of preaggregated ( 24 hr ) but not monomeric ( 0 hr ) Aβ1-42 to primary hippocampal neurons causes within 24 hr an increase in tau phospho-occupancy at AT8 and AT180 phosphorylation sites , in the absence of an effect on total tau levels ( K9JA ) . ( C ) Addition of SST or CST bioactive peptides in the absence of Aβ1-42 , leads to a reduction in tau phospho-occupancy . ( D ) Addition of SST14 or CST17 , but not negative control cyclic peptides , can potentiate tau phospho-occupancy in the presence of monomeric Aβ1-42 . ( E ) Tau hyperphosphorylation observed following addition of SST14 and mAβ1-42 to the cell culture appears to depend on conducive SST14 pre-incubation conditions . In all panels red and green font designates parameters ( pH and pre-aggregation periods ) applied to the preparation of Aβ1-42 and other neuropeptides , respectively . Peptides were added separately to the cell culture medium except for samples 3-11 , Panel E , when peptides were mixed and pre-incubated for an additional 1 or 3 hr . Exposures of cells to peptides were for 24 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 28401 . 017 When primary hippocampal neurons were exposed to SST14 alone , levels of tau phospho-occupancy at pSer202/pThr205 ( AT8 ) were slightly reduced relative to vehicle-treated control cells . A similar reduction in phospho-occupancy was seen when SST14 was replaced with SST28 ( Figure 8C , lanes 2-5 ) , the longer physiologically relevant and bioactive version of this peptide , or the dominant preprocortistatin-derived regulatory peptides , CST17 or CST29 . The addition of these peptides together with oAβ1-42 did not prevent the anticipated tau hyperphosphosphorylation-inducing effect of oAβ1-42 exposure ( Figure 8C , lanes 8-11 ) . Interestingly , the SST- dependent reduction in tau phosphorylation at the AT8 site , which was also observed when SST was subjected to a pre-aggregation step before its addition to primary hippocampal neurons ( Figure 8D , lane 2 ) , was not seen in the presence of other identically treated control neuropeptides , including AVP , bombesin and oxytocin ( Figure 8D , lanes 3-5 ) . Most remarkably , however , both SST14 and CST17 , but not the aforementioned control peptides , were observed to potentiate tau hyperphosphorylation under the latter conditions when Aβ1-42 was concomitantly added in monomeric form to the cell culture medium ( Figure 8D , lanes 7 and 8 ) . This observation was notable on several grounds: ( i ) it indicated that SST14 and CST17 might be equivalent also in this experimental paradigm; and ( ii ) it revealed a novel way in which tau hyperphosphorylation can be stimulated that , remarkably , was reliant on the co-addition of monomeric , not oligomeric , Aβ1-42 to the cell culture medium . The objective of a series of subsequent experiments was to better define the experimental conditions that needed to be met for this SST-dependent tau hyperphosphorylation to occur . This line of investigation was pursued because , in contrast to the oAβ1-42-induced tau hyperphosphorylation phenotype , which was robustly observed , SST-dependent tau hyperphosphorylation following co-administration of monomeric Aβ1-42 was not observed every time . Several plausible explanations can be invoked to explain such an experimental behavior , including reliance on experimental paradigms that are governed by threshold-effects of critical parameters or involve a stochastic component . Our current results argue against a stochastic limitation and are instead consistent with the view that once conducive conditions are met , the phenotype can be reliably observed . This view is , for instance , corroborated by one series of experiments , in which SST14 pre-aggregation was allowed to proceed for 24 or 48 hr . This experiment led to a tau hyperphosphorylation phenotype in three independent biological replicates only when the 48 hr pre-aggregated SST14 was added to the cell culture dish ( Figure 8E , lanes 5 , 8 and 11 ) . As before , this molecular phenotype depended on the co-addition of monomeric Aβ1-42 and was again not observed when monomeric Aβ1-42 was added to the cell culture medium alone ( Figure 8E , lanes 3 , 6 and 9 ) . The current study was conducted with the intent to produce an in-depth inventory of proteins within the human brain that oAβ1-42 binds to . The study made use of a hypothesis-free discovery approach that capitalized on advanced workflows for the high-pressure nanoflow reversed-phase separation and relative quantitation of peptides , as well as recent improvements to mass spectrometry instrumentation . Taken together , these advances afforded an unprecedented depth of analysis of oAβ1-42 interactors in three independent interactome datasets , each undertaken with three biological replicates of samples and controls . More specifically , these analyses facilitated the direct comparisons of binders to oAβ1-42-baits that were tethered to affinity capture matrices through N- or C-terminal biotin moieties , or had been captured in oligomeric versus monomeric form . From the large amount of data the study generated , several observations stand out: ( 1 ) despite its small size , oAβ1-42 baits were observed to bind reproducibly to more than one hundred proteins in all analyses undertaken; ( 2 ) in particular , when oAβ1-42 was conjugated to the affinity matrix through an N-terminal biotin moiety , a large number of candidate interactors co-enriched with the bait , corroborating a scenario whereby the hydrophobic C-terminal amino acid sequence stretch within Aβ1-42 is probably mostly protein- or lipid-bound; ( 3 ) the small cyclic peptide somatostatin , which has previously been implicated in the etiology of AD but had not been shown to interact with Aβ , was observed to bind directly to oAβ1-42 in several orthogonal biochemical assays; ( 4 ) the presence of SST14 was observed to compromise antibody-based detection of Aβ1-42; ( 5 ) distinct assemblies of 50-60 kDa , reminiscent of the previously reported Aβ*56 , were robustly generated in the presence of SST in reaction mixtures with Aβ1-42 but not with Aβ1-40; and ( 6 ) whereas Aβ alone does not induce tau hyperphosphorylation in a primary hippocampal neuron assay , unless it is added to the cell culture medium in a pre-aggregated form , the concomitant exposure of cells to monomeric Aβ and pre-aggregated SST can induce tau hyperphosphorylation at key AD-associated phosphorylation sites . Several of the oAβ1-42 binders observed in this work , including clusterin ( Narayan et al . , 2011; Ghiso et al . , 1993 ) , 17-β-hydroxysteroid dehydrogenase X ( HSD10 ) ( Yan et al . , 1997 ) , the ATP synthase complex ( Schmidt et al . , 2008 ) , GAPDH ( Verdier et al . , 2005 ) , and the prion protein ( Laurén et al . , 2009 ) , had previously been proposed to represent Aβ interactors . HSD10 , also known as ERAB , 3-hydroxyacyl-CoA dehydrogenase type-2 or Aβ-binding alcohol dehydrogenase ( ABAD ) , initially emerged from a yeast two-hybrid screen as an Aβ binding protein ( Yan et al . , 1997 ) . Follow-up validation work , which preceded this study , generated evidence that this interaction may contribute to Aβ-mediated mitochondrial toxicity ( Lustbader et al . , 2004; Yao et al . , 2011; Valaasani et al . , 2014 ) . The prion protein , which has been described as a major cell surface receptor for oligomeric Aβ ( Laurén et al . , 2009; Chen et al . , 2010 ) , was in this report found to co-enrich with oAβ1-42 tethered to the affinity matrix at its N-terminus but did not interact with C-terminally tethered oAβ1-42 . A few candidate oAβ1-42 interactors revealed in this study , including the ubiquitin carboxy-terminal hydrolase L1 ( Uch-L1 ) and the ras-related nuclear protein ( RAN ) , had previously been proposed to play roles in the molecular etiology of AD but the mechanisms of their involvement have remained speculative . More specifically , both Uch-L1 and RAN were reported to be downregulated in AD ( Poon et al . , 2013; Choi et al . , 2004; Mastroeni et al . , 2013 ) , and overexpression Uch-L1 was observed to restore synaptic function in hippocampal slices treated with oAβ ( Gong et al . , 2006 ) . It will be interesting to explore the possible contribution of a physical interaction between oAβ1-42 and these proteins to these observations . Finally , this study revealed many new candidate interactors that had neither been known to interact with Aβ nor had otherwise been tied to the etiology of AD . The significance of these interactions is not known at this time . It is hoped that this inventory of interactors will expedite knowledge regarding the molecular underpinnings of the disease . In extensive follow-up experiments , we validated the surprising interaction of Aβ1-42 with SST14 , also known as somatotropin release-inhibiting factor ( SRIF ) . A conclusion whether CST17 was also present in the initial oAβ1-42 interactome dataset could not be made because , in contrast to the situation for SST14 , no peptide was observed that uniquely identified CST17 . However , the possibility that binding of oAβ1-42 may extend to CST17 is supported by our observation that , although not yielding qualitatively different results in any of the binding assays , this peptide exerted even more pronounced effects on Aβ aggregation than SST14 . Not only did the oAβ-SST14 interaction stand out as a rare example of an interaction between two relatively small peptides but it also was remarkable in its dependence on a freely accessible N-terminus of Aβ1-42 . Interest in binders to the N-terminus is heightened by the recognition that the C-terminus of Aβ1-42 is rarely accessible in vivo due to its hydrophobic nature . Although initially captured at pH 8 . 0 , a widely-used pH for interactome analyses , owed to its wide-spread use in radioimmunoprecipitation assay ( RIPA ) buffer formulations , the oAβ-SST14 interaction was in subsequent validation experiments observed to influence Aβ-dependent aggregation characteristics in a buffer intended to mimic the physiological environment , and also withstood the presence of a lower pH during FRET analyses . To our knowledge , the smallest protein Aβ had previously been known to bind to is MT-RNR2 , also known as humanin , a 21-24 amino acid peptide coded within the mitochondrial 16S ribosomal RNA gene ( Hashimoto et al . , 2001 ) . Humanin has no resemblance to bioactive SST14 at the amino acid level , has been shown to acquire a three-turn α-helical fold , and interacts with an Aβ binding epitope encompassing residues 17-28 ( Maftei et al . , 2012 ) , a segment of Aβ not sufficient for binding to SST ( Figure 2D ) . Several previous reports have shown that synthetic macrocycles can assemble into oligomeric assemblies ( Levin and Nowick , 2007 ) characterized by tetrameric assemblies in which β-sheet dimers pair through complementary dry interfaces ( Levin and Nowick , 2007; Liu et al . , 2011; Pham et al . , 2014 ) and have the ability to serve as building blocks of larger fibril-like supramolecular assemblies ( Pham et al . , 2014 ) . In the context of their shared predominant release into the extracellular synaptic clefts in the brain , the robust influence of SST14 on Aβ aggregation observed in this work recommends SST14 as a natural Aβ binding partner that could contribute to alternative Aβ conformers and the existence of Aβ strains . An intriguing data point in this work has been the observation that Aβ1-42 , but not Aβ1-40 , assembles into distinct complexes of 50-60 kDa in the presence of SST ( or CST ) . Aβ complexes of similar size , most often referred to as Aβ*56 , have been observed in transgenic APP mouse models ( Lesné et al . , 2006 ) and human AD brains ( Lesné et al . , 2013 ) , and their levels have been proposed to correlate with cognitive impairment and toxicity . Consistent failures to generate Aβ*56 in vitro have led to the proposition that its formation may require an unknown co-factor ( Larson and Lesné , 2012 ) . While the relationship of Aβ*56 to SST-dependent oAβ complexes of similar mass remains unresolved , the ability to generate Aβ assemblies of this size in vitro should aid their structural characterization and facilitate the raising of antibodies with the intent to evaluate their presence in patient samples . Moreover , the ability of SST to prevent the fibrillization of Aβ1-42 in vitro may have implications for the generation of toxic Aβ1-42 oligomers in AD patients . We further documented that the presence of SST profoundly diminished the detection of Aβ by some of the most frequently used antibodies . More work is needed to understand the molecular basis for this finding and gauge the extent to which its existence may have undermined Aβ detection in studies that preceded this work . SST , originally identified as a factor that inhibits the secretion of pituitary growth hormone ( Brazeau et al . , 1973 ) , is not a neophyte in the field of AD; a PubMed query conducted with the search terms ‘somatostatin’ and ‘Alzheimer’ returns more than 300 entries , including 60 review articles ( e . g . , ( Epelbaum et al . , 2009; Burgos-Ramos et al . , 2008 ) . Several observations stand out amongst the connections of SST to AD and should be considered in future efforts to understand the possible significance of the direct interaction between SST and Aβ1-42: ( 1 ) Already in 1980 , it was observed that SST levels are conspicuously reduced in AD-afflicted brains ( Davies et al . , 1980 ) , a finding which has been repeatedly confirmed by others ( Beal et al . , 1985; Gahete et al . , 2010; Gabriel et al . , 1993 ) . More recently , it was shown that SST gene transcripts are amongst a relatively small number of transcripts whose abundance levels decline persistently throughout adult life ( Lu et al . , 2004 ) ; ( 2 ) A 1985 study reported a striking co-localization of SST-filled neurons with sites of amyloid plaque formation in AD ( Morrison et al . , 1985 ) ; ( 3 ) a manuscript published back-to-back documented that the expression of SST characterized a subclass of neurons in the brain , which were most vulnerable to forming neurofibrillary tangles and undergoing destruction in AD ( Roberts et al . , 1985 ) ; ( 4 ) two studies undertaken with Finnish and Chinese patient cohorts reached the conclusion that an SST gene polymorphism may increase the risk for AD ( Xue et al . , 2009; Vepsäläinen et al . , 2007 ) and , ( 5 ) independent investigations revealed that SST can regulate Aβ catabolism by modulating the expression of neprilysin ( Saito et al . , 2005 ) and insulin-degrading enzyme ( Tundo et al . , 2012 ) , two critical players in the best-characterized Aβ clearance pathways . More specifically , Aβ levels were reported to be approximately 1 . 5-fold higher in SST-deficient mice ( Saito et al . , 2005 ) , an increase that did not lead to spontaneous plaque formation , thereby leaving the possible involvement of SST in this process currently untested . Once AD models become available in which SST has been eliminated , it will be informative to assess how their Aβ-related pathobiology is altered . It will also be interesting to investigate the constraints of the SST-oAβ interaction in more detail . Our current data suggest that this interaction depends on the presence of tryptophan 8 within SST in the context of flanking residues 5-11 and as such strongly overlaps the epitope within SST known to mediate binding to its cell surface receptors ( Veber et al . , 1979 ) . Our results further suggest that binding does not occur between monomers of both peptides but may require at least one of the two binding partners to be present in oligomeric ( pre-aggregated ) form . This limitation is likely to pose technical difficulties for its detailed study but also offers the tantalizing prospect that future insights into the binding interface can be used to design compounds , which may selectively bind to the respective oligomeric forms of Aβ . Such compounds might find application in disease diagnosis , might be used to prevent the SST-Aβ interaction from forming , or could possibly be derivatized in ways that target the respective conformers , and possibly amyloid plaques , for destruction . Synthetic human Aβ1-42 ( catalog number AS-24224 ) , Aβ1-40 ( catalog number AS-24236 ) , SST14 ( catalog number AS-24277 ) , [Arg8]-Vasopressin ( catalog number AS-24289 ) , Bombesin ( catalog number AS-20665 ) , Oxytocin ( catalog number AS-24275 ) , SST28 ( catalog number AS-22902 ) and amylin1-37 ( catalog number AS-60804 ) were purchased from Anaspec , Inc . ( Fremont , CA , USA ) . The alternative synthetic human SST14 peptide ( catalog number H-1490 ) sourced from a different vendor ( see Figure 4B ) , CST17 ( catalog number H-5536 ) and CST29 ( catalog number H-6458 ) were obtained from Bachem Americas , Inc . ( Torrance , CA , USA ) . The biotinylated Aβ peptides , including biotin-Aβ1-42 , Aβ1-42-biotin , Aβ17-42-biotin , biotin-SST14 and truncated or mutant SST peptides were synthesized by LifeTein LLC ( Hillsborough , NJ , USA ) . The Edans-SST donor and Aβ1-42-TMR acceptor peptides were in-house synthesized as described previously ( Bateman and Chakrabartty , 2011 ) . Primary antibodies used in this study were the anti-Aβ1-16 antibody ( 6E10 ) ( catalog number 803015 ) , the anti-Aβ17-24 antibody ( 4G8 ) ( catalog number 800701 ) , and the anti-Aβ1-42 antibody ( 12F4 ) ( catalog number 805501 ) , which were all sourced from BioLegend ( San Diego , CA , USA ) ( note that epitope ranges embedded in vendor names for these antibodies are imprecise , and validated epitopes for the above antibodies are depicted in Figure 5A ) . The polyclonal antibody recognizing total Tau ( K9JA ) ( catalog number A0024 ) was a Dako product ( Agilent Technologies Canada , Mississauga , ON , Canada ) . The monoclonal antibodies recognizing Tau phosphorylated at Ser202/ Thr205 ( AT8 ) ( catalog number MN1020 ) or at Thr231 ( AT180 ) ( catalog number MN1040 ) were from Thermo Fisher Scientific ( Burlington , ON , Canada ) . All Western blot reagents were purchased from Thermo Fisher Scientific ( Burlington , ON , Canada ) . For denaturing gels ( Figures 3 , 5 and 6 ) , peptide samples were mixed with Bolt LDS Sample Buffer ( catalog number B0007 ) in the presence of 2 . 5% 2-mercaptoethanol and boiled at 70 ˚C for 10 min before loading . The samples were separated on Bolt 12% Bis-Tris Plus gels ( catalog number NW00125BOX ) in MES SDS Running Buffer ( catalog number NP0002 ) at 100 to 120 V for 1 . 5 to 2 hr . For native gels ( Figure 6B ) , the peptide samples were mixed with Novex Tris-Glycine Native Sample Buffer ( catalog number LC2673 ) without boiling before loading . The samples were separated on SDS-free 4-20% Tris-Glycine Mini Gels ( catalog number XP04202BOX ) in Novex Tris-Glycine Native Running Buffer ( catalog number LC2672 ) at 150 V for 1 . 5 hr . 0 . 3 to 0 . 5 µg of Aβ peptide were loaded to each lane for both denaturing and native gels . For the analysis of hippocampal neuron lysates , samples were prepared and separated in the same way as described for denaturing gels except that 4% to 12% Bis-Tris gradient gels were used and 20 to 30 µg of total protein were loaded to each lane . For all immunoblot analyses , peptides were transferred to polyvinylidene difluoride ( PVDF ) membranes at 50 V in Tris-Glycine buffer containing 10-20% methanol for 1 . 5 to 2 hr . Membranes were blocked for 2 hr in conventional tris-buffered saline and 0 . 1% Tween 20 ( TBST ) containing 5% fat-free milk and probed overnight with the respective primary antibodies . After at least three washes with TBST , membranes were incubated for 2 hr with 1: 2000 to 1: 5000 diluted anti-mouse or anti-rabbit horseradish peroxidase-conjugated secondary antibodies ( Bio-Rad Laboratories , Inc . , Hercules , CA , USA ) . The band signals were visualized using enhanced chemiluminescence reagents ( catalog number 4500875; GE Health Care Canada , Inc . , Mississauga , ON , Canada ) and X-ray films . Silver staining of gels was done with reagents from Pierce Silver Stain Kit ( catalog number 24612 ) ( Thermo Fisher Scientific , Burlington , ON , Canada ) following the protocol provided by the manufacturer . Aβ peptides were dissolved in 1 , 1 , 1 , 3 , 3 , 3-hexafluoro-2-propanol ( HFIP ) at a concentration of 1 mg peptide per mL for one hour at room temperature , then dried in a centrifugal evaporator and stored at −80°C until use . Monomeric Aβ peptides were prepared by dissolving peptides in dimethyl sulfoxide ( DMSO ) at a concentration of 2 mM and further diluting them in phosphate buffered saline ( PBS , pH 7 . 4 ) to 100 μM , followed by centrifugation ( 14 , 000 g , 20 min ) to remove traces of insoluble aggregates . Oligomeric Aβ was created by incubating the monomeric preparation at 4°C for 24 hr . The Aβ oligomers were then purified by centrifugation ( 14 , 000 g , 20 min ) . The supernatant containing Aβ oligomers was further diluted as required and indicated in figure legends describing specific experiments . When high-purity monomeric Aβ peptides at concentrations below 10 μM were required for interactome and ThT assay analyses , the solubilized Aβ preparations were passed through a size-exclusion column ( see below for details ) . Pre-aggregated SST was generated by incubating 1 mM SST in PBS at 37°C for 1 hr with shaking at the speed of 700 rpm . For SST and other neuropeptides prepared for the hippocampal neuron culture assay , the peptides were dissolved in PBS ( at pH 4 . 5 , 6 . 5 or 7 . 4 ) at 400 µM without incubation ( Figure 8C ) or were incubated at room temperature for 24 or 48 hr ( Figure 8D and E ) . 50 μM Aβ1-42 and 40 μM SST/other neuropeptides were mixed ( Figure 8D ) and incubated for 1 or 3 hr ( Figure 8E ) before they were added to cell culture medium . The final concentration of Aβ1-42 and SST/other neuropeptides in the medium were 5 and 4 µM , respectively . The biotinylated Aβ oligomers or monomers were captured on Streptavidin UltraLink Resin beads ( Thermo Fisher Scientific , Burlington , ON , Canada ) by overnight incubation in PBS at 4°C and continuous agitation on a slow-moving turning wheel . Additional negative control samples were generated by saturation of Streptavidin UltraLink Resin ( catalog number 53113 , Thermo Fisher Scientific , Inc . ) with biotin . Subsequently , the bait peptide- or biotin-saturated beads were washed with Lysis Buffer ( 0 . 15% digitonin , 150 mM NaCl , 100 mM Tris , pH 8 . 0 ) . Human frontal lobe tissue samples from individuals ( two males and two females ) who had died in their early 70s of non-dementia causes served as the biological source material . These samples were adopted from a former Canadian Brain Tissue Bank at the Toronto Western Hospital and are held in −80°C freezers in the biobank of the Tanz Centre for Research in Neurodegenerative Diseases . 1 g pieces each of these brain tissue samples were combined and homogenized in Lysis Buffer supplemented with Complete protease inhibitor cocktail ( Roche , Mississauga , ON , Canada ) . Following the removal of insoluble debris by centrifugation for 30 min at 14 , 000 g , the protein concentration was adjusted to 2 mg/mL before the brain homogenates were added to the pre-saturated affinity capture beads for overnight incubation at 4°C . Following the affinity capture step , the affinity capture beads ( 100 μL per biological replicate ) were extensively washed in three consecutive wash steps with a total of 150 mL of Lysis Buffer . Subsequently , the beads were additionally washed with 50 mL of 20 mM Hepes , pH 7 . 0 , and transferred to Pierce Spin columns ( catalog number 69705 , Thermo Fisher Scientific , Inc . ) to remove primary amines stemming from the Tris buffer and to prepare the samples for elution . Captured proteins were finally eluted by rapid acidification mediated by a solution comprising 0 . 2% trifluoroacetic acid and 20% acetonitrile in deionized water ( pH 1 . 9 ) . N-terminally biotinylated SST14 was captured on Streptavidin UltraLink Resin beads ( Thermo Fisher Scientific , Burlington , ON , Canada ) at a concentration of 50 µM by overnight incubation in PBS at 4°C with continuous agitation on a turning wheel . Next , the biotin-SST14 saturated streptavidin agarose beads were incubated at room temperature for 4 hr with monomeric or oligomeric Aβ1-42 preparations ( 2 . 5 µM ) , prepared as described in section ‘Preparation of monomeric versus oligomeric Aβ or SST’ . Before and after the incubation , the beads were extensively washed in Lysis Buffer ( 0 . 15% digitonin , 150 mM NaCl , 100 mM Tris , pH 8 . 0 ) . Finally , bound peptides were eluted from the affinity capture beads by 15 min boiling in Laemmli sample buffer and analyzed by Western blotting using NuPAGE gels as described in section ‘Western blot and silver staining’ . 200 µg of monomeric Aβ1-42 were generated as described in the section ‘Preparation of monomeric versus oligomeric Aβ or SST/other neuropeptides’ . To further remove residual traces of Aβ1-42 oligomers , the soluble centrifugation supernatant was subsequently subjected to size-exclusion chromatography on a Superdex 75 10/300 GL column ( GE Healthcare Life Sciences , Mississauga , ON , Canada ) in ThT assay buffer ( 20 mM sodium phosphate , pH 8 . 0 , 200 μM EDTA and 0 . 02% NaN3 ) at the flow rate of 0 . 5 mL/min . Late eluting fractions , containing Aβ1-42 monomers , were collected on ice and the Aβ1-42 concentration in these fractions was determined by Western blot analysis , using known amounts of synthetic Aβ1-42 as signal intensity calibrants . Affinity-capture eluates were essentially processed as described before ( Gunawardana et al . , 2015; Mehrabian et al . , 2014; Jeon and Schmitt-Ulms , 2012 ) . Briefly , sample tubes were moved to a centrifugal evaporator to remove the organic solvent . Additional acidity of the sample was removed following the addition of water and continuous evaporation . Subsequently , protein solutions were denatured by the addition of 9 M urea ( to achieve a final concentration of 6 M urea ) and 10 min incubation at room temperature . Next , the pH was raised by the addition of 100 mM HEPES , pH 8 . 0 , and proteins were reduced for 30 min at 60°C in the presence of 5 mM tris ( 2-carboxyethyl ) phosphine ( TCEP ) , and alkylated for 1 hr at room temperature in the presence of 10 mM 4-vinylpyiridine ( 4-VP ) . To ensure that the residual urea concentration did not exceed 1 . 5 M , protein mixtures were diluted with 50 mM tetraethylammonium bromide ( TEAB ) , pH 8 . 0 , to a total volume of 100 μL . Samples were then digested with side-chain-modified porcine trypsin ( Thermo Fisher Scientific , Burlington , ON , Canada ) overnight at 37°C . The covalent modifications of primary amines with isobaric labels provided in the form of tandem mass tag ( TMT ) reagents ( Thermo Fisher Scientific , Inc . ) or isobaric tagging for relative and absolute quantitation ( iTRAQ ) reagents ( Applied Biosystems , Foster City , CA , USA ) followed instructions provided by the manufacturers . Equal amounts of the labeled digests were pooled into a master mixture and purified with C18 ( catalog number A5700310 ) or SCX ( catalog number A5700410 ) Bond Elut OMIX tips ( Agilent Technologies , Inc . , Mississauga , ON , Canada ) using manufacturer instructions . Peptide mixtures were finally reconstituted in 0 . 1% formic acid and analyzed by tandem mass spectrometry analysis on a Tribrid Orbitrap Fusion instrument . Instrument parameters during the data acquisition were as described in detail before ( Gunawardana et al . , 2015 ) . The post-acquisition data analyses of interactome data sets was conducted against the human international protein index ( IPI ) database ( Version 3 . 87 ) which was queried with Mascot ( Version 2 . 4; Matrix Science Ltd , London , UK ) and Sequest HT search engines within Proteome Discoverer software ( Version 1 . 4; Thermo Fisher Scientific , Burlington , ON , Canada ) . Spectra exceeding a stringent false discovery rate ( FDR ) target of ∆Cn of 0 . 05 for input data and a FDR of 0 . 01 for the decoy database search were detected and removed by the Percolator algorithm ( Käll et al . , 2007 ) as described before ( Gunawardana et al . , 2015 ) . PEAKS Studio software ( Version 6 . 0; Bioinformatics Solutions Inc . , Waterloo , ON , Canada ) was used to assess the reproducibility of nano-HPLC separations . A maximum of two missed tryptic cleavages and naturally occurring variable phosphorylations of serines , theonines and tyrosines were considered . Other posttranslational modifications considered were carbamylations , oxidation of methionines and deamidation of glutamines or asparagines . Mass spectrometry data sets have been deposited to the ProteomeXchange Consortium ( Vizcaíno et al . , 2014 ) via the PRIDE partner repository ( Vizcaíno et al . , 2013 ) with the dataset identifier PXD004867 and have been made fully accessible . All fluorescence spectra were acquired with a Photon Technology International fluorescence spectrophotometer ( QuantaMaster , HORIBA Scientific , London , ON , Canada ) using a 1 cm quartz cuvette with an excitation wavelength of 335 nm and a 2 nm slit width at room temperature . Solutions containing 20 μM concentrations of Edans-SST14 , Aβ1-42-TMR or a mixture of both fluorescently labeled peptides at pH 8 . 5 were rapidly adjusted to pH 5 . 2 and the resulting reaction mixtures incubated overnight . Fluorescence emission spectra were recorded in a continuous window spanning 350 nm to 650 nm . FRET competition assays were performed using the fluorescence spectrophotometer described above . A solution of 2 μM Aβ1-42-TMR was mixed with increasing concentrations of SST14 or AVP ( 0-50 μM ) at pH 8 . 5 followed by dropping the pH to 5 . 2 and incubating the resulting solutions overnight . To compete for binding to Aβ1-42-TMR , a solution of 2 μM Edans-SST was then added to each of the solutions at a final volume of 150 μL at continuous incubation of peptide mixtures at pH 5 . 2 overnight . Fluorescence spectra were recorded and binding curves were plotted against the concentration of unlabeled peptides . The curves were fitted with assistance of OriginPro 8 . 5 software ( OriginLab Corporation , Northampton , MA ) using nonlinear least squares fitting given by the equation:y= ( ymax−ymin/ ( 1+ ( x/EC50 ) n ) +ymin with y = observed donor fluorescence , x = concentration , and n = Hill coefficient . Assay procedures were largely based on a protocol described by Sarah Linse’s group ( Hellstrand et al . , 2010 ) . Briefly , Aβ1-42 , Aβ1-40 and/or SST14 were prepared in assay buffer ( 20 mM sodium phosphate , pH 8 . 0 , 200 μM EDTA and 0 . 02% NaN3 ) or in PBS , pH 7 . 4 , at concentrations specified in the individual figures . Human amylin1-37 was initially dissolved in deionized water but then assayed under identical conditions as Aβ1-42 . 100 μL of peptide solutions were supplemented with 25 μM thioflavin T ( ThT ) ( catalog number T3516 , Sigma-Aldrich Canada , Oakville , ON , Canada ) and loaded into 96-Well Half-Area Microplates ( catalog number 675096 , Greiner Bio One International , Kremsmünster , Austria ) . The subsequent plate incubation proceeded at 37°C with shaking at 700 rpm for 4 of every 5 min in a microplate reader ( CLARIOstar , BMG Labtech , Guelph , ON , Canada ) for overall durations specified in individual figures . ThT fluorescence was measured every 5 min at excitation and emission wavelengths of 444 nm and 485 nm , respectively . Particle size was analyzed using a NanoSight NS300 ( Malvern Instruments , Southborough , MA , USA ) . Samples were vortexed for 40 s before being loaded into a luer lock syringe and infused at a constant rate through the sample chamber using an attached syringe pump ( Malvern ) . Particle movement was captured for 60 s x 5 acquisitions at a rate of 25 frames per second , at room temperature . Data were analyzed by nanoparticle tracking analysis ( NTA 3 . 2 software , Malvern ) with detection threshold held constant between replicates . Polystyrene latex microsphere beads of 100 nm ( Malvern ) were used as a size standard . Data are shown as mean diameter ± standard deviation . Five-microliter samples were adsorbed for 1 min onto freshly glow-discharged Formvar-carbon-coated 400-mesh copper grids . The grids were washed with 0 . 1 M and 0 . 01 M ammonium acetate buffer ( pH 7 . 4 ) , stained with freshly filtered 1% uranyl acetate solution , and excess stain removed with filter paper . The grids were allowed to dry overnight and then viewed with a Tecnai F20 electron microscope ( FEI Company / Thermo Fisher Scientific , Eindhoven , The Netherlands ) at an acceleration voltage of 200 kV . Electron micrographs were recorded on an Eagle 4K charge-coupled device camera ( FEI Company / Thermo Fisher Scientific , Eindhoven , The Netherlands ) . Primary hippocampal cultures were generated from E16 ~18 C57BL/6 mouse embryos . The hippocampi were dissected while immersed in Hank’s Balanced Salt Solution buffered with 10 mM HEPES and cells were dissociated with 0 . 125% trypsin ( Invitrogen Canada , Inc . , Burlington , ON , Canada ) for 10 ~ 15 min at 37°C , followed by trituration . Dissociated cells were plated at a density of 1 . 0 ~ 1 . 5 × 105 cells/cm2 in 6-well plates pre-coated with poly-D lysine and grown in Neurobasal medium with B-27 supplement and Glutamax ( Invitrogen ) . Half the medium was refreshed every 3 days . The treatment of the primary hippocampal neurons with Aβ and other peptides was started on day 18 . After treatment , cells were lysed in 0 . 5% SDS , 1% Nonidet P-40 , 2 mM EDTA , 100 mM NaCl , 100 mM Tris , pH 7 . 6 , supplemented with protease inhibitor and phosphatase inhibitor cocktails ( Roche , Mississauga , ON , Canada ) for Western blot analyses .

Treating Alzheimer’s disease and related dementias is one of the major challenges currently facing healthcare providers worldwide . A hallmark of the disease is the formation of large deposits of a specific molecule , known as amyloid beta ( Aβ ) , in the brain . However , more and more research suggests that smaller and particularly toxic amyloid beta clumps – often referred to as oligomeric Aβ – appear as an early sign of Alzheimer’s disease . To understand how the formation of these smaller amyloid beta clumps triggers other aspects of the disease , it is important to identify molecules in the human brain that oligomeric Aβ binds to . To this end , Wang et al . attached amyloid beta or oligomeric Aβ molecules to microscopically small beads . The beads were then exposed to human brain extracts in a test tube , which allowed molecules in the extracts to bind to the amyloid beta or oligomeric Aβ . The samples were then spun at high speed , meaning that the beads and any other molecules bound to them sunk and formed pellets at the bottom of the tubes . Each pellet was then analyzed to see which molecules it contained . The experiments identified more than a hundred human brain proteins that can bind to amyloid beta . One of them , known as somatostatin , selectively binds to oligomeric Aβ . Wang et al . were able to determine the structural features of somatostatin that control this binding . Finally , in further experiments performed in test tubes , Wang et al . noticed that smaller oligomeric Aβ clumps were more likely to form than larger amyloid beta deposits when somatostatin was present . This could signify a previously unrecognized role of somatostatin in the development of Alzheimer’s disease . Further studies are now needed to confirm whether the presence of somatostatin in the brain favors the formation of smaller , toxic oligomeric Aβ clumps over large innocuous amyloid beta deposits . If so , new treatments could be developed that aim to reduce oligomeric Aβ levels in the brain by preventing somatostatin from interacting with amyloid beta molecules . Wang et al . also suggest that somatostatin could be used in diagnostic tests to detect abnormal levels of oligomeric Aβ in the brain or body fluids of people who have Alzheimer’s disease .

2017

Somatostatin binds to the human amyloid β peptide and favors the formation of distinct oligomers

Cystic fibrosis ( CF ) is caused by mutations in CF transmembrane conductance regulator ( CFTR ) . The most frequent mutation ( F508del-CFTR ) results in altered proteostasis , that is , in the misfolding and intracellular degradation of the protein . The F508del-CFTR proteostasis machinery and its homeostatic regulation are well studied , while the question whether ‘classical’ signalling pathways and phosphorylation cascades might control proteostasis remains barely explored . Here , we have unravelled signalling cascades acting selectively on the F508del-CFTR folding-trafficking defects by analysing the mechanisms of action of F508del-CFTR proteostasis regulator drugs through an approach based on transcriptional profiling followed by deconvolution of their gene signatures . Targeting multiple components of these signalling pathways resulted in potent and specific correction of F508del-CFTR proteostasis and in synergy with pharmacochaperones . These results provide new insights into the physiology of cellular proteostasis and a rational basis for developing effective pharmacological correctors of the F508del-CFTR defect . Cystic fibrosis ( CF ) is the most common lethal genetic disease in Caucasians . It is caused by mutations in the CF transmembrane conductance regulator ( CFTR ) gene that encodes a chloride channel localised to the apical membrane of several epithelial cells . Mutations that cause CFTR loss of function impair the transepithelial movement of salts at the cell surface , resulting in pleiotropic organ pathology and , in the lungs , in chronic bacterial infections that eventually lead to organ fibrosis and failure ( Riordan , 2008 ) . The CFTR protein comprises two membrane-spanning domains , two cytosolic nucleotide-binding domains , and a regulatory domain , folded together into a channel ( Riordan , 2008 ) . Folding occurs in the endoplasmic reticulum ( ER ) through the sequential action of multiple chaperone complexes ( Loo et al . , 1998; Meacham et al . , 1999; Rosser et al . , 2008 ) and is followed by export out of the ER and glycosylation in the Golgi before arrival at the plasma membrane ( PM ) , where CFTR undergoes several cycles of endocytosis before degradation in the lysosomes ( Gentzsch et al . , 2004 ) . The most frequent mutant , which is present in ~90% of the patients with CF , misses a phenylalanine at position 508 ( F508del-CFTR ) and folds in a kinetically and thermodynamically impaired fashion into a conformation that is recognized as defective by the ER quality control ( ERQC ) system . It is thus retained in the ER and targeted for ER-associated degradation ( ERAD ) by the ubiquitin–proteasome machinery ( Jensen et al . , 1995; Ward et al . , 1995 ) . A small fraction of F508del-CFTR may escape degradation in the ER and reach the PM , where it can function as a channel . This might have therapeutic relevance because patients that express even low levels of functional channel have milder symptoms ( Amaral , 2005 ) . However , at the PM , F508del-CFTR is recognized by the peripheral ( or PM-associated ) quality control ( PQC ) system and is rapidly degraded in the lysosomes ( Okiyoneda et al . , 2010 ) . In contrast to the fairly extensive knowledge about the machinery involved in the proteostasis ( or protein homeostasis ) of F508del-CFTR , the regulatory mechanisms that operate on the F508del-CFTR proteostasis machinery remain relatively less explored . Notable exceptions are the recent studies on the effects of the unfolded protein response and heat shock response ( UPR and HSR , respectively ) on the proteostasis of F508del-CFTR . The UPR and HSR operate as homeostatic reactions that tend to redress the imbalances between the load of unfolded proteins and the folding capacity of a cell essentially by enhancing the transcription of the cellular folding machinery . Investigators have therefore sought to induce these reactions by pharmacological means with the aim to rescue the F508del-CFTR folding/transport defect , with partial success ( Roth et al . , 2014; Ryno et al . , 2013 ) . Very little is known instead about the regulation of proteostasis by the ‘classical’ signalling networks composed of GTPases , second messengers , kinases , etc . that are usually activated by PM receptors and control most , if not all , of the cellular functions . We and others have previously shown that constitutive trafficking along the secretory pathway is potently controlled by such signalling networks triggered by both extra- and intracellular stimuli ( Cancino et al . , 2014; Chia et al . , 2012; De Matteis et al . , 1993; Farhan et al . , 2010; Giannotta et al . , 2012; Pulvirenti et al . , 2008; Simpson et al . , 2012 ) . This suggests that the machinery of proteostasis viz . protein synthesis , folding , and degradation , is also likely to be controlled by similar signalling systems . Identifying the relevant regulatory components of these systems would not only enhance our understanding of the physiology of proteostasis , but also have significant impact on future therapeutic developments , because components of the signalling cascades , such as membrane receptors and kinases , are generally druggable , and are , in fact , the main targets of most known drugs . Thus , this study aims to uncover signalling pathways that control proteostasis of F508del-CFTR . To this end , we have developed a strategy based on the analysis of the mechanisms of action ( MOAs ) of drugs that regulate the proteostasis of F508del-CFTR . The choice of this strategy over more traditional approaches such as kinome-wide screenings was based on the rationale that since many of the successful drugs target multiple molecular pathways simultaneously ( Lu et al . , 2012 ) and with limited toxicity , elucidating the MOAs of these drugs might lead to uncovering molecular networks that regulate proteostasis in a synergistic and relatively ‘safe’ manner . Several drugs that regulate the proteostasis of F508del-CFTR ( hereinafter referred to as proteostasis regulators ) and enhance its ability to reach the PM have been identified over the years , largely through screening campaigns ( Calamini et al . , 2012; Carlile et al . , 2012; Hutt et al . , 2010 ) . In addition , molecules that bind directly to F508del-CFTR and facilitate its folding have also been characterized ( pharmacochaperones ) ( Calamini et al . , 2012; Kalid et al . , 2010; Odolczyk et al . , 2013; Pedemonte et al . , 2005; Sampson et al . , 2011; Van Goor et al . , 2006; Wang et al . , 2007 ) . Both these groups of drugs that enhance the ability of F508del-CFTR to reach the PM are referred to as correctors . The MOA of the pharmacochaperones has been partially understood ( Farinha et al . , 2013a; Okiyoneda et al . , 2013 ) , and they are approaching the level of effectiveness required for clinical use ( [Wainwright et al . , 2015] and see also http://www . fda . gov/NewsEvents/Newsroom/PressAnnouncements/ucm453565 . htm ) , while the proteostasis regulators are presently too ineffective to be of clinical interest . Here , we have analysed the MOAs corrector drugs that are proteostasis regulators by deconvolving their transcriptional effects . Changes in gene expression are significant components of the MOAs of many drugs ( Popescu , 2003; Santagata et al . , 2013 ) , and the analysis of transcriptional MOAs is a growing research area ( Iorio et al . , 2010; Iskar et al . , 2013 ) . However , a major difficulty here is that the available proteostasis regulator drugs include representatives of diverse pharmacological classes such as histone deacetylase inhibitors ( Hutt et al . , 2010 ) , poly ( ADP-ribose ) polymerase inhibitors ( Anjos et al . , 2012; Carlile et al . , 2012 ) , hormone receptor activators ( Caohuy et al . , 2009 ) , cardiac glycosides ( Zhang et al . , 2012 ) , and others . Thus , the effects of the available F508del-CFTR correctors are most probably not mediated by their heterogeneous principal MOA , but by some unknown weak secondary MOAs ( ‘side effects’ ) that these drugs share . In order to extract the transcriptional changes that are correction-related from those that are due to the ( correction-irrelevant ) principal MOAs of the corrector drugs we have developed a new approach based on the ‘fuzzy’ intersection of gene expression profiles . This method , applied to a set of proteostatic correctors will identify genes that are commonly modified by these drugs and should therefore correspond to the correction-related pathways and not to their heterogeneous primary effects . Using this strategy , we harvested a group of few hundred genes that are regulated by most of the proteostatic correctors , and then derived a series of molecular networks from this gene pool through bioinformatic and experimental approaches . Several of these networks are signalling pathways made up of druggable receptors and kinases . Silencing or targeting these pathways with chemical blockers inhibit the degradation in the ER and enhance the transport of F508del-CFTR to the PM . Moreover , the large pool of ER-localised foldable F508del-CFTR that results from the inhibition of ER degradation can be acted upon by pharmacochaperones , to further enhance correction . These findings build on previous screening studies and on the accumulated knowledge of F508del-CFTR proteostasis to start to define the network of signalling pathways that control F508del-CFTR proteostasis , and thereby provide a rational basis for the development of novel , potent and specific proteostasis corrector treatments for CF . Proteostasis regulators share the ability to correct ( albeit weakly ) the F508del-CFTR folding-trafficking defect but have principal pharmacological effects not related to F508del-CFTR correction . If the correction-related MOAs of these drugs are transcription-dependent ( see Materials and methods for evidence that they are ) , then the gene signatures of these drugs should comprise both genes related to F508del-CFTR correction and genes related to their heterogeneous primary effects . Thus , we sought to analyse the transcriptional MOAs of correctors ( 24 drugs/conditions altogether ) with different chemical structures and pharmacological activities ( Table 1 ) , excluding known pharmacochaperones . The gene signatures of 13 correctors were obtained in our laboratories using immortalised CF bronchial epithelial ( CFBE41o- ) cells ( Kunzelmann et al . , 1993 ) on an Agilent microarray platform ( CFBE dataset; see Supplementary file 1 and GEO accession number GSE67698 for the expression profiles ) . Another 11 signatures were extracted from the MANTRA ( Mode of action by network analysis , [Iorio et al . , 2010]; MANTRA dataset ) that were based on the Affymetrix platform . Two drugs , glafenine and ouabain , were present in both datasets . Even though the signatures of each drug in the two platforms were obtained from different cell lines , they were similar enough ( not shown ) to suggest that two datasets can be treated together . To extract the correction-related transcriptional effects from those due to primary effects of the correctors , we first attempted to cluster the drugs based on commonalities in their transcriptional profiles . These attempts using classical and alternate clustering methods did not yield meaningful results ( see Figure 1—figure supplement 1 and Supplementary file 2 ) , possibly because the strong transcriptional effects of the heterogeneous principal MOAs of these drugs obscures the potential clustering of drug signatures based on their secondary correction-relevant MOAs . In order to detect these weak but common transcriptional signatures we developed a method based on the fuzzy intersection of transcriptional profiles ( FIT ) ( Figure 1A ) . Here , the corrector gene signatures are ‘intersected’ to identify their commonalities , and this returned a pool of genes that are potentially correction-related ( CORE ) and are modulated by most of the correctors . The intersections among the majority of the signatures should include the correction-related ( CORE ) genes but exclude genes related to the heterogeneous principal effects of the drugs . The method thus captures common MOAs but not MOAs specific for individual drugs or small groups of drugs . 10 . 7554/eLife . 10365 . 003Table 1 . The list of corrector drugs used in this study with their corresponding known primary MOAs ( related to Figure 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10365 . 003Drugs of the CFBE dataset ( Reference for correction activity ) Primary Use/Class 4-AN , PARP1 inhibitor ( Anjos et al . , 2012 ) PARP1 inhibitorABT888 ( Anjos et al . , 2012 ) A poly ( ADP-ribose ) polymerase ( PARP ) -1 and -2 inhibitor with chemosensitizing and antitumor activities . ABT-888 inhibits PARPs , thereby inhibiting DNA repair and potentiating the cytotoxicity of DNA-damaging agents . Glafenine ( Robert et al . , 2010 ) An anthranilic acid derivative with analgesic properties used for the relief of all types of pain ( 1 ) GSK339 ( DY Thomas lab , unpublished ) Androgen receptor ligand ( Norris et al . , 2009 ) . Ibuprofen ( Carlile et al . , 2015 ) Ibuprofen is a nonsteroidal anti-inflammatory drug . It is a non-selective inhibitor of cyclooxygenase . JFD03094PARP inhibitorKM11060 ( Robert et al . , 2008 ) PDE5 inhibitor ( an analog of sildenafil ) . Latonduine ( Carlile et al . , 2012 ) PARP3 inhibitorMinocycline H ( D Y Thomas lab unpublished ) A tetracycline analog that inhibits protein synthesis in bacteria . Also known to inhibit 5-lipooxygenase in the brain ( 2 ) . Ouabagenin ( Zhang et al . , 2012 ) A cardiaoactive glycoside obtained from the seeds of Strophanthus gratus . Acts by inhibiting Na+/K+-ATPase , resulting in an increase in intracellular sodium and calcium concentrations ( 2 ) . Ouabain ( Zhang et al . , 2012 ) A cardiaoactive glycoside obtained from the seeds of Strophanthus gratus . Acts by inhibiting Na+/K+-ATPase , resulting in an increase in intracellular sodium and calcium concentrations ( 2 ) . PJ34 ( Anjos et al . , 2012 ) PARP1 inhibitorLow temperature ( Denning et al . , 1992 ) Chloramphenicol ( Carlile et al . , 2007 ) Inhibitor bacterial protein synthesis by binding to 23S rRNA and preventing peptidyl transferase activity ( 2 ) . Chlorzoxazone ( Carlile et al . , 2007 ) Muscle relaxant . Acts by inhibiting degranulation of mast cells and preventing the release of histamine and slow-reacting substance of anaphylaxis . It acts at the level of the spinal cord and subcortical areas of the brain where it inhibits multi-synaptic reflex arcs involved in producing and maintaining skeletal muscle spasm ( 2 ) . Dexamethasone ( Caohuy et al . , 2009 ) A synthetic glucocorticoid agonist . Its anti-inflammatory properties are thought to involve phospholipase A2 inhibitory proteins , lipocortins ( 2 ) . Doxorubicin ( Maitra et al . , 2001 ) DNA intercalator that inhibits topoisomerase II activity by stabilizing the DNA-topoisomerase II complex ( 2 ) . Glafenine ( Robert et al . , 2010 ) An anthranilic acid derivative with analgesic properties used for the relief of all types of pain ( 1 ) . Liothyronine ( Carlile et al . , 2007 ) L-triiodothyronine ( T3 , liothyronine ) thyroid hormone is normally synthesized and secreted by the thyroid gland . Most T3 is derived from peripheral monodeiodination of T4 ( L-tetraiodothyronine , levothyroxine , L-thyroxine ) . The hormone finally delivered and used by the tissues is mainly T3 . Liothyronine acts on the body to increase the basal metabolic rate , affect protein synthesis and increase the body's sensitivity to catecholamines ( such as adrenaline ) . It is used to treat hypothyroidism ( 2 ) . MS-275 ( Hutt et al . , 2010 ) Also known as Entinostat . An inhibitor of Class Ihistone deacetylases ( preferentially HDAC 1 , also HDAC 3 ) ( Hu et al . , 2003 ) . Scriptaid ( Hutt et al . , 2010 ) An inhibitor of Class I histone deacetylases ( HDAC1 , HDAC3 and HDAC8 ) ( Hu et al . , 2003 ) . Strophanthidin ( Carlile et al . , 2007 ) A cardioactive glycoside that inhibits Na+/K+-ATPase . Also known to inhibit the interaction of MDM2 and MDMX ( 1 ) . Thapsigargin ( Egan et al . , 2002 ) A sesquiterpene lactone found in roots of Thapsia garganica . A non-competitive inhibitor of sarco/endoplasmic Ca2+-ATPase ( SERCA ) ( 1 ) . Trichostatin-A ( Hutt et al . , 2010 ) An inhibitor of histone deacetylases ( HDAC1 , HDAC3 , HDAC8 and HDAC7 ) ( Hu et al . , 2003 ) . ( 1 ) http://pubchem . ncbi . nlm . nih . gov/ ( 2 ) www . drugbank . ca 10 . 7554/eLife . 10365 . 004Figure 1 . Corrector drugs modulate a set of CORE genes . ( A ) Schema of the FIT method . The upregulated ( red ) and downregulated genes ( blue ) were fuzzy intersected to identify CORE genes . ( B ) The number of probe sets in the corrector drug profiles ( MANTRA dataset ) as well as random profiles from MANTRA database were intersected with variable fuzzy cut-offs ( represented as number of drugs out of 11 ) to obtain optimal fuzzy cut-off for the analysis . The enlargement ( inset ) shows that at the optimal fuzzy cut-off ( 0 . 7; 8 out of 11 drugs ) , the signal-to-noise ratio was close to 3 ( 108 probe-sets in the corrector drug intersection vs 32 in the random drug intersection ) . ( C ) At a fuzzy cut-off of 0 . 7 , the number of random drug profiles used was varied , and the number of probe-sets present in the intersection is shown . ( D ) Using the optimal parameters ( see A , C ) the FIT analysis resulted in 402 upregulated and 219 downregulated CORE genes . ( E ) The number of CORE genes associated with the enriched GO terms is shown . Those genes that did not associate with enriched GO terms were excluded from the chart . ( F ) Protein-protein interactions between the CORE and the proteostasis genes ( restricted to those that connect the two groups ) are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 10365 . 00410 . 7554/eLife . 10365 . 005Figure 1—figure supplement 1 . Corrector drugs ( MANTRA dataset ) have diverse transcriptional responses corresponding to their primary MOA . As noted earlier , the correctors had few , if any , common principal MOA . First , we wished to identify sub-groups of correctors , if any , which might have shared correction-relevant MOA . The use of prototype ranked lists ( PRLs ) for the correctors present in the MANTRA dataset ( Iorio et al . , 2010 ) results in the loss of information about fold changes in transcript levels and corresponding p-values . Thus , these lists are not suitable for comparisons and cluster analyses made with traditional metrics ( such as Correlation or Euclidian distance ) and classic partitional or agglomerative/hierarchical clustering methods . However , we attempted to overcome this issue by a state-of-the-art method to 'cluster' the profiles . In the past , in an effort to cluster the drugs present in the connectivity map database ( Iorio et al . , 2010 ) based on their transcriptional response similarity , we had designed a novel gene set enrichment analysis ( GSEA: [Subramanian et al . , 2005] ) based metric to obtain similarity scores between ranked lists ( see [Iorio et al . , 2010] for details ) . This study resulted in a network that contains clusters of densely interconnected drugs ( named drug communities ) , whose corresponding ranked-lists had high 'similarity' scores . Since this network is basically the output of a state-of-the-art method to cluster the ranked-lists of genes derived from connectivity map , we mined this network in an attempt to cluster the F508del-CFTR corrector drugs . Shown here is the drug similarity network assembled , by iterative affinity propagation clustering . The topology of this network is hierarchical and derived from a gene set enrichment analysis ( GSEA ) -derived similarity metric applied to drug-induced transcriptional profiles ( see [Iorio et al . , 2010] for details ) . Each node represents a drug . An edge connects two nodes if the corresponding consensual transcriptional responses are significantly similar and its length is inversely proportional to the similarity of the response . The clusters ( referred to as drug-communities ) correspond to drugs with highly similar transcriptional responses . The F508del-CFTR correctors belonging to this network ( from MANTRA dataset ) are highlighted together with the identifier of the drug-community they belong to . The nodes ( corrector drugs ) are color-coded such that drugs belonging to the same community share the same color . The F508del-CFTR correctors do not cluster together , and most of them belong to drug-communities enriched for their primary MOA ( Supplementary file 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10365 . 00510 . 7554/eLife . 10365 . 006Figure 1—figure supplement 2 . IPA based analysis uncovers networks of CORE genes . The CORE genes were analyzed by the IPA-Core analysis function both together and individually as up and downregulated gene sets . Among the networks built by IPA using the downregulated CORE genes set , four significant ones are shown . The downregulated CORE genes are shown in gray while the nodes that were predicted by IPA to be part of the network are shown in white . The nodes predicted by IPA include NF-KB , AKT , VEGF , and PI3K complexes . The differences in the shape of the nodes indicate the protein families to which the nodes belong ( see the panel legend for details ) . The blue circles indicate the network hubs that were tested in the screening assay ( see Figure 2A–D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10365 . 006 The main parameters of the FIT analysis ( number of correctors; number of genes to be analysed in each signature , and cut-off threshold for inclusion in the correction-relevant gene pool; see Materials and methods and Figure 1B–C ) were selected to identify a sufficiently large CORE gene pool for pathway analysis , and also to minimise the number of ‘false’ CORE genes . The FIT analysis of the gene signatures resulted in 219 downregulated and 402 upregulated CORE genes ( Supplementary file 3; Figure 1D; see also Materials and methods ) . Each of these CORE genes were shared by 70% of the corrector signatures . The number of CORE genes were threefold higher than that expected on a random basis ( see Materials and methods ) . This indicates that common transcriptional programmes that might be correction-relevant are indeed embedded in the signatures of proteostasis correctors . To understand the relation of CORE genes to CFTR proteostasis , we built a dataset of known F508del-CFTR proteostasis-relevant genes by assembling literature data ( Supplementary file 4 ) and mapped their interactions with the CORE pool using STRING ( Franceschini et al . , 2013 ) . We found extensive and statistically significant ( see Materials and methods ) protein-protein interactions among the nodes of the union of these two datasets ( Figure 1F ) , indicating that ( at least a fraction of ) the CORE genes are related to CFTR proteostasis . Significant interactions were also found between the CORE genes from CFBE and the MANTRA datasets ( not shown ) confirming they are related and thus can be analysed together . We next applied standard bioinformatic tools to the CORE gene pool to identify functionally coherent pathways/networks/groups . A search proteostasis components among CORE genes retrieved 48 folding/degradation and 24 transport-machinery components ( Supplementary file 5 ) , some of which are known to be involved in F508del-CFTR proteostasis . However , perhaps surprisingly , they were not significantly enriched , indicating that changes in the expression levels of significant numbers of the proteostasis machinery genes is not part of the MOA of the proteostasis regulator drugs . Potential explanations for this could be that some proteostasis genes that are crucial enough to have a strong effect on F508del-CFTR proteostasis are indeed regulated by the correctors , but are not numerous enough to result in a statistical enrichment of this gene group; or that the corrector drugs act by modulating the expression of regulatory genes/pathways that act post-translationally on the proteostasis machinery ( see results from the screening below ) . Since our interest was in the identification of signaling networks that regulate proteostasis we also searched for the presence of signalling molecules among CORE genes and found 24 kinases and 6 phosphatases ( Supplementary file 5 ) . Further , Ingenuity pathway analysis ( IPA ) tool identified several statistically significant signalling networks . The IPA networks comprised also ( predicted ) interactors of CORE genes , some of which were network hubs ( Figure 1—figure supplement 2 ) . Such hubs were often constituents of signalling pathways such as growth-factor-mediated pathways ( e . g . , receptors for vascular endothelial growth factor [VEGF] and platelet-derived growth factor [PDGF] , phosphatidylinositol 3-kinase [PI3K] , and mitogen-activated protein kinases [MAPKs] ) , inflammation-associated pathways ( NF-κB subunits , Toll-like receptor 4 [TLR4] ) , stress-activated protein kinase ( SAPK ) pathways [MAP2K3/6 ( MKK3/6 ) , MAP2K4/7 ( MKK4/7 ) ] , and casein-kinase pathway ( CSNK2A1/ CKII ) . These hubs might control the CORE genes . Of note , many of the hubs were frequently present in the gene signatures of the individual correctors , although below the fuzzy cut-off threshold of 0 . 7 required for inclusion in the CORE gene pool itself ( not shown ) . Analysis of the promoters of CORE genes aimed at the identification of upstream transcription factors did not generate interpretable results . We then turned to experimental validation of the role of CORE genes in the regulation of F508del-CFTR proteostasis . Experiments were carried out using a characterised biochemical assay ( See Materials and methods and Figure 2—figure supplement 1 for details ) that detects both the amount of core-glycosylated CFTR trapped in the ER ( band B with western blotting ) and the amount of CFTR fully glycosylated in the Golgi ( most of which presumably resides at the PM; band C with western blotting ) . As a model system , we used non-polarised CFBE41o-cells stably expressing F508del-CFTR ( Bebok et al . , 2005 ) ( hereafter referred to as CFBE ) ; but many experiments were carried out also in HeLa , BHK and polarized CFBE cells , with results that were in good qualitative agreement with the CFBE data . While this assay is not suitable for large-scale screening , it provides quantitative information on the main proteostasis parameters including CFTR accumulation in the ER , ER-associated CFTR degradation , and transport and processing in the Golgi complex . Moreover , this assay is specific for proteostasis as it separates the effects on the F508del-CFTR protein from the effects on conductance as revealed by faster chloride-permeability assays ( Pedemonte et al . , 2005 ) . Experimental validation was restricted to a limited set of genes: downregulated CORE genes ( to exploit the availability of siRNA-based downregulation and of small-molecule inhibitors ) that showed functional coherence , that is , were found in protein-protein interaction networks or in enriched GO groups; or were network hubs from Ingenuity analysis , or ubiquitin ligases and signalling molecules . In total , this resulted in a group of 108 genes ( Supplementary file 3 ) . Notably , these genes had no previously reported role in the regulation of F508del-CFTR proteostasis . CFBE cells were treated with siRNAs against these genes and the effects on both bands B and C were monitored . As a reference for correction , we used the investigational drug VX-809 ( Van Goor et al . , 2006 ) , a robust corrector that acts as a pharmacochaperone . VX-809 treatment increased band C levels by four- to fivefold over control in most experiments . In all , 47 out of the 108 genes tested were found to be active in regulating F508del-CFTR proteostasis ( Figure 2A–D ) . Of these , 32 genes ( when depleted ) enhanced the levels of bands B and C by 1 . 5-fold to more than 10-fold over controls , while 15 genes decreased bands B and C by 20–80% of the control levels . We refer to these as anti-correction and pro-correction genes , respectively . Among these active genes , 30 were CORE genes and 17 were hubs in IPA networks . Notably , the correction that was induced by the depletion of many anti-correction genes was greater than that achieved by VX-809 ( Figure 2A ) , or by the corrector drugs originally included in the study ( see Figure 2—figure supplement 2A ) . This was in particular the case for a group of four poorly characterised ubiquitin ligases ( RNF215 , UBXO5 , ASB8 , FBXO7 ) that were not known to regulate F508del-CFTR proteostasis . RNF215 depletion increased the levels of bands C to over 10-fold the control levels ( Figure 2A ) . Given these strong effects , RNF215 is a worthy candidate for further studies as a potential ERAD machinery component . Also notably , the depletion of many anti-correction genes not only enhanced the bands B and C but also markedly increased ( to different extents ) the band C/band B ratio ( Figure 2C ) , suggesting that these genes affect the efficiency of export of F508del-CFTR protein from the ER and/or the stability of this protein after export . Given that siRNA treatments often have off-target effects , several controls were performed to ensure the specificity of the observed effects on proteostasis ( see Methods section for details ) . It is also to be noted that the downregulation of anti-correction genes did not change the levels of F508del-CFTR mRNA ( Figure 2—figure supplement 2B ) , suggesting that the observed effect is not due to increased F508del-CFTR synthesis , although these data cannot by themselves exclude an effect on translation . Further experiments confirm that the effect of the downregulation of these genes is mostly ( if not completely ) on F508del-CFTR degradation and folding/export ( see below ) . 10 . 7554/eLife . 10365 . 007Figure 2 . Validation of the selected CORE genes . ( A–D ) CFBE cells were treated with siRNAs targeting CORE genes and changes in F508del-CFTR proteostasis monitored by western blotting . The fold change in the levels of band C obtained by downregulating anti-correction ( A ) and pro-correction ( D ) genes and the fold change in levels of band B ( B ) and band C/band B ratio ( C ) after downregulation of the anti-correction genes are shown . The effects of negative control siRNAs ( dashed line ) and VX-809 ( green ) are indicated . ( E ) The validated CORE genes ( blue – pro-correction hits , orange – anti-correction hits , and gray – no action ) were assembled into coherent networks based on information from databases . Non-directional interactions denote protein-protein interaction , directional interactions represent phosphorylation cascades and dashed arrows indicate indirect connections through intermediaries . ( F ) Western blot of CFBE cells treated with mitoxantrone ( 2 . 5-20 μM for 48 hr ) , a potential corrector identified using downregulation of anti-corrector genes as selection criteria . Mitoxantrone increased the levels of both band C and band B . ( G ) Treatment of CFBE cells with the indicated combinations of siRNAs targeting CORE genes led to a synergistic increase in the band C levels . Changes in the levels of band C were quantitated from western blotting are presented as mean ± SEM ( n > 3 ) . The representative blots are shown in the insert , the upper panel corresponds to a blot with lower exposure where the differences in band C levels can be easily appreciated , while the bottom panel corresponds to a blot with higher exposure where a faint band C can be seen even under control siRNA treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 10365 . 00710 . 7554/eLife . 10365 . 008Figure 2—figure supplement 1 . Characterization of biochemical assay to monitor F508del-CFTR correction . ( A ) The changes in the F508del-CFTR proteostasis were monitored by a biochemical assay . This assay is sensitive and robust enough to detect even small changes in the proteostasis of CFTR . CFTR is resolved by SDS-PAGE into two distinct bands - one at ~140 kDa ( band B; core glycosylated protein present in the ER ) and another at ~170 kDa ( band C; processed by Golgi-localized glycosylation enzymes ) . In the case of wild type CFTR , the majority of the protein is in the form of band C ( not shown ) while in the case of mutant F508del-CFTR the major form is band B . Sometimes a lower molecular weight band , termed band A , is also observed that probably corresponds to non-glycosylated CFTR . For the quantitation’s described in this manuscript , bands B and A were considered together as band B . Quantitation of the changes in the levels of bands B and C were carried out as outlined here in this example . CFBE cells were treated with indicated corrector drugs for 48 hr and were lysed and processed for western blotting as described in the Materials and methods . Treatment with corrector drugs increases the levels of both band B and band C of F508del-CFTR . Na+/K+-ATPase was used as loading control . To quantitate the levels of bands B and C , the X-ray films obtained were scanned using a flatbed scanner ( Hewlett-Packard ) at a resolution of 200–600 dpi and exported as 8-bit tiff images . ( B ) The scanned images were then imported into ImageJ for quantitation . First , the images were converted to grayscale and background of the image was removed using the 'subtract background' function with a rolling ball radius of 50–150 pixels . Next , using the rectangular selection tool , C bands ( shown here ) or B bands were selected . ( C ) The gel/blot analysis tool ( Analyze>Gels ) was then used for quantitation of the selected bands . First the 'select first lane' function was executed followed by the 'plot lane' function . This resulted in a plot as shown here . The area under the curve ( as a measure of the density of bands ) was then quantitated using the magic wand tool to select the peaks individually and exporting the density into a text file . The quantitated density values are indicated below the plot . We note that the method described here is semi-quantitative; nevertheless , the values obtained adequately describe the biological phenomenon observed . ( D ) The fold change in band C levels calculated using the measured values of density in ( C ) is plotted . ( E , F ) To test whether the semi-quantitative process described here is within the linear range of detection , we resolved indicated quantities of cell lysates prepared from CFBE cells by SDS-PAGE followed by western blotting , and the blots were processed as described before . The exposure time of the blots to the X-ray films was varied to get different intensity levels of the bands . Only one exposure of 30 s is shown here ( E ) for clarity . The intensity of the bands was calculated using ImageJ ( see above ) and expressed as fold change in band B ( E ) relative to the minimum value . As can be noted , the amounts of band B quantitated are linear in the range from 10 to 80 μg of cell lysate loaded . Similar results were obtained for band C . The linearity was also preserved for exposures that were twice or half of the time shown here . For experiments described in this manuscript the total amount of protein loaded ranged from 25 to 40 μg and exposure time of blots ranged from 30 to 45 s . DOI: http://dx . doi . org/10 . 7554/eLife . 10365 . 00810 . 7554/eLife . 10365 . 009Figure 2—figure supplement 2 . Downregulation of CORE genes rescues F508del-CFTR more efficiently than the corrector drugs used originally , without altering the F508del-CFTR mRNA levels . ( A ) CFBE cells were treated with indicated corrector drugs for 48 hr and then lysed and prepared for western blotting , to assay the rescue of F508del-CFTR from ERQC . The changes in the levels of band C after drug treatment are shown as mean ± SEM ( n > 3 ) . ( B ) CFBE cells were treated with indicated siRNAs ( targeting the anti-correction genes ) for 72 hr , and then total RNA from the cells was purified . The levels of CFTR mRNA were quantitated by RT-PCR . The data are presented as mRNA levels relative to the negative control siRNAs . The values are expressed as mean ± SEM ( n = 4 ) . ( C ) Representative blot used for quantitations represented in Figure 2A–C . ( D ) Representative blot used for quantitations represented in Figure 2D DOI: http://dx . doi . org/10 . 7554/eLife . 10365 . 00910 . 7554/eLife . 10365 . 010Figure 2—figure supplement 3 . The siRNAs efficiently reduce the transcript levels of their target genes . ( A ) CFBE cells were treated with indicated siRNAs for 72 hr . The efficiency of silencing for each gene was evaluated by quantitative RT-PCR and expressed as fold change represented as mean ± SEM ( n = 3 ) . The siRNA treatment reduced the mRNA levels of the target genes to ~10–30% of the levels present in control cells treated with non-targeting siRNA . ( B ) CFBE cells were treated with the indicated doses of siRNAs targeting the CORE genes for 72 hr . The change in the mRNA levels of the CORE genes was quantitated by RT-PCR and expressed as fold change represented as mean ± SEM ( n > 3 ) . The effect on F508del-CFTR proteostasis was also monitored by western blotting and shown in the insert . There was a positive correlation between the reduction in the levels of CORE genes and the change in proteostasis of F508del-CFTR . DOI: http://dx . doi . org/10 . 7554/eLife . 10365 . 010 It might appear surprising to find both pro-correction and anti-correction genes within the downregulated CORE gene pool . However , these genes are , presumably , components of complex transcriptional modules whose role is to control cellular functions in a balanced manner . To this end , the concomitant operation of regulatory systems of opposite signs is probably necessary ( Hart and Alon , 2013 ) . These observations are therefore likely to be a reflection of the organization of the transcriptional programs that regulate proteostasis . Based on the above results , we sought to identify putative pathways/networks/groups ( collectively networks ) within the 47 active CORE gene pool , using literature data and pathway building tools ( see Figure 2E for details ) . This resulted in several small potential networks ( each comprising two to six connected elements ) , four of which were composed of signalling molecules and will be referred to by the name of their ‘central’ components: MLK3 ( MAP3K11 ) , PI3K , and CKII ( with predominantly anti-correction activity ) , and ERBB4 ( with pro-correction activity ) . A recent kinome-wide screening ( published while this manuscript was being submitted ) identified several kinases that regulate the rescue of F508del-CFTR ( Trzcinska-Daneluti et al . , 2015 ) , with no overlap with the hits identified here ( possibly due to the different functional assays and cell types used in that study versus ours ) . Other three of the networks shown in Figure 2 comprised spliceosome , centromere and mediator complex components , and two were groups of ubiquitin-ligases and kinases . We next sought to verify whether the effects of correctors on the CORE genes might explain the action of these drugs . We first analysed the frequency of the active CORE genes among the genes downregulated by the corrector drugs . The CORE genes were ~3-fold enriched in the signatures of correctors compared to those of other ~200 drugs taken at random from the MANTRA database ( not shown ) . We next searched for MANTRA drugs that significantly downregulate the CORE genes ( anti-correctors ) using GSEA ( specifically two-tailed symmetric GSEA as implemented in MANTRA; http://mantra . tigem . it/ ) . The top 25 hits included three of the correctors that we had used for the FIT analysis . From the remaining 22 , we selected eight drugs ( based on availability ) for testing in the correction assay . Among these , mitoxantrone was found to potently increase both band C and band B . ( Figure 2F ) ; in addition , among the top five hits was Vorinostat , an HDAC inhibitor that was shown to act as a corrector ( Hutt et al . , 2010 ) . Thus , at least 20% ( 5 out of 22 ) of the short-listed drugs were correctors , while among a large number ( >20 ) of randomly selected drugs none showed correction activity ( not shown ) . These data suggest that the downregulation of CORE genes is a useful criterion to identify correctors . We then extended our analysis of the top hits by comparing the five active drugs with those that failed to correct and also by examining upregulated genes in their gene expression profiles . The correctors showed a high frequency ( two- to threefold more than non-corrector drugs ) of upregulation of the potent pro-corrector genes MKK1 , MKK3 and FGFBP1 , while the non-corrector drugs upregulated more frequently ( two- to threefold more than corrector drugs ) the anti-corrector genes NF-κB2 and MKK7 . These results thus suggest that considering also the upregulation of CORE genes will help in further defining the search space for new correctors . Altogether , the above data indicate that the drug-induced modulation of CORE genes is a significant component of the MOAs of corrector drugs . Thousands of gene signatures of drugs and perturbagens are being deposited in specialized databases ( http://www . lincscloud . org/ ) . These and a more extensive search for CORE genes will provide useful tools for a more refined bioinformatic identification of new correctors . As described earlier , an advantage of using this approach ( deconvolution of drug MOA ) to identify regulatory pathways is the possibility of discovering synergistic pathways . Thus , in order to explore the possible epistatic interactions between the CORE networks/pathways , siRNAs against selected targets were combined and tested on F508del-CFTR rescue . These candidates were chosen for their potential druggability and/or strong effects on correction . Strong synergistic interactions were observed between various combinations of siRNAs against CKII , CAMKK2 , MLK3 , and NUP50 ( a spliceosomal network component ) ( Figure 2G ) , thus validating our choice of the approach . As a note of caution here , the efficacy of the combined siRNA treatments was more variable than that observed with single siRNAs . In our experience , this is because siRNAs in combinations are less effective than the individual siRNAs in depleting their target proteins , and a depletion threshold must be reached to achieve synergy . We conclude that , using the FIT technique and a series of bioinformatic and experimental filters , we have identified a set of synergistic molecular networks that show strong control over F508del-CFTR proteostasis . Next , we sought to define the composition and the role in correction of two representative CORE-networks , namely , the MLK3 and the CAMKK2 pathways . MLK3 ( or MAP3K11 ) is part of a group of 14 MAP3 kinases that act through cascades of MAP2K and MAPK enzymes . MLK3 can be activated by various PM receptors , including the TNF-α , TGF-β , VEGF , and PDGF receptors , through at least two MAP4Ks ( haematopoietic progenitor kinase [HPK]1 and germinal centre kinase [GCK] ) and glycogen synthase kinase ( GSK ) 3β , or via the CDC42/Rac family ( summarised in [Schachter et al . , 2006] ) . MLK3 can also be activated by stress , e . g . , oxidative stress ( Lee et al . , 2014 ) ( i . e . , it is a Stress Activated Protein Kinase , or SAPK ) and it can , in turn , trigger three main kinases: p38 MAPK , c-Jun N-terminal kinase ( JNK ) , and extracellular signal regulated kinase ( ERK ) , depending on cell type and conditions , through the intermediate kinases MKK3/6 , MKK4/7 and MKK1/2 , respectively ( Schachter et al . , 2006 ) . MLK3 is also known to be an upstream activator of NF-kB ( Hehner et al . , 2000 ) . We thus sought to determine which components of the MLK3 pathway have roles in F508del-CFTR correction . The VEGF and PDGF receptors , MKK7 , and NF-κB2 , like MLK3 , appear to be components of the correction-relevant branch of the MLK3 pathway , as indicated by the screening data in Figure 2A . Among the components upstream of MLK3 , we found TGF receptors , CDC42 , Rac2 , and HPK1 to be active in correction ( i . e . their depletion induced correction ) ( Figure 3A ) . Within the cascade downstream of MLK3 , MKK7 ( Figure 2A ) and further downstream , JNK2 ( Figure 3B ) were active components ( JNK2 is highly expressed in bronchial epithelial cells [http://biogps . org] ) . The p38 MAPK , also downstream of MLK3 ( through MAP2K3 and MAP2K6 ) was inactive in CFBE but moderately active in HeLa cells , indicating some cell type-dependent specificity in the effects of these kinases ( Figure 3—figure supplement 1A ) . Thus , altogether , suppression of the MKK7-JNK2 branch of the MLK3 pathway induced F508del-CFTR correction . Conversely , when the activity of the MLK3 pathway was enhanced by transfection of the MLK3 activator CDC42 or MKK7 or JNK2 into the CFBE cells , the levels of both bands B and C dropped markedly ( Figure 3C ) , confirming that the MLK3 pathway has a tonic negative effect on the proteostasis of F508del-CFTR . 10 . 7554/eLife . 10365 . 011Figure 3 . Delineation of the MLK3 pathway branch that controls F508del-CFTR proteostasis . ( A ) CFBE cells were treated with indicated siRNAs targeting the upstream activators of MLK3 and their effect on F508del-CFTR proteostasis monitored by western blotting . The fold change in band C levels is shown as mean ± SEM ( n > 3 ) . Reduction in TGF receptor , HPK , CDC42 and RAC2 levels rescued F508del-CFTR from ERQC . The rescue obtained with TNFR2 siRNA was quite variable and thus was not considered further . ( B ) JNK isoforms were tested for their effect on F508del-CFTR proteostasis after siRNA-mediated downregulation of their levels . Downregulation of JNK2 leads to efficient rescue of F508del-CFTR that is comparable to that obtained with MLK3 . The fold change in band C levels is presented as mean ± SEM ( n > 3 ) with western blot in the insert . ( C ) CFBE cells were transfected with activators of the MLK3 pathway to study their effect on F508del-CFTR proteostasis . The fold change in the band B levels is shown as mean ± SEM ( n > 3 ) with western blot in the insert . All of them reduced the levels of both band B and band C ( not shown ) of F508del-CFTR . The corresponding increase in the levels of phospho-c-jun indicates an increase activation of the MLK3 pathway activity . ( D , E ) Schematic representation of the proposed MLK3 ( D ) and CAMKK2 ( E ) pathways that regulate F508del-CFTR proteostasis . The directional interactions proposed between the components of the pathways are based on published literature . DOI: http://dx . doi . org/10 . 7554/eLife . 10365 . 01110 . 7554/eLife . 10365 . 012Figure 3—figure supplement 1 . Delineation of the MLK3 and CAMKK2 pathway branches that regulate F508del-CFTR proteostasis . ( A ) HeLa cells [HeLa cells stably expressing HA-tagged F508del-CFTR] were treated with indicated siRNAs targeting MLK3 pathway components including p38 MAPK ( mix of siRNAs targeting all four isoforms ) and JNK ( mix of siRNAs targeting all three JNKs ) . The effect on F508del-CFTR proteostasis monitored by western blotting . Fold change in the levels of band C was quantitated and presented as mean ± SEM ( n > 3 ) with a representative blot shown in the insert . The downregulation of the MLK3 pathway components ( including p38 MAPK ) leads to the rescue of F508del-CFTR in HeLa cells . SiRNAs targeting Rma1 and Aha1 used as positive controls for rescue of F508del-CFTR . ( B ) Screening for F508del-CFTR proteostasis regulators among the CORE genes led to the identification of CAMKK2 as an anti-correction hit . Three downstream components and nine upstream components of the CAMKK2 signaling pathway ( as derived from literature mining ) were tested , by siRNA-mediated downregulation , for their role in regulation of F508del-CFTR proteostasis . CFBE cells were treated with the indicated siRNAs for 72 hr and their effect on F508del-CFTR proteostasis monitored by western blotting . Four of them ( CALML5 , ITPR2 , CAMK4 , and AMPK [by a mix of siRNAs targeting PRKAA1 and PRKAA2] ) rescued F508del-CFTR from ERQC as seen by an increase in band C levels . ( C ) The changes in the levels of band C from ( B ) were quantitated and are represented as mean ± SEM ( n > 3 ) . See Figure 3 for a representation of the derived CAMKK2 pathway that regulates F508del-CFTR proteostasis . DOI: http://dx . doi . org/10 . 7554/eLife . 10365 . 012 In sum , as shown in Figure 3D , a signal regulating F508del-CFTR proteostasis flows from the ligands and receptors upstream of MLK3 , through HPK1 and CDC42/Rac2 , to impinge on MLK3 and is then passed on through the JNK2 arm . NF-κB2 is also a probable downstream component of this proteostasis regulatory pathway . We also tested ( again by siRNA silencing ) seven other MAP3Ks ( including TAK1/MAP3K7 , see below ) that can activate JNK or p38 , for their effect on F508del-CFTR proteostasis . They had no effect ( not shown ) . This highlights the remarkable specificity of MLK3 in the regulation of proteostasis , possibly due to spatial/temporal compartmentalization of the MAPK networks ( Engstrom et al . , 2010 ) . A similar series of experiments were performed to characterise the CAMKK2 cascade in F508del-CFTR correction . The results are reported in detail in Figure 3—figure supplement 1B , C ( see also Figure 3E ) , and indicate that the CAMKK2 pathway has negative effects on F508del-CFTR proteostasis similar to those found for the MLK3 pathway . The increase in the levels of band B induced by inhibition of the MLK3 pathway might be due to increased synthesis or due to decreased degradation of F508del-CFTR . Downregulation of MLK3 did not increase the CFTR mRNA levels ( Figure 2—figure supplement 2B ) , speaking against the former possibility , although an effect of this pathway on the translational efficiency cannot be excluded . We then examined the degradation of band B using both a cycloheximide ( CHX ) chase and a radioactive pulse-chase assay . Downregulation of the MLK3 pathway markedly slowed the degradation of band B when measured by CHX chase assay ( Figure 4A , B ) and similar effects were obtained with the radioactive pulse-chase method ( Figure 4—figure supplement 1A , B ) . We also examined the effects of enhancing the activity of MLK3 pathway by overexpressing CDC42 or MKK7 or JNK2: under these conditions the rate of degradation of band B increased twofold ( Figure 4C , D; see also [Ferru-Clement et al . , 2015] ) . Notably , the ubiquitin-proteasome system itself was not detectably affected by the modulation of the MLK3 pathway activity , as judged by the lack of effects on both the proteasome sensor-ZsProsensor-1 ( Figure 4—figure supplement 1C ) and the accumulation of poly-ubiquitinated proteins ( Figure 4—figure supplement 1D ) . Thus , the MLK3 pathway appears to regulate the ERQC/ERAD of F508del-CFTR at a step prior to proteasomal digestion . 10 . 7554/eLife . 10365 . 013Figure 4 . MLK3 pathway regulates the degradation of F508del-CFTR . ( A , B ) CFBE cells pretreated with siRNAs were treated with CHX ( 50 μg/mL ) for indicated times and the levels of band B of F508del-CFTR was monitored ( A ) . The levels were quantitated and represented in ( B ) as mean ± SEM ( n > 3 ) . Downregulation of MLK3 or JNK2 reduced the kinetics of reduction of band B of F508del-CFTR . ( C , D ) . CHX chase assay ( see above ) after overexpression of the activators of MLK3 pathway . The activation of MLK3 pathway increases the rate of degradation of band B ( C ) . Quantitation of the blot is shown in ( D ) as mean ± SEM ( n > 3 ) . ( E , F ) CFBE cells were treated with indicated siRNAs followed by incubation at 26°C for 6 hr followed by shift to 37°C for the indicated time periods . The changes in band C levels were monitored as measure of PQC ( C ) . See ( F ) for quantitation of band C levels represented as mean ± SEM ( n > 3 ) . ( G , H ) . PQC assay ( see above ) after overexpression of CDC42 or JNK2 shows an increased rate of degradation of band C ( G ) upon CDC42 overexpression . JNK2 overexpression has no effect on the PQC of F508del-CFTR . The blots were quantified and presented in ( H ) as mean ± SEM ( n > 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10365 . 01310 . 7554/eLife . 10365 . 014Figure 4—figure supplement 1 . Characterization of the mode of action of the MLK3 pathway on F508del-CFTR proteostasis . ( A , B ) CFBE cells treated with MLK3 siRNA were pulsed with radioactive [35S]-cysteine and methionine for 15 min , and then chased for the indicated times . CFTR was immunoprecipitated and processed for autoradiography ( A ) . The signals corresponding to band B from ( A ) were quantitated and presented in ( B ) . The data are representative of two independent experiments . Note the reduced degradation of F508del-CFTR upon downregulation of MLK3 . ( C ) Downregulation of the MLK3-JNK pathway does not affect the activity of proteasomes . CFBE cells treated with MLK3 or JNK2 siRNA for 72 hr were transfected with Proteasome ZsProsensor-1 for the final 24 hr , and the levels Proteasome ZsProsensor-1 monitored by fluorescence microscopy . Treatment with MG132 ( 20 μg/ml for 3 hr ) , a proteasomal inhibitor , was used as a positive control . While treatment with MG132 increases the fluorescence levels of ZsProsensor-1 compared to untreated cells indicating a reduced proteasome activity , downregulation of MLK3 or JNK2 did not change the levels of fluorescence , suggesting that proteasome activity is not changed under these conditions . ( D ) CFBE cells were treated with MLK3 or JNK2 siRNA and processed for western blotting to monitor the accumulation of poly-ubiquitinated proteins . There was no change in the levels of poly ubiquitinated proteins suggesting that these treatments do not affect proteasome activity . ( E ) Downregulation of MLK3 does not affect the folding of F508del-CFTR . CFBE cells expressing wild type CFTR or F508del-CFTR were treated with MLK3 siRNA as indicated . Untreated CFBE cells incubated at 26°C for 24 hr were used as a positive control for the promotion of folding . Membrane fractions from the cells were isolated and subjected to trypsin digestion for 10 min on ice , followed by western blotting with M3A7 antibody that recognizes the NBD2 domain of F508del-CFTR , or with 3G11 antibody that recognizes NBD1 . The wild-type CFTR and its NBD domains show more resistance to trypsin digestion compared to F508del-CFTR . There was no change in the stability of F508del-CFTR or its NBD domains upon downregulation of MLK3 , while the low temperature treatment enhanced the stability of F508del-CFTR and its NBD1 domain . The western blots are representative of at least three different experiments . ( F ) Downregulation of MLK3 does not affect the exit of cargo from the ER . CFBE cells were treated with MLK3 siRNA for 72 hr , and during the final 16 hr they were transfected with VSVG-GFP and incubated at 40°C . The cells were then shifted to 32°C to allow the exit of cargoes from the ER . The cells were fixed at the indicated times after shift to 32°C and labeled for GM130 ( red ) to mark the Golgi apparatus and DAPI ( blue ) to stain the nucleus . The rate of transport of VSVG-GFP ( green ) either from ER to Golgi or from Golgi to the PM was not significantly affected by the downregulation of MLK3 . Bar = 20 μm . G . HeLa cells were treated with JNKi II ( 2 . 5 μM ) for 24 hr and F508del-CFTR was immunoprecipitated . The immunoprecipitate was subjected to western blotting to reveal the amount of associated HOP . The blots ( like in the insert ) were quantitated and normalized to the levels of F508del-CFTR ( not shown ) . The values are presented as mean ± SD ( n>3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10365 . 014 In addition , silencing of the MLK3 pathway ( and of several CORE genes ) increased also the band C/band B ratio ( see Figure 2C ) . This is not explained by reduced ERAD alone and suggested that the MLK3 pathway might have additional effects on the folding/ export of F508del-CFTR , or on the stability of band C at the PM ( or both ) . The trypsin susceptibility assay to assess the folding status of F508del-CFTR and an assay for protein transport out of the ER using vesicular stomatitis virus G protein ( VSVG ) , a classical probe to study secretory trafficking , ruled out large effects of the MLK3 pathway on F508del-CFTR folding or on the general ER-export machinery ( Figure 4—figure supplement 1E , F ) . Nevertheless , we note that these assays ( trypsin susceptibility or VSVG export ) are limited in their scope and do not capture the wide spectrum of subtle regulations that can influence the outcome of proteostasis ( see below for a discussion of the effect of the MLK3 pathway on folding/export ) . We next tested the effect of MLK3 on the stability of F508del-CFTR at the PM . We depleted MLK3 and exposed the cells to low temperature ( 26°C ) , to accumulate F508del-CFTR at the cell surface , and then shifted the cells back to 37°C , a temperature at which the F508del-CFTR at the PM is subjected to accelerated ubiquitination and degradation ( Okiyoneda et al . , 2010 ) . Under these conditions , the depletion of MLK3 slowed the degradation rate of band C , increasing the t1/2 from ~2 to ~4 hr ( Figure 4E , F ) , whereas overexpression of CDC42 to activate MLK3 enhanced the band C degradation rate ( Figure 4G , H ) . These data suggest that also the peripheral QC of F508del-CFTR is regulated by MLK3 . In contrast , the knockdown of JNK2 ( or its overexpression ) did not change the degradation kinetics of band C , although it increased the band C/band B ratio ( not shown ) , suggesting that JNK2 may have additional effects , for instance on the folding and/or export of F508del-CFTR ( see above and Discussion ) . Similar effects on F508del-CFTR folding seem likely to be induced also by many of the CORE genes whose depletion greatly increases the band C/band B ratio , in some cases up to 4-fold over control levels ( Figure 2C ) . In conclusion here , depletion of the receptor and stress-activated MLK3 signalling pathway markedly inhibited both the ER-associated and peripheral degradation processes of F508del-CFTR while possibly at the same time increasing the efficiency of F508del-CFTR folding/ER export . Consequently , inhibition of this pathway results in large increases in the levels of the Golgi-processed mature form of F508del-CFTR . We next examined the effects on F508del-CFTR proteostasis of agents known to activate MLK3 such as TNF-α , TGF-β ( Schachter et al . , 2006 ) and reactive oxygen species ( ROS ) ( Lee et al . , 2014 ) . TNF-α and TGF-β have been proposed to be genetic modifiers of CF ( Cutting , 2010 ) , and ROS have been reported to be enhanced in CF cells ( Luciani et al . , 2010 ) and to be massively produced by neutrophils during the inflammatory reactions that are common in CF patients ( Witko-Sarsat et al . , 1995 ) . We treated CFBE cells with TNF-α , TGF-β or H2O2 ( to increase ROS ) , and monitored the effects on F508del-CFTR . The effects of H2O2 at non-toxic concentrations were dramatic , with a marked drop of the F508del-CFTR levels within 30 min ( Figure 5A , B ) . Also TNF-α and TGF-β induced rapid , although less complete ( 50% ) decreases in levels of F508del-CFTR ( Figure 5C , D ) . Under these conditions , the reduction in F508del-CFTR levels was completely abolished by MLK3 downregulation ( Figure 5A–D ) , confirming the crucial role of MLK3 pathway in F508del-CFTR QC/degradation . 10 . 7554/eLife . 10365 . 015Figure 5 . ROS and inflammatory cytokines control F508del-CFTR proteostasis via the MLK3 pathway . ( A ) CFBE cells ( pre-treated with MLK3 siRNA ) were treated with 1 mM H2O2 for 30 min and levels of band B monitored by western blotting . Control cells show drastic reduction in F508del-CFTR levels upon treatment with H2O2 that is prevented by the downregulation of MLK3 . ( B ) Quantitation of band B levels from ( A ) represented as mean ± SEM ( n > 3 ) . ( C ) CFBE cells ( pre-treated with MLK3 siRNA ) were treated with 50 ng/mL of TNF-α or TGF-β for 15 min and levels of band B monitored by western blotting . MLK3 downregulation prevents the decrease in band B levels brought about by treatment with the cytokines . ( D ) Quantitation of the band B levels from ( C ) represented as mean ± SEM ( n > 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10365 . 015 These results , and in particular the effects of H2O2 , provide evidence for the existence of a rapid and potent mechanisms of protein degradation that depend on the MLK3 pathway and act on F508del-CFTR ( and presumably on other misfolded mutant proteins ) . These regulatory mechanisms might have pathological relevance , as discussed below . We next tested the effect of selected kinase inhibitors on F508del-CFTR proteostasis in CFBE cells . A well-known characteristic of the kinase inhibitors is their promiscuity . In our experience , inhibitors that nominally target the same kinase can cause divergent effects on correction ( see below ) , most likely because they target other kinases with different or competing effects . We sought to overcome this difficulty by selecting kinase inhibitors with different structures and modes of action , and by using information from the KINOMEscan library ( http://lincs . hms . harvard . edu/data/kinomescan/ ) . For JNK , we tested a set of 10 reported JNK inhibitors ( JNKi ) , three of which led to robust increases in the levels of band B and band C ( Figure 6A–D; JNKi II , JNKi IX , JNKi XI ) at concentrations that were required for JNK inhibition in CFBE cells ( Figure 6—figure supplement 1A ) . These JNK inhibitors have different chemical structures; moreover , while JNKi II and JNKi IX are ATP-competitive inhibitors of JNK , JNKi XI is an inhibitor of substrate/scaffold binding to JNK . Therefore , they appear to be reliable tools to correct F508del-CFTR by targeting JNK . A previously proposed MLK3 inhibitor ( K252a ) had no clear effects on correction , perhaps because it inhibits MLK3 weakly and has diverging effects on other kinases ( see http://www . kinase-screen . mrc . ac . uk/screening-compounds/345892 ) . We thus searched the KINOMEscan library for a molecule that had a suitable inhibitory pattern on the MLK3 pathway . ( 5Z ) -7-oxozeaenol ( hereafter referred to as oxozeaenol ) ( Ninomiya-Tsuji et al . , 2003 ) potently inhibits the MLK3 pathway members VEGF and PDGF receptor kinases and ( less potently ) MLK3 itself and MKK7 , as well as , more weakly , a few kinases with antagonistic effects on correction ( http://lincs . hms . harvard . edu/db/datasets/20211/ ) . Oxozeaenol markedly increased the bands B and C of F508del-CFTR ( Figure 6A–C ) . This drug had been identified as a corrector in an earlier screening study , and proposed to have F508del-CFTR corrector properties as an inhibitor of TAK1 ( MAP3K7 ) ( Trzcinska-Daneluti et al . , 2012 ) . However , the downregulation of TAK1 itself had no effect on correction ( Figure 6—figure supplement 1B ) . Although the pharmacological inhibition and the siRNA knockdown of kinases can sometimes have distinct consequences , these data suggest that oxozeaenol likely acts through MLK3 pathway kinases to affect proteostasis rather that through TAK1 . In line with this notion , the corrective effects of oxozeaenol were not additive with MLK3 knockdown ( Figure 6—figure supplement 1C ) and were accompanied by a reduction in phospho c-jun levels ( c-jun phosphorylation is diagnostic of JNK activity ) ( Figure 6—figure supplement 1D ) . 10 . 7554/eLife . 10365 . 016Figure 6 . Inhibitors of the MLK3 pathway rescue F508del-CFTR . ( A–D ) CFBE cells were treated with the indicated inhibitors of the MLK3 pathway or VX-809 for 48 h , and the rescue of F508del-CFTR from ERQC was monitored ( A ) . MLK3 pathway inhibitors rescue F508del-CFTR to levels comparable to or more than that achieved by VX-809 . Changes in the levels of band C ( B ) , band B ( C ) and the band C/band B ratio ( D ) were quantitated and shown mean ± SEM ( n > 3 ) . Normalized concentration ( abscissa in panels B–D ) refers to concentration ( VX-809 , JNKi IX and Oxozeaenol [1 . 25 , 2 . 5 , 5 , 10 μM] , JNKi II [6 . 25 , 12 . 5 , 25 , 50 μM] , JNKi XI [3 . 12 , 6 . 25 , 12 . 5 , 25 μM] ) values that were normalized to the maximum used concentrations of the respective drugs . Refer panel A for concentrations ( µM ) of the drugs used . ( E ) CFBE cells were treated with inhibitors of the MLK3 pathway and/or VX-809 ( 5 μM ) for 48 hr and changes in band C levels monitored . The concentrations of the MLK3 pathway inhibitors used were: JNKi II ( 12 . 5 μM ) , JNKi IX ( 5 μM ) , JNKi XI ( 25 μM ) and oxozeaenol ( 5 μM ) . Wild-type CFTR ( wt-CFTR ) was used as a control . ( F ) Quantitation of band C levels from ( E ) normalized to the levels of band C after VX-809 treatment are shown as mean ± SEM ( n > 3 ) . The results show that synergy obtained between the MLK3 pathway inhibitors and VX-809 brings the levels of band C to ~40% of the wild-type levels . DOI: http://dx . doi . org/10 . 7554/eLife . 10365 . 01610 . 7554/eLife . 10365 . 017Figure 6—figure supplement 1 . Small-molecule inhibitors of the MLK3 pathway rescue F508del-CFTR and other structurally related mutant proteins from degradation . ( A ) CFBE cells were treated with indicated JNK inhibitors for 24 hr and processed for western blotting . The levels of phospho-c-jun as a measure of JNK inhibition was monitored . MLK3 pathway inhibitors reduce phospho-c-jun levels efficiently indicating a strong reduction in the activity of JNK and hence presumably of the MLK3 pathway . ( B ) CFBE cells were treated with TAK1 or MLK3 siRNA as indicated and changes in F508del-CFTR proteostasis were monitored by western blotting . TAK1 does not regulate F508del-CFTR proteostasis , as evidenced by the absence of change in the levels of bands C or B . The fold change in the band C levels was quantitated and plotted as mean ± SD ( n = 2 ) . ( C ) CFBE cells were treated with 5 μM oxozeaenol for 48 hr , or with MLK3 siRNA , or with both , and the correction of the F508del-CFTR folding/trafficking defect was monitored by changes in the levels of band C . There was no additive effect observed with the combination of MLK3 downregulation and oxozeaenol treatment . The quantitated band C levels are expressed as mean ± SD ( n > 3 ) . ( D ) CFBE cells were treated with 5 μM oxozeaenol for 24 hr , and the activity of the JNK pathway was measured by western blotting for phospho c-jun levels and F508del-CFTR . The levels of phospho c-jun were reduced , suggesting that oxozeaenol leads to a reduction in the activity of JNK . The increase in band C levels of F508del-CFTR show that the reduction in the activity of JNK is accompanied by a rescue of F508del-CFTR from ERQC . ( E ) CFBE cells were treated with flunarizine ( at concentrations 6 . 25–50 μM ) targeting the CAMKK2 pathway for 48 hr and the effect on F508del-CFTR proteostasis measured by western blotting . Treatment with flunarizine increased the levels of band C of F508del-CFTR . Other small molecules known to inhibit the CAMKK2 pathway components ( verapamil and STO-609 ) did not show any effect on correction of F508del-CFTR . ( F ) CFBE cells transiently transfected with the P-glycoprotein mutant ( P-gp DY490 ) , the NCC mutant ( R948X ) , or the hERG mutant ( G601S ) were treated with JNKi II for 24 hr , and the effect of the drug on their proteostasis monitored by western blotting . While the trafficking of P-gp DY490 out of the ER was enhanced by this treatment ( seen as an increase in the Golgi-associated band C , indicated by arrows ) , other mutants are subjected to enhanced degradation upon drug treatment , as shown by a decrease in the levels of both bands B and C . ( G ) CFBE cells were treated with JNKi II at indicated concentrations ( 25 , 50 µM ) for 48 hr and the changes in the proteostasis of non-mutant endogenous proteins like E-cadherin , IGF1Rß and EGFR were monitored by western blotting . There was no significant change in the proteostasis of these proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 10365 . 01710 . 7554/eLife . 10365 . 018Figure 6—figure supplement 2 . Small molecule inhibitors of MLK3 pathway rescue channel function of F508del-CFTR . ( A ) CFBE-YFP cells were treated with MLK3 pathway inhibitors and/or VX-809 for 48 hr , and the anion transport measured as described in the Materials and methods . The rate constants of the decrease in YFP fluorescence ( K ) , a measure of anion conductance , after inhibitor treatments are shown . The data are expressed as mean ± SEM ( n > 3 ) . ( B ) Anion transport was measured in CFBE-YFP cells after downregulating the MLK3 pathway activity by siRNA-mediated knockdown MLK3 or JNK2 . The rate constants of the decrease in YFP fluorescence ( K ) , a measure of anion conductance , after downregulation of the indicated MLK3 pathway components are shown . The data are expressed as mean ± SEM ( n > 3 ) . Treatment with VX-809 was used as a positive control for the rescue . ( C ) CFBE41o- cells were grown under polarizing conditions before addition of oxozeaenol at indicated concentrations for 48 hr , followed by the measurement of short circuit currents using Ussing chamber assays ( see Materials and methods ) . The columns show the measured values of the short-circuit current after oxozeaenol treatment at the indicated concentrations . The values are mean ± SEM ( n>3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10365 . 018 In sum , selected chemical blockers of the MLK3 ( and CAMKK2; Figure 6—figure supplement 1E ) pathway potently increase the levels of band B and band C as well as the band C/band B ratio ( Figure 6A–D ) . The level of correction obtained with these inhibitors is higher than the effects of the corrector compounds from which the pathways were deduced ( Figure 2—figure supplement 2A ) , and similar to , or higher than , the effects of VX-809 . We next considered that , while inhibitors of MLK3 pathway lead to large increases in the ER-localised band B , the pharmacochaperone VX-809 increases F508del-CFTR band C levels with only a limited effect on band B ( Figure 6B–D ) , presumably because it primarily enhances F508del-CFTR folding . This suggested that if the large pool of band B protein accumulating in the ER following the inhibition of the MLK3 pathway is in a foldable state , VX-809 should act on such pool to enhance its folding , and greatly increase the generation of the band C mature protein . Indeed , when we added both MLK3 pathway inhibitors and VX-809 , there was potent synergy between them ( Figure 6E , F ) with increases in levels of band C that were over 20-fold the basal band C level and fourfold over those obtained with VX-809 alone . Given the promising results of VX-809 in combination therapies in recent clinical trials ( Wainwright et al . , 2015 ) and in experimental settings ( Okiyoneda et al . , 2013; Phuan et al . , 2014 ) , the observed additive/synergistic effects of VX-809 combined with MLK3 pathway inhibitors might provide a potential therapeutic opportunity . We next examined the effects of the MLK3 pathway inhibition on the proteostasis of other conformational disease mutants . We transfected CFBE ( and HeLa ) cells with different conformational mutants ( i . e . , Sodium-chloride symporter [NCC , R948X mutant]; P-glycoprotein , [P-gp , G268V and DY490 mutants]; human Ether-à-go-go-Related Gene [hERG , G601S mutant]; Wilson’s disease associated protein [ATP7B , H1069Q and R778L mutants] ) and then treated cells with JNKi II . The effects on their proteostasis was monitored by assessing changes in their glycosylation patterns ( NCC , P-gp , hERG mutants ) or in their intracellular movement from the ER to the Golgi complex ( ATP7B mutants ) . JNKi II rescued some of these mutants ( P-gp DY490 and ATP7B mutants ) , while it had no effects or had ‘negative’ effects , on others ( Figure 6—figure supplement 1F ) . The effects on ATP7B were large and led to almost complete correction , as we report elsewhere ( Chesi et al . , 2015 ) . We also tested the effect of JNKi II on other endogenous non-mutant proteins ( apart from Na+/K+-ATPase that was a reference in many of our treatments ) including , E-cadherin , IGF1R-β and EGFR ( Figure 6 and Figure 6—figure supplement 1G ) and found no effect on their proteostasis . Both P-glycoprotein and ATP7B , like CFTR , have two groups of transmembrane domains with an interconnecting nucleotide-binding domain . Moreover , the mutations tested ( DY490 and H1069Q ) are located in the nucleotide-binding domains of these proteins , and result from either a loss or substitution of aromatic amino acids , as for F508del-CFTR . These similarities suggest that a common proteostatic machinery might be involved in the processing of these mutants and can be targeted by the MLK3 pathway in a selective fashion . Based on literature data , interaction databases and our own experimental findings , the correction-relevant components we identified can be organised into five signalling cascades , which , for brevity , we refer to as the MLK3 , CAMKK2 , PI3K , CKII , and ERBB4 networks . Other networks were made up of constituents of the spliceosome , centromere and mediator ( transcriptional ) complexes , or of ubiquitin ligases . The physiological role of these CORE signalling systems might be to regulate the stringency of the QC and degradation processes . Most of the CORE pathways enhance the efficiency of QC and degradation . This is the case of the MLK3 pathway , which can be activated by selected cytokines and by cellular stresses . The ERBB4 pathway , in contrast , is activated under growth conditions , and appears to have the effect of suppressing the QC and degradation processes . It may be speculated that cells under stress need to reduce the toxic burden of certain classes of unfolded proteins to survive , while growing cells might need to ‘tolerate’ higher levels of folding/unfolded proteins to proliferate , and that the CORE pathways regulate the proteostasis machinery according to needs . A further possibility is that at least part of the CORE pathways might function as part of an internal control system ( Cancino et al . , 2014; Luini et al . , 2014 ) that is activated by the presence of unfolded/misfolded proteins . Interestingly in this regard , MLK3 interacts directly with ( and might be activated by ) HSP90 ( Zhang et al . , 2004 ) , a component of the F508del-CFTR folding and QC machinery . Under pathological conditions such as cystic fibrosis and similar diseases , the activity of MLK3 and other core pathways can become deleterious , as it enhances the degradation of protein mutants that retain the potential to function ( such as F508del-CFTR ) . Also importantly , they can be hyperactivated under pathological conditions , leading to vicious circles . For example , large amounts of ROS are produced by neutrophils in the inflamed lungs of CF patients ( Witko-Sarsat et al . , 1995 ) , and elevated serum VEGF are detected in some CF patients ( McColley et al . , 2000 ) . Both of these molecules act via the MLK3 pathway to enhance the degradation of F508del-CFTR , and in particular the ROS do so with striking efficacy and speed ( Figure 5A , B ) . These effects most probably result in lowering F508del-CFTR to levels below those that would be determined by the primary folding defect , which might be harmful because even low residual levels of F508del-CFTR may help to improve the CF phenotype in the long term ( Amaral , 2005 ) . Blocking the MLK3 pathway is thus probably important to stop maladaptive processes that can adversely affect therapeutic efforts . Similar considerations apply to the CAMKK2 and other CORE-derived pathways . The ER quality control relies on chaperones such as HSP90 and HSC70 that are also involved in folding and can switch between folding and quality control/degradation roles depending on their dwell-time on the folding client proteins ( Zhang et al . , 2013 ) . The simplest interpretation of the data is therefore that inhibition of the MLK3 pathway regulates this folding/degradation switch by impairing the entry of F508del-CFTR into the degradation pathway and giving the mutant more time to fold and exit the ER . Although MLK3 does not measurably affect the folding of F508del-CFTR as measured by the trypsin susceptibility assay , it cannot be excluded , however , that MLK3 ( and other CORE genes ) might exert subtle direct actions on the folding and/or ER export mechanisms . This is supported by the strong effects of some of the CORE pathways on the band C/band B ratio , and also by the observation that the inhibition of MLK3 markedly stimulates the efficiency of export of a mutant of ATP7B ( similar in structure to CFTR ) from the ER ( Chesi et al . , 2015 ) . At the molecular level , the mechanisms underlying these rescue effects remain unclear . Some initial insight might come from our observation that the phosphoprotein HOP co-precipitates much less efficiently with F508del-CFTR in cells treated with JNK inhibitors than in control cells ( Figure 4—figure supplement 1G ) , while its amount remains unchanged . HOP serves as a link between HSC70 and HSP90 in the F508del-CFTR folding process , and its depletion induces rescue of F508del-CFTR ( Marozkina et al . , 2010 ) , possibly by acting on the folding/ERQC switch discussed above . Thus , a reduced interaction of HOP with F508del-CFTR–associated QC/folding complex might be one of the possible modes of action of MLK3 on F508del-CFTR rescue . This might be an example of how corrector drugs may act by modulating the ‘activity’ , rather than the levels , of relevant machinery proteins . A complete analysis of the effects of the MLK3 pathway on the interactions and posttranslational modifications of the ERQC/ERAD machinery components remains a task for future work . As already noted , this study does not aim to generate efficient correctors ready for clinical use . Rather , it aims to elucidate the regulatory signalling networks that control a central element of the disease viz . the proteostasis machinery acting on F508del-CFTR , with a view to providing a rational basis to identify relevant pharmacological targets and , in the long run , more effective F508del-CFTR correctors . This goal appears realistic , because signalling cascades are eminently druggable ( the majority of the known drug targets are signalling components [Imming et al . , 2006] ) , and an enormous repertoire of drugs directed at kinases and other related molecules has been developed by the pharmaceutical industry for the therapy of major diseases . For instance , over 120 inhibitors against the correction-related kinases identified in this study ( not shown ) are currently in clinical trial . Moreover , as shown for the case of oxozeaenol ( Figure 6A–D ) , suitable kinase inhibitors can be selected in a rational fashion by matching the list of CORE kinases with the kinase inhibitory patterns of the many available drugs of this class , according to polypharmacology principles ( Aggarwal et al . , 2007 ) . It is thus quite possible that some of these drugs may be repositioned for the CF therapy . In addition , the CORE ubiquitin ligases , particularly RNF215 , are also attractive targets in view of their potent effect on F508del-CFTR correction ( Figure 2A ) . Although the technology for developing ubiquitin ligase inhibitors is still in its early stages , robust progress is being made in this direction ( Goldenberg et al . , 2010 ) . A further consideration is that the inhibitors of the CORE pathways show corrective effects that are ( partially ) selective for F508del-CFTR ( and structurally related mutants ) ( see Figure 6—figure supplement 1F ) ; and that these effects are complementary and synergic with those of the pharmacochaperone VX-809 . Since these synergies lead to levels of correction that are several-fold higher than those achieved by VX-809 alone , it is possible that they result in combination therapies of clinical interest . Also of note is that the MOA-based approach used here can be exploited further in the future to identify more CORE pathways as well as more effective and specific correctors . A remaining obstacle is the need to combine the correction of proteostasis , which is the focus of this manuscript , with the restoration of the activity of the channel . The restitution of chloride channel conductance at the PM that we observe with the inhibitors of MLK3 pathway does not always match up to the level of rescue obtained on proteostasis alone ( see also Figure 6—figure supplement 2 ) . Notably , a similar quantitative discrepancy between the effect on proteostasis and the effect on chloride channel activity has been observed before ( see [Hutt et al . , 2010] ) . Regulation of the chloride channel activity is a complex phenomenon and is determined by several factors including the input from signaling pathways ( for e . g . PKA pathway and others ) , and the presence ( or absence ) of regulators such as NHERF and cytoskeletal components . Further studies that reveal the regulatory network controlling the activity of the channel at the PM will complement our study on proteostasis and will help in the rational design of pharmaceutical approaches . In addition , a key requirement for translating our findings towards clinical treatments is the conservation of the CORE pathways in bronchial epithelial cells in situ . We have observed fundamentally similar roles of the CORE networks across several human and mammalian cell lines , both under polarized and non-polarized conditions , suggesting that these networks are well conserved . Moreover , JNK has been reported to be hyperactive in the lungs of a mice model of CF ( Grassme et al . , 2014 ) , as is p38 MAPK ( also activated by MLK3 ) in the lungs of CF patients ( Berube et al . , 2010 ) , indicating that a SAPK pathway is activated under these conditions . Also notably , the MLK3 pathway inhibitor oxozeaenol has been reported to be effective in correcting the F508del-CFTR proteostasis defect in the primary human bronchial epithelial cells ( Trzcinska-Daneluti et al . , 2012 ) . These observations , together with the fact that the CF genetic modifiers TNF-α and TGF-β potently affect F508del-CFTR proteostasis , support the notion that a regulatory network similar to that uncovered in CFBE cells operates on the proteostasis machinery in bronchial epithelial cells in CF patients . In sum , this study builds on previous screening studies and on the accumulated knowledge about the F508del-CFTR proteostasis machinery ( Balch et al . , 2011; Farinha et al . , 2013b; Lukacs and Verkman , 2012; Turnbull et al . , 2007 ) to identify signalling pathways acting on F508del-CFTR proteostasis . This provides new insight into the physiopathology of F508del-CFTR and opens new possibilities to pharmacologically correct the folding and trafficking defect of this mutant protein . To establish the efficacy of these interventions in human bronchial epithelia and relevant animal models ( Yan et al . , 2015 ) will be the next stage towards the rational development of effective F508del-CFTR proteostasis regulators for patients with CF . Cell lines used in this study are characterized and reported CF model systems . These include CFBE cells stably expressing wild type CFTR or F508del-CFTR ( Bebok et al . , 2005 ) and stably expressing halide sensitive YFP ( Pedemonte et al . , 2005 ) and HeLa cells stably expressing HA-tagged F508del-CFTR ( Okiyoneda et al . , 2010 ) that were obtained after material transfer agreements from the respective laboratories . These cell lines are not commercial and their STR status is unknown . Mycoplasma contamination was not observed in the cell cultures . CFBE cells were cultured in Minimal Essential Medium supplemented with 10% foetal bovine serum , non-essential amino acids , glutamine , penicillin/streptomycin and 2 µg/ml puromycin . This media additionally supplemented with 50 μg/ml G418 was used for the CFBE-YFP cells . HeLa cells were cultured in Dulbecco’s modified Eagle’s medium ( DMEM ) supplemented with 10% foetal bovine serum , glutamine , penicillin/streptomycin and 1 µg/ml puromycin . The antibodies used were: anti-phospho-c-jun , EGFR , IGF1R-β ( Cell Signaling Technology , Danvers , MA ) , E-cadherin ( Abcam , UK ) , monoclonal anti-HA , anti-actin and anti-tubulin ( Sigma , St . Louis , MO ) , rat anti-CFTR ( 3G11; CFTR Folding Consortium ) , mouse monoclonal anti-CFTR ( M3A7 ) , HRP-conjugated anti-mouse , rabbit and rat IgG ( Merck Millipore , Germany ) and anti-Na/KATPase α1 ( Thermo Fisher Scientific , Waltham , MA ) . The plasmids used were: JNK2 ( pCDNA3 Flag MKK7B2Jnk2a2; Addgene plasmid #19727 ) and MKK7 ( pCDNA3 Flag MKK7b1; Addgene plasmid #14622 ) from Roger Davis ( University of Massachusetts Medical School , Worcester , USA ) , ZsProSensor-1 proteasome sensor ( Clontech , Mountain View , CA ) , VSVG tagged with GFP ( Jennifer Lippincott-Schwartz , NICHD , NIH , Bethesda , USA ) , Cdc42 ( A . Hall , Sloan-Kettering Institute , New York , NY , USA ) , P-glycoprotein wild type , G268V and DY490 mutants ( David M . Clarke , University of Toronto , Canada ) and hERG wild type and G601S mutant ( Alvin Shrier , McGill University , Montreal , Canada ) . The reagents used include: VX-809 ( Selleckchem , Germany ) , JNKi II ( SP600125 ) , JNKi IX and JNKi XI ( Merck Millipore , Germany ) , oxozeaenol ( Tocris Bioscience , UK ) , siRNAs ( see Supplementary file 6 ) , lipofectamine 2000 ( Thermo Fisher Scientific ) and ECL ( Luminata crescendo from Merck Millipore ) . To understand if the proteostatic correctors have a transcriptional component in their corrector MOA , the sensitivity of their corrector action to actinomycin D , an inhibitor of transcription , was tested . To this end , we first analysed the kinetics of action of the corrector drugs in modulating F508del-CFTR proteostasis . HeLa cells stably expressing HA-tagged F508del-CFTR ( Okiyoneda et al . , 2010 ) were treated with selected corrector drugs ( from MANTRA dataset ) that resulted in a detectable increase in the intensity of band B already after 3 hr , while the effects on band C levels were detectable only after 12–24 hr of drug treatment ( not shown ) . In order to minimize the toxic effects of actinomycin D on cell physiology and since the increase in band C was always preceded by an increase in band B , we decided to monitor the actinomycin D sensitivity of the corrector drugs towards the regulation of F508del-CFTR proteostasis by monitoring the early changes in band B levels . To this end , HeLa cells stably expressing HA-tagged F508del-CFTR were treated either with corrector drugs alone ( [Chlorzoxazone [50 µM] , Glafenine [50 µM] , Trichostatin-A [500 nM] , Dexamethasone [500 nM] , Doxorubicin [500 nM] ) or along with 10 ug/ml actinomycin-D for 3 hr and the levels of band B were determined by western blotting . While actinomycin D treatment did not have any effect on the increase in band B levels resulting from treatment with VX-325 or Corr-4a , it significantly reduced band B levels in other cases ( not shown ) . This suggests that the corrector drugs ( except VX-325 and Corr-4a ) are proteostatic regulators that act by inducing transcriptional changes in the cell . Polarised CFBE41o-cells cultured at the air–liquid interface were treated with the corrector drugs of interest ( CFBE dataset , Table 1 ) for 24 hr . Total RNA was extracted and hybridization was carried out on to Whole Human Genome 44 K arrays ( Agilent Technologies , product G4112A ) following the manufacturer’s protocol . See ( Zhang et al . , 2012 ) for experimental details . The microarray data for ouabain and low temperature treatments have been published elsewhere ( Zhang et al . , 2012 ) . The microarrays from the connectivity map database ( https://www . broadinstitute . org/cmap/ ) were processed to produce prototype ranked lists ( PRLs ) ( Iorio et al . , 2010 ) . In these PRLs , cell line-specific responses are diluted , thus summarising consensual transcriptional responses to drug treatment . In each PRL , microarray probe-sets are ordered from the most upregulated to most downregulated one . We downloaded PRLs for the whole panel of small molecules in the connectivity map ( www . connectivitymap . org ) from which the MANTRA database is derived ( http://mantra . tigem . it/ ) . We used these in conjunction with ranked lists of probe sets based on fold changes ( and assembled following the guidelines provided in [Iorio et al . , 2010] ) from microarray profiles that we generated in house ( CFBE dataset ) . The FIT analysis identifies microarray probe-sets that tend to respond consistently to a group of drugs ( see also [Iorio et al . , 2010] for description of a similar method ) . The top and bottom 20% of the probe-sets ( corresponding to the up- and downregulated probe-sets , respectively ) were used for the analysis . The 20% cut-off was used since the merging of individual gene expression profiles into PRLs precludes the application of other thresholds based on fold change ( or p-value ) to identify significantly differentially expressed genes . To build a null model against which the significance of the final genes sets can be tested ( as detailed below ) , a fixed number of PRLs ( N ) from the MANTRA dataset were randomly selected and the upregulated or downregulated probe-sets from this selection were intersected by varying the fuzzy cut-off threshold ( i . e . the ratio of drugs that a given probe-set should transcriptionally respond to , in order to be considered ‘consistently’ regulated , hence to be included in the fuzzy intersection ) . After 1000 of these iterations , we derived an empirical null distribution of the number of probes included in the resulting fuzzy intersections and used it for p-value assignments ( Figure 1B ) . For the CFBE dataset ( generated on an Agilent platform , which is different from that used for the connectivity map and MANTRA database ) , we derived this null distribution by randomly permuting all the individual probes . Finally , we determined the optimal fuzzy cut-off values for the transcriptional profiles elicited by the corrector drugs ( 11 contained in MANTRA and 13 in the CFBE dataset ) . Briefly , we selected the value such that the number of probes present in the final fuzzy intersection was at least threefold higher than that expected by random chance and its p-value < 0 . 05 ( according to the computed null models ) . By using this method , no significantly upregulated probes from the MANTRA dataset were identified across all of the range of tested fuzzy cut-offs . For the downregulated probe-sets a fuzzy cut-off of 8 ( out of 11 corrector drugs ) or above produced significant fuzzy intersection of probe-sets . For the CFBE dataset , a significant cut-off of 6 drugs ( out of 13 ) and above was identified . To optimise the selection of these cut-offs further , we chose the maximal cut-off yielding a fuzzy intersection of probe-sets enriched in one or more Gene Ontology terms . With this criterion , we obtained a final cut-off value of 8 for the MANTRA downregulated probe set and cut-off of 9 for the CFBE dataset . Intersecting the corrector-induced gene expression profiles using this optimal fuzzy cut-off resulted in 541 upregulated probe-sets ( mapping 402 unique genes ) and 191 downregulated probe-sets ( mapping to 117 unique genes ) for the CFBE dataset , and 108 downregulated probe-sets ( mapping 102 genes ) for the MANTRA dataset . Note that most of the CORE genes ( 519 out of the 621 CORE genes ) are derived from the CFBE dataset . This , we suppose , is due to the use of PRLs in the case of cMAP dataset and use of data derived from a single cell line in the case of CFBE dataset . The use of single cell line-derived data can potentially lead to high number of false positives since perturbation-independent response of cell lines to treatments is usually stronger than the perturbation-dependent response ( Iorio et al . , 2010 ) . We finally validated the optimal number of drugs that need to be considered for a fuzzy cut-off of 70% ( corresponding to 8 out 11 drugs cut-off from the MANTRA dataset ) , providing a minimum number of false positives in the intersection ( i . e . genes expected to be contained in the resulting intersections by random chance ) . This was performed by a permutation test where , in a series of iterations , the fuzzy cut-off is kept constant and the number of randomly selected drugs varied within a given range ( specifically from 1 to 20 ) . At each of these iterations , we computed the cardinality of the resulting fuzzy intersections , observing that this value reached a plateau at 10 drugs ( Figure 1C ) , which suggests that the number of drugs that was used in the analysis ( i . e . 11 drugs in the cMAP dataset ) was fairly close to the optimal level . The protein-protein interactions were downloaded from the STRING database ( http://string-db . org/ ) ( Franceschini et al . , 2013 ) , and those with a confidence level of >0 . 7 were used for the analysis . To build the proteostasis gene ( PG ) dataset , we included known proteostatic regulators of CFTR , that is , proteins where their expression/activity level changes have been shown to affect CFTR proteostasis . We also included the interactors of CFTR and CF pathology related genes/proteins present in GeneGO Metaminer Cystic Fibrosis database ( see Supplementary file 4 for the list of the proteostasis genes ) . The number of interactions observed among the CORE gene dataset and the proteostasis gene dataset as well as among the CORE gene dataset were more than expected on a random basis and were statistically significant . For details on the statistical test used see ( Franceschini et al . , 2013 ) . The gene sets were analyzed using the CORE analysis application of the Ingenuity pathway analysis , a web-based software application . The default settings of the analysis were used . Each network had an assigned significance score based on the p-value ( calculated using Fischer's exact test ) for the probability of finding the focus genes in a set of genes randomly selected from the global molecular network . The upregulated and downregulated genes of the CFBE dataset and the downregulated genes of the cMAP dataset were analyzed separately and also together , to infer common pathways or networks embedded among them . Cells were washed three times in ice-cold Dulbecco’s phosphate-buffered saline , and lysed in RIPA buffer ( 150 mM NaCl , 1% Triton X-100 , 0 . 5% deoxycholic acid , 0 . 1% SDS , 20 mM Tris-HCl , pH 7 . 4 ) , supplemented with protease inhibitor cocktail and phosphatase inhibitors . The lysates were clarified by centrifugation at 15 , 000 x g for 15 min , and the supernatants were resolved by SDS-PAGE . BCA Protein Assay kit ( Pierce ) was used to quantitate protein levels before loading . The western blots were developed with appropriate antibodies and using ECL . The blots were then exposed to x-ray films and exposure time was varied to obtain optimal signal . The X-ray films were then scanned and the bands were quantitated using ImageJ gel-analysis tool ( see Figure 2—figure supplement 2A–D ) . The protein concentration and the exposures used for quantitation of the blots were optimized to be in a linear range of detection ( Figure 2—figure supplement 1E , F ) . Each gene was targeted by three siRNAs and as control non-targeting siRNAs provided by the manufacturer were used ( see Supplementary file 6 for list of siRNAs used ) . A gene was considered as active if: ( 1 ) at least two different siRNAs targeting a gene gave concordant changes in the levels of band C that was >2 SD from the mean value of the control siRNAs and ( 2 ) the change in band C levels was ± 20% of the level of band C obtained with the control siRNAs . Those genes that increased band C levels significantly upon their downregulation were termed anti-correction genes and those that decreased band C levels were termed pro-correction genes . A potential problem in siRNA-based experiments is the possibility of off-target effects . The specificity of the observed effects on F508del-CFTR proteostasis are supported by the following lines of evidence: ( 1 ) The quantitation of the on-target effect of the siRNAs by RT-PCR ( see Figure 2—figure supplement 3 ) , which shows that treatment with siRNAs brings down the transcript levels of the target genes to ~10–30% of the levels present in cells treated with non-targeting siRNAs; ( 2 ) The use of at least three different siRNAs ( individually ) for the screenings , the majority of which showed concordant results on proteostasis . Moreover , for selected genes ( MLK3 , CAMKK2 , RNF215 , NUP50 and CD2BP2 ) , the findings from the screening studies were reinforced using additional siRNAs ( for instance , five additional siRNAs for MLK3; see Supplementary file 6 for details of the siRNAs used ) that showed similar effects on proteostasis of F508del-CFTR ( not shown ) . ( 3 ) The coherent effects of the siRNA targeting different genes of signaling pathways supports the functional significance of the effects of each siRNA ( see Figure 3A , B and Figure 3—figure supplement 1B ) . ( 4 ) Overexpression of identified hits show an opposite effect on proteostasis to that of the siRNAs ( see below Figure 3C ) . ( 5 ) Finally , there was a positive correlation between the concentration of siRNAs used and the effect on proteostasis ( Figure 2—figure supplement 3 ) . All these evidence together confirm the specificity of the observed effect of the siRNAs on proteostasis . HeLa cells cultured in 10-cm plates ( 80% confluence ) were treated with appropriate corrector drugs for 24 hr . The cells then were washed three times in ice-cold Dulbecco’s phosphate-buffered saline , and lysed in immunoprecipitation buffer ( 150 mM NaCl , 1% Triton X-100 , 20 mM Tris-HCl , pH 7 . 4 ) on ice for 30 min . The lysates were clarified by centrifugation at 15 , 000 x g for 15 min , and the protein content of the supernatants quantitated by BCA Protein Assay kit ( Pierce ) . Equal amounts of proteins from control and treated cell lysates were incubated with Protein-G sepharose beads conjugated with anti-HA antibody ( Sigma ) overnight at 4°C . The beads were then washed in the immunoprecipitation buffer five times and the bound proteins eluted with HA-peptide ( Sigma ) at a concentration of 100 μg/ml . The eluted proteins were then resolved by SDS-PAGE and immunoblotted . The trypsin digestion assay was similar to that described previously ( Zhang et al . , 1998 ) . Cells were grown in a 10-cm plate and post-treatment were washed three times with 10 mL phosphate-buffered saline ( PBS ) . They were then scraped in 5 mL PBS , and pelleted at 500 x g for 5 min in 4°C . The cell pellet was resuspended in 1 mL of hypertonic buffer ( 250 mM sucrose , 10 mM Hepes , pH 7 . 2 ) and the cells were then homogenized using a ball bearing homogenizer . The nuclei and unbroken cells were removed by centrifugation at 600 x g for 15 min . The membranes were then pelleted by centrifugation at 100 , 000 x g for 30 min , and then resuspended in digestion buffer ( 40 mM Tris pH 7 . 4 , 2 mM MgCl2 , 0 . 1 mM EDTA ) . Then membranes corresponding to 50 μg of protein were incubated with different concentrations of trypsin ( 1 to 50 μg/ml ) on ice for 15 min . The reactions were stopped with the addition of soya bean trypsin inhibitor ( Sigma ) to a final concentration of 1 mM , and the samples were immediately denatured in sample buffer ( 62 . 5 mM Tris-HCL , pH 6 . 8 , 2% SDS , 10% glycerol , 0 . 001% bromophenol , 125 mM dithiothreitol ) at 37°C for 30 min . The samples were resolved on 4-16% gradient SDS-PAGE ( Tris-glycine ) and transferred onto nitrocellulose membranes . These membranes were developed with the 3G11 anti-CFTR antibodies ( that recognize nucleotide-binding domain 1 - NBD1 ) or the M3A7 clone ( that recognizes nucleotide-binding domain 2 -NBD2 ) . The PQC assay was essentially as described previously ( Okiyoneda et al . , 2010 ) . CFBE cells were untreated or treated with siRNAs for 72 hr and for the final 31 hr they were kept at low temperature ( 26°C ) and for an additional 5 hr at 26°C with CHX ( 100 µg/ml ) . Then , the cells were shifted to 37°C for 1 . 5 hr with 100 µg/ml CHX before the turnover measurements started at 37°C . The cells were lysed at 0 , 1 , 3 and 5 hr and the kinetics of degradation of band C was examined by immunoblotting . Twenty-four hours after plating , the CFBE cells that stably expressed halide sensitive YFP were incubated with the test compounds at 37°C for 48 hr . At the time of the assay , the cells were washed with PBS ( containing 137 mM NaCl , 2 . 7 mM KCl , 8 . 1 mM Na2HPO4 , 1 . 5 mM KH2PO4 , 1 mM CaCl2 , 0 . 5 mM MgCl2 ) and stimulated for 30 min with 20 μm forskolin and 50 μm genistein . The cells were then transferred to a Zeiss LSM700 confocal microscope , where the images were acquired with a 20x objective ( 0 . 50 NA ) and with an open pinhole ( 459 μm ) at a rate of 330 ms/frame ( each frame corresponding to 159 . 42 μm x 159 . 42 μm ) , at ambient temperature . The excitation laser line 488 nm was used at 2% efficiency coupled to a dual beam splitter ( 621 nm ) for detection . The images ( 8-bit ) were acquired in a 512x512 format with no averaging to maximize the speed of acquisition . Each assay consisted of a continuous 300-s fluorescence reading with 30 s before and the rest after injection of an iodide-containing solution ( PBS with Cl− replaced by I−; final I− concentration in the well 100 mM ) . To determine the fluorescence-quenching rate associated with I− influx , the final 200 s of the data for each well were fitted with a mono-exponential decay , and the decay constant K was calculated using GraphPad Prism software . Short-circuit current ( Isc ) was measured across monolayers in modified Ussing chambers . CFBE41o- cells ( 1x106 ) were seededonto 12-mm fibronectin-coated Snapwell inserts ( Corning ) and the apical medium was removed after 24 hr to establish an air-liquid interface . Transepithelial resistance was monitored using an EVOM epithelial volt-ohm meter and cells were used when the transepithelial resistance was 300–400 Ω . cm2 . CFBE41o- monolayers were treated on both sides with optiMEM medium containing 2% ( v/v ) FBS and one of the following compounds: 0 . 1% DMSO ( negative control ) , or compounds at the stated dosage for 48 hr before being mounted in EasyMount chambers and voltage-clamped using a VCCMC6 multichannel current-voltage clamp ( Physiologic Instruments ) . The apical membrane conductance was functionally isolated by permeabilising the basolateral membrane with 200 µg/ml nystatin and imposing an apical-to-basolateral Cl- gradient . The basolateral bathing solution contained 1 . 2 mM NaCl , 115 mM Na-gluconate , 25 mM NaHCO3 , 1 . 2 mM MgCl2 , 4 mM CaCl2 , 2 . 4 mM KH2PO4 , 1 . 24 mM K2HPO4 and 10 mM glucose ( pH 7 . 4 ) . The CaCl2concentration was increased to 4mM to compensate for the chelation of calcium by gluconate . The apical bathing solution contained 115 mM NaCl , 25 mM NaHCO3 , 1 . 2 mM MgCl2 , 1 . 2 mM CaCl2 , 2 . 4 mM KH2PO4 , 1 . 24 mM K2HPO4 and 10 mM mannitol ( pH 7 . 4 ) . The apical solution contained mannitol instead of glucose to eliminate currents mediated by Na-glucose co-transport . Successful permeabilization of the basolateral membrane was obvious from the reversal of Isc under these conditions . Solutions were continuously gassed and stirred with 95% O2-5% CO2 and maintained at 37°C . Ag/AgCl reference electrodes were used to measure transepithelial voltage and pass current . Pulses ( 1 mV amplitude , 1 s duration ) were delivered every 90 s to monitor resistance . The voltage clamps were connectedto a PowerLab/8SP interface for data collection . CFTR was activated by adding 10 µM forskolin to the apical bathing solution .

Cystic fibrosis is a genetic disease that commonly affects people of European descent . The condition is caused by mutations in the gene encoding a protein called “cystic fibrosis transmembrane conductance regulator” ( or CFTR for short ) . CFTR forms a channel in the membrane of cells in the lungs that help transport salt across the membrane . Mutated versions of the protein are not as efficient at transporting salts , and eventually this damages the lung tissue . As the damage progresses , individuals become very vulnerable to bacterial infections that further damage the lungs and may eventually lead to death . One of the reasons CFTR mutations are harmful is that they cause the protein to fold up incorrectly and remain trapped inside the cell . Cells have quality control systems that recognize and destroy poorly folded proteins , and so only a few of the mutated CFTR proteins ever make it to the membrane to move salts . New therapies have been developed that improve folding of the protein and/or help the CFTR proteins that make it to the membrane work better . But more and better treatment options are needed . Hegde , Parashuraman et al . have now tested drugs that control how proteins fold and move to the membrane to see how they affect gene expression in cells with the most common cystic fibrosis-causing mutation . These drugs are known to improve the activity of the CFTR mutant , but do so too weakly to be of clinical interest . The experiments revealed that the expression of a few hundred genes was changed in response the drugs . Many of these genes were involved in major signalling pathways that control how CFTR is folded and trafficked within cells . Next , Hegde , Parashuraman et al . tested drugs that inhibit these signalling pathways to see if they improve salt handling in the mutated cells . The experiments demonstrated that these inhibitor drugs efficiently block the breakdown of misfolded CFTR , or boost the likelihood of CFTR making it to the membrane , helping improve salt trafficking in the cells . The inhibitors produced even better results when used in combination with a known CFTR-protecting drug . The results suggest that identifying and targeting signalling pathways involved in the folding , trafficking , and breakdown of CFTR may prove a promising way to treat cystic fibrosis .

2015

Unravelling druggable signalling networks that control F508del-CFTR proteostasis

Certain forms of translational regulation , and translation itself , rely on long-range interactions between proteins bound to the different ends of mRNAs . A widespread assumption is that such interactions occur only in cis , between the two ends of a single transcript . However , certain translational regulatory defects of the Drosophila oskar ( osk ) mRNA can be rescued in trans . We proposed that inter-transcript interactions , promoted by assembly of the mRNAs in particles , allow regulatory elements to act in trans . Here we confirm predictions of that model and show that disruption of PTB-dependent particle assembly inhibits rescue in trans . Communication between transcripts is not limited to different osk mRNAs , as regulation imposed by cis-acting elements embedded in the osk mRNA spreads to gurken mRNA . We conclude that community effects exist in translational regulation . Translation happens when a ribosome assembles productively on an mRNA . A number of molecular interactions influence the proper convergence of these parts , often involving RNA elements or factors at the opposite ends of the mRNA ( Jackson et al . , 2010 ) . For example , Poly ( A ) binding protein ( PABP ) binds to the poly ( A ) tail at the 3’ end of an mRNA ( Mangus et al . , 2003 ) and enhances translation through its interaction with eIF4G , a protein associated with the 5’ cap of the mRNA ( Tarun et al . , 1997; Sonenberg and Dever , 2003 ) . Similarly , many repressors of translation bind to elements in the 3’ untranslated regions ( 3’ UTRs ) of mRNAs and interfere with the normal action of initiation factors at the 5’ end of the mRNA ( Jackson et al . , 2010; Gebauer and Hentze , 2004 ) . Such long-range protein-protein interactions , spanning the full length of the mRNA , are generally assumed to occur only in cis: PABP bound to the tail of one molecule of mRNA is expected to bind only the eIF4G associated with the other end of the same mRNA . In principle , however , inter-transcript interactions are also possible , and their likelihood depends on the concentration of the mRNAs . When mRNAs are dilute , intra-transcript interactions will predominate , but as the local density of mRNAs increases , inter-transcript interactions become more likely . Previously , we described an example involving the Drosophila oskar ( osk ) mRNA , in which control elements on one transcript influenced translation of another transcript ( Reveal et al . , 2010 ) . osk mRNA is expressed during oogenesis , and is subject to multiple forms of post-transcriptional control . Following transcription in the nurse cells of each egg chamber , osk mRNA is transported through cytoplasmic connections to the transcriptionally-silent oocyte . Eventually , as oogenesis proceeds , osk mRNA localizes to the posterior pole of the oocyte , and only then does Osk protein begin to accumulate . Translation of osk mRNA is highly regulated: repression prevents premature translation from unlocalized mRNA , and activation turns on translation of the localized mRNA ( Kim-Ha et al . , 1995; Markussen et al . , 1995; Rongo et al . , 1995 ) . Both types of regulation rely on binding sites for Bruno ( Bru ) ( Reveal et al . , 2010; Kim-Ha et al . , 1995; Webster et al . , 1997; Reveal et al . , 2011 ) . These sites , BREs and others , reside in two clusters - the AB and C regions - near the opposite ends of the 3’ UTR ( Figure 1A ) . All the BREs contribute to translational repression ( Kim-Ha et al . , 1995 ) , and the C region BREs also play a role in translational activation ( Reveal et al . , 2010 ) . Strikingly , defects in either repression ( from mutation of all BREs in osk ABC BRE- ) or activation ( from mutation of the C BREs in osk C BRE- ) can be rescued by the presence of another osk mRNA , one that itself cannot make Osk protein but has all regulatory elements intact ( Reveal et al . , 2010 ) . 10 . 7554/eLife . 10965 . 003Figure 1 . Rescue of osk expression in trans is not limited to Bru-dependent regulation . ( A ) Diagram of the osk mRNA showing sites of mutations discussed in the text . The UTRs are shown as thick lines and the coding region as a rectangle . Because two different start codons are used , portions of the 5’ region can be both UTR and coding region . Not all the mutations at the 3’ end of the mRNA are shown . ( B ) Embryonic patterning assays to monitor osk activity . For the upper panel , the only source of osk mRNA was a genomic osk transgene with the indicated mutation . For the lower panel , osk54 mRNA was also present . Embryos from these mothers were were scored for cuticular patterning defects . Wild-type embryos have eight abdominal segments . Lower osk activity results in fewer abdominal segments . n values are the number of embryos scored . ( C ) Rescue in trans of the Osk expression defect caused by mutation 3’1004–1008 . At top is a diagram of the mRNA from the genomic osk transgene whose activity was monitored . All sequences are from osk , except for the inserted GFP . Below are images of stage 10 egg chambers ( left ) and the posterior ends of late stage egg chambers ( right ) . Scale bars , 25 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 10965 . 003 To explain rescue in trans , we proposed that long-range protein-protein interactions bridging the two ends of an mRNA could also bridge two different mRNAs , to enable 'regulation in trans' . A proposed mechanism for translational repression by Bru relies on such a long-range interaction . Bru binds Cup , a protein that competes with eIF4G for binding to eIF4E , the 5’ cap-binding protein . Cup engagement of eIF4E at the expense of the eIF4E/eIF4G interaction blocks translation initiation ( Nakamura et al . , 2004 ) . In our model , the Bru/Cup complex bound to the rescuing mRNA would bridge to eIF4E bound to the cap of the mRNA unable to bind Bru , thereby conferring translational repression . We hypothesized that assembly of osk mRNAs into ribonucleoprotein particles ( RNPs ) places the mRNAs in close proximity , thus increasing their concentration and facilitating inter-transcript interactions to allow 'regulation in trans' ( Reveal et al . , 2010 ) . If this hypothesis is correct , several features of 'regulation in trans' are expected . First , the rescuing ( donor ) mRNA should have the regulatory element that the rescued ( recipient ) mRNA lacks . Second , rescue in trans should not be Bru-specific , but should extend to other forms of regulation that also rely on long-range protein-protein interactions . Finally , rescue in trans would depend on the cis-acting osk mRNA elements and trans-acting proteins that mediate RNP assembly . Several participants in osk RNP assembly are known - Polypyrimidine tract binding protein ( PTB ) , Bru and an RNA dimerization motif - and all are candidates to support rescue in trans . PTB binds osk mRNA to form RNPs in vitro ( Besse et al . , 2009 ) . PTB also mediates association of osk mRNAs in vivo , as shown by the effects of loss of PTB on hitchhiking ( Besse et al . , 2009 ) , the phenomenon in which localization-competent osk mRNAs can assist other mRNAs in moving to the posterior pole of the oocyte ( Hachet and Ephrussi , 2004 ) . Similarly , Bru mediates in vitro oligomerization of RNAs with an unusually high density of Bru binding sites and could also oligomerize osk mRNA in vivo ( Chekulaeva et al . , 2006 ) . Finally , osk mRNA undergoes direct RNA dimerization , relying on a dimerization motif in the 3’ UTR ( Jambor et al . , 2011 ) . Elimination of one or more types of osk RNP assembly would be expected to block rescue in trans if our hypothesis is correct . Here , we show that all these expectations for the 'regulation in trans' model are met . Disruption of individual contributions to RNP assembly reveals that PTB has a critical role in rescue in trans , with direct RNA dimerization being much less important . Furthermore , we provide evidence that regulation imposed by embedded cis-acting elements in osk mRNA spreads to the another mRNA bound by PTB , the gurken ( grk ) mRNA , thus revealing a community effect in translation . Previously , we found that translational regulatory defects associated with mutation of Bru binding sites were rescued in trans . To determine if rescue in trans is a property unique to Bru-dependent regulation , or more general , we have now identified and tested additional regulatory mutants . The osk gene produces two Osk isoforms , Long Osk and Short Osk , from two translation start codons ( Figure 1A ) ( Markussen et al . , 1995 ) . The oskM1R∆121–150 mutant lacks the start codon for Long Osk and has a 30 nt deletion in the 5’ UTR for Short Osk ( Figure 1A ) . This mutant is largely defective in Short Osk protein production , in addition to making no Long Osk because of the M1R mutation ( Kanke and Macdonald , 2015 ) . Although coexpression with osk54 ( a protein null mutant with a short DNA insertion , introducing a stop codon at the end of the first exon ) ( Kim-Ha et al . , 1991 ) strongly rescues the activation defect of osk C BRE- ( Figure 1B ) , no significant rescue of the oskM1R∆121–150 activation defect was found ( Figure 1B ) . Therefore , the oskM1R∆121–150 mutant is not sensitive to rescue in trans . Mutations positioned close to the 3’ end of the osk mRNA also disrupt Osk expression . Mutants osk3’977–981 and osk3’984–988 ( numbering indicates position in the osk 3’ UTR ) have almost no osk patterning activity and are strongly defective in Osk protein production ( Ryu and Macdonald , 2015 ) . Mutant osk3’1004–1008 , in which an A-rich sequence is disrupted ( Kanke et al . , 2015 ) ( Figure 1A ) , also had strongly reduced osk patterning activity ( Figure 1B ) . Although these mutations are positioned close to Bru-binding sites in the osk 3’ UTR C region , they have little or no effect on Bru binding ( Kim-Ha et al . , 1995; Ryu and Macdonald , 2015 ) . Instead , the 977–980 and 984–988 mutations disrupt binding of BSF protein ( Ryu and Macdonald , 2015 ) , while the sequence affected by mutation 1004–1008 is a binding site for Poly ( A ) binding protein ( PABP ) ( Vazquez-Pianzola et al . , 2011 ) . The patterning defects of all three of these mutants were strongly rescued by coexpression of osk54 mRNA ( Figure 1B , [Ryu and Macdonald , 2015] ) . Similarly , a more direct assay of Osk protein expression , using an osk::GFP transgene bearing the osk3’1004–1008 mutation , showed rescue in trans ( Figure 1C ) . From analysis of these four additional regulatory mutants , we find a correlation ( albeit with few examples ) between the type of regulatory defect and its ability to be rescued in trans: regulatory elements likely to engage in long-range interactions ( i . e . those in the 3’ UTR ) are sensitive to rescue in trans , while a regulatory element positioned near the site of translational initiation and thus less likely to participate in a long-range interaction is not . Notably , by the model of 'regulation in trans' only the former type of regulatory element should be sensitive to rescue in trans , consistent with the results . Thus far , osk54 has been used as the donor for rescue in trans of osk mRNA regulatory defects . To determine if the ability to rescue is typical , or is an unusual property of osk54 , additional osk mutants unable to provide Osk patterning activity were tested . In oskM139L , the translation start codon for the Short Osk protein isoform is mutated . Because Short Osk is the isoform with embryonic patterning activity , oskM139L lacks that activity ( Vanzo and Ephrussi , 2002 ) . Like osk54 , oskM139L strongly rescued the translational activation defect of osk C BRE- and restored normal embryonic patterning ( Figure 2A ) . Likewise , oskM1R∆121–150 , the mutant whose regulatory defect is not rescued in trans , was an effective donor for rescue of the osk C BRE- mutant in trans ( Figure 2A ) . Rescue was also assayed by examination of Osk protein production , using the osk::GFP C BRE- mRNA as the recipient and monitoring GFP fluorescence ( this allows the mRNA producing the protein to be unambiguously identified , as the donor mRNAs do not include GFP ) . Both oskM139L and oskM1R∆121–150 restored Osk::GFP protein expression to wild-type levels , just as with the osk54 donor ( Figure 2C ) . 10 . 7554/eLife . 10965 . 004Figure 2 . Identification of sequences required in donor mRNAs for rescue in trans . ( A ) Embryonic patterning assays to monitor rescue of the translational activation defect of osk C BRE- . The osk C BRE- mRNA was expressed as the only osk mRNA , or in combination with donor osk mRNAs as indicated . ( B ) Levels of transgenic osk mRNAs . Values are relative to an osk+ transgene that fully rescues an osk mutant . Error bars indicate standard deviations . The osk C BRE- mRNA used in panel C is not included but has similar abundance ( Reveal et al . , 2010 ) . ( C ) Detection of GFP signal from osk::GFP C BRE- , expressed as the only osk mRNA or in combination with the donor osk mRNAs indicated . ( D ) Embryonic patterning assays to monitor rescue of the translational activation defect of osk C BRE- . The osk C BRE- mRNA was expressed as the only osk mRNA ( in the oskA87/osk0 background ) , together with a version of UAS-GFP expressed under GAL4 transcriptional control . Each UAS-GFP transgene contains osk 3’ UTR sequences as indicated ( positions of the AB and C regions are shown in Figure 1A ) . p values from student’s t tests were determined as described in 'Experimental Procedures' , with comparison sets highlighting the incremental increases in rescuing activity resulting from the presence of the C region together with additional parts of the 3’ UTR . Values for effect size and power were: AB vs C , 0 . 444 and 1 . 0; C vs AB+C , 0 . 448 and 1 . 0; AB+C vs osk 3’ UTR , 2 . 21 and 1 . 0 . Additional student’s t tests for comparison of GFP alone to GFP plus the C , AB+C and 3’ UTR parts all had p values of <10–10 . ( E ) Levels of mRNAs from D . Error bars indicate standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 10965 . 004 These results , and others ( below ) , show that a variety of osk mutants can act as donors for rescue in trans; this property is not unique to osk54 . Furthermore , rescue by oskM1R∆121–150 is consistent with 'regulation in trans' , as the oskM1R∆121–150 mRNA includes the regulatory element that the osk::GFP C BRE- mRNA lacks . The model for 'regulation in trans' predicts that the regulatory element missing from the recipient mRNA must be provided by the donor . This idea can be tested with osk::GFP C BRE- as the recipient , and osk C BRE- as the donor , as both carry the same regulatory mutation . Although the activation defect of osk::GFP C BRE- was strongly rescued by osk54 ( and by other donors with intact C region BREs; Figure 2C , below ) , little or no GFP signal was detected with osk C BRE- as the donor ( Figure 2C ) . Having shown that rescue of a regulatory defect in the recipient mRNA requires the cognate , wild type regulatory element in the donor mRNA , we next asked if the presence of that element was sufficient for rescue , or if other sequences ( e . g . those that might contribute to RNP assembly ) were also required . Portions of the osk mRNA 3’ UTR were added to a UAS-GFP transgene , and expressed using the UAS/GAL4 system ( Brand and Perrimon , 1993 ) . The osk 3’ UTR AB and C regions contain the 5’ and 3’ clusters of Bru binding sites , respectively ( Figure 1A ) . These regions were tested alone , or in combination . Also tested was the entire osk 3’ UTR . The UAS-GFP donor did not rescue the activation defect of osk C BRE- ( Figure 2D , compare to Figure 2A ) . Similarly , the donor with the osk 3’ UTR AB region provided no rescue . A donor carrying the osk 3’ UTR C region , and thus the activation element mutated in the osk C BRE- recipient , provided a low level of rescue . Rescue was increased when both the AB and C regions were present , and the highest level of rescue was obtained when the GFP mRNA included the entire osk 3’ UTR . The different levels of rescue cannot be attributed to differences in the amounts of the various donor mRNAs ( Figure 2E ) . The results of these experiments are consistent with 'regulation in trans' . The C region must be present to provide the missing translational activation element , while other Bru binding sites that do not contribute to activation ( the AB region sites ) are not sufficient . The weak rescue from the C region alone is consistent with the notion that rescue requires both the missing element and sequences which mediate assembly of osk RNPs . We suggest that by itself the C region supports weak association of GFP-C mRNAs with osk C BRE- transcripts , and this association is strengthened moderately by addition of the AB region and more strongly by addition of the full 3’ UTR . To ask if RNP assembly is required for rescue in trans , we wished to inhibit or abolish each of the known forms of RNP assembly . Removing assembly factors - PTB and Bru - is not a good approach , as these proteins bind to many mRNAs and indirect effects on osk mRNA regulation are possible . In a more selective approach , the osk mRNA is modified , mutating elements involved in RNP assembly . With donor osk transgenes defective in one or more types of RNP assembly , any change in the ability to rescue the activation defect of the osk::GFP C BRE- recipient mRNA can be monitored . Testing the role of Bru-mediated osk mRNA oligomerization ( Chekulaeva et al . , 2006 ) is problematic , as the Bru sites are also required to provide the missing regulatory element for osk C BRE- ( above ) . However , the other contributions to assembly - mRNA dimerization and PTB-dependent RNP assembly - can be manipulated ( Figure 3A ) . Direct osk mRNA dimerization requires a RNA element ( Jambor et al . , 2011 ) , which is deleted in mutant osk3’∆665–685 ( labeled as dimer- in Figure 3 ) . Testing the role of PTB-mediated assembly could be done by mutating PTB-binding sites . These sites are spread throughout the osk mRNA 3’ UTR ( as shown by RNA binding assays ) , although individual sites have not been defined ( Besse et al . , 2009 ) . However , because the binding specificity of Drosophila PTB is likely to be very similar or identical to that of the highly conserved mammalian homolog , for which binding sites have been characterized by a combination of approaches ( Pérez et al . , 1997; Xue et al . , 2009 ) , candidate sites can be identified ( Figure 3—figure supplement 1 ) . Mutating all the candidate sites ( consisting of pyrimidine tracts or PTs ) is the simplest strategy to disrupt PTB binding , but introducing so many mutations increases the probability of an inadvertent effect on rescue in trans . To strike a balance between these design considerations , a subset of the PT sequences was mutated ( osk3’PTmut ) ( Figure 3—figure supplement 1 ) , focusing on those with features associated with the strongest sites for mammalian PTB binding , while avoiding mutations that might also disrupt osk mRNA dimerization or Bru binding . A UV crosslinking competition binding assay was used to determine if the PT mutations affected PTB binding ( Figure 3B ) . PTB bound the radiolabeled osk mRNA 3’ UTR probe mRNA ( left lane ) . Increasing amounts of unlabeled osk 3’ UTR RNA ( osk+ ) effectively competed for binding . When the competitor RNA included the PT mutations ( 3’PTmut ) , competition was less effective , demonstrating that the mutations do affect PTB-binding sites . 10 . 7554/eLife . 10965 . 005Figure 3 . Mutation of pyrimidine tracts in the osk mRNA 3’ UTR disrupts rescue in trans . ( A ) Diagram of osk mRNAs used in the assays . The dimer- mRNA has a deletion of positions 665–685 of the 3’ UTR . The 3’PTmut mRNA has multiple mutations ( Figure 3—figure supplement 1 ) , indicated figuratively by the red dots . Each donor osk transgene for rescue has the M139L mutation to prevent translation of the Short Osk isoform . Only Short Osk has patterning activity ( Vanzo and Ephrussi , 2002 ) , and we eliminated this isoform to address the possibility that some mutations might cause inappropriate Osk expression and dominant maternal-effect lethality from excess osk patterning activity . ( B ) PT mutations disrupt PTB binding . UV crosslinking assay of MBP-PTB binding to the osk 3’ UTR . The left lane shows binding to the radiolabeled osk 3’ UTR RNA in the absence of competitor RNA . For the remaining lanes , increasing amounts ( 3 , 10 , 30x molar excess ) of unlableled competitor RNAs were included in the binding assays . The competitor RNAs were the wild type osk 3’ UTR ( osk+ ) , the osk 3’ UTR with the PT mutations included in the osk 3’PTmut transgene ( 3’PTmut ) , or a nonspecific control RNA corresponding to a portion of the bicoid mRNA 3’ UTR ( segment 3R: 6756590–6756463 , r6 . 08 ) with no strong predicted PTB binding sites ( Control ) . ( C ) Examples of oocytes expressing osk::GFP C BRE- alone ( and thus strongly defective in translational activation ) , or in combination with a donor osk transgene as indicated . Late stage oocytes were used because the translational activation defect caused by the C BRE- mutations is not fully penetrant at earlier stages . Although only a small fraction of stage 9/10 osk C BRE-oocytes have detectable Osk protein ( Reveal et al . , 2010 ) , we wanted to remove this variability for experiments in which there could be small differences between different genotypes . Note that the imaging conditions used to provide these examples of different levels of rescue are not the same as those used for quantitation in panel E . The conditions for these images were chosen to highlight trace levels of GFP , and as a consequence the signal in some samples is saturated . ( D ) Levels of donor mRNAs . Values are in comparison to a rescuing osk transgene as in Figure 2 . Error bars indicate standard deviations . ( E ) Posterior GFP fluorescence from osk::GFP C BRE- when expressed in combination with donor osk mRNAs bearing the indicated mutations . Each dot represents one late stage oocyte , with averages indicated by horizontal lines . All samples were from flies grown , fixed , processed and imaged in parallel ( see 'Experimental Procedures' and Figure 3—figure supplements 2 and 3 for details of quantitation ) . The student’s t test was used to test for significance of differences relative to the osk+ donor mRNA ( left ) , with the p values shown above . DOI: http://dx . doi . org/10 . 7554/eLife . 10965 . 00510 . 7554/eLife . 10965 . 006Figure 3—figure supplement 1 . Predicted PTB-binding sites in the osk mRNA 3’ UTR . The complete osk 3’ UTR sequence is shown , with runs of 4 or more mixed pyrimidines ( Pyrimidine Tracts , or PTs ) highlighted with blue shading . The mutations included in the osk3’PTmut transgene are indicated in red . The subsets of these mutations included in the derivative transgenes are shown using overlying brackets , with each mutant named to the left of the bracket . DOI: http://dx . doi . org/10 . 7554/eLife . 10965 . 00610 . 7554/eLife . 10965 . 007Figure 3—figure supplement 2 . Quantitation of fluorescence at the posterior pole of oocytes . A and B . Measuring the amounts of Osk::GFP fluorescence in different z positions at the posterior pole of an oocyte to determine the variability introduced by selecting a single focal plane for analysis . A series of images were captured using the Series function of the Leica SP software , and sums of posterior fluorescence were determined with FIJI ( 'Experimental Procedures' ) and plotted with Kaleidagraph ( below the images ) . Deviations from the maximal values are shown ( above the images ) . For both A and B , the z position selected visually as being brightest had an intensity less than 2% different than the true maximum . Thus , there is only a small potential error from choice of focal plane for imaging , with the magnitude being much less than the differences observed in Figures 3–4 . DOI: http://dx . doi . org/10 . 7554/eLife . 10965 . 00710 . 7554/eLife . 10965 . 008Figure 3—figure supplement 3 . Consistency of rescue of Osk::GFP expression . Measuring the amounts of Osk::GFP fluorescence at the posterior pole of different oocytes of the same genotype . Samples from a single experiment ( with osk::GFP C BRE- in combination with an osk+ donor ) were mounted on two slides . Set 1 images were from one slide , and Set 2 images from the other slide . The student’s t test was used to test for significance of differences between the two data sets , with the p value shown above . The results are statistically indistinguishable , indicating that sampling errors are unlikely to explain to differences observed for different genotypes in Figures 3–4 . DOI: http://dx . doi . org/10 . 7554/eLife . 10965 . 00810 . 7554/eLife . 10965 . 009Figure 3—figure supplement 4 . Qualitative differences in weak rescue of osk::GFP C BRE- expression . Examples of posterior GFP signal from osk::GFP C BRE- recipient mRNA in combination with the donor mRNAs indicated at left . The examples show the tract levels of posterior GFP sometimes found with the 3’PTmut donor , but never seen with the 3’PTmut dimer-donor . DOI: http://dx . doi . org/10 . 7554/eLife . 10965 . 00910 . 7554/eLife . 10965 . 010Figure 3—figure supplement 5 . PT and dimer-mutations in the osk mRNA 3’ UTR do not disrupt posterior localization . Transgenes with the indicated mutations were expressed in flies lacking any other osk mRNA , and the transgenic mRNAs were detected by in situ hybridization . In all cases , the transgenic mRNAs appear in a crescent at the posterior pole of the oocyte . DOI: http://dx . doi . org/10 . 7554/eLife . 10965 . 010 We coexpressed different donor mRNAs ( all present at similar levels; Figure 3D ) together with the activation-defective osk::GFP C BRE- recipient mRNA , and measured posterior GFP levels in late stage oocytes to determine how effectively the activation defect was rescued ( Figure 3C ) . The wild-type donor provides strong rescue of activation , as does the donor lacking the dimerization motif ( dimer- ) ( Figure 3E ) . By contrast , the osk3’PTmut and osk3’PTmut dimer- donors provided almost no rescue of activation . Although no statistically significant difference between GFP levels with these latter two donors was detected , some oocytes had trace levels with the osk3’PTmut donor , while none were above background with the osk3’PTmut dimer- donor , consistent with an additive effect of the two types of mutations ( Figure 3E ) . Examples of the trace posterior GFP signals with the osk3’PTmut donor , and their absence with the osk3’PTmut dimer- donor , are shown in Figure 3—figure supplement 4 . The loss of rescue in trans cannot be attributed to failure of the donor mRNAs to localize to the posterior pole of the oocyte , as the PT mutations do not interfere with posterior localization ( Figure 3—figure supplement 5 ) . Mutation of candidate PTB-binding sites in the osk mRNA clearly disrupts rescue in trans . Loss of PTB-dependent RNP assembly is one explanation of the results , and is consistent with 'regulation in trans' . Alternatively , inadvertent mutation of the binding site of some other factor could make the osk3’PTmut mRNA a poor donor . To distinguish between these options two approaches were used . For one approach , we reasoned that if osk3’PTmut failed to rescue because of reduced PTB binding , this would likely be an additive defect that depends on mutation of multiple different PTs . Thus , mutating only subsets of the PTs might yield intermediate levels of rescue . By contrast , if loss of rescue was due to inadvertent mutation of a binding site for another factor , then , assuming that this other factor does not have the same binding specificity as PTB , this effect would likely be due to a specific PT mutation . If so , we would expect that only a single subset of PT mutations would disrupt rescue in trans . The mutations in osk3’PTmut were subdivided into four groups ( Figure 4A , Figure 3—figure supplement 1 ) . To confirm that each subset of PT mutations reduced PTB binding , short segments of the osk mRNA 3’ UTR containing the subsets of mutations were tested in UV crosslinking assays . Each of the four 3’ UTR segments bound PTB , and for each the binding was substantially lower when the RNA contained the PT mutations ( Figure 4B ) . 10 . 7554/eLife . 10965 . 011Figure 4 . Evidence that PTB is required for rescue in trans . ( A ) Diagram of the osk mRNA 3’ UTR . Above the black line ( the 3’ UTR ) are indicated candidate PTB-binding sites ( blue ) , consisting of any tract of mixed pyrimidines at least 4 nt in length . Mutations within pyrimidine tracts ( PTs ) that are present in the osk 3’PTmut transgene are indicated in red . Subsets of PT mutations , incorporated into genomic osk transgenes used in panel B , are indicated above ( numbers correspond to positions in the osk 3’ UTR ) . Below the black line , the Bru sites and position of the stem-loop structure containing the RNA dimerization motif are indicated . ( B ) Subsets of PT mutations disrupt PTB binding to short segments of the osk mRNA 3’ UTR . Short segments of the osk 3’ UTR , each encompassing a subset of PT sites noted in panel A , were used as RNA substrates for UV crosslinking assays with MBP-PTB . The extent of each RNA is indicated above ( numbers correspond to positions in the osk 3’ UTR ) . Radiolabeled RNAs were prepared in wild type form ( + ) , or with the PT mutations from that region ( i . e . the same mutations as indicated in red in A ) ( mut ) . ( C ) Subsets of PT mutations disrupt PTB binding to the complete osk mRNA 3’ UTR . UV crosslinking assays with unlabeled competitor RNAs were performed as in Figure 3E , again using the complete osk 3’ UTR as a radiolabeled-binding substrate . The competitor RNAs are indicated above , and consist of the complete osk 3’ UTR with the designated PT mutations . The bar graph below shows the level of residual binding to the radiolabeled substrate RNA at the highest level of competitor ( 30x molar excess ) . Smaller values indicate stronger binding of the competitor RNA . Assays were performed three times , and the error bars indicate standard deviations . The 375–540 PTmut RNA was a significantly less effective competitor than the other RNAs with subsets of mutations ( p<0 . 05 for all ) . ( D ) . Posterior GFP fluorescence from osk::GFP C BRE- when expressed in combination with donor osk mRNAs bearing the indicated mutations . Data are presented as in Figure 3 . The student’s t test was used to test for significance of differences relative to the osk+ donor mRNA ( left ) , with the p values shown above . Additional statistical tests evaluated effect size ( Cohen’s d ) and power . Effect size and power , respectively , for the different donors relative to osk+ were: 3’PTmut , 5 . 01 and 1 . 0; 3’159–300 PTmut , 1 . 53 and >0 . 99; 3’375–540 PTmut , 4 . 12 and 1 . 0; 3’746–789 PTmut , 0 . 58 and 0 . 56; 3’853–923 PTmut , 0 . 88 and 0 . 91 . The statistical significance of the reduced rescuing activity of the 3’746–789 PTmut donor is less compelling than for the other mutants , but this does not invalidate the overall conclusion that the PT mutations have an additive effect on loss of rescuing activity . ( E ) Levels of donor mRNAs . Values are in comparison to a rescuing osk transgene as in Figure 2 . Error bars indicate standard deviations . ( F ) Posterior GFP fluorescence from osk::GFP ( left ) or osk::GFP C BRE- ( right ) when expressed in combination with osk+ . Flies were either heph+ or heterozygous for heph1545 , as indicated below . The student’s t test was used to test for significance of differences between the two samples in each panel , with the p values shown above . Additional statistical tests evaluated effect size ( Cohen’s d ) and power . Effect size and power , respectively , were 0 . 169 and 0 . 134 ( left panel ) and 1 . 12 and >0 . 99 ( right panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10965 . 011 The same subsets of mutations were also introduced into osk donor transgenes ( Figure 4A , Figure 3—figure supplement 1 ) . Just as for the osk3’PTmut mRNA , all the new osk donor transgenes were properly localized to the posterior pole of the oocyte ( Figure 3—figure supplement 5 ) . Each of these donors was impaired for rescue of the translational activation defect of the osk::GFP C BRE- recipient , although none was as defective as the osk3’PTmut donor ( Figure 4D ) . Thus , the PT mutations had an additive effect , and some subsets of the mutations inhibited rescue in trans more than others . The differences in rescuing activity did not correlate with the amounts of the different donor mRNAs ( Figure 4E ) . To determine if reduced donor activity displayed a correlation with reduced PTB binding activity , PTB binding was measured in competition-binding assays ( Figure 4C ) . Binding of PTB to a radiolabeled wild type osk 3’ UTR RNA was performed in the presence of increasing amounts of unlabeled osk 3’ UTR competitor RNAs . Competitors with subsets of PT mutations had intermediate levels of competition , not as strong as the wild type osk 3’ UTR but stronger than the osk 3’ UTR with all PT mutations . Notably , among this group , the osk 3’ UTR RNA with the 3’375–540 subset of PT mutations was the weakest competitor , just as the osk3’375–540 PTmut transgene was the least effective donor for rescue in trans . Thus , we did observe a correlation between strength of PTB binding and strength of rescue in trans . For the other three subsets of PT mutations , their effects on PTB binding were too similar to one another to draw clear distinctions among them ( Figure 4C ) , much like the effects of these mutations on rescue in trans ( Figure 4D ) . As a second approach to ask if PTB is required for rescue in trans , no PTB binding sites in either donor or recipient osk mRNAs were mutated , but instead the level of PTB protein was reduced . The osk+ donor and osk::GFP C BRE- recipient mRNAs were coexpressed in flies either wild type or heterozygous mutant for heph ( heph encodes PTB ) ( Figure 4F , right ) . Reducing the level of PTB activity in the heph heterozygotes impaired rescue of the translation activation defect , causing a lower level of Osk::GFP . By contrast , in a control experiment using a recipient mRNA with no translation defect , osk::GFP , reducing the level of PTB activity had no effect on Osk::GFP levels ( Figure 4F , left ) . Thus , the results of both kinds of tests to determine the role of PTB - mutating candidate PTB-binding sites or reducing PTB activity - point to the same conclusion: PTB-dependent assembly of osk mRNA into RNPs is required for rescue in trans , supporting the model of 'regulation in trans' . If assembly of osk transcripts in RNPs promotes 'regulation in trans' , would translation of other mRNAs present in the same RNPs also be influenced by the translational control elements in the osk mRNA ? Candidates for such trans regulation are expected to share two features with osk mRNA: reliance on a shared factor for RNP assembly , and overlap in subcellular distribution . Because PTB is most strongly implicated in rescue of osk mRNA regulatory defects in trans , we focused on mRNAs with candidate PTB-binding sites . Among ovarian mRNAs , one of the most highly enriched for candidate PTB-binding sites is gurken ( grk ) ( Supplementary file 1 ) . Furthermore , PTB associates with grk mRNA , both in vitro and in vivo ( McDermott et al . , 2012; McDermott and Davis , 2013 ) . Notably , grk mRNA is , like osk , highly enriched in the oocyte during previtellogenic stages ( Figure 5A , B ) ( Neuman-Silberberg and Schüpbach , 1993 ) . Later , at stage 8 , both mRNAs align at the anterior margin of the oocyte , but then go their separate ways within the oocyte: grk becomes restricted to the anterodorsal corner , while osk moves to the posterior ( Kim-Ha et al . , 1991; Neuman-Silberberg and Schüpbach , 1993; Ephrussi et al . , 1991 ) . Thus , grk and osk mRNAs have similar distributions for part of oogenesis and have the potential to reside in the same RNPs during that period . Consequently , each mRNA could influence the translation of the other . Because translation of osk mRNA is strongly repressed for the entire period of shared localization , the most likely trans effect is for osk mRNA to confer some degree of translational repression on grk mRNA . 10 . 7554/eLife . 10965 . 012Figure 5 . osk mRNA influences the level of Grk protein . A and B . Distribution of osk ( A ) and grk ( B ) mRNAs in egg chambers early in oogenesis . mRNAs were detected by in situ hybridization . For each egg chamber , the cell at right with intense signal is the oocyte . ( C ) Distribution of Grk protein at stage 6 of oogenesis . C shows both Grk ( green ) and nuclei ( red ) , and C’ shows only Grk ( white ) . Grk is most strongly expressed in the oocyte , and is excluded from the oocyte nucleus . ( D ) Experimental design . Each rectangle represents an oocyte , showing the grk and osk mRNAs present . Both have multiple PTs , indicated by the yellow dashes . The osk 3’PTmut mRNA has many PTs mutated , and thus fewer yellow dashes . ( E and F ) Immunodetection of Osk protein , showing no detectable precocious expression of Osk protein from either the control osk mRNA ( E ) or the osk3’PTmut mRNA ( F ) . Nuclei are red . ( G ) Levels of donor mRNAs . Values are in comparison to a rescuing osk transgene as in Figure 2 . Error bars indicate standard deviations . ( H ) Levels of Grk protein in oocytes when expressed in the presence of both wild type and control osk mRNAs ( left ) or only the control osk mRNA ( right ) . The control osk mRNA is from the osk 11-13- transgene ( Figure 6 ) which has no defect in Osk expression or rescue in trans . See 'Experimental Procedures' and Figure 5—figure supplement 1 for details of quantitation . ( I ) Levels of Grk protein in oocytes when expressed in the presence of osk mRNAs as indicated at bottom . In the left panel , osk 3’PTmut line 1 was used , and in the right panel osk 3’PTmut line 2 was used . Additional statistical tests evaluated effect size ( Cohen’s d ) and power , respectively: 1 . 51 and 0 . 94 ( left panel ) and 1 . 91 and >0 . 99 ( right panel ) . For comparison , the values for the control experiment in panel H were 0 . 45 ( effect size ) and 0 . 22 ( power ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10965 . 01210 . 7554/eLife . 10965 . 013Figure 5—figure supplement 1 . Quantitation of fluorescence in Stage 5/6 oocytes . Measuring the amounts of Grk immunofluorescence in different z positions within an oocyte to determine the variability introduced by selecting a single focal plane for analysis . A series of images were captured using the Series function of the Leica SP software , covering the region identified visually as having the maximum brightness , and average fluorescence levels were determined with FIJI ( 'Experimental Procedures' ) and plotted with Kaleidagraph ( below the images ) . Variation in levels across the entire z series was less than 5% , much less than the differences observed in Figure 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 10965 . 013 To address the possibility that osk transcripts influence grk mRNA translation , we monitored Grk protein levels in the oocytes of previtellogenic stage 5/6 egg chambers ( Figure 5C ) . If wild-type osk mRNAs are present ( Figure 5D upper ) , then 'regulation in trans' is possible . If osk3’ PTmut mRNA is the only osk mRNA present ( Figure 5D lower; loss of PTs is indicated by fewer yellow dashes ) , then any PTB-dependent association of osk and grk mRNAs should be impaired , and 'regulation in trans' could be reduced . Remarkably , the type of osk mRNA present did affect Grk protein levels . When only osk3’PTmut mRNA was present , Grk protein levels were higher than when osk+ mRNA was also present ( Figure 5I ) . This effect was observed in each of two separate experiments , making use of different lines of the osk3’PTmut transgene . As a control , we performed the same type of experiment with an osk transgene that is a strong donor for rescue of osk mRNA regulatory defects and found no change in Grk protein levels ( Figure 5H ) . The differing behaviors of the osk3’PTmut and osk control transgenes cannot be attributed to differences in mRNA levels ( Figure 5G ) . Importantly , none of the osk mRNAs in these experiments produce any Osk protein at these stages of oogenesis ( Figure 5E , F ) . Therefore , we conclude that osk mRNA , not Osk protein , influences Grk expression . The increase in Grk levels when osk mRNA is impaired in PTB-dependent RNP assembly is consistent with the strong translational repression of osk mRNAs acting in trans to confer some degree of translational repression on grk mRNAs present in the same RNPs . As described above , we found that each of several osk mutants with defects in 3’ UTR regulatory elements could be rescued in trans , consistent with the notion that rescue relies on long-range interactions that bridge between donor and recipient mRNAs . However , the osk IBE-mutant is an exception to this rule . The IBE sequence - UUUAY ( Y is pyrimidine ) - was identified as a binding site for Imp protein ( Munro et al . , 2006 ) . Thirteen copies of the IBE appear in the osk mRNA , all in the 3' UTR . The osk IBEs were mutated in subsets to test their role . Three of the mutants ( subsets A , C and D; Figure 6C ) fail to accumulate Osk despite having normal localization of osk mRNA , while the fourth mutant ( subset B ) has normal Osk expression . Mutating the A , C , or D subsets of IBEs disrupts colocalization of Imp protein with osk mRNA at the posterior pole of stage 9/10 oocytes , supporting the notion that Imp binds osk mRNA through these elements and mediates activation of osk mRNA translation ( Munro et al . , 2006 ) . However , mutants lacking Imp protein have normal osk expression , suggesting that the IBEs also bind another factor , with this second factor contributing to activation of osk translation ( Munro et al . , 2006; Geng and Macdonald , 2006 ) . 10 . 7554/eLife . 10965 . 014Figure 6 . IBE mutations disrupt rescue in trans and have gain-of-function properties . ( A ) Embryonic patterning assays to monitor rescue of the translational activation defect of osk C BRE- . ( B ) Levels of transgenic osk mRNAs . Values are relative to an osk+ transgene that fully rescues an osk mutant . Error bars indicate standard deviations . ( C ) Absence of Osk expression from the osk IBE- transgene ( left ) ( stage 9 egg chamber ) and positions of the IBEs in the osk 3’ UTR [diagram adapted from Figure 6A of ( Munro et al . , 2006 ) ] . The osk IBE- transgene used has the A subset of mutations , and makes no detectable Osk protein ( the same phenotype as osk mutants with the C or D subsets of IBE- mutations ) ( Munro et al . , 2006 ) . ( D ) Mutation and phenotype of IBEs 11–13 . Left , Osk expression from the osk 11-13- transgene mRNA ( stage 9 egg chamber ) . The osk 11-13- transgene has an HA epitope tag , which was used for immunodetection ( an osk+ transgene with the same tag is expressed at similar levels; not shown ) . Right , sequences of the osk 3’ UTR region containing IBEs 11–13 . Mutated bases are in red . The IBE- mutations are those used by Munro et al . ( 2006 ) . ( E ) Mutation and phenotype of IBEs 1–3 . Left , Osk expression from the osk3’∆83–116 transgene mRNA ( numbering indicates position in the osk mRNA 3’ UTR ) ( stage 9 egg chamber ) . Right , sequences of the osk 3’ UTR region containing IBEs 1–3 . Mutated bases are in red , and the region deleted is indicated by underscoring . The IBE- mutations are those used by Munro et al . ( 2006 ) . ( F ) Osk::GFP expression in stage 10 egg chambers ( posterior portion only ) from females expressing osk::GFP C BRE- in combination with the donor osk mRNAs indicated . Fluorescence intensities were quantitated for the graph at right . DOI: http://dx . doi . org/10 . 7554/eLife . 10965 . 014 Previously , we found that the osk IBE- mutant was not strongly rescued in trans ( Reveal et al . , 2010 ) . Now , we asked if the osk IBE- mutant can act as a donor for rescue in trans . Neither of two transgenes with mutated IBEs strongly rescued the translational activation defect of the osk C BRE-mutant ( Figure 6A ) . Similarly , neither IBE mutant restored protein expression of the osk::GFP C BRE- mutant ( Figure 6F , data not shown ) . Failure to rescue strongly was not due to insufficient mRNA , as both IBE mutant osk mRNAs were present at high levels ( Figure 6B ) . Thus , mutation of the IBEs disrupted rescue in trans , with the mutant osk mRNAs unable to act as either donor or recipient . Several features of the IBE mutants , notably their inability to participate in rescue in trans , as well as a report that Imp protein has only weak affinity for osk mRNA ( Geng and Macdonald , 2006 ) , prompted further analysis of the IBEs . We made two new mutants ( Figure 6D , E ) , each affecting a subset of IBEs implicated in activation of osk translation ( Munro et al . , 2006 ) . In mutant osk 11-13- , the 3’-most IBEs ( subset D ) were altered . Instead of mutating the central position in the UUUAY pentamer from U to G ( as in the original osk IBE-mutants ) , each nucleotide of each pentamer was changed , and so a stronger defect in binding - to Imp or to the unknown factor - is expected . Notably , the osk 11-13- mutant had normal Osk accumulation ( Figure 6D ) . For mutant osk 3'∆83–116 , the entire region encompassing the 5’-most IBEs ( subset A ) was deleted . Again , this mutant is expected to more strongly disrupt Imp/unknown binding than the corresponding osk IBE- mutant , but it had normal Osk accumulation ( Figure 6E ) . Thus , the original IBE mutations appear to disrupt translational activation of osk mRNA not by removing a binding site , but by creating a novel type of element , such as a binding site for another factor . The new mutants lacking subsets of IBEs were also tested for the ability to act as donors for rescue in trans . The activation-defective osk:GFP C BRE- was used as the recipient , and we monitored GFP to determine if activation was rescued . The original osk IBE- mutant did not substantially rescue translation of the recipient mRNA . By contrast , both osk 3'∆83–116 and osk 11-13- mutants strongly rescued translational activation ( Figure 6F ) . It seems likely , then , that the gain-of-function property of the original IBE- mutations not only disrupts osk mRNA translation , but also interferes with the ability of the mRNA to participate in rescue in trans . Female flies of the genotype to be tested , together with sibling males , were aged in well-yeasted vials for 3–4 days after eclosion , and then transferred to small population cages for egg laying on yeasted apple juice plates . The plates were changed at intervals ( ~12–24 hr ) , the embryos aged for a further 20–24 hr , and cuticles were prepared for mounting in Hoyer’s medium ( Wieschaus et al . , 1986 ) . A wild-type embryo has eight abdominal ventral denticle belts . When osk activity is reduced , fewer denticle belts form with the number of belts proportional to the level of osk activity . Conversely , if there is excess osk activity , anterior structures are lost . For the transgenes studied here , only reduced osk activity was detected . Each embryo was scored for the number of complete abdominal denticle belts . The results presented in Figs . 1B , 2A and 6A represent the embryos from three or more sequential embryo collections . For the results of Figure 2D , the results from two separate experiments ( i . e . independent sets of crosses to generate the flies ) were combined to increase sample sizes . RNA levels were determined by RNase protection assays , using rp49 for normalization of sample amounts ( Reveal et al . , 2010 ) . At least three assays were done for each mRNA . RNA distributions were detected by in situ hybridization ( Jambor et al . , 2011 ) . Proteins were imaged in whole mount samples by confocal microscopy with a Leica TCS-SP ( Reveal et al . , 2010; Kim-Ha et al . , 1995 ) . Antibodies for immunodetection: rabbit anti-Osk ( Reveal et al . , 2010 ) , 1/3000; mouse anti-HA ( Covance ) , 1/300; mouse anti-Grk 1D12 [Developmental Studies Hybridoma Bank ( DSHB ) ] , 1/10; and mouse anti-Hts 1B1 ( DSHB ) , 1/1 . TO-PRO-3 Iodide ( Invitrogen ) ( 1:1 , 000 ) was used to stain nuclei . For immunodetection , and for detection of GFP from fusion proteins , samples in comparison groups ( corresponding to sets of data in outlined boxes in Figures 3–5 ) were from flies grown in parallel and aged in well-yeasted vials . They were dissected , fixed , and stained in parallel , and imaged in a single session using identical confocal settings in which the fluorescence signals were not saturated . For quantitation of posterior Osk::GFP in late stage oocytes , we first determined how reliably an optical section could be visually identified as having the maximum signal intensity . Note that all late stage oocytes , having a high anterior/posterior to dorsal/ventral axial ratio , are positioned with the anterior/posterior axis parallel to the plane of the slide . This orientation ensures that a z series of images is comparable for all oocytes . A z position appearing to have the highest level of Osk::GFP signal was noted , and then a z series of images roughly centered on that position was collected . Signal intensities were determined ( as described below ) , and plotted across the z dimension . As shown in the two oocyte examples in Figure 3—figure supplement 2 , this approach was adequate to identify a z position having a signal intensity within 2% of the maximum measured intensity . To measure signal intensity of the posterior crescent of Osk::GFP , the posterior peripheral region of each oocyte was traced in FIJI using a Wacom Intuos tablet . Because oocytes differ in shape , a single standardized trace outline could not be used . When the Osk::GFP signal intensity was strong , the region to be traced was readily identified , with the boundaries of the traces extending well beyond the anterior-most signal . When Osk::GFP signal intensity was low or undetectable , a region similar in shape and size was traced . Within the traced region for each oocyte , the total signal was measured . Other options for quantitation of the posterior crescent of Osk::GFP are to determine average or maximum signal intensity . Neither provides a reliable estimate of total Osk::GFP . For an average intensity to be reliable , the area scored must be identical for all oocytes . Given the variation in oocyte shape , tracing areas identical in size is not practical . For a maximum signal intensity to be reliable , there must be a point source of Osk::GFP . This is not the case . To evaluate the consistency of the assay , a single biological sample was divided into two portion , and each was mounted on a separate slide . Both slides were imaged in a single session . Quantitation using the approach described showed a high level of consistency ( Figure 3—figure supplement 3 ) . Quantitation of Grk protein in early oocytes presents different challenges and a different approach was used . Within early oocytes Grk protein is roughly uniform in distribution , except that the protein is excluded from the nucleus . Because all samples in these experiments contain readily detectable Grk , the position of the nucleus ( unstained ) is obvious and the cytoplasm can be easily traced . Given the uniform distribution of Grk in the cytoplasm , measuring an average signal intensity is appropriate . By contrast , measuring the total amount of signal does not provide a reliable estimate of Grk levels , as the area of the oocyte varies substantially depending on z position . For these images , additional steps were taken to obtain similar data from all samples . First , only egg chambers oriented in a roughly horizontal position were analyzed . This ensures that the position of the nucleus ( lacking Grk ) , and the region of the oocyte it occupies , are similar for all samples . Second , the focal plane was chosen to include the oocyte nucleus , thus avoiding the peripheral regions of the oocyte . Near the periphery , only a small portion of the oocyte is included in an optical section , which causes high variability in average signal intensity . Limiting the analysis to the more central region yielded only minor variation in measured signal intensities ( Figure 5—figure supplement 1 ) , much less than the differences observed in the experiments of Figure 5 . A maltose-binding protein-PTB fusion protein ( Besse et al . , 2009 ) was expressed in E . coli BL21/pLysS and affinity purified on amylose resin ( New England Biolabs ) according to the manufacturers protocol . UV crosslinking assays were performed as described ( Macdonald et al . , 1995 ) . Radiolabeled RNAs were synthesized by in vitro transcription ( MaxiScript; Ambion ) with alpha 32-P UTP from plasmid templates containing either the full osk mRNA 3’ UTR ( Figure 3 and Figure 4C ) , or from plasmid templates containing short segments of the osk 3’ UTR ( Figure 4B ) . Competition-binding experiments were performed as described ( Macdonald et al . , 1995 ) . For all assays , phosphorimaging of dried gels was used to visualize and quantitate binding . All assays were performed at least three times , with no significant variation in results . The patterning activity and imaging assays can be divided into two groups: those with dramatically different outcomes and those with graded differences which can be small . For the first group , such as the patterning assays in Figure 2A and the imaging results in Figure 2C , no statistical analysis is included . For the second group , statistical tests were used to evaluate the significance of any observed differences . For comparisons between two samples , the student’s t test was used , with p values provided in each figure . Determinations of effect size and power ( post hoc calculation ) were performed using G*Power ( Faul et al . , 2007; Faul et al . , 2009 ) . To facilitate comparisons in patterning assays of Figure 2D , embryos in different phenotypic classes were assigned numbers of abdominal denticle bands as follows: 6–7 , 6 . 5; 3–5 , 4; and 0–2 , 1 ( embryos scored with 8 bands retained the 8 value ) . The assigned values were used to calculate average number of denticles bands and standard deviations , from which effect size and power were determined in G*Power . These values were also used in student’s t tests .

Genes encode the instructions needed to make proteins and other molecules . To make a protein , the DNA within a gene is copied to produce molecules of messenger ribonucleic acid ( mRNA ) that are then used as templates to build proteins via a process called translation . This process – which involves protein machines called ribosomes binding to the start of the mRNA – is tightly regulated to control the amounts of particular proteins in cells . For example , in fruit fly ovaries , a protein called Bruno both represses and activates the translation of a gene known as oskar . To achieve this , Bruno binds to regions near the end of the oskar RNA known as Bruno response elements . It is not clear how Bruno acts to control translation . However , because ribosomes begin translation near the start of the mRNA , while Bruno is bound to regions near the end of the mRNA , there must be long-range interactions between the two ends of the mRNA . It is generally assumed that such long-range interactions only occur between proteins that are bound to the same mRNA molecule . However , in 2010 , researchers observed that Bruno response elements within one oskar mRNA could influence the translation of other oskar mRNAs . This is known as “regulation in trans” . Here , Macdonald et al . – including some of the researchers from the earlier work – investigated this observation in more detail in fruit flies . In cells , multiple mRNA molecules and their associated proteins can assemble into particles . Macdonald et al . proposed that the close proximity of many mRNA molecules in these particles could allow trans regulation to take place . Indeed , the experiments found that blocking the assembly of oskar mRNA into particles inhibited trans regulation as expected . Macdonald et al . also asked if trans regulation can occur between mRNAs that encode different proteins . The experiments show that oskar mRNA could block the translation of an mRNA produced by the gurken gene , even when oskar mRNA was not being translated . More work is needed to find out how widely trans regulation is used to control translation .

2016

Community effects in regulation of translation

Design of complex alpha-beta protein topologies poses a challenge because of the large number of alternative packing arrangements . A similar challenge presumably limited the emergence of large and complex protein topologies in evolution . Here , we demonstrate that protein topologies with six and seven-stranded beta sheets can be designed by insertion of one de novo designed beta sheet containing protein into another such that the two beta sheets are merged to form a single extended sheet , followed by amino acid sequence optimization at the newly formed strand-strand , strand-helix , and helix-helix interfaces . Crystal structures of two such designs closely match the computational design models . Searches for similar structures in the SCOP protein domain database yield only weak matches with different beta sheet connectivities . A similar beta sheet fusion mechanism may have contributed to the emergence of complex beta sheets during natural protein evolution . Modular domains constitute the primary structural and functional units of natural proteins . Multi-domain proteins likely evolved through simple linear concatenation of successive domains onto the polypeptide chain or through the insertion of one or more continuous sequences into the middle of another , now discontinuous domain ( Aroul-Selvam et al . , 2004; Berrondo et al . , 2008; Lupas et al . , 2001; Pandya et al . , 2013 ) . By analogy , new proteins have been engineered from existing domains by simple linear concatenation or insertion of one domain into another ( Ay et al . , 1998; Collinet et al . , 2000; Cutler et al . , 2009; Doi and Yanagawa , 1999; Edwards et al . , 2008; Guntas and Ostermeier , 2004; Ostermeier , 2005 ) . How individual domains evolved , in contrast , is much less clear . Both experimental and computational analyses have suggested that new folds can evolve by insertion of one fold into another ( Lupas et al . , 2001; Grishin , 2001; Söding and Lupas , 2003; Krishna and Grishin , 2005; Friedberg and Godzik , 2005; Ben-Tal and Kolodny , 2014 ) , but to our knowledge , there is no evidence that complex beta sheet topologies can be formed in this manner . On the protein design front , there has been progress in de novo design of idealized helical bundles ( Park et al . , 2015 ) and alpha beta protein structures with up to 5 strands ( Koga et al . , 2012 ) , and although new folds have been generated by tandem fusion of natural protein domains followed by introduction of additional stabilizing mutations ( Hocker et al . , 2004; Shanmugaratnam et al . , 2012 ) , assembly of large and complex beta sheets poses a challenge for de novo protein design . One possible route to the large and complex beta sheet topologies found in many native protein domains is recombination of two smaller beta sheet domains . Here , we explore the viability of such a mechanism by inserting one de novo designed alpha beta protein into another such that the two beta sheets are combined into one . The backbone geometry at the junctions between the original domains is regularized , and the sequence at the newly formed interface is optimized to stabilize the single integrated domain structure . Crystal structures of two such proteins demonstrate that complex beta sheet structures can be designed with considerable accuracy using this approach and provide a proof-of-concept for the hypothesis that complex beta topologies in natural proteins may have evolved from simpler beta sheet structures in a similar manner . A first extended sheet protein was created by inserting a designed ferredoxin domain into a beta turn of the designed TOP7 protein to create a half-barrel structure , with the two sheets fused into a single seven strand sheet flanked by four helices ( Figure 1A ) . The CD spectra show both alpha and beta structures ( Figure 2—figure supplement 1 ) . Two crystal structures ( NESG target OR327 ) were solved by molecular replacement and refined to 2 . 49 Å ( PDB entry 4KYZ ) and 2 . 96 Å ( PDB entry 4KY3 ) resolutions . Further analysis refers only to the higher resolution structure ( 4KYZ ) . The structure shows excellent agreement with the design model ( Figure 2A ) , particularly in low B-factor regions , with C-alpha RMSD ranging from 1 . 76 to 1 . 85 Å among the four protomers in the crystal . The relative orientation of the strands packed against the helices is close to that in the design model , and core sidechains at the designed interfaces are in very similar conformations in the design model and crystal ( Figure 2B , C ) . 10 . 7554/eLife . 11012 . 003Figure 1 . Generation of protein domains with single extended beta sheets by inserting one beta sheet containing protein into another . ( A ) Insertion of a ferrrodoxin domain ( purple ) into TOP7 ( red ) . ( B ) Insertion of one ferrodoxin domain into another . In both cases , two beta strands from each partner ( red and purple ) are concatenated to form the central strand pair of the fusion protein ( pink ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11012 . 00310 . 7554/eLife . 11012 . 004Figure 2 . Comparison of the crystal structure of ferredoxin-TOP7 fusion to design model . ( A ) Backbone superposition of the crystal structure of ferredoxin-TOP7 ( 4KYZ , chain A ) with the design model . The backbones of the two proteins are nearly identical . ( B , C ) The core sidechain packing in the ferrodoxin-TOP7 fusion is very similar in the crystal structure and design model both in the insert ( B ) and host ( C ) domains . The crystal structure is colored by B-factor and the design model is in gray . DOI: http://dx . doi . org/10 . 7554/eLife . 11012 . 00410 . 7554/eLife . 11012 . 005Figure 2—figure supplement 1 . The circular dichroism spectrum of ferrodoxin-TOP7 has the shape expected for an alpha/beta protein . DOI: http://dx . doi . org/10 . 7554/eLife . 11012 . 005 A second extended sheet protein was created by inserting one designed ferredoxin domain into another to create a half-barrel structure with four alpha helices and six beta strands ( Figure 1B ) . A beta turn segment between two beta strands of the host ferredoxin was removed and the resulting cut-points in the host beta strands were linked to two beta strand cut-points in the insert , fusing the two strand pairs into a single , longer pair at the center of a six-stranded beta sheet . CD spectra show that the protein contains both alpha and beta structures ( Figure 3—figure supplement 1 ) . Crystals were obtained which diffracted to 3 . 3Å resolution . Molecular replacement using the computational design models ( DiMaio et al . , 2013 ) yielded a solution for which the refinement statistics are shown in Supplementary file 1 ( PDB entry 5CW9 ) . Attempts to improve these statistics by rebuilding portions of the model proved unsuccessful , possibly due to a register shift or dynamic fluctuations in the structure ( perhaps corresponding to slightly 'molten-globule'-like behavior ) that are difficult to computationally model . However , unbiased low-resolution omit maps suggest that the overall topology is correct ( Figure 3—figure supplement 2 ) . In the model that displays the best refinement statistics , the protein backbone was similar to the design model with a C-alpha RMSD value of 2 Å ( Figure 3A , B ) . The fused beta sheet aligns with the design model , while the inter-domain helices shift slightly to accommodate the inter-domain interface . The sidechain packing between the newly juxtaposed beta strands succeeded in anchoring the secondary structure elements in their intended orientations , but the low resolution of the crystal structure prevents evaluation of the atomic-level accuracy of the design ( Figure 3—figure supplement 2 ) . 10 . 7554/eLife . 11012 . 006Figure 3 . Comparison of the crystal structure of the ferredoxin-ferredoxin fusion to the design model . The crystal structure ( 5CW9 ) aligns well with the design model over both the helices ( A ) and the fused beta sheet ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11012 . 00610 . 7554/eLife . 11012 . 007Figure 3—figure supplement 1 . Circular dichroism spectra of ferrrodoxin-ferrodoxin at 25°C . DOI: http://dx . doi . org/10 . 7554/eLife . 11012 . 00710 . 7554/eLife . 11012 . 008Figure 3—figure supplement 2 . Ferredoxin-Ferredoxin 2Fo-Fc omit map superimposed with crystal structure shows core packing of host ( A ) and insert ( B ) domains . DOI: http://dx . doi . org/10 . 7554/eLife . 11012 . 008 To compare the folds of these designed proteins to those in the SCOP v . 1 . 75 domain database ( Murzin et al . , 1995 ) , the TMalign structure-structure comparison method was used to search a 70% sequence non-redundant set of SCOP domains ( Ben-Tal and Kolodny , 2014 ) for structure alignments containing a minimum 75% overlap with the designed proteins . The most similar SCOP domains had weak TM-align scores ( 0 . 54 and 0 . 51 ) , and the sheets in these matched structures have different connectivities than those of the designs , suggesting that the two designed proteins have novel folds ( Figure 4 ) . While there are no domains with globally similar folds , both designed proteins are similar to a number of SCOP domains over the ferrodoxin-like substructure ( s ) ( maps of the proteins to the domain network of Nepomnyachiy et al . ( Ben-Tal and Kolodny , 2014 ) are shown in Figure 4—figure supplement 1 ) . The mutations introduced at the redesign stage of the domain insertion design protocol are compatible with the parent fold structures with minimal perturbation of the protein backbone ( Figure 4—figure supplement 2 ) suggesting the designed folds would have the potential to evolve from insertion followed by neutral mutational drift of the parent structures . 10 . 7554/eLife . 11012 . 009Figure 4 . Top two SCOP domain structural homologues for Fd-Top7 ( A ) and Fd-Fd ( B ) designed domain found in TM-align searches . Ribbon diagrams are shown on left , the strand connectivity , at the right . The beta strand connectivity is quite different in the designs than in these closest structural matches . DOI: http://dx . doi . org/10 . 7554/eLife . 11012 . 00910 . 7554/eLife . 11012 . 010Figure 4—figure supplement 1 . Parent domain PDB structures ( 2KL8 , 1QYS ) and daughter designed folds ( 5CW9 , 4KYZ ) ( pink ) mapped into the α+β region of the SCOP domains network of Nepomnyachi et al . ( A ) and zoomed region ( B ) highlighting parent , designed , and first neighbor folds . DOI: http://dx . doi . org/10 . 7554/eLife . 11012 . 01010 . 7554/eLife . 11012 . 011Figure 4—figure supplement 2 . Neutral drift mutant models , relative changes to predicted free energy of folding in REU ( Rosetta Energy Units ) , and multiple sequence alignment of parent and designed sequences , showing mutations in ferredoxin-top7 ( A ) and ferredoxin-ferredoxin ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11012 . 011 We have shown that single designed protein domains can be combined into larger domains with complex beta sheet topologies . This mechanism provides a straightforward route to designing large and complex beta sheet structures capable of scaffolding the pockets and cavities essential for future design of protein functions . Our success in designing larger beta sheet domains by recombining smaller independently folded beta sheet proteins suggests a similar mechanism could have played a role in the evolution of naturally occurring complex beta sheet proteins . Structures have been deposited in the Protein Data Bank as entries 5CW9 , 4KYZ , and 4KY3 .

A protein is made up of a sequence of amino acids and must fold into a specific three-dimensional structure if it is to work correctly . The structure is formed by segments of the protein adopting specific shapes , the two most common shapes being alpha helices and beta strands . Beta strands commonly interact with each other to form regions called beta sheets . Researchers trying to design proteins with new abilities have managed to create proteins that contain up to five beta strands and four alpha helices . Larger and more complex proteins are more challenging to make because there are many different ways that a protein can fold . It is also difficult to understand how complex structures such as large beta sheets emerged naturally , over the course of evolution . King et al . have now used computer modeling to explore how a large , complex beta sheet might form . In the model , one small , newly designed protein was inserted into another so that their beta sheets merged to form a single extended sheet . The model then stabilized this structure by changing the amino acids found at the points where the two proteins met . King et al . were then able to synthesize these new proteins in bacteria and use a technique called X-ray crystallography to determine the structure of two of them . The structures closely matched the computer models; one protein contained a six-stranded beta sheet , and the other had a seven-stranded beta sheet . The folds of the two designed proteins were then compared with those found in a database that classifies proteins on the basis of their structure . The beta sheets in the designed proteins did not match the protein structures in the database , which suggests that the designed proteins contained new types of folds . In the future , the technique used by King et al . could be used to design other large and complex beta sheet structures . Furthermore , the results suggest that such large structures could have evolved naturally through the combination of smaller , less complex proteins .

2015

Precise assembly of complex beta sheet topologies from de novo designed building blocks

Cytoplasmic microtubules ( cMT ) control mitotic spindle positioning in many organisms , and are therefore pivotal for successful cell division . Despite its importance , the temporal control of cMT formation remains poorly understood . Here we show that unlike the best-studied yeast Saccharomyces cerevisiae , position of pre-anaphase nucleus is not strongly biased toward bud neck in Ogataea polymorpha and the regulation of spindle positioning becomes active only shortly before anaphase . This is likely due to the unstable property of cMTs compared to those in S . cerevisiae . Furthermore , we show that cMT nucleation/anchoring is restricted at the level of recruitment of the γ-tubulin complex receptor , Spc72 , to spindle pole body ( SPB ) , which is regulated by the polo-like kinase Cdc5 . Additionally , electron microscopy revealed that the cytoplasmic side of SPB is structurally different between G1 and anaphase . Thus , polo-like kinase dependent recruitment of γ-tubulin receptor to SPBs determines the timing of spindle orientation in O . polymorpha . Segregation of sister chromatids into two daughter cells is pivotal to the proliferation of eukaryotic cells . Chromosome segregation is followed by cytokinesis , which results in physical separation of two daughter cells . In many organisms , the position of the mitotic spindle dictates the site of cytokinesis , which ensures the inheritance and maintenance of genomic information in the daughter cells . Astral microtubules or cytoplasmic microtubules ( cMTs ) , which emanate from the spindle poles and extend to the cell cortex , have a principle role in positioning and orienting the spindle with respect to the polarity cues of the cell type . Mechanisms governing the spindle positioning/orientation have been studied in a number of systems . However , regulations that determine the timing of establishing the spindle orientation , or the position of the centrosome , the primary MT organizing centre ( MTOC ) , in interphase , are not well understood ( Kiyomitsu , 2015; Woyke et al . , 2002 ) . Spindle positioning is of particular importance in the budding yeast Saccharomyces cerevisiae , where the cleavage site is determined at the start of the cell cycle independently of the position of the mitotic spindle . Therefore , cells position the pre-anaphase spindle close to the bud neck and orient it along the mother-bud axis . As the spindle elongates in anaphase , one spindle pole translocates into the bud to accomplish segregation of one set of chromosomes into the daughter cell ( Pereira and Yamashita , 2011; Markus et al . , 2012; Winey and Bloom , 2012 ) . In S . cerevisiae , the nuclear positioning and spindle orientation are regulated by two redundant pathways acting on cMTs , the Kar9 and dynein pathways ( Li et al . , 1993; Miller and Rose , 1998; Winey and Bloom , 2012 ) . Concomitant deletions in components of both pathways result in lethality , whereas loss of one pathway can be compensated by the function of the other with moderate spindle orientation defects ( Miller and Rose , 1998 ) . Survival of single deletion mutants largely relies on the function of the spindle orientation checkpoint ( SPOC ) that retains cells in anaphase until the spindle orientation is corrected ( Bardin et al . , 2000; Pereira et al . , 2000; Caydasi and Pereira , 2012 ) . Furthermore , MTs in S . cerevisiae are organized exclusively from the spindle pole body ( SPB ) , which is the functional equivalent of animal centrosome . The SPB is a multilayered cylindrical organelle that is embedded in the nuclear envelope ( NE ) throughout the cell cycle ( Byers and Goetsch , 1974; Byers and Goetsch , 1975 ) The outer plaque faces the cytoplasm and nucleates cMTs , whereas the inner plaque is inside the nucleus and organizes the nuclear MTs . The central plaque anchors and interconnects the outer and inner plaques ( O'Toole et al . , 1999; Jaspersen and Winey , 2004 ) . In G1 phase , some fractions of the cMTs are organized from a modified region of the NE associated with one side of the SPB known as the half-bridge ( Byers and Goetsch , 1974; Byers and Goetsch , 1975 ) . Spc72 , a γ-tubulin complex ( γ-TuSC ) receptor , is required for nucleating MTs at both the outer plaque and the half-bridge ( Chen et al . , 1998; Knop and Schiebel , 1998; Wigge et al . , 1998; Souès and Adams , 1998 ) . Localisation of Spc72 at the outer plaque is mediated by binding to Nud1 , whereas Kar1 serves as a G1 specific binding site of Spc72 at the half-bridge ( Pereira et al . , 1999; Gruneberg et al . , 2000 ) . Spc72 also has a structural role as an integral part of the outer layer and as such localisation of Spc72 to the SPB and the ability to nucleate cMTs persist through the entire cell cycle ( Shaw et al . , 1997; Pereira et al . , 1999; Kosco et al . , 2001 ) . Importantly , Spc72 , and hence cMTs , is not recruited for the formation of the SPB . New SPB acquires Spc72 and cMTs after the formation of a 1 µm long spindle ( Shaw et al . , 1997; Segal et al . , 2000; Juanes et al . , 2013 ) . In addition to the γ-tubulin complexes , Spc72 exerts a role in recruiting several other proteins to SPBs including Stu2 , a microtubule-associated protein ( MAP ) of the XMAP215/Dis1 family , the SPOC kinase Kin4 , as well as polo-like kinase Cdc5 ( Chen et al . , 1998; Usui et al . , 2003; Maekawa et al . , 2007; Snead et al . , 2007 ) . Cdc5 regulates multiple cellular functions including SPB duplication , progression through G2/M phase , promoting mitotic exit , and cytokinesis ( Shirayama et al . , 1998; Hu et al . , 2001; Song and Lee , 2001; Archambault and Glover , 2009; Elserafy et al . , 2014 ) . Cdc5 is also involved in the regulation of spindle orientation in pre-anaphase and migration of the anaphase spindle ( Snead et al . , 2007; Park et al . , 2008 ) . Although Spc72 becomes highly phosphorylated during mitosis in a Cdc5-dependent manner , it is unclear whether this phosphorylation has a regulatory effect on Spc72 and/or cMTs ( Maekawa et al . , 2007; Snead et al . , 2007 ) . The molecular mechanisms that control spindle orientation in S . cerevisiae have been well established . However , other species that employ the budding mode of cell division may have adopted different strategies . In the pathogenic yeast Candida albicans , the nucleus is located away from the bud neck in pre-anaphase cells ( Martin et al . , 2004; Finley et al . , 2008 ) . C . albicans and probably some of other species in Saccharomycotena may therefore have different mechanisms and regulations in this fundamental biological process . Ogataea polymorpha ( previously Hansenula polymorpha ) is extensively used in industrial biotechnology , for the production of various pharmaceuticals in particular for its advantageous characteristics including methylotrophy , nitrate assimilation , availability of strong promoters , and low amount of secreted proteins ( Gellissen et al . , 2005; Stöckmann et al . , 2009 ) . Another attractive property of O . polymorpha is its thermotolerant nature ( up to approximately 50°C ) , which may reduce the cost of cooling in , for instance , bioethanol production that requires the treatment of raw materials at high temperature prior to fermentation . However , despite its importance , cell biology research on this organism remains limited . A better understanding of the molecular physiology of O . polymorpha is beneficial towards improving the abilities and characteristics of this yeast for a wide variety of applications . Here , we describe cMT organization and its regulation during the cell cycle of the methylotrophic yeast O . polymorpha . Unlike S . cerevisiae , the pre-anaphase spindle is not readily positioned and oriented in O . polymorpha owing to the poorly organized cMTs at early cell cycle stages . The bottleneck of cMT nucleation/anchoring at SPBs occurs at the level of Spc72 recruitment to the SPBs , for which the polo-like kinase Cdc5 plays a crucial role . Consistent with the cell cycle dependent activity of cMTs , SPB structure also undergoes cell cycle dependent modification . Thus , our study shed light on the divergent nature of the temporal control of the cMT formation in yeast species . The nucleus is positioned close to the bud neck in large budded pre-anaphase cells of S . cerevisiae ( Figure 1A ) . Similar organization was observed in other budding yeast species including Candida glabrata , Kluyveromyces lactis , Pichia pastoris , and Yarrowia lipolytica ( Figure 1—figure supplement 1 ) . Notably in O . polymorpha , however , nuclear position was not biased to the bud neck , although it remained in the mother cell body ( Figure 1A , Figure 1—figure supplement 1 ) . The phenotype resembled , but was more exaggerated than , that in C . albicans where the nucleus is located with a distance from the bud neck in pre-anaphase cells ( Martin et al . , 2004; Finley et al . , 2008 ) . A similar phenotype was observed in species closely related to O . polymorpha or C . albicans ( Figure 1—figure supplement 1 ) . Close examination of O . polymorpha revealed that the nucleus was located in the cell centre in 76 . 7% of G2/M cells whereas in the remainder of the cells it was off-centred with no bias towards the bud neck ( Figure 1A , Figure 1—figure supplement 2 ) . These results suggest that the nuclear position is not determined before anaphase onset in O . polymorpha . As a consequence , the early stages of anaphase occurred in the mother cell body ( Figure 1B ) . This was further confirmed in cells expressing the α-tubulin gene ( Tub1 ) -GFP and histone H3 gene ( Hht1 ) -mRFP ( monomeric red fluorescent protein ) to visualize MTs and chromosomal DNA , respectively . Majority of the early anaphase spindles as judged according to their length ( <5 µm ) as well as a stretched DNA mass were located entirely in the mother cell body ( 97 . 4% , Figure 1C , Figure 1—figure supplement 3A ) . However , 94 . 7% of the early anaphase spindles were aligned along the bud-mother axis and almost all of the late anaphase spindles with two segregated DNAs were inserted into the bud ( Figure 1—figure supplement 3 ) , suggesting immediate and efficient orientation of the spindle during anaphase . SPB position during the cell cycle was examined to clarify the spindle cycle relative to bud size in cells expressing the SPB marker ( Mps3-GFP ) ( Figure 1D ) . Two SPB signals appeared in some of small/medium budded cells ( Figure 1D , panel c , lower cell ) , suggesting that SPBs were duplicated at the timing of bud emergence or later , which is similar to that in S . cerevisiae although the precise cell cycle stage should be carefully determined . In the rest of small budded cells , one SPB signal was evident until large budded cells ( Figure 1D , panel d ) . This may be because duplicated SPBs remained in a close proximity and could not be resolved by our fluorescence microscopy . Consistent with this , intensity of Mps3-GFP in medium/large-budded cells with a single SPB was much higher than that in unbudded G1 cells or cells with separated SPBs ( Figure 1—figure supplement 4 ) . Moreover , SPB in G1 cells as well as small budded cells was not in the defined position within the mother cell body ( Figure 1D ) . Subsequent time lapse analysis revealed that after spindle assembly , ~1 µm long spindles remained at their central positions and loosely oriented toward the bud neck until shortly before anaphase onset ( defined by the rapidly increase of pole-to-pole distance ) ( Figure 1—figure supplement 5 ) . Anaphase initiated in the mother cell body ( Figure 1E , 11 min , Figure 1—figure supplement 6 , 3–4 min ) . These observation defined cells with 2SPBs in a < 2 µm distance as pre-anaphase cells . Spindle alignment was corrected around the time of ( or shortly after ) spindle elongation , followed by SPB insertion into the bud . After spindle breakdown , the SPB moved vigorously with no relationship to the polarity axis ( Figure 1E , Figure 1—figure supplement 6 ) . Lack of nuclear positioning and spindle orientation in pre-anaphase cells may indicate a very low number of cMTs at SPBs during this cell cycle window . To test this notion , we first investigated the MT organization during the cell cycle in GFP-TUB1 HHT1-mCherry cells ( Figure 2A and B ) . Cell cycle stages were judged by the bud size and the number of DNA masses . cMTs were observed in all anaphase cells , whereas less than 50% of G1 and pre-anaphase cells carried cMTs . Furthermore , cMTs that apparently did not associate with the SPB were observed in 13 . 8% of cells prior to anaphase ( Figure 2C ) . Time lapse analysis revealed that detached cMTs remained in the cytoplasm for only a short period of time before depolymerized ( Figure 2—figure supplement 1 ) . This situation is in stark contrast to that in S . cerevisiae where almost all cells exhibit cMTs that are stably associated with the SPB during the cell cycle ( Shaw et al . , 1997; Kosco et al . , 2001 ) . The small number of observed cMTs might have arisen because of reduced cMT nucleation . Another and not mutually exclusive possibility considers that cMTs might not be stably anchored to the SPB and thus might not persist over long periods . To test these possibilities , we performed time lapse experiments with cells expressing GFP-TUB1 , in which cMTs were observed during 20 . 6% of the recorded time points although the majority ( >80% ) did not persist longer than 30 s ( Figure 2D , cMT persistence , Figure 2—figure supplement 2 ) . These results suggest that cMTs are short-lived during early stages of the cell cycle . Once cMTs were lost , a relatively long time was required until new MTs appeared at the SPB ( Figure 2D , cMT re-establishment; median value 90 . 0 s , average 173 . 8 ± 192 . 4 s ) . Thus , cMTs are less frequently nucleated and unstable at early stages of the cell cycle . Acquired cMTs efficiently corrected the spindle orientation in pre-anaphase cells , suggesting that the spindle orientation is regulated largely at the level of cMT acquisition ( Figure 2—figure supplement 3 ) . Next , we examined the SPB structure in G1 and anaphase by electron microscopy . An electron dense SPB-like structure was evident in all cells examined , while a half-bridge-like structure—which plays an important role in cMT organization in G1 of S . cerevisiae—was not clearly observed ( Figure 3 ) . Anaphase SPBs had an additional thin layer in the cytoplasm that resembled the outer plaque of S . cerevisiae SPB ( Figure 3A and B ) ( Byers and Goetsch , 1974; Byers and Goetsch , 1975 ) . In contrast , there was no detectable outer plaque in G1 SPBs ( Figure 3C and D ) . These results suggested that O . polymorpha SPBs undergo structural cycling in every cell cycle . Our attempt to arrest cells in late G1 by introducing cdc28-as allele , which arrest S . cerevisiae cells in late G1 with a single SPB and satellite , was failed probably because of insufficient inhibition of the kinase ( Figure 3—figure supplement 1 ) . However , inhibitor addition delayed cell cycle progression leading to the accumulation of cells with unseparated SPBs . This allowed us to examine the structure of side-by-side SPBs by EM ( Figure 3—figure supplement 2 ) . All six side-by-side SPBs had outer plaques which were similar to that in nocodazole arrested cells , albeit some of those were somewhat fuzzy . The result suggested that the SPB structure on cytoplasmic side is reconstructed before spindle formation . An additional electron dense cloud was observed on the cytoplasmic side of nuclear envelope between two SPB bodies , which resembled the half-bridge/bridge structure of S . cerevisiae SPB ( Figure 3—figure supplement 2A , B and E; orange arrowheads ) . It was not clear whether this structure was present at other cell cycle stages . A better synchronization method is required to determine the fine structure and the precise timing of emergence/disappearance of the outer plaque during the cell cycle . Lack of the outer plaque in G1 prompted us to search for SPB components whose association with SPBs was cell cycle dependent . In S . cerevisiae , core SPB components are found at SPBs throughout the mitotic cell cycle including central plaque components ( Spc42 , Spc29 ) , outer plaque components ( Cnm67 , Nud1 ) , half-bridge components ( Sfi1 , Cdc31 , Mps3 , and Kar1 ) , and membrane anchors ( Ndc1 , Nbp1 , Mps2 , and Bbp1 ) ( Winey et al . , 1991; Spang et al . , 1995; Bullitt et al . , 1997; Brachat et al . , 1998; Wigge et al . , 1998; Chial et al . , 1998; Adams and Kilmartin , 1999; Elliott et al . , 1999; Schramm et al . , 2000; Kilmartin , 2003; Jaspersen et al . , 2002; Araki et al . , 2006 ) . The γ-TuSC recruiting factors Spc110 and Spc72 also represent core components of SPB in the inner and outer plaques , respectively ( Knop and Schiebel , 1997; Knop and Schiebel , 1998 ) . BLAST and HMMER searches have identified putative orthologues of genes for Mps3 , Sfi1 , Spc72 , Spc110 , and Nud1 as well as γ-TuSC components , Tub4 , Spc97 , and Spc98 in the O . polymorpha genome ( Altschul et al . , 1990; Sobel and Snyder , 1995; Geissler et al . , 1996; Knop and Schiebel , 1997; Eddy , 1998; Maekawa and Kaneko , 2014; Riley et al . , 2016 ) . SPB-like localisation was verified by expressing GFP or mRFP-fused version of these proteins . GFP or mRFP dot-like signals of Tub4 , Spc98 , Mps3 , Sfi1 , and Nud1 were observed in most of the cells , suggesting that they represent constitutive components of the SPB throughout the cell cycle ( Figure 4A , Figure 4—figure supplement 1 ) . In contrast , the Spc72 signal was either weak or absent in cells at early cell cycle stages , whereas all anaphase cells carried two strong SPB signals ( Figure 4A , Figure 4—figure supplement 1 ) . Deletion of SPC72 in S . cerevisiae results in severe growth defects or lethality depending on the strain background . To evaluate the effect of SPC72 deletion in O . polymorpha , SPC72/spc72Δ::natNT2 heterozygous diploid cells were subjected to tetrad dissection analysis . Notably , 21 out of 29 tetrads yielded one or two viable colonies , all of which were sensitive to nourseothricin ( Figure 4—figure supplement 2 ) . Microscopic inspection revealed that 91 . 7% of spores with spc72Δ::natNT2 genotype derived from tetrads that gave two viable nourseothricin-sensitive colonies were germinated . These results suggested that SPC72 is essential for growth in O . polymorpha . To more precisely evaluate the amount of Spc72 at the SPB , images were obtained in logarithmically growing wild-type cells carrying SPC72-GFP MPS3-mRFP and the GFP intensity at the SPB was quantified ( Figure 4B ) . The GFP signal was 2 . 5 times weaker in cells with a short spindle than in anaphase cells ( p<0 . 001 ) . In contrast , Nud1 , which comprises the putative binding site of Spc72 on the outer plaque as suggested by the direct interaction between orthologues of these proteins in S . cerevisiae , did not show this trend , nor did Sfi1 , a half-bridge component ( Figure 4B ) ( Gruneberg et al . , 2000; Kilmartin , 2003 ) . High intensity of Sfi1-GFP signal in S/G2 cells most likely arose from SPBs that were duplicated but not yet separated . These results suggest that Spc72 is cell cycle regulated and the incorporation of Spc72 into SPBs may be the key step to stabilize cMTs . To further confirm this notion , time-lapse microscopy was carried out to determine the timing of Spc72 association with SPBs ( Figure 4C , Figure 4—figure supplement 3 ) . In all cells that progressed into anaphase , an Spc72-GFP signal became detectable <4 min prior to the initiation of anaphase ( average 3 . 68 ± 1 . 74 min , n = 14 ) ( Figure 4C , orange arrowheads ) . Within 5 min after appearance of the Spc72-GFP signal , spindle orientation was corrected when it had not done already ( Figure 4—figure supplement 3 , average 3 . 50 ± 1 . 61 min , n = 12 ) ; therefore , one half part of an anaphase nucleus was successfully inserted into the bud . Thus , Spc72 accumulates at SPB in early mitosis , most likely in metaphase , and remains high during anaphase . As cells exit from mitosis and entre the next cell cycle , Spc72-GFP signal was gradually decreased at SPBs with the timing that varied from cell to cell . This difference of timing may explain the relatively high and variable intensity of Spc72-GFP at SPB in G1 cells ( Figure 4B ) . However , in all cases , Spc72-GFP levels reached a minimum well before short spindle was formed ( Figure 4 , Figure 4—figure supplement 4 ) . If low abundance of Spc72 at the SPB is the reason underlying cMT instability , higher expression of Spc72 might increase the level of Spc72 at the SPB and consequently raise the number of cMTs , and thereby promote positioning the nucleus close to the bud neck at early stages of the cell cycle . We expressed the SPC72-GFP gene from a strong constitutive TEF1 promoter in cells whose endogenous SPC72 was also fused to GFP and examined the position of the SPB relative to the bud neck in pre-anaphase cells as a readout of cMT function ( Kiel et al . , 2007 ) . Overexpressed Spc72-GFP was efficiently targeted to SPB because a strong Spc72-GFP signal was observed in cells carrying the PTEF1-SPC72-GFP gene but not in wild-type cells during G1 and G2/M phases ( Figure 5A , Figure 5—figure supplement 1 ) . In cells overexpressing SPC72-GFP , SPB was positioned close to the bud neck , which is reminiscent of the SPB position in S . cerevisiae ( Figures 1A and 5B ) , and cMTs were more often observed ( Figure 5C and D ) . Together , these results strongly supported our hypothesis that cMT organization is regulated at the level of repeated Spc72 recruitment to the SPB in every cell cycle . cMT play important roles in yeast mating and karyogamy , which are initiated in G1 . Because mating is triggered by nutrient starvation in O . polymorpha , we examined cMTs and Spc72 in nutrient starved cells . Interestingly , while Spc72 was accumulated at SPBs , cMTs were not observed ( Figure 5—figure supplement 2 ) . Thus , specific mechanism may regulate Spc72 and cMT organization under such conditions . Spc72 might be regulated at the level of protein expression . To synchronize cells for monitoring changes in protein levels during the cell cycle , we transferred the recently developed auxin-inducible degradation ( AID ) -degron system to O . polymorpha ( Nishimura and Kanemaki , 2014 ) . CDC5 encodes the only polo-like kinase in yeast . It is thus essential for growth in S . cerevisiae , and its inactivation causes cell cycle arrest in late anaphase ( Kitada et al . , 1993 ) . Similarly , a single CDC5 orthologue was identified in the O . polymorpha genome ( OpCDC5 ) . Logarithmically growing cells carrying a 3mAID-tagged version of CDC5 were arrested as large budded cells by incubation in the presence of auxin and then released into fresh medium without auxin to resume the cell cycle ( Figure 6A ) . Spc72 protein abundance did not fluctuate as cells entered into anaphase and proceeded into the following cell cycle ( Figure 6A , Figure 6—figure supplement 1A ) . Furthermore , the Spc72 band migrated slower in nocodazole-arrested cells than that in asynchronous cells ( Figure 6—figure supplement 1B ) . These results suggested that either post-translational modification of Spc72 or regulation of Spc72 binding proteins might be utilized to achieve cell cycle dependency of SPB localisation . Furthermore , we noticed that the GFP intensity of Spc72-GFP at the SPB was significantly lower in cells arrested by Cdc5-depletion than that in cells after re-accumulation of Cdc5 in both metaphase ( SPB distance <2 µm , p<0 . 0001 ) and anaphase ( SPB distance >4 µm , p<0 . 0001 ) ( Figure 6B and C ) . Strong dependency of SPB binding of Spc72 on Cdc5 was further confirmed in metaphase-arrested cells by nocodazole ( Figure 6D , Figure 6—figure supplement 2 ) . These results suggested that the stable association of Spc72 requires Cdc5 kinase , the activity of which is likely cell cycle-regulated . Spc72 is phosphorylated by Cdc5 kinase in S . cerevisiae ( Maekawa et al . , 2007; Snead et al . , 2007 ) , which prompted us to investigate whether Spc72 is subjected to a Cdc5-dependent phosphorylation in O . polymorpha . Our gel electrophoresis analyses of nocodazole-arrested cells suggested that Spc72 of O . polymorpha is subjected to post-translational modifications in a Cdc5-dependent manner because the Spc72 band was more smeared and migrated slightly slower in wild type and CDC5-overexpressing cells as compared to that in Cdc5-depleted cells ( Figure 6—figure supplement 3A ) . Although this difference was largely lost during preparation of cell extract , the λ-phosphatase treatment revealed in vivo phosphorylation of Spc72 that was independent of Cdc5 ( Figure 6—figure supplement 3B ) , suggesting that Cdc5 contributes to a subset of phosphorylations of Spc72 . Thus , Spc72 is phosphorylated at multiple sites , only some of which depend on Cdc5 . We next examined the localisation of Cdc5 during the cell cycle . Distinctive localization of Cdc5-GFP became apparent after S phase and was lost prior to or during the following G1 phase ( Figure 7A and B , Figure 7—figure supplement 1 ) . Nuclear and NE localisation appeared at early stages of the cell cycle and persisted until the end of mitosis . In addition , a fraction of the GFP signals appeared to overlap with those of SPBs during mitosis . Neither NE nor SPB-associated GFP signals were detected in unbudded or small budded cells . Notably , SPB signal may arise from the SPB itself , NE surrounding the SPB , or kinetochores that cluster close to the SPB during interphase and anaphase in yeast ( Jin et al . , 1998 ) . Therefore , in order to clarify on which side of SPB the Cdc5-GFP signal resided , we employed structured illumination microscopy ( SIM ) . Localisation was investigated in metaphase-arrested cells with nocodazole where Cdc5-GFP signal was observed at SPB as well as in nucleus ( Figure 7—figure supplements 1B and 2 ) , and Spc72 , and Spc110 were used as references for the cytoplasmic and nuclear side of SPB , respectively . Spc72-GFP and Spc110-tdTomato signals were clearly distinguished in 58% of the cells , which verified that our method could discriminate signals in the cytoplasmic and the nuclear side of SPB in >50% of cells ( Figure 7C ) . The resolution of both signals probably depends on the orientation of the SPB ( top versus side view ) . Only the SPB side view will resolve Spc72-GFP and Spc110-tdTomato signals at SPBs . Cdc5-GFP overlapped with Spc110-tdTomato in 53% of cells , which is similar to the degree of co-localisation observed between Spc72 and Spc110 . In contrast , the Cdc5-GFP signal overlapped with Spc72-tdTomato in 87% of cells ( Figure 7D ) . These results suggest that Cdc5-GFP locates at the position on SPBs closer to Spc72 than to Spc110 , indicating that Cdc5-GFP signal arises from the cytoplasmic side of SPB . Thus , Cdc5 likely becomes first localised to the nucleus and the NE in G2 , and then in mitosis to the cytoplasmic side of SPBs . The timing of Cdc5 binding to SPBs coincides well with the recruitment of Spc72 to SPBs . To further confirm the significance of Cdc5 kinase in the recruitment of Spc72 to SPBs , we constitutively expressed the CDC5 gene at high level . While Cdc5 expression showed no effect on the protein level of Spc72-GFP ( Figure 8—figure supplement 1 ) , Spc72-GFP intensity at the SPB was higher in metaphase-arrested cells following ectopic expression of CDC5 from the TEF1 promoter than in wild-type cells ( p<0 . 0001 ) ( Figure 8A and B ) . In the similar analysis performed in asynchronously growing cells , accumulation of Spc72-GFP at SPB was significantly higher at all stages of the cell cycle in cells overexpressing CDC5 than in wild type cells ( p<0 . 01 for G1 cells; p<0 . 0001 for S/G2 , G2/M , and anaphase cells ) , with the strongest effect observed in G2/M phase ( Figure 8C and D ) , and cMTs were more prevalent ( Figure 8E and F ) . As a consequence , the SPB positioned closer to the bud neck ( Figure 8G ) and the spindle was at an angle within 30° with respect to the mother-bud axis in 83% of pre-anaphase cells overexpressing CDC5 compared with 48% in wild type cells ( Figure 8—figure supplement 2 ) . These results suggest that the stable SPB association of Spc72 is restricted to the time period where Cdc5 kinase activity is sufficiently high . The mode of cell division by budding represents a type of asymmetric cell division . The mechanism to achieve a high-fidelity of chromosome segregation in such a situation has been a focus of interest and has been investigated intensively in S . cerevisiae . These process was recently studied in the other ascomycetous yeast C . albicans as well as in the yeasts Cryptococcus neoformans and Ustilago maydis in another phylum of fungi , Basidiomycota ( McCully and Robinow , 1972a; McCully and Robinow , 1972b; Kopecká et al . , 2001; Steinberg et al . , 2001; Woyke et al . , 2002; Straube et al . , 2003; Martin et al . , 2004; Finley et al . , 2008 ) . Although the movement of the nucleus during the cell cycle differs between ascomycetous and basidiomycetous yeasts , it is commonly positioned close to the bud neck in both phyla prior to chromosome segregation . We report here that the ascomycetous yeast O . polymorpha does not follow the same strategy . In O . polymorpha , the nucleus generally locates centrally within the mother cell body and the spindle is not aligned properly along mother-bud axis until anaphase onset . Consequently , spindle elongation in early anaphase occurs entirely in the mother often with an inappropriate angle against the polarity axis . Despite this potential complication , one nucleus penetrates successfully into the bud during anaphase , which may largely rely on an immediate correction of the orientation of the spindle and on SPOC activity . Those SPB movements are in contrast to S . cerevisiae in which spindle is aligned during metaphase and therefore SPB translocation into the bud coincides with spindle elongation . Currently molecular mechanism ( s ) that regulate spindle orientation is unknown . However , although the timing of spindle orientation relative to cell cycle progression appears to be different from that of other yeasts , two redundant molecular mechanisms of spindle orientation , one requiring dynein and the other Kar9 , may be conserved in O . polymorpha , because putative orthologs of KAR9 and dynein were identified in O . polymorpha genome sequences ( Li et al . , 1993; Miller and Rose , 1998; Maekawa and Kaneko , 2014; Nordberg et al . , 2014 ) . In S . cerevisiae , Spc72 is stably incorporated into SPBs once it is recruited and organise cMTs throughout the cell cycle . In O . polymorpha , the strong SPB association of OpSpc72 in anaphase becomes weakened as cells enter into the following G1 phase , whereas they re-accumulate later in the cell cycle . The timing of OpSpc72 recruitment to SPBs during early mitosis appears to primarily dictate the organization of cMTs and hence nuclear position ( Figure 9 ) . As the polo-like kinase Cdc5 protein plays an important role in this regulation , a question arises regarding the substrates of Cdc5 kinase in this process ( Archambault and Glover , 2009 ) . Because ScSpc72 binds to and is phosphorylated by Cdc5 in S . cerevisiae , OpSpc72 represents an obvious candidate ( Maekawa et al . , 2007; Snead et al . , 2007 ) . Our electrophoresis analyses indeed detected Cdc5-dependent phosphorylation of Spc72 . However , OpCdc5 failed to phosphorylate recombinant Spc72 in vitro . Further analyses are required to verify whether Cdc5 directly phosphorylates Spc72 and the potential effects of such modifications on the regulation of cMT . Additionally , it may also be important to identify other kinase ( s ) that are responsible for the Cdc5-independent phosphorylation of Spc72 observed in our analyses . Cdc5 may have another important substrates other than Spc72 . Notably , a polo box binding site present in ScSpc72 is missing in OpSpc72 . Cdc5 might therefore phosphorylate other SPB proteins such as Nud1 , which has one site matching the consensus sequences of polo box binding site ( S-Sp/Tp-P ) , and thereby indirectly influence the affinity of Spc72 towards the SPB . It is unclear what brought highly expressed Spc72-GFP to SPBs at early cell cycle stages when Cdc5 activity was low . Weak Spc72-GFP signal in the absence of Cdc5 may be because of low affinity or unstable association to SPBs . Increased Spc72 protein levels may simply result in the increased number of Spc72 protein at SPBs at any given time . Alternatively , overexpression may overcome a negative regulation that normally maintains the low level of Spc72 at SPBs during early stages of the cell cycle ( Figure 4 ) . Such Cdc5-independent regulation is consistent with the observation that Spc72 was gradually lost from SPBs at early stages of the cell cycle . Examining properties of Spc72 protein when overexpressed such as post-translational modifications , protein stability , and molecular dynamics at SPBs would clarify this point . In S . cerevisiae , SPB duplication initiates in late G1 phase by forming a satellite at the distal end of an extended half-bridge of the pre-existing ‘old’ SPB , which is then inserted into the nuclear envelope . In contrast to the old SPB which maintains cMTs from the previous mitosis , this ‘new’ SPB acquires the MT nucleation activity in the inner plaque prior to spindle assembly while it is still connected to the old SPB ( side-by-side SPBs ) . On the other hand , a recent report suggested that the acquisition of ScSpc72 to the outer plaque , and hence of cMTs , occurs only after SPB separation and spindle assembly ( Juanes et al . , 2013 ) . In O . polymorpha , the cMTs acquisition is also regulated at the level of Spc72 recruitment to SPBs . OpSpc72 dissociate from SPBs at the end of mitosis and recruited to both old and new SPBs shortly before anaphase of the following cell cycle . This suggests that the cell cycle regulation of Spc72 recruitment may be applied to both old and new SPBs in O . polymorpha . Even though the Spc72 recruitment is cell cycle regulated in both species , its timing seems to be different: while it occurs in early G2 phase in S . cerevisiae , it does in metaphase in O . polymorpha . Moreover , their regulatory mechanisms are likely different because their acquisition is Cdc5-dependent in O . polymorpha , but not in S . cerevisiae ( Juanes et al . , 2013 ) . Inhibition of Cdc5 kinase in S . cerevisiae causes the misaligned spindle phenotype , which indicates a role of Cdc5 in cMT functions ( Snead et al . , 2007 ) . However , it is unlikely that Cdc5 acts at the level of Spc72 recruitment since both SPBs of the misaligned spindle caused by Cdc5 inhibition carried cMTs ( Snead et al . , 2007 ) . Alternatively , it is possible that the Cdc5-dependent regulation of ScSpc72 function on cMT nucleation/anchoring has been overlooked . Growth suppression of cnm67∆ cells by CDC5 overexpression might point towards this possibility ( Park et al . , 2004 ) . It might also be because of the redundancy with the SPB-anchored Stu2 function ( Usui et al . , 2003 ) . The Stu2 binding domain of Spc72 ( aa176–230 in ScSpc72 , Figure 4—figure supplement 5 ) is conserved in ascomycetous yeasts including O . polymorpha but it differs somewhat in OpSpc72 . In particular , the detached cMTs observed in approximately 14% of O . polymorpha pre-anaphase cells are reminiscent of the spc72∆Stu2 phenotype in S . cerevisiae ( Usui et al . , 2003 ) . Thus , the potential binding of Stu2 to Spc72 in O . polymorpha should be investigated . In O . polymorpha , structure of the cytoplasmic side of SPB , along with cMT nucleation competence , is modified as the cell cycle progresses ( Figure 9 ) . Notably , our BLAST search failed to identify orthologs of several SPB core components identified in S . cerevisiae in genomes of outside of Saccharomycetaceae: the central plaque components Spc42 and Spc29 , the membrane anchors Nbp1 and Bbp1 , and the half-bridge component Kar1 . The failure of identification may be due to the highly divergent nature of amino acid sequences of these coiled-coil proteins , or alternatively the function of these SPB components is not required in O . polymorpha . The absence of these proteins could be one of the reasons underlying the poor appearance of the central plaque and the half-bridge in our electron microscopy anlayses . Our observation that the outer plaque in the cytoplasm was evident in anaphase SPBs but not in G1 SPBs may suggest the outer plaque from the previous mitosis is removed in the following G1 phase . However , given that Nud1 , a putative outer plaque component , is present at the SPB in G1 , it is more likely that the outer plaque is only partially disassembled . Furthermore , the appearance of the outer plaque in EM analysis proceeded that of Spc72 in fluorescent microscopy ( Figure 9 ) . Electron microscopy analysis using another fixation method that better preserves the SPB structure might clarify this point . Equally important is the analysis of the half-bridge which organizes cMTs in G1 in S . cerevisiae . If the half-bridge plays the same role in O . polymorpha SPB as that in S . cerevisiae , the loss of Spc72 from outer plaque alone should not cause the loss of cMTs . In addition , in S . cerevisiae , SPBs are segregated in a defined mode where the old SPB normally migrates into the daughter cell while the new SPB remains in the mother cell ( Pereira et al . , 2001 ) . The SPB history in the outer plaque was proposed to primarily determine the destination of SPBs and to bias spindle asymmetry via Nud1 ( Hotz et al . , 2012; Juanes et al . , 2013 ) . Whether the mode of SPB inheritance is conserved in O . polymorpha and other yeast species is also worthy of further study . It is unlikely that the mode of nuclear positioning in O . polymorpha is ancestral , given that Y . lipolytica and P . pastoris , who are diverted from the common ancestor earlier than O . polymorph , share the same nuclear organization with S . cerevisiae . However , several yeast species relatively close to O . polymorpha or C . albicans exhibited a similar nuclear position in pre-anaphase cells ( Figure 1—figure supplement 1 ) . Thus , the Spc72 recruitment mechanism described in O . polymorpha may be widely utilized in several Clades in Saccharomycotina , including Ogataea , Ambrosiozyma , and Nakazawaea . Yeast strains and plasmids used in this study are listed in Table 1 . Unless otherwise indicated , all O . polymorpha strains were derived from NCYC 495 and were generated by PCR-based methods ( Lu et al . , 2000; Janke et al . , 2004; Saraya et al . , 2012 ) . GFP , mRFP , and 5flag tagged alleles were generated in ku80Δ or ku70Δ cells and then crossed with auxotrophic wild-type strains to obtain KU80+ or KU70+ cells carrying the tagged allele ( Maekawa and Kaneko , 2014 ) . O . polymorpha cells were transformed by electroporation ( Faber et al . , 1994 ) . The 500 bp sequences up- and downstream of the OpTEF1 open reading frame ( ORF ) were used as the OpTEF1 promoter and terminator , respectively and those of the OpCDC28 ORF were used as the OpCDC28 promoter and terminator ( Kiel et al . , 2007 ) . For overexpression of CDC5 , we expressed an N-terminal-truncated version of CDC5 ( CDC5∆53 ) that is equivalent to the ScCDC5∆N70 allele which is resistant to APC-dependent ubiquitination ( Shirayama et al . , 1998 ) . The CDC5∆53 ORF was amplified by PCR and inserted into pHM949 . The resulting plasmid pHM950 was digested and a zeocin resistance marker was inserted ( pHM956 ) . To obtain the dyn1∆ strain , tetrad dissection plates were incubated at room temperature for >5 days until colonies were formed because the dyn1∆ cells grew extremely slowly . Glycerol stocks were prepared from the initial master plate of tetrad analysis . YPDS liquid medium was inoculated with either the initial colonies from tetrad dissection or glycerol stocks , and the resulting cells were subjected to analyses . Yeast strains were grown either in YPD medium containing 200 mg/l adenine , leucine , and uracil ( YPDS ) or in synthetic/defined ( SD ) medium supplemented with appropriate amino acids and nucleotides ( Sherman , 1991 ) . Cells were grown at 30°C unless otherwise indicated . To depolymerize MTs , cells were incubated in either YPDS medium or SD medium containing 1 . 5 μg/ml nocodazole at 30°C for 1 . 5 hr . Yeast cells carrying GFP-TUB1 , HHT1-mCherry , SPC72-GFP , GFP-NUD1 , SFI1-GFP , SPC98-GFP , TUB4-GFP , SPC110-GFP , CDC5-GFP , or MPS3-mRFP were immediately analysed by fluorescence microscopy without washing or fixation in Figures 1A , 2 , 4 , 5 , 6B–E , 7 and 8A , Figure 4—figure supplement 1 , Figure 5—figure supplement 1 . For the visualisation of DNA with 4'6 , -diamidino-2-phenylindole ( DAPI ) , cells were fixed with 70% ethanol , washed with phosphate buffered saline ( PBS ) , and incubated in PBS containing 1 µg/ml DAPI . Z-series images of 0 . 4 μm steps were captured with DeltaVision ( Applied Precision , Issaquah , WA , USA ) equipped with GFP and TRITC filters ( Chroma Technology Corp . , Bellows Falls , VT , USA ) , a 100 × NA 1 . 4 UPlanSApo oil immersion objective ( IX71; Olympus , Tokyo , Japan ) , and a camera ( CoolSNAP HQ; Roper Scientific , Trenton , NJ , USA ) and were quantified/processed with SoftWoRx 3 . 5 . 0 ( Applied Precision , Issaquah , WA , USA ) or Prism4 . 3 . 0 software ( Chen et al . , 1992; Chen et al . , 1996 ) . Deconvolved and projected images are shown . The fluorescence intensity of Spc72-GFP was measured on a plane that has an SPB in focus . Time-lapse experiments of Figures 1A and 2D and that of Figure 4C were carried out in YPDS and SD complete medium respectively on a glass-bottom dish ( MatTek , Ashland , MA , USA ) coated with concanavalin A ( 037–08771 , Wako , Japan ) at room temperature . Z series at 0 . 4 μm steps were acquired every 3 min for Figures 1A and 4C , or every 30 s for Figure 2D . For Figure 1B , cells were fixed with 70% ethanol , washed with PBS , and incubated in PBS containing 1 mg/ml DAPI to visualize the DNA ( Maekawa and Kaneko , 2014 ) . ImageJ 1 . 47 ( NIH , Bethesda , MD , USA ) and Photoshop ( Adobe Systems , San Jose , CA , USA ) were used to mount the images and to produce merged colour images . No manipulations other than contrast and brightness adjustments were used . To exclude cells that were non-proliferating from the GFP intensity measurements in Figure 4B , cells were first incubated in YPDS containing Alexa 594 conjugated concanavalin A ( Thermo Fisher Scientific , Waltham , MA , USA ) until all cells were labelled and then washed once with YPDS and incubated in YPDS for 1 hr prior to image capture . Cells that had lost the label or had a bud with no label were subject to the analyses . Cells were mounted on a glass-bottom dish ( MatTek ) coated with concanavalin A and covered with fixative [2% glutaraldehyde in 0 . 1 M sodium phosphate buffer ( pH 7 . 2 ) ] . After 1 min , cells were further fixed with fresh fixative for 2 hr at 4°C . After washing with buffer , low melting agarose was applied onto the cells to prevent loss of cells during subsequent procedures . Zymolyase solution ( 0 . 4 mg⁄ml zymolyase 100T , Seikagaku Co . , Tokyo , Japan ) was applied on top of the agarose for 60 min at 37°C , postfixed with 2% OsO4 for 2 hr at room temperature , stained with 1% uranyl acetate for 1 hr , dehydrated with acetone in an ascending series from 50% to 100% , and embedded in epoxy resin . Serial sections of 80 nm thicknesses were obtained , poststained with uranyl acetate and lead citrate , and analysed using a Zeiss EM900 Transmission Electron Microscope at Central Unit Electron Microscopy in the German Cancer Research Center ( DKFZ ) ( ZEISS , Oberkochen , Germany ) or a Hitachi H-7500 Transmission Electron Microscope at Research Centre for Ultra-High Voltage Electron Microscopy at Osaka University ( Hitachi , Tokyo , Japan ) . For synchronization , CDC5-3mAID cells were incubated in YPDS containing indole-3-acetic acid ( IAA ) ( 45533 , Sigma-Aldrich , St . Louis , MO , USA ) ) for 2 . 5 hr at 30°C until >80% of cells had a large sized bud to deplete Cdc5 ( Nishimura and Kanemaki , 2014 ) . Cells were then washed with pre-warmed growth medium to remove IAA and re-suspended in YPAD medium at 30°C . Whole cell extracts were prepared for SDS-PAGE and immunoblotting ( Knop et al . , 1999; Janke et al . , 2004; Meitinger et al . , 2016 ) . Samples representing 1–2 OD600 of liquid culture were resuspended in 950 μl 0 . 29 M NaOH and incubated on ice for 10 min . Then , 150 μl 55% ( w/v ) trichloroacetic acid was added and the solutions were mixed and incubated for 10 min on ice . After centrifugation the supernatant was removed . The protein pellet was resuspended in high urea buffer ( 8 M urea , 5% SDS , 200 mM NaPO3 pH 6 . 8 , 0 . 1 mM EDTA , 100 mM dithiothreitol , and bromophenol blue ) and heated at 65°C for 10 min . A sample comprising one-fifth of the total sample amount was loaded for SDS-PAGE ( Figure 6A ) and western blotting was performed using a standard protocol . For immunoprecipitation , total cell extracts were prepared from logarithmically growing cells in immunoprecipitation buffer ( 100 mM Tris , pH 8 . 0 , 10 mM EDTA , 150 mM NaCl , 5% glycerol , 0 . 2 mM NaVO3100 mM β-glycerophosphate , 50 mM NaF , 1 mM PMSF , 1 mM DTT , 1% NP-40 , and Complete EDTA-free protease inhibitor cocktail [Roche] ) . 10 mg of total cell extract was incubated with M2-bound magnetic beads ( M8823 , Sigma ) for 2 hr at 4°C . The beads were washed three times with immunoprecipitation buffer . The bound proteins were subjected to λ phosphatase treatment and then eluted in 30 μl of SDS-PAGE sample buffer by incubated at 37°C for 30 min . 3 μl of eluates were loaded on a Mini-PROTEAN TGX Precast Gels ( 4561021 , BIO-RAD Laboratories , Hercules , CA , USA ) and western blotting was performed using a standard protocol . Monoclonal antibodies JL-8 ( 632381 , TaKaRa Bio Clontech , Shiga , Japan ) and M2 ( F1804 , Sigma ) were used to detect GFP- and flag-tagged proteins respectively . Plot profile function of ImageJ was used to plot intensity value across a line in Figure 6—figure supplement 2 . Cells were arrested for 2 . 5 hr with 1 . 5 µg/ml nocadazole and fixed for 15 min in 4% paraformaldehyde/2% sucrose in phosphate-buffered saline ( PBS ) solution followed by extensive washing in PBS . The cells were immobilized on a concanavalin A ( Sigma-Aldrich , MO , USA ) - coated 35 mm glass bottom dish ( MatTek , P35G-1 . 5–14C ) and maintained in PBS for the duration of the imaging process in PBS . The samples were imaged in the 2D-SIM mode on a Nikon N-SIM system ( Tokyo , Japan ) equipped with a TIRF Apochromat 100x/1 . 49 NA oil immersion objective and a single photon detection EM-CCD camera ( Andor iXon3 DU-897E; Belfast , UK ) . The 488 nm and 561 nm laser lines were used for excitation of yeGFP and tdTomato , respectively , combined with emission band pass filter 520/45 and 610/60 . Images were taken sequentially within a small z-stack and in consideration of imaging SPBs close to the coverslip to minimise spherical aberrations . Subsequently the reconstruction and channel alignment was performed using the NIS imaging and image analysis software ( Nikon ) . For the xyz chromatic shift correction we used in a reference sample tetraspeck beads in a reference sample . All images show a single stack of the z-slices .

Before a cell divides , it needs to duplicate its genetic material to provide the new daughter cell with a full set of genetic information . To do so , the cell forms a complex of proteins called the spindle apparatus , which is made up of string-like microtubules that divide the chromosomes evenly . In many organisms , the position of the spindle determines where in the cell this separation happens . However , in baker’s yeast , the location where the cell will divide is determined well before the spindle is formed . Unlike many other eukaryotic cells , these yeast cells divide asymmetrically and create buds that will form the new daughter cells . The position of this bud determines where the spindle should be located and where the chromosomes separate . The spindle itself is then organised by a structure called the spindle pole body , which connects to microtubules inside the cell nucleus and microtubules in the cell plasma . Several proteins control where and how the spindle forms , including a protein called the spindle pole component 72 , or Spc72 for short , and an enzyme called Cdc5 . However , until now it was unclear how spindle formation is timed and controlled in other yeast species . Now , Maekawa et al . have used fluorescent markers and time lapse microscopy to examine how the spindle forms in the yeast species Ogataea polymorpha , an important industrial yeast used to produce medicines and alcohol . The results show that in O . polymorpha , the positioning and orientation of the spindle only occurred very late in the cell cycle and the microtubules in the cell plasma remained unstable until the chromosomes were about to separate . This was linked to changes in the level of Spc72 , which increased at the spindle pole body before the chromosomes separated and then dropped again . This was controlled by Cdc5 . Understanding when and where microtubules are formed is an important step in understanding how cells divide . This is the first example of a budding yeast that creates new microtubules in the cell plasma every time the cell divides . Unravelling the molecular differences between yeast species could lead to new ways to optimise the use of industrial yeasts like O . polymorpha , or to combat disease-causing ones .

2017

Polo-like kinase Cdc5 regulates Spc72 recruitment to spindle pole body in the methylotrophic yeast Ogataea polymorpha

Centrioles play critical roles in organizing the assembly of the mitotic spindle and templating the formation of primary cilia . Centriole duplication occurs once per cell cycle and is regulated by Polo-like kinase 4 ( PLK4 ) . Although significant progress has been made in understanding centriole composition , we have limited knowledge of how PLK4 activity controls specific steps in centriole formation . Here , we show that PLK4 phosphorylates its centriole substrate STIL on a conserved site , S428 , to promote STIL binding to CPAP . This phospho-dependent binding interaction is conserved in Drosophila and facilitates the stable incorporation of both STIL and CPAP into the centriole . We propose that procentriole assembly requires PLK4 to phosphorylate STIL in two different regions: phosphorylation of residues in the STAN motif allow STIL to bind SAS6 and initiate cartwheel assembly , while phosphorylation of S428 promotes the binding of STIL to CPAP , linking the cartwheel to microtubules of the centriole wall . Centrioles are microtubule-based structures that recruit a surrounding pericentriolar material ( PCM ) to form the centrosome ( Nigg and Holland , 2018; Gönczy , 2012 ) . Centrosomes nucleate the formation of the microtubule cytoskeleton in interphase cells and form the poles of the mitotic spindle during cell division . In quiescent cells , centrioles dock at the plasma membrane and act as basal bodies that template the formation of cilia and flagella ( Breslow and Holland , 2019 ) . Cycling cells tightly couple centriole biogenesis with cell cycle progression . Centriole duplication begins at the G1-S phase transition when a new procentriole grows perpendicularly from a single site at the proximal end of each of the two parent centrioles . In late G2 phase , the two centriole pairs separate and increase PCM recruitment to promote the formation of the mitotic spindle . At the end of mitosis , the centrosomes are equally partitioned so that each daughter cell inherits a pair of centrioles . Defects in centriole biogenesis can result in the formation of supernumerary centrosomes which promote mitotic errors that can contribute to tumorigenesis ( Levine et al . , 2017; Levine and Holland , 2018; Basto et al . , 2008; Serçin et al . , 2016; Coelho et al . , 2015 ) . Moreover , mutations in centriole proteins are linked to growth retardation syndromes and autosomal recessive primary microcephaly ( MPCH ) in human patients ( Nigg and Raff , 2009; Chavali et al . , 2014 ) . The initiation of centriole duplication requires a conserved set of five core proteins: PLK4 ( ZYG-1 in C . elegans ) , CEP192 ( SPD-2 in C . elegans and Spd-2 in Drosophila ) , CPAP ( also known as CENPJ , SAS-4 in C . elegans and Sas-4 in Drosophila ) , STIL ( SAS-5 in C . elegans and Ana-2 in Drosophila ) and SAS6 ( Leidel et al . , 2005; Leidel and Gönczy , 2003; Dammermann et al . , 2004; Delattre et al . , 2004; Kemp et al . , 2004; O'Connell et al . , 2001; Pelletier et al . , 2004; Kirkham et al . , 2003 ) . Of these components , PLK4 has been identified as the central regulator of centriole assembly ( Habedanck et al . , 2005; Bettencourt-Dias et al . , 2005 ) . In mammalian cells , PLK4 is recruited to the centriole during G1 phase through binding to its centriole receptors CEP152 and CEP192 , which encircle the proximal end of the parent centriole ( Cizmecioglu et al . , 2010; Hatch et al . , 2010; Kim et al . , 2013; Park et al . , 2014; Sonnen et al . , 2013 ) . At G1-S phase transition , PLK4 transforms from a ring-like localization to a single focus on the wall of the parent centriole that marks the site of procentriole formation ( Kim et al . , 2013; Ohta et al . , 2014; Sonnen et al . , 2012; Dzhindzhev et al . , 2017 ) . This transition bears features of a symmetry breaking reaction and can be recreated in silico using two positive feedback loops that act on PLK4 ( Goryachev and Leda , 2017; Leda et al . , 2018 ) . Binding of PLK4 to its centriole substrate STIL promotes activation of the kinase and is required for its ring-to-dot transformation ( Ohta et al . , 2014; Moyer et al . , 2015; Lopes et al . , 2015 ) . PLK4 phosphorylates STIL in a conserved STAN motif to promote the binding and recruitment of SAS6 ( Ohta et al . , 2014; Moyer et al . , 2015; Dzhindzhev et al . , 2014; Kratz et al . , 2015 ) . SAS6 homo-oligomerizes to organize the central cartwheel , a stack of ring-like assemblies with nine-fold symmetry that provides the structural foundation for the procentriole ( Kitagawa et al . , 2011a; van Breugel et al . , 2011; van Breugel et al . , 2014; Cottee et al . , 2015; Guichard et al . , 2017 ) . Following cartwheel assembly , the centriole protein CPAP plays a critical role in the formation and stabilization of the triplet microtubule blades that make up the procentriole wall . CPAP interacts with multiple centriole proteins including STIL ( Cottee et al . , 2013; Tang et al . , 2011; Vulprecht et al . , 2012 ) , CEP152 ( Cizmecioglu et al . , 2010; Dzhindzhev et al . , 2010 ) , CEP120 ( Lin et al . , 2013a ) , CEP135 ( Lin et al . , 2013b ) , and Centrobin ( Gudi et al . , 2015 ) . The C-terminal TCP domain of CPAP directly binds to a highly conserved PRP motif in STIL ( Cottee et al . , 2013; Tang et al . , 2011; Vulprecht et al . , 2012 ) , while the N-terminal domain of CPAP interacts with α/β-tubulin heterodimers ( Sharma et al . , 2016; Zheng et al . , 2016; Hung et al . , 2004 ) . These interactions allow CPAP to act as a molecular link between the cartwheel and the triplet microtubule blades of the centriole wall . Importantly , an MCPH mutation in the CPAP TCP domain weakens the STIL-CPAP interaction , highlighting the importance of this complex in centriole assembly and function ( Cottee et al . , 2013; Tang et al . , 2011; Bond et al . , 2005 ) . CPAP positively regulates centriolar microtubule growth and , consequently , overexpression of CPAP leads to the formation of overly long centrioles in human cells ( Tang et al . , 2009; Schmidt et al . , 2009; Kohlmaier et al . , 2009 ) . In addition to its role in controlling microtubule growth , CPAP also functions in recruiting PCM , either through the direct tethering of PCM proteins or by recruiting Plk1/Polo which is critical in promoting PCM assembly ( Zheng et al . , 2014; Gopalakrishnan et al . , 2011; Chou et al . , 2016; Novak et al . , 2016 ) . At present , the binding of SAS6 to the STAN motif of STIL is the only interaction known to be controlled by PLK4 kinase activity . While the regulation of this assembly step is conserved in humans and flies ( Ohta et al . , 2014; Moyer et al . , 2015; Dzhindzhev et al . , 2014; Kratz et al . , 2015 ) , it is unclear whether this takes place in C . elegans , where ZYG-1/PLK4 directly binds and recruits SAS6 to promote cartwheel assembly ( Lettman et al . , 2013 ) . Moreover , although phosphomimetic mutations in the crucial PLK4 phosphorylation sites in the STAN motif of STIL are functional , they cannot support centriole duplication in the absence of PLK4 kinase activity ( Kim et al . , 2016 ) . This suggests that PLK4 must phosphorylate STIL , or other substrates , at additional sites to promote centriole assembly . Indeed , recent work in Drosophila identified additional PLK4 phosphorylation sites required for centriole biogenesis in the N-terminus of Ana2/STIL , but exactly how these phosphorylation events contribute to centriole formation remains unclear ( Dzhindzhev et al . , 2017; McLamarrah et al . , 2018 ) . In this manuscript , we identify a conserved PLK4 phosphorylation site on STIL that promotes binding to CPAP in vitro and in vivo . This phospho-dependent binding interaction is conserved in flies and allows STIL to link the growing cartwheel to the outer microtubule wall of the centriole . Together , our findings offer insight into a novel step in centriole assembly that is regulated by PLK4 kinase activity . PLK4 phosphorylates conserved residues in the STIL STAN motif to promote binding to SAS6 ( Ohta et al . , 2014; Moyer et al . , 2015; Dzhindzhev et al . , 2014 ) . To determine whether phosphorylation of STIL by PLK4 might affect the interaction of STIL with other components of the centriole duplication machinery , we tested the ability of Myc-GFP-STIL to interact with its known centriolar binding partners in the presence of kinase active ( PLK4WT ) or kinase dead ( PLK4KD ) PLK4 . Active PLK4 triggers its own degradation and thus , we used a PLK4∆24 mutant that stabilizes the active kinase by preventing PLK4-induced autodestruction ( Holland et al . , 2010 ) . Expression of kinase active PLK4∆24-mCherry increased the binding of STIL to SAS6 in cells ( Figure 1A ) , but did not increase binding to the STIL-interacting partners RTTN ( Chen et al . , 2017 ) or CEP85 ( Figure 1B , C ) ( Liu et al . , 2018 ) . Unexpectedly , we observed that PLK4 kinase activity promoted a robust increase in STIL binding to CPAP , suggesting that PLK4 kinase activity also controls the interaction of CPAP with STIL ( Figure 1D ) . To determine how PLK4 phosphorylation promotes binding of CPAP to STIL , we mapped in vitro PLK4 phosphorylation sites on STIL using mass spectrometry . Recombinant full-length GST-STIL was phosphorylated with the His-PLK4 kinase domain in vitro . Of the 84 in vitro phosphorylation sites we identified on STIL , S428 was of particular interest as it is highly conserved , matches the PLK4 consensus phosphorylation sequence and is positioned close to the known CPAP binding region on STIL ( Figure 2A , Figure 2—figure supplement 1 ) ( Cottee et al . , 2013; Kettenbach et al . , 2012; Johnson et al . , 2007; Hatzopoulos et al . , 2013 ) . To determine if phosphorylation of STIL S428 was responsible for enhancing the binding of CPAP to STIL , we co-expressed FLAG-CPAP and a wild type ( WT ) or S428A mutant of Myc-GFP-STIL in the presence of kinase active or inactive PLK4∆24-mCherry . The expression of kinase active PLK4 promoted a > 7 fold increase in the amount of CPAP bound to WT STIL , but this increased binding was not observed with STIL S428A ( Figure 2B ) . To test if this phospho-regulated binding interaction can be reconstituted with purified components , we performed GST-pull down experiments on recombinant WT or S428A GST-STIL that had been phosphorylated with the His-PLK4 kinase domain and then incubated with a recombinant Flag-CPAP TCP domain . Phosphorylation of WT GST-STIL with PLK4 increased binding to the Flag-TCP domain by ~2 . 5 fold , but this increased binding was not observed with STIL S428A ( Figure 2C ) . The use of the CPAP TCP domain rather than full-length protein may explain the more modest increase in CPAP binding to STIL in vitro compared to in vivo . These data show that phosphorylation of STIL S428 promotes CPAP binding to STIL in vitro and in vivo . To demonstrate STIL S428 is a bona fide PLK4 phosphorylation site , we raised a phospho-specific antibody to this site . The affinity-purified pS428 antibody recognized recombinant GST-STIL in the presence of ATP and the His-PLK4 kinase domain , but not in the absence of ATP ( Figure 2D ) . Moreover , recognition of phosphorylated GST-STIL by the pS428 antibody was abolished by the S428A mutation , demonstrating the specificity of the pS428 antibody ( Figure 2D ) . To determine if PLK4 can phosphorylate STIL S428 in cells , we co-expressed WT or a S428A mutant of Myc-GFP-STIL with kinase active or inactive PLK4∆24-mCherry ( Figure 2E ) . The pS428 antibody recognized WT STIL in the presence of kinase active , but not kinase inactive , PLK4 , showing that PLK4 phosphorylates STIL at S428 in vitro and in vivo . D . melanogaster PLK4 ( DmPLK4 ) was recently shown to phosphorylate the STIL homolog Ana2 at S38 , a residue equivalent to S428 in the human STIL ( Dzhindzhev et al . , 2017; McLamarrah et al . , 2018 ) ( Figure 2A ) . Phosphorylation of S38 was shown to be required for Ana2 recruitment ( Dzhindzhev et al . , 2017 ) but was reported to not alter Ana2 binding to the CPAP orthologue DmSas4 ( McLamarrah et al . , 2018 ) . We reasoned that this discrepancy might arise because PLK4 is a low-abundance protein that is activated at the centriole and is unable to efficiently phosphorylate a significant fraction of the transfected Ana2 . To test this possibility , we transfected D . melanogaster S2 cells with either WT or S38A Myc-GFP-Ana2 and Flag-DmSas4 in the presence or absence of a stabilized version of DmPLK4SBM-mCherry ( Rogers et al . , 2009 ) . DmPLK4SBM promoted a > 6 fold increase in the amount of DmSas4 bound to WT Ana2 , but this increased binding was not observed with Ana2 S38A ( Figure 2F ) . These data suggest that the increased binding of CPAP/DmSas4 to phosphorylated STIL/Ana2 is conserved between human and flies . To test the requirement of STIL S428 phosphorylation by PLK4 for centriole biogenesis , we integrated doxycycline-inducible , siRNA-resistant , WT or S428A Myc-GFP-STIL transgenes a pre-defined genomic locus in a DLD-1 host cell line . As a control , we also generated DLD-1 cells expressing a Myc-GFP-STIL S1116A transgene , which contains a mutation at a conserved PLK4 phosphorylation site in the STIL STAN motif required for efficient binding to SAS6 ( Ohta et al . , 2014; Moyer et al . , 2015; Dzhindzhev et al . , 2014 ) . WT , S1116A and S428A Myc-GFP-STIL transgenes were all expressed to similar levels in cells ( Figure 3—figure supplement 1A , B ) , and at a level ~2–3 fold higher than endogenous STIL . Depletion of STIL by siRNA for 48 hours resulted in 100% of mitotic cells with ≤2 centrioles , and this effect was almost completely rescued by expression of the WT STIL transgene ( Figure 3A , Figure 3—figure supplement 1C ) without promoting substantial centriole overduplication . By contrast , expression of either STIL S428A or S1116A only led to a partial rescue of centriole duplication ( 61% S428A cells and 70% S1116A cells contain ≤2 centrioles in mitosis ) ( Figure 3A ) . Importantly , preventing S428 phosphorylation did not affect the ability of STIL to bind to or stimulate PLK4 kinase activity , suggesting that a failure to activate PLK4 kinase activity is not responsible for the failure of centriole duplication ( Figure 3—figure supplement 2A ) ( Moyer et al . , 2015 ) . These data show that phosphorylation of STIL S428 by PLK4 promotes centriole duplication . CPAP interacts with STIL via a short proline-rich region containing a highly conserved PRxxPxP motif ( Figure 2A ) ( Cottee et al . , 2013 ) . To test whether defective CPAP binding causes the failure to rescue centriole duplication with STIL S428A , we created a PRP mutant of STIL by mutating the conserved PRPIPSP CPAP binding motif to AAPIASP ( P404A , R405A , P408A ) . This mutation did not affect PLK4’s ability to phosphorylate STIL on S428 ( Figure 3—figure supplement 2B ) . As expected , the STIL PRP mutant showed impaired binding to CPAP ( Figure 3E and Figure 3—figure supplement 2B ) . Expression of Myc-GFP-STIL PRP led to only a partial rescue of centriole duplication in cells depleted of endogenous STIL by siRNA , similar to that of the STIL S428A mutation ( Figure 3A ) . Combining the S428A or PRP mutation with the S1116A mutation in the STIL STAN motif that impairs binding to SAS6 prevented any rescue of centriole duplication ( Figure 3A , B ) . By contrast , a Myc-GFP-STIL transgene containing both the S428A and PRP mutation rescued centriole duplication to a level similar to that observed with a STIL transgene that contained either mutation on its own ( Figure 3A , B ) . Collectively , these data suggest that S428 phosphorylation and the PRxxPxP motif of STIL function in the same pathway to promote CPAP binding to STIL , and that they both are in a separate pathway from mutations that disrupt SAS6 binding to STIL . To evaluate the impact of disrupting CPAP and SAS6 binding on the centriole targeting of STIL , we measured the levels of Myc-GFP-STIL transgenes at the centriole in S/G2 phase cells depleted of endogenous STIL . Preventing phosphorylation at S428 reduced the localization of Myc-GFP-STIL by 28% , compared to that of the WT transgene ( Figure 3C , D ) . Preventing phosphorylation at S1116 in the STAN motif reduced the abundance of STIL at the centriole by ~2 fold , as previously reported ( Figure 3C ) ( Moyer et al . , 2015 ) . However , preventing phosphorylation of both sites reduced the centriole localization of STIL by 80% , suggesting that stable incorporation of STIL into the centriole requires strong binding to both CPAP and SAS6 ( Figure 3B , C , D , Figure 3—figure supplement 3 ) . Combining the S428A and PRP mutations did not diminish the levels of centriolar STIL below that observed with either mutation alone ( Figure 3C ) . This provides further evidence that the S428A and PRP mutations function in the same pathway and act to reduce the stability of CPAP binding to STIL . Experiments in flies showed that DmPLK4 phosphorylates S38 to promote Ana2 recruitment to the centriole and then phosphorylates conserved residues in the STAN motif to enable SAS6 recruitment ( Dzhindzhev et al . , 2017; McLamarrah et al . , 2018 ) . However , we found that phosphorylation of STIL S428 in human cells does not abolish the centriolar recruitment of STIL ( Figure 3C , D ) . To determine whether phosphorylation of the STAN motif requires phosphorylation of STIL S428 , or vice versa , we monitored phosphorylation of STIL S428 and S1116 using phospho-specific antibodies . Expression of kinase active PLK4∆24-mCherry promoted phosphorylation of a Myc-GFP-STIL transgene at both S428 and S1116 , and mutation of either site did not prevent phosphorylation of the other ( Figure 3—figure supplement 4A ) . To test if PLK4-mediated phosphorylation of the STIL STAN motif at the centriole requires phosphorylation of S428 , we monitored phosphorylation of S1108 in the STAN motif using a phospho-specific antibody ( Moyer et al . , 2015 ) . Although treatment with the PLK4 inhibitor centrinone abolished STIL S1108 phosphorylation , the S428A or PRP motif mutation did not affect phosphorylation of STIL S1108 ( Figure 3—figure supplement 4B ) . These data show that phosphorylation of the STIL STAN motif by PLK4 does not require prior phosphorylation of STIL S428 . To examine how S428 phosphorylation affects the binding dynamics of centriolar STIL , we performed Fluorescence Recovery after Photobleaching ( FRAP ) in cells depleted of endogenous STIL and expressing Myc-GFP-STIL transgenes . Myc-GFP-STIL WT partially recovered following bleaching , showing that STIL exists in both mobile and immobile pool at the procentriole ( Figure 4A , Myc-GFP-STIL WT , recovery percentage ( R% ) =38% ) . Consistent with previous observations , mutation of the S1116 phosphorylation site increased the mobile fraction of centriolar STIL ( Figure 4A , Myc-GFP-STIL S1116A , R% = 71% ) ( Moyer et al . , 2015 ) . Importantly , the S428A and PRP mutants of STIL also showed an increased recovery of centriolar STIL ( Figure 4A , Figure 4—figure supplement 1 ) ; Myc-GFP-STIL S428A , R% = 57%; Myc-GFP-STIL PRP , R% = 58% ) , suggesting that CPAP binding allows more stable incorporation of STIL into the procentriole . To understand how mutations in STIL affect centrosomal CPAP dynamics , we monitored centrosomal GFP-CPAP turnover using FRAP by knocking down endogenous CPAP and expressing a siRNA-resistant Myc-GFP-CPAP transgene . As previously reported , Myc-GFP-CPAP partially recovered after photobleaching ( Figure 4—figure supplement 2A ) , Myc-GFP-CPAP , R% = 51% ) ( Kitagawa et al . , 2011b ) . Surprisingly , performing the same measurements in cells depleted of STIL led to an almost complete turnover of Myc-GFP-CPAP ( Figure 4—figure supplement 2A , Myc-GFP-CPAP , STIL siRNA , R% = 96% ) . While STIL is uniquely localized to the procentriole , CPAP is present at both the parent centriole and procentriole . In addition , CPAP has been reported to localize in the PCM material ( Sonnen et al . , 2012 ) , consistent with a proposed role in recruiting PCM proteins ( Zheng et al . , 2014; Gopalakrishnan et al . , 2011; Chou et al . , 2016; Novak et al . , 2016 ) . Since we bleach all of the pools of CPAP in the FRAP experiments , we cannot distinguish which centrosomal populations of CPAP are dynamic and which are stably bound . Nevertheless , given STIL localizes exclusively to the procentriole , one interpretation of our data is that STIL is required for the stable incorporation of CPAP at the procentriole , while the parental centriole and PCM pool of CPAP are dynamic and display a more transient association with the centrosome . To test the role of STIL S428 phosphorylation in modulating CPAP turnover at the centrosome , we integrated into DLD1 cells a Myc-STIL-T2A-GFP-CPAP transgene in which siRNA-resistant Myc-STIL and siRNA-resistant GFP-CPAP were both expressed from the same doxycycline-inducible promoter . As expected , Myc-STIL expression significantly suppressed the increased turnover of GFP-CPAP observed in cells depleted of endogenous CPAP and STIL ( Figure 4—figure supplement 2A , GFP-CPAP , Myc-STIL WT background , R% = 50% ) . However , mutation of S428 or the PRP motif on STIL increased the turnover of GFP-CPAP compared with WT STIL ( Figure 4B and Figure 4—figure supplement 2B , GFP-CPAP , Myc-STIL S428A background , R% = 69%; GFP-CPAP , Myc-STIL PRP background , R% = 67% ) . Expression of Myc-STIL S1116A did not increase GFP-CPAP turnover , indicating that the increase in GFP-CPAP turnover in a Myc-STIL S428A background reflects a specific defect in the STIL/CPAP interaction ( Figure 4B , GFP-CPAP , Myc-STIL S1116A background , R% = 55% ) . These data suggest that STIL binding allows a more stable incorporation of CPAP into the centrosome , possibly by facilitating interactions with CPAP at the procentriole . However , since the depletion of STIL resulted in a higher level of CPAP turnover than specifically disrupting the STIL/CPAP interaction , it is likely STIL recruits additional proteins that collectively act to stabilize the incorporation of CPAP into the centrosome . Together , our data show that the interaction of CPAP with STIL allows both proteins to incorporate more stably into the centrosome . To better understand the requirement of the STIL/CPAP interaction in centriole duplication , we constructed an RNAi-replacement system in DLD-1 cells where endogenous CPAP was depleted by siRNA and replaced with near physiological levels ( ~2–3 fold higher ) of an siRNA-resistant Myc-GFP-CPAP transgene ( Figure 5A , Figure 5B ) . Knockdown of CPAP by siRNA resulted in 81% of mitotic cells with ≤2 centrioles , and this was largely rescued by expression of the CPAP WT transgene ( Figure 5C ) without promoting centriole overduplication . By contrast , expression of a CPAP transgene with mutations in the TCP domain that reduced binding to STIL ( F1229A or E1235K ) , led to a partial rescue of centriole duplication ( 58% of F1229A cells and 48% of E1235K cells contain ≤2 centrioles in mitosis ) ( Figure 5C , Figure 5D ) ( Cottee et al . , 2013; Hatzopoulos et al . , 2013 ) . Importantly , the presence of these TCP domain mutations increased the turnover of Myc-GFP-CPAP at the centrosome ( Figure 5E , Myc-GFP-CPAP R% = 51%; Myc-GFP-CPAP F1229A , R% = 76%; Myc-GFP-CPAP E1235K , R% = 67% ) , but did not alter the abundance of the Myc-GFP-CPAP or STIL at the centrosome ( Figure 5F–H ) . Collectively , these data support the conclusion that the STIL-CPAP interaction facilitates the stable centrosomal integration of CPAP , but that this interaction does not have a major impact on the level to which CPAP accumulates at the centrosome . To further test whether the overall level of CPAP present at the centrosome depends on STIL , we measured centrosomal CPAP levels in S/G2 cells depleted of STIL by siRNA . STIL knockdown reduced the level of CPAP at the centrosome by ~30% ( Figure 5—figure supplement 1A–C ) . Importantly , we observed that centrosomal CPAP levels were similar following depletion of endogenous STIL and expression of a WT or mutant Myc-GFP-STIL transgene ( Figure 5—figure supplement 1D , E ) . This suggests a model in which CPAP is localized to the parent centriole independently of STIL , while procentriole localized CPAP requires STIL for stable binding . However , our data argue that the role of STIL in recruiting CPAP to the centrosome is largely independent of the STIL/CPAP interaction and likely depends on the recruitment of other proteins . This is supported by our FRAP analysis which showed that depletion of STIL increased the turnover of CPAP at the centriole from 51% to 96% , while disruption of STIL binding to CPAP increased CPAP turnover to only ~70% . We conclude that the recruitment of the majority of CPAP present at the centrosome does not require CPAP binding to STIL . The multiple populations of CPAP present at the centrosome prevented us from specifically testing the requirement of STIL S428 phosphorylation for CPAP recruitment to the procentriole . To analyze the role of STIL S428 phosphorylation in recruiting CPAP specifically to assembling procentrioles , we induced the formation of freestanding de novo centrioles in cells expressing various STIL mutants . DLD-1 cells expressing a Myc-GFP-STIL transgene were chronically treated with the PLK4 inhibitor centrinone to remove centrioles ( Wong et al . , 2015 ) . Acentriolar cell lines were then depleted of endogenous STIL using siRNA for twenty-four hours , and centrinone was removed to induce the formation of freestanding de novo centrioles ( Figure 6A ) . De novo centrioles were defined as foci marked by both PLK4 and Centrin ( Figure 6—figure supplement 1 ) . As expected , depletion of STIL suppressed de novo centriole assembly , and this was rescued by expression of a WT Myc-GFP-STIL transgene ( 48% and 87% of cells expressing WT STIL contained de novo centrioles at 24 and 72 hr after centrinone washout , respectively ) ( Figure 6B ) . Acentriolar cell lines expressing a S428A , S1116A , or PRP Myc-GFP-STIL transgene were all deficient in assembling de novo centrioles , with only 25% , 12% , and 33% of cells containing PLK4 and Centrin positive foci at 72 hr after centrinone removal , respectively . While cells expressing S1116A Myc-GFP-STIL formed very few GFP-STIL/PLK4 foci , the number of foci observed in cells expressing S428A and PRP Myc-GFP-STIL was comparable to that observed with the WT STIL transgene ( Figure 6C , Figure 6—figure supplement 1 ) . This suggests that cells expressing the S428A and PRP mutant STIL fail centriole assembly at a later stage than those expressing the S1116A mutant of STIL . To determine why the S428A and PRP mutations fail de novo centriole formation after forming STIL/PLK4 foci , we measured the recruitment of SAS6 and CPAP to the newly formed GFP-STIL/PLK4 foci . WT Myc-GFP-STIL recruited both CPAP and SAS6 to as expected ( Figure 6D–H ) . While recruitment of SAS6 was identical in WT , S428A , and PRP Myc-GFP-STIL , the S428A and PRP mutations resulted in a > 90% reduction in the amount of CPAP recruitment to the GFP-STIL/PLK4 foci ( Figure 6D–H ) . Together , these data suggest that S428A and PRP mutant STIL bind to SAS6 and assemble a cartwheel but fail to recruit CPAP to complete the formation of de novo centrioles . Significant progress has been made in understanding the composition of centrioles and how protein interactions can direct centriole assembly ( Andersen et al . , 2003; Jakobsen et al . , 2011; Firat-Karalar et al . , 2014; Galletta et al . , 2016; Gupta et al . , 2015 ) . However , we have a limited understanding of which assembly steps are controlled by PLK4 to maintain the number of centrioles in cycling cells . Our data now establish that PLK4 phosphorylates its centriole substrate STIL on a conserved site close to the PRP motif to promote STIL binding to CPAP in vitro and in vivo . The STIL/CPAP complex is only the second binding interaction shown to be controlled by PLK4 and highlights a new regulated step in centriole assembly . Our data lead us to propose a model whereby active PLK4 phosphorylates STIL at the site of procentriole assembly in two different regions with distinct functional consequences ( Figure 7 ) : phosphorylation of multiple residues in the STAN motif , most notably S1116 , allows STIL binding to SAS6 to promote cartwheel assembly . Second , phosphorylation of S428 promotes the binding of the STIL PRP motif to CPAP , thereby linking the growing cartwheel to the triplet microtubules of the centriole wall . This model is consistent with our analysis of de novo centriole assembly , which showed that phosphorylation of the STIL STAN motif is required to recruit SAS6 to the site of procentriole assembly while phosphorylation of STIL S428 is required at a later stage to recruit CPAP to the cartwheel . Moreover , super-resolution imaging of Drosophila centrioles has revealed that the C-terminal region of Ana2/STIL containing the STAN motif is located closer to the cartwheel hub , while the N-terminal region containing the PRP motif is positioned close to the C-terminus of Sas-4/CPAP at the periphery of the cartwheel ( Gartenmann et al . , 2017 ) . A mutation in the CPAP TCP domain that causes microcephaly in humans has been shown to decrease the affinity of CPAP to STIL ( Cottee et al . , 2013; Tang et al . , 2011; Bond et al . , 2005 ) . Also , mutations in STIL that reside in the CPAP and SAS6 interacting motifs have also been identified in patients with microcephaly , although the significance of these alterations remains to be determined ( Cristofoli et al . , 2017 ) . Although recruitment of CPAP to assembling de novo centrioles requires STIL S428 phosphorylation , this modification is not required for recruiting the bulk of CPAP to the centrosome . We envisage two possible explanations for this discrepancy . First , although sharing obvious similarities , de novo centriole assembly may have some distinct requirements compared with canonical centriole biogenesis . For example , the presence of a parent centriole may help direct CPAP recruitment to the site of procentriole assembly; CPAP localized in the PCM could be recruited to the procentriole by some of CPAP’s other interacting partners in the absence of STIL S428 phosphorylation . A second possibility is that multiple pools of CPAP at the centrosome ( parent centriole , procentriole and PCM associated ) may obscure the ability to accurately monitor the role of STIL phosphorylation in CPAP recruitment at the nascent procentriole . In any case , it is clear that even if the STIL-CPAP interaction is not strictly necessary for the recruitment of either protein to canonically duplicating centrioles , it does allow for the more stable integration of these proteins into the centrosome . A previous study solved the structure of the CPAP TCP domain bound to a short STIL peptide ( residues 395–416 ) containing the PRP motif but lacking the S428 phosphorylation site ( Cottee et al . , 2013 ) . A central question that now emerges is how phosphorylation of S428 , which is positioned ~20 amino acids downstream of the core PRP interaction motif in STIL , promotes binding to CPAP in cells . We envisage two , non-mutually exclusive possibilities . First , phosphorylation creates an extended binding interface that increases the affinity of STIL to CPAP . Indeed , sequence conservation in the CPAP TCP domain is not confined to the region that directly interacts with the PRP motif of STIL , but extends further along the surface of the TCP domain beta sheet , suggesting that additional contacts with STIL may occur in this region ( Cottee et al . , 2013 ) . Moreover , there is high conservation around the S428 phosphorylation site on STIL , and this conserved motif was proposed to be well positioned to form an extended interaction interface with conserved residues in the CPAP TCP domain ( Cottee et al . , 2013 ) . Phosphorylation of STIL S428 could , therefore , seed the binding of this conserved region and cooperatively enhance the binding of STIL to CPAP . An alternative hypothesis is that S428 phosphorylation generates a conformational change that unmasks the PRP motif in STIL and exposes it for binding to CPAP . In support of this model , work in Drosophila has shown that phosphorylation of the homologous site ( S38 ) on Ana2/STIL leads to a dramatic mobility shift in an SDS page gel that is likely to reflect a significant conformational change in Ana2 ( Dzhindzhev et al . , 2017 ) . It is notable that C . elegans SAS-5/STIL lacks an obvious PRP motif , but directly binds to SAS-4/CPAP through a disordered region ( Cottee et al . , 2013 ) . Moreover , the SAS-4/CPAP TCP domain is required for the incorporation of SAS-4 into the centriole in C . elegans . SAS-5 has also been shown to bind to microtubules through a region that overlaps with the SAS-4 binding domain , suggesting that the interaction of SAS-5 with microtubules and SAS-4 may be mutually exclusive ( Bianchi et al . , 2018 ) . In the future , it will be interesting to investigate if ZYG-1/PLK4 kinase activity regulates the critical SAS-5/SAS-4 interaction in C . elegans by switching SAS-5 from binding microtubules to an association with SAS-4 . Human Centrin2 ( a . a . 1–172 ) was cloned into a pET-23b bacterial expression vector ( Novagen ) containing an N-terminal 6xHis-SUMO1 tag . Recombinant protein was purified from E . coli using Ni–NTA beads ( BioRad ) , cleaved from beads with SUMO protease and used for immunization in rabbits ( ProSci ) . An N-terminal CEP192 fragment ( a . a . 1–211 ) was cloned into a pGEX GST bacterial expression vector containing an N-terminal GST tag . Recombinant protein was purified from E . coli using glutathione sepharose beads ( GoldBio ) , cleaved from beads with PreScission protease ( GE ) and used for immunization in goats . Rabbit and goat immune sera were affinity-purified using standard procedures . Affinity-purified antibodies were directly conjugated to Alexa Fluor 488 , DyLight 550 , or DyLight 647 fluorophores ( Thermo Scientific ) for use in immunofluorescence . A synthetic phospho-peptide based on the human STIL sequence flanking serine 428 [CSVPEL ( pS ) LVDG] was synthesized , coupled to KLH and injected into rabbits ( ProSci ) . Polyclonal pS428 antibodies were affinity-purified using the hSTIL phosphopeptide coupled to a SulfoLink Coupling Resin ( Thermo Scientific ) . Additionally , a synthetic phospho-peptide based on the human PLK4 sequence flanking threonine 170 [HEKHY ( pT ) LCGTC] was synthesized , coupled to KLH and injected into rabbits ( ProSci ) . Polyclonal pT170 antibodies were affinity-purified using the human PLK4 phosphopeptide coupled to a SulfoLink Coupling Resin ( Thermo Scientific ) . Mammalian cell culture was performed as previously described ( Moyer et al . , 2015 ) . Cells were maintained at 37°C in a 5% CO2 atmosphere with 21% oxygen . Cells were grown in Dulbecco's Modified Eagle Medium ( DMEM ) containing 10% FB Essence serum ( VWR ) , 100 U/mL penicillin , 100 U/mL streptomycin and 2 mM L-glutamine . HEK293FT cells were used in co-transfection experiments , while Flp-In TRex-DLD-1 cells ( a kind gift from Stephen Taylor , the University of Manchester , UK ) were used in all other experiments . Flp-In TRex-DLD-1 cells were engineered using the Flp-In TRex Core Kit ( Life Technologies ) to stably express the Tetracycline repressor protein and contain a single , genomic FRT/lacZeo site . Centrinone ( a kind gift from Karen Oegema ) was dissolved in DMSO and used at a final concentration of 500 nM . DLD-1 cell lines were authenticated using STR genotyping . All cell lines were determined to be free from mycoplasma contamination using DAPI staining . Drosophila S2 cells ( a kind gift from Ji Hoon Kim ) were cultured at room temperature in vented T-25 flasks with Schnedier media ( Gibco ) containing 10% Fetal Bovine Serum ( Sigma ) . 1 . 5 × 106 cells were seeded in a 6-well plate in 2 mL media . The following day cells were transfected with the indicated constructs using Effectene ( QIAGEN ) according to the manufacturer’s protocol . Cells were harvested 48 hr later and subjected to co-immunoprecipitation ( procedure below ) . Molecular cloning was performed as previously described ( Moyer et al . , 2015 ) . All constructs were cloned into a pcDNA5/FRT/TO vector backbone ( Life Technologies ) and expressed from a CMV promoter . DNA constructs for Drosophila S2 transfection were cloned into a pAc vector backbone ( Invitrogen ) . Stable cell lines and siRNA treatment was performed as previously described ( Moyer et al . , 2015 ) . Stable , isogenic cell lines expressing indicated constructs from a CMV promoter under the control of two Tetracycline operator sites were generated according to the manufacturer’s recommendation using the FRT/Flp-mediated recombination in Flp-In TRex-DLD-1 cells ( Life Technologies Flp-In TRex Core Kit ) . Expression of Myc-GFP-STIL was induced with 1 µg/mL Tetracycline ( Sigma ) . Expression of Myc-GFP-CPAP was induced with 2 ng/mL Tetracycline . Expression of Myc-STIL-T2A-GFP-CPAP constructs was induced with 1 µg/mL Tetracycline . For RNA interference , 2 × 105 cells were seeded in a 6-well plate , and duplexed siRNAs were introduced using RNAiMAX ( Life Technologies ) 24 hr later . siRNA directed against STIL ( 5’-GCUCCAAACAGUUUCUGCUGGAAU-3’ ) or CPAP ( 5’-AGAAUUAGCUCGAAUAGAAUU-3’ ) was purchased from Dharmacon , and control siRNA ( Universal Negative Control #1 ) was purchased from Sigma . 24 hr after transfection , Tetracycline was added to induce expression of RNAi-resistant Myc-GFP-STIL or Myc-STIL-T2A-GFP-CPAP . Expression of the Myc-GFP-CPAP transgene was induced concurrently with siRNA treatment . Cells were harvested and processed for immunoblotting or fixed for immunofluorescence 24 hr later . Acentriolar Flp-In TRex-DLD-1 cell lines were generated by culturing lines in centrinone at 500 nM for seven or more days . Cells were subjected to the siRNA protocol ( above ) . 24 hr later , centrinone was washed out where noted by replacing media twice on cells for ten minutes each and then resuspending cells across multiple coverslips and adding Tetracycline at given concentrations depending on the cell line ( concentrations above ) . Cells were fixed at 24 , 48 , and 72 hr post-washout as described below . Co-immunoprecipitation was performed as previously described ( Moyer et al . , 2015 ) . 2 × 106 293 FT cells were seeded into 10 cm2 dishes and 24 hr later were transfected with 3 µg of plasmid DNA . 48 hr later , transfected cells were lysed in lysis buffer [10 mM Tris pH 7 . 5 , 0 . 1% Triton X-100 , 250 mM NaCl , 1 mM EDTA , 50 mM NaF , 50 mM β-glycerophosphate , 1 mM DTT , 500 nM microcystin , 1 mM PMSF and EDTA-free protease inhibitor tablet ( Roche ) ] , sonicated and soluble extracts prepared . The supernatant was incubated with beads coupled to GFP-binding protein ( Rothbauer et al . , 2008 ) . Beads were washed three times in lysis buffer , and immunopurified protein was analyzed by immunoblot . Immunoblotting and immunofluorescence were performed as previously described ( Moyer et al . , 2015 ) . For immunoblot analysis , protein samples were separated by SDS-PAGE , transferred onto nitrocellulose membranes with a Trans-Blot Turbo Transfer System ( BioRad ) and probed with the following antibodies: tubulin ( mouse DM1A anti-α-tubulin , Sigma , T6199 , 1:5000 ) , STIL ( rabbit , Bethyl , A302-441A , 1:2500 ) , FLAG M2 ( mouse , Sigma , F1804 , 1:1000 ) , CPAP ( rabbit , a kind gift from Karen Oegema , 1:1000 ) , HA ( rat , 3F10 , Roche , 1:1000 ) , Myc 4A6 ( mouse , EMD Millipore , 1:1000 ) , SAS6 ( mouse , Santa Cruz , sc-81431 , 1:1000 ) , Plk4 pT170 ( rabbit , this study , 1:1000 ) ( Nakamura et al . , 2013 ) , Plk4 ( rabbit , this study , 1:3200 ) , mCherry ( rabbit , a kind gift from Joo Soek-Han , 1:1000 ) , STIL pS1116 ( ( Moyer et al . , 2015 ) , 1:500 ) and STIL pS428 ( rabbit , this study , 1:1000 ) . Blots were blocked with 3% BSA in PBST and washed with PBST . Phospho-antibody blots were blocked in TBS Starting Block ( Thermo ) supplemented with 0 . 05% Tween 20 and washed in TBST . Antibodies were diluted in respective blocking buffers . For immunofluorescence , cells were grown on 18 mm glass coverslips and fixed in 100%–20C methanol for 10 min . Cells were blocked in 2 . 5% FBS , 200 mM glycine , and 0 . 1% Triton X-100 in PBS for 1 hr . Antibody incubations were conducted in the blocking solution for 1 hr . DNA was detected using DAPI , and cells were mounted in Prolong Antifade ( Invitrogen ) . Staining was performed with the following primary antibodies: Centrin ( rabbit , directly-conjugated , this study , 1:1000 ) , Plk4 ( rabbit , directly-labeled , Moyer et al . , 2015 , 1:1000 ) , STIL ( rabbit , directly-conjugated , Moyer et al . , 2015 , 1:250 ) , STIL pS1108 ( rabbit , Moyer et al . , 2015 , 1:250 ) , SAS6 ( mouse , sc-81–431 Santa Cruz , 1:1000 ) , CEP192 ( goat , directly-labeled , this study , 1:000 ) , CPAP-Cy3 ( directly-labeled rabbit , a kind gift from Karen Oegema , Ludwig Institute for Cancer Research , CA , 1:1000 ) and CENP-F ( sheep , raised against CENP-F a . a . 1363–1640 , a kind gift from Stephen Taylor , the University of Manchester , UK , 1:1000 ) . Secondary donkey antibodies were conjugated to Alexa Fluor 488 , 555 or 650 ( Life Technologies ) . Immunofluorescence images were collected using a Deltavision Elite system ( GE Healthcare ) controlling a Scientific CMOS camera ( pco . edge 5 . 5 ) . Acquisition parameters were controlled by SoftWoRx suite ( GE Healthcare ) . Images were collected at room temperature using an Olympus 60 × 1 . 42 NA or Olympus 100 × 1 . 4 NA oil objective at 0 . 2 μM z-sections and subsequently deconvolved in SoftWoRx suite . Images were acquired using Applied Precision immersion oil ( N = 1 . 516 ) . For quantification of signal intensity at the centrosome , deconvolved 2D maximum intensity projections were saved as 16-bit TIFF images . Signal intensity was determined using ImageJ by drawing a circular region of interest ( ROI ) around the centriole ( ROI S ) . A larger concentric circle ( ROI L ) was drawn around ROI S . ROI S and L were transferred to the channel of interest , and the signal in ROI S was calculated using the following formula: IS – [ ( IL-IS/AL-AS ) x AS] A = Area , I = Integrated pixel intensity . Fluorescence recovery after photobleaching was performed as previously described ( Moyer et al . , 2015 ) . Cells were seeded into 4-chamber , 35 mm glass bottom culture dishes ( Greiner ) and maintained in cell culture medium at 37°C and 5% CO2 in an environmental control station . Images were collected using a Zeiss 40 × 1 . 4 NA PlanApochromat oil-immersion objective on a Zeiss LSM 780 confocal equipped with a solid-state 488 nm laser and a spectral GaAsP detector . Images were acquired using Carl Zeiss immersion oil ( N = 1 . 518 ) . Acquisition parameters , shutters , and focus were controlled by Zen black software ( Zeiss ) . 11 × 0 . 5 µM z-sections were acquired for EGFP at each time point . Two consecutive pre-bleach scans were collected at 5% of the maximum ATOF value . Centrosome localized Myc-EGFP-STIL or EGFP-CPAP was bleached within a circular region encompassing the centrosome ( ~3 µM in diameter ) at 100% laser power with 100 µsec dwell time . Post-bleach scans were performed at 20 s time intervals for a total period of 400 s . Maximum intensity projections were created using Zen black . The integrated intensity value within a circular region of interest in the cytosol of the cell was subtracted from an identically sized region of interest drawn around the bleached centrosome . Recovery values were plotted relative to the difference between the fluorescence pre- and post-bleach . Recombinant protein expression and purification was performed as previously described ( Moyer et al . , 2015 ) . GFP-binding protein ( GBP ) , His-hPlk4 ( a . a . 1–416 ) , and GST-Flag-CPAP TCP domain ( aa 1142–1338 ) were expressed and purified from E . coli [strain Rosetta ( DE3 ) ] using standard procedures . Recombinant GST-hSTIL was expressed and purified from SF9 insect cells ( Invitrogen ) using the Bac-to-Bac expression system ( Invitrogen ) . Infected cell pellets were suspended in lysis buffer ( 10 mM PO43- pH 7 . 4 , 137 mM NaCl , 2 . 7 mM KCl , 10% glycerol , 2 mM MgCl2 , 5 mM DTT , 100 nM Microcystin , 1 mM Na3V04 , 250 U of Benzonaze nuclease ( Sigma ) , 1 mM PMSF and EDTA-free protease inhibitor tablet ( Roche ) ) and lysed by sonication . After centrifugation at 15 , 000 rpm for 30 min , the supernatant was supplemented with 110 mM KCl and 0 . 1% Triton X-100 and incubated with Glutathione Sepharose beads ( GE Healthcare ) for 4 hr at 4°C . Beads were washed extensively in wash buffer [10 mM PO43- pH 7 . 4 , 137 mM NaCl , 2 . 7 mM KCl , 10% glycerol , 5 mM DTT , 0 . 1% Triton X-100 , 100 mM KCl , 1 mM PMSF and EDTA-free protease inhibitor tablet ( Roche ) ] and protein eluted in elution buffer ( 10 mM PO43- pH 7 . 4 , 137 mM NaCl , 2 . 7 mM KCl , 10% glycerol , with 40 mM reduced glutathione and 5 mM DTT ) . Protein was dialyzed into a final buffer of 10 mM PO43- pH 7 . 4 , 137 mM NaCl , 2 . 7 mM KCl , and 10% glycerol . When necessary , the GST tag was removed by overnight incubation with GST-PreScission protease ( GE Healthcare ) . In vitro kinase assays were performed as previously described ( Moyer et al . , 2015 ) . Assays were conducted for 30 min at 30°C in 20 mM Tris pH 7 . 5 , 25 mM KCl , 1 mM DTT , and in the presence of 10 mM MgCl2 and 100 µM ATP . 2 µg of substrate was incubated with 1 µg of His-hPlk4 ( a . a . 1–416 ) . Kinase reactions were stopped with sample buffer and analyzed by SDS-PAGE and western blotting . Recombinant GST-hSTIL was bound to GSH resin ( GoldBio ) for four hours at 4°C in PBS , 10% glycerol , and 1 mM DTT . Beads were washed twice in kinase buffer ( see above ) supplemented with 10% glycerol and incubated with 6xHis-hPlk4 ( a . a . 1–416 ) in kinase buffer with or without 10 µM cold ATP at 33°C for two hours . Reactions were then spun down and washed twice with cold binding buffer ( 50 mM Na-HEPES pH 7 . 5 , 400 mM NaCl , 2 mM MgCl2 , 1 mM EGTA , 1 mM DTT , 0 . 15% Triton-X 100 , 100 nM Microcystin ( Calbiochem ) and 0 . 5 mg/ml BSA ) and spun in binding buffer at 4°C for 2 hr in the presence of recombinant Flag-CPAP TCP domain . Beads were washed three times in binding buffer without BSA , and then proteins were eluted in SDS sample buffer and immunoblotted . Mass spectrometry was performed as previously described ( Moyer et al . , 2015 ) . In-solution protein digestion was carried out using ‘Filter Assisted Sample Preparation’ ( FASP ) method ( Wiśniewski et al . , 2009 ) . Data- dependent MS/MS analysis of peptides was carried out on the LTQ-Orbitrap Velos ( www . thermoscientific . com ) interfaced with Eksigent 2D nanoflow liquid chromatography system ( www . eksigent . com system ) . Peptides were enriched on a 2 cm trap column ( YMC gel ODS-A S-10µm ) , fractionated on a 75 µm x 15 cm column packed with 5 µm , 100 Å Magic AQ C18 material ( Michrom Bioresources ) , and electrosprayed through a 15 µm emitter ( PF3360-75-15-N-5 , New Objective ) . Reversed-phase solvent gradient consisted of 0 . 1% formic acid with increasing levels of 0 . 1% formic acid , 90% acetonitrile over a period of 90 min . LTQ orbitrap Velos was set at 2 . 0 kV spray voltage , full MS survey scan range was set at 350–1800 m/z , data-dependent HCD MS/MS analysis set for top eight precursors with minimum signal of 2000 . Other parameters include peptide isolation width of m/z 1 . 9; dynamic exclusion limit 30 s and normalized collision energy 35; precursor and the fragment ions resolutions were 30 , 000 and 15 , 000 , respectively . Internal mass calibration was applied using lock mass ion m/z = 371 . 101230 . Mass spectrometry raw files were automatically processed through Proteome Discoverer 1 . 4 software . Raw MS and MS/MS data were isotopically resolved with deconvolution and deisotoping using Thermo Scientific Xtract and MS2-processor software in addition to default spectrum selector node . The data were searched in Refseq human entries using Mascot ( v2 . 2 . 6 , Matrix Sciences ) search engine interfaced with different processing nodes of Proteome Discoverer 1 . 4 . Mass tolerances on precursor and fragment masses were set to 15 ppm and 0 . 03 Da , respectively . Peptide validator node was used for identification confidence and 1% false discovery rate cutoff was used to filter the peptides . Phosphorylation site probability was analyzed using phosphoRS 3 . 0 node in Proteome discoverer software ( Taus et al . , 2011 ) .

A cell’s DNA is the chemical instruction manual for everything it does . Each cell in our bodies contains over two meters of DNA , which is divided into 46 packages of information called chromosomes . When the body needs to make more cells , for example during growth or repair , existing cells divide in two in order to replicate themselves . This means that they also need to copy all of their DNA and then deliver identical sets of chromosomes to each new cell . Animal cells use structures called centrioles to help them divide their sets of chromosomes accurately . When cells are about to divide , they make a new set of centrioles by assembling a variety of proteins . This assembly process must be carefully controlled; if too many or too few centrioles are built , cell division errors can occur that lead to the generation of new cells with abnormal numbers of chromosomes . The enzyme PLK4 helps to assemble centrioles , but its exact role in the construction process has remained largely unknown . For example , how it might modify different components of the centriole , and why this matters , is poorly understood . By performing cell biological and biochemical experiments using human cells , Moyer and Holland show that PLK4 interacts with a protein called STIL that is found in the central part of the centriole . The modification of STIL at a specific location by PLK4 was needed to link it to another protein in the outer wall of the centriole , and was also necessary for the cells to build new centrioles . Cells in which PLK4 was unable to modify STIL had too few centrioles when they were beginning to divide . Testing the activity of PLK4 in fruit flies revealed that it plays a similar role as in human cells . This suggests that the modification of STIL by PLK4 is important for normal cell division across different species . The results presented by Moyer and Holland help us to understand how dividing cells build the complex machinery that enables them to pass on their genetic material accurately . Future work that builds on these findings could provide insight into human diseases , such as brain development disorders and cancer , where centrioles are either defective or present in the wrong number .

2019

PLK4 promotes centriole duplication by phosphorylating STIL to link the procentriole cartwheel to the microtubule wall

Transcription initiation of archaeal RNA polymerase ( RNAP ) and eukaryotic RNAPII is assisted by conserved basal transcription factors . The eukaryotic transcription factor TFIIE consists of α and β subunits . Here we have identified and characterised the function of the TFIIEβ homologue in archaea that on the primary sequence level is related to the RNAPIII subunit hRPC39 . Both archaeal TFEβ and hRPC39 harbour a cubane 4Fe-4S cluster , which is crucial for heterodimerization of TFEα/β and its engagement with the RNAP clamp . TFEα/β stabilises the preinitiation complex , enhances DNA melting , and stimulates abortive and productive transcription . These activities are strictly dependent on the β subunit and the promoter sequence . Our results suggest that archaeal TFEα/β is likely to represent the evolutionary ancestor of TFIIE-like factors in extant eukaryotes . The conserved core of the archaeal and eukaryotic transcription machineries encompasses a highly complex multisubunit RNAP as well as evolutionary conserved transcription factors that govern its activities through the transcription cycle . The minimal requirements for promoter-directed and start site-specific transcription are identical for the RNAPII system in eukaryotes ( Parvin and Sharp , 1993 ) and the RNAP of archaea ( Hausner et al . , 1996; Qureshi et al . , 1997; Werner and Weinzierl , 2002 ) . Two basal transcription factors , TBP ( TATA-binding protein ) and TFB ( TFIIB in eukaryotes ) , are necessary and sufficient to initiate transcription in archaea in vitro . TBP and TFB facilitate promoter recognition and the recruitment of RNAP ( Werner and Grohmann , 2011 ) . In archaea as well as eukaryotes a third factor TFE ( TFIIE ) enhances the next step in initiation , the transition of the closed to the open complex ( Holstege et al . , 1995 , 1996; Werner and Weinzierl , 2005 ) . Eukaryotic TFIIE is a heterodimer composed of subunits TFIIEα and TFIIEβ in humans ( Tfa1 and 2 in yeast ) . Archaeal TFE ( hereafter referred to as TFEα ) is monomeric and homologous to TFIIEα , but lacks its C-terminal acidic domain ( Bell et al . , 2001; Hanzelka et al . , 2001 ) . To date no archaeal homologue of TFIIEβ has been identified . During open complex formation the DNA strands are separated ( melted ) and the template strand is loaded into the active centre of RNAP . Similar to TFIIE , TFEα facilitates open complex formation by directly interacting with the non-template DNA strand ( NTS ) , and via an allosteric mechanism that is likely to involve structural changes in the RNAP clamp and stalk ( Grohmann et al . , 2011 ) . While in archaea the closed-to-open complex transition occurs spontaneously ( Werner and Weinzierl , 2002 ) , on most eukaryotic RNAPII promoters it is dependent on the translocase activity of TFIIH ( Guzman and Lis , 1999; Kim et al . , 2000; Fishburn et al . , 2015 ) . However , the dependency on TFIIH changes with DNA template topology . Strand separation and subsequent initiation from linear DNA templates strictly depends on TFIIH , but for some promoters this requirement can be overcome by using negatively supercoiled DNA templates ( Parvin and Sharp , 1993 ) . Under these conditions DNA melting is weakly stimulated by TFIIE , an effect that is obscured by the strong melting activity of TFIIH ( Holstege et al . , 1995 ) . Since TFIIH is not strictly essential for RNAPII initiation per se , similar molecular mechanisms are likely to operate during open complex formation of archaeal and eukaryotic RNAPs . The archaeal model systems provide a distinct advantage in this respect allowing us to study how TFIIE/TFE facilitates DNA melting in the absence of a TFIIH-like factor . Archaea , like bacteria , utilise one type of RNAP to execute their genetic programmes , while the transcription space of eukaryotes is partitioned into distinct and non-overlapping subsets of genes that are transcribed by 3 and 5 specialised types of RNAPs in animals and plants , respectively . The common past of all types of nuclear eukaryotic RNAP systems is reflected in the sequence , structure and function of RNAP subunits and associated basal transcription factors ( Vannini and Cramer , 2012 ) . TFIIE is a prominent example of this relationship . The human RNAPIII subunits hRPC62/39 ( C82/34 in yeast ) are positioned at the periphery of RNAPIII nearly identical to the binding site of TFIIE on RNAPII ( Vannini and Cramer , 2012 ) . TFEα , TFIIEα and hRPC62 contain homologous winged helix ( WH ) domains ( Carter and Drouin , 2010; Lefèvre et al . , 2011 ) ( Figure 1A ) . TFIIEβ and hRPC39 both contain tandem WH domains suggesting their paralogous nature ( Vannini and Cramer , 2012 ) , although no significant sequence homology was found ( Carter and Drouin , 2010 ) . The origin and evolution of these factors and their role in the multiplication of parallel transcription systems in eukaryotes has remained elusive . 10 . 7554/eLife . 08378 . 003Figure 1 . Identification of a dimeric TFE factor in S . solfataricus . ( A ) Conserved and acquired domains of TFIIEα and TFIIEβ-related proteins in Eukaryotes and Archaea derived from ( Blombach et al . , 2009; Vannini and Cramer , 2012 ) . ( B ) Homologous expression and nickel affinity purification of Sso0944 as C-terminal His10-tag fusion . Immunodetection was used to detect co-purification of Sso RNAP ( subunit Rpo2 ) and the basal transcription factors . MF—membrane fraction . I—input ( soluble fraction ) loaded onto column . F—flow-through fraction . ( C ) Immunodetection of RNA polymerase ( RNAP ) , TFEα and Sso0944 ( TFEβ ) in S . solfataricus P2 cell lysate fractionated by size exclusion chromatography . ( D ) Multiplex immunodetection of Sso RNAP ( stalk subunits Rpo4/7 ) and the basal transcription factors in S . solfataricus P2 during different growth phases . Samples were taken at the indicated time points ( blue ) . 18 µg of lysed cells ( total soluble protein content ) was loaded into each lane . Immunodetection of the chromatin protein Alba served as loading control . Immunodetection of Rpo4/7 yielded a strong signal for Rpo7 at around 20 kDa , but only faint signal for Rpo4 ( around 13 kDa ) . The experiment was carried out in triplicate and a typical result is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08378 . 00310 . 7554/eLife . 08378 . 004Figure 1—figure supplement 1 . Sso0944 and TFEα form a dimeric complex . Sso0944 was isolated as C-terminal His10-tag fusion from S . solfataricus M16 transformed with pMJ0503-Sso0944 under more stringent conditions . The salt concentration during the Ni-affinity purification was raised to 500 mM KCl and a 20 CV wash with 50 mM imidazole was introduced prior to elution . The upper panel shows an SDS-gel with the Ni-affinity elution fractions with total protein stained with SYPRO Ruby . The lower panel shows the immunodetection with Sso0944 and TFEα antisera of the Ni-affinity elution fractions . DOI: http://dx . doi . org/10 . 7554/eLife . 08378 . 00410 . 7554/eLife . 08378 . 005Figure 1—figure supplement 2 . Quantitative immunodetection of TFEα , TFEβ and TBP in S . solfataricus P2 cell lysates during exponential growth phase ( n = 3 ) . 18 µg of cell lysate ( total soluble protein content ) was loaded into each lane . Total protein-staining of the Western blots was performed using Ponceau S to verify loading of equal protein amounts . In parallel increasing amounts of recombinant dimeric TFEα/β ( non-His-tagged ) or TBP were loaded in order to generate a calibration curve ( black crosses ) that was used to determine the expression levels of TFEα , TFEβ and TBP ( red diamond , average value , error bars show 1× standard deviation ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08378 . 005 All crenarchaeal and most euryarchaeal genomes encode a gene that bears resemblance to the RNAPIII subunit hRPC39 ( Blombach et al . , 2009 ) . Two regions of this gene show homology to two separate and distinct domains of eukaryotic hRPC39 proteins , namely the second of its three WH domains and a C-terminal domain that includes four highly-conserved cysteine residues ( Figure 1A ) . We have conducted a comprehensive structure-function characterisation of the archaeal hRPC39-like gene product using both in vivo and in vitro approaches . Our results show that the hRPC39-like gene from the archaeon Sulfolobus solfataricus is the bona fide homologue of eukaryotic TFIIEβ . We chose the hRPC39 homologue Sso0944 from the archaeon S . solfataricus ( Sso ) as model protein because the gene is a good representative of its kind ( Blombach et al . , 2009 ) . To identify interaction partners of Sso0944 we expressed His-tagged Sso0944 in S . solfataricus M16 and probed the presence of co-purifying components of the basal transcription apparatus following metal-affinity chromatography by immunodetection . While we found no evidence that the RNAP , TBP or TFB1 co-purified with Sso0944 , TFEα co-eluted with Sso0944 , indicating that TFEα and Sso0944 are associated in vivo ( Figure 1B ) . Sypro Ruby-stained SDS-PAGE of the affinity-purified material demonstrates that the polypeptides form a dimeric complex , that is , their association is not dependent on additional factors ( Figure 1—figure supplement 1 ) . To rule out the possibility that the affinity tag of Sso0944 prevented its stable association with the RNAP we fractionated a wild type S . solfataricus P2 cell lysate by size exclusion chromatography and analysed the fractions using immunodetection . The elution profile of endogenous Sso0944 overlapped with that of TFEα consistent with a heterodimeric TFEα/Sso0944 complex of 36 . 1 kDa . The elution profile of TFEα was somewhat broader indicating that part of TFEα might be present in the monomeric form . RNAP eluted in earlier fractions corresponding to its molecular weight of approximately 400 kDa ( Figure 1C ) . As Sso0944 is not stably incorporated into RNAP in contrast to the related RNAPIII subunit hRPC39 , but rather forms a complex with TFEα , we renamed the archaeal protein TFEβ . Like most genes that encode basic components of the transcription apparatus , both RPC34 and TFA2 , the genes encoding C34 and Tfa2 , respectively , are essential for cell viability in Saccharomyces cerevisiae ( Stettler et al . , 1992; Feaver et al . , 1994 ) . In order to test whether the gene encoding TFEβ is essential in Sulfolobus we attempted to delete the gene ( Saci_1342 ) in the uracil-auxotroph Sulfolobus acidocaldarius strain MW001 using a pop-in/pop-out strategy and selection marker pyrEF ( Wagner et al . , 2012 ) . Following genomic integration of the Saci_1342 deletion construct , counter-selection using 5-fluoroorotic acid all clones reverted to wild type via reciprocal excision ( 80 clones tested ) ( Table 1 and Table 1—source data 1 ) . In contrast , deletion of Saci_1342 was readily achieved in strain MW001 Saci_1162::Saci_1342 where we introduced a second copy of Saci_1342 replacing the non-essential gene Saci_1162 ( 4 out of 20 clones ) ( Table 1 and Table 1—source data 1 ) . This ultimately demonstrates that the gene encoding TFEβ is essential in S . acidocaldarius . 10 . 7554/eLife . 08378 . 006Table 1 . Genetic experiments showing that the TFEβ encoding gene Saci_1342 is essentialDOI: http://dx . doi . org/10 . 7554/eLife . 08378 . 00610 . 7554/eLife . 08378 . 007Table 1—source data 1 . Genetic experiments showing that the TFEβ encoding gene Saci_1342 is essentialDOI: http://dx . doi . org/10 . 7554/eLife . 08378 . 007Parental strainPlasmid integration relative to Saci_1342Number of clones testedClones with Saci_1342 deletion obtainedMW001upstream400downstream400MW001 Saci_1162::Saci_1342upstream104downstream100 In order to compare and characterise the steady-state levels of TFEα and TFEβ during exponential and stationary growth of S . solfataricus we carried out quantitative Western blotting . TFEα and TFEβ levels are near stoichiometric during exponential growth ( 24 ± 3 pmol/mg soluble protein and 27 ± 3 pmol/mg soluble protein , respectively ) and about sevenfold lower than TBP levels ( 184 ± 17 pmol/mg ) ( Figure 1—figure supplement 2 ) . TFEβ levels are decreased when cells enter stationary phase while RNAP , TBP , TFB and TFEα remain largely unaffected ( Figure 1D ) . Our results demonstrate that dimeric TFEα/β is the predominant form of the factor in exponentially growing cells , and that the steady-state levels of the complex vary as a function of the growth cycle . In order to carry out a structure-function analysis of TFEα/β we expressed and purified a recombinant form of the TFEα/β complex in in Escherichia coli . Concentrated recombinant TFEα/β has a dark brown colour and its absorption spectrum displays a shoulder at 410 nm that is characteristic for iron-sulphur cluster harbouring proteins ( Figure 2A ) . In order to define the origin of this absorption more precisely we recorded continuous-wave electron paramagnetic resonance ( cw-EPR ) spectra . The profile of the cw-EPR spectrum with a g-value of 2 . 01 , its sensitivity to the reducing agent sodium dithionite and the sharp decrease in signal at temperatures above 20 K are consistent with a signal corresponding to a cubane [3Fe-4S]+ cluster ( Figure 2B and Figure 2—figure supplement 1 ) . Double integration and comparison with a spin standard indicate a low cluster-occupancy of approximately 4% , which is in contrast to the strong 410 nm signal in the absorption spectrum . It is common that before reduction aerobically prepared proteins containing [4Fe-4S] clusters display a [3Fe-4S]+ signal due to oxidative impairment ( Beinert et al . , 1996 ) . Addition of a reducing agent often results in the disappearance of the signal of the [3Fe-4S]+ cluster concomitant with the appearance of the signal from the intact [4Fe-4S]+ cluster . Here , however , only the former effect is observed: [4Fe-4S]2+ clusters may not be reduced—and remain EPR-silent—if the redox potential is lower than that of the reducing agent ( dithionite ) , or if the rate of reduction is too slow . Alternatively the reduced [4Fe-4S]+ cluster could be present in a high spin state , which could give rise to a signal too broad to be detected . To investigate the presence of an EPR-silent [4Fe-4S]+ cluster , we recorded native mass spectra ( MS ) of TFEα/β ( Figure 2C and Figure 2—source data 1 ) . The spectrum contained two major species one with a mass of 21 , 236 . 6 Da and the other with a mass of 36 , 530 . 8 Da . The former corresponds to the expected mass for TFEα with a Zn2+ ion bound by its zinc ribbon ( ZR ) domain while the later corresponds to the mass of the TFEα/β harbouring a Zn2+ ion and a [4Fe-4S]2+ cluster . Given that the TFEα ZR harbours a Zn2+ ion , the [4Fe-4S]2+ cluster must be coordinated by the four cysteines in the C-terminal domain of TFEβ . As this domain is highly conserved in human hRPC39 ( Figure 2G ) , we examined the human protein for the presence of an equivalent FeS cluster . Similar to archaeal TFEβ , the human hRPC62/C39 complex shows an absorption spectrum with a shoulder at 410 nm ( Figure 2D ) ; its cw-EPR signature is characteristic for a [3Fe-4S]+ cluster with low occupancy ( <3% ) ( Figure 2E and Figure 2—figure supplement 1 ) . The native mass spectrum this time revealed three major species with masses of 35 , 555 . 7 , 62 , 516 . 3 and 98 , 455 . 5 Da corresponding to hRPC39 , hRPC62 and hRPC62/C39 complex bound to a [4Fe-4S]2+ respectively ( Figure 2F ) . Tandem MS experiments revealed that the cluster was bound to hRPC39 ( Figure 2—figure supplement 2 and Figure 2—source data 1 ) . 10 . 7554/eLife . 08378 . 008Figure 2 . The C-terminal domain of Sso TFEβ harbours a 4Fe-4S cluster that is conserved in human RNAPIII subcomplex hRPC62/39 . ( A ) UV-vis spectrum of TFEα/β-His . ( B ) Cw-EPR spectra of TFEα/β prepared in the presence of 1 mM DTT with or without the addition of 20 mM Na-dithionite . Besides the main [3Fe-4S]+ cluster signal at g = 2 . 01 , a small amount of spurious high spin Fe3+ at g = 4 . 3 was also detected . The asterisks denote a background signal . ( C ) Nano-electrospray ionization ( nESI ) mass spectrum of TFEα/β-His . Filled circles indicate the different charge state series: TFEα/β-His + Zn ion + 4Fe-4S cluster ( blue ) , TFEα + Zn ion ( green ) . ( D ) UV-vis spectrum of recombinant human RNAP III subcomplex hRPC62/C39 . ( E ) Cw-EPR spectra of hRPC62/39 prepared in the presence of the reducing agent Na-dithionite . The asterisks denote a background signal . ( F ) nESI mass spectrum of recombinant human hRPC62/C39 . Filled circles indicate the different charge state series: hRPC62/C39 complex + 4Fe-4S cluster ( blue ) , monomeric hRPC62 ( green ) , monomeric hRPC39 ( yellow ) . ( G ) Sequence alignment of the C-terminal domains of S . solfataricus TFEβ ( gene id: 1455187 ) , H . sapiens hRPC39 ( gene id: 10621 ) , Schizosaccharomyces pombe C34 ( gene id: 2538992 ) , and S . cerevisiae C34 ( gene id: 855737 ) . Identical and similar residues are shaded in black and grey , respectively . The four cysteine residues coordinating the FeS cluster are highlighted in yellow . DOI: http://dx . doi . org/10 . 7554/eLife . 08378 . 00810 . 7554/eLife . 08378 . 009Figure 2—source data 1 . Theoretical and experimentally calculated masses of proteins and protein complexes . DOI: http://dx . doi . org/10 . 7554/eLife . 08378 . 00910 . 7554/eLife . 08378 . 010Figure 2—figure supplement 1 . Temperature dependence of the cw-EPR spectra of Sso TFEαβ and hRPC62/39 . The decrease in signal at temperatures above 20 K is characteristic of cubane-like FeS clusters . DOI: http://dx . doi . org/10 . 7554/eLife . 08378 . 01010 . 7554/eLife . 08378 . 011Figure 2—figure supplement 2 . Nano-electrospray ionization mass spectrum of human hRPC62/C39 ( top ) . The +22 charge state , corresponding to the hRPC62/C39 + [4Fe-4S] complex was subjected to collision induced dissociation . At the low m/z region of the tandem mass spectrum , two charge state series corresponding to hRPC39 ( yellow ) and hRPC39 + [4Fe-4S] ( red ) are observed ( bottom and Figure 2—source data 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08378 . 011 In summary , our results reveal the presence of an [4Fe-4S] cluster in the C-terminal domains of archaeal TFEβ and the human RNAPIII subunit hRPC39 , thus demonstrating that this feature has been conserved through evolution . The four cysteines coordinating the [4Fe-4S] cluster show high conservation in TFEβ/hRPC39 homologs with some exceptions such as haloarchaea ( Blombach et al . , 2009 ) and the yeast S . cerevisiae C34 , where all four cysteine residues have been lost ( Figure 2G ) . TFEα and β both have a bipartite domain architectures consisting of a WH and a ZR domain , and a WH and FeS domain , respectively ( Figures 1A , 3A ) . In order to characterise the interaction network between the domains in the TFEα/β heterodimer , we tested the dimerization properties of domain deletion and substitution variants using a metal affinity co-purification approach and a His-tagged TFEβ variant ( Figure 3B ) . The input fractions show that all mutant variants were expressed in a soluble and heat-stable form with the exception of TFEβ Δ1–84 . The TFEα WH domain is essential for dimerization ( Δ1–110 ) , while the ZR domain is dispensable ( Δ114–147 , Figure 3B ) . However , the WH alone is not sufficient for dimerization ( Δ111–178 ) , but requires the C-terminal tail ( Δ148–178 ) for interaction with TFEβ . The ZR domain of Sso TFEα includes only three of the four conserved cysteine residues that coordinate the Zn ion in archaeal TFEα and eukaryotic TFIIEα . In line with the ZR being dispensable , mutation of the unpaired cysteine in TFEα ( C117S ) does not impair complex formation with TFEβ ( Figure 3B ) . In yeast TFIIE , the N-terminal tip of α-helix 3 of the Tfa1 WH domain is essential for dimerization ( Grünberg et al . , 2012 ) . We tested whether the corresponding region in the TFEα WH domain is essential for TFEα/β dimerization by introducing point mutations ( K46E , D49T , R51E/K52E ) or deleting the entire region ( Δ46–52 ) ( Figure 3B and Figure 3—figure supplement 1 ) . None of these mutations abrogated dimerization with TFEβ , which reflects that the interaction network differs between yeast TFIIE and TFEα/β . 10 . 7554/eLife . 08378 . 012Figure 3 . Characterisation of the TFEα/β heterodimerization interface . ( A ) Amino acid sequences and secondary structure of TFEα and TFEβ . The α-helices in the TFEα WH domain are numbered according to ( Meinhart et al . , 2003 ) . Deletion ( red boxes ) and substitution deletions ( highlighted in red ) are indicated . ( B , C ) TFE α/β interaction analysis using co-purification of TFE alpha and his-tagged TFE beta . TFEα ( B ) and TFEβ-His ( C ) substitution- and domain deletion variants were co-expressed in E . coli and purified using Nickel-affinity chromatography . Input ( heat-stable cell lysate , 25% ) ( I ) and elution fractions ( E ) were analysed by SDS-PAGE and Coomassie staining . Red and blue triangles indicate the position of the respective mutant variants of TFEα and TFEβ-His , respectively . The question mark denotes that the TFEβ Δ1–84 variant is probably instable . Note that the TFEα expression levels are higher than TFEβ , and that the expression levels of TFEβ mutants are lower than WT , which in some cases make it difficult to discern in the input fractions . ( D ) Minimal heterodimeric TFEα/β complex consisting of TFEα Δ114–178 and TFEβ Δ1–73 . DOI: http://dx . doi . org/10 . 7554/eLife . 08378 . 01210 . 7554/eLife . 08378 . 013Figure 3—figure supplement 1 . Additional mutants of the N-terminal tip of α-helix 3 of TFEα and their effect on dimerization with TFEβNi-affinity co-purification assays conducted for TFEβ-His co-expressed with TFEα ( WT ) , TFEα K46E , TFEα D49T , or TFEα R51E/K52E . Red and blue triangles indicate the position of the respective mutant variants of TFEα and TFEβ-His , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 08378 . 013 We subsequently determined the TFEβ domains required for TFEα binding . As the exact domain boundary in TFEβ is unknown we tested different deletion variants including or excluding linker residues 74–84 in dimerization experiments . The WH domain of TFEβ is not required for dimerization ( Δ1–73 ) , while the FeS domain is vital ( Δ85–125 , Figure 3C ) . In order to investigate the role of the FeS cluster-chelating cysteine residues in TFEβ we produced cysteine to serine mutants at positions C92 , C95 , C101 and C112 . Following metal affinity chromatography of TFEα/β , the C92S , C95S and C112S variants rapidly lost the FeS cluster based on the absorbance spectra ( Figure 4 ) . Concomitant with the loss of the Fe-S cluster no complexes with TFEα were obtained . In contrast , the fourth mutation C101S had little effect on FeS cluster stability , yield and dimerization . 10 . 7554/eLife . 08378 . 014Figure 4 . TFEαβ dimerization depends on the integrity of the Fe-S cluster . Ni-affinity chromatography of TFEβ-His ( WT ) , TFEβ-His C92S , TFEβ-His C95S , TFEβ-His C101S or TFEβ-His C112S co-expressed with TFEα . The graphs show the absorption profile monitored at 260 nm , 280 nm , 350 nm , and 420 nm . Absorption in the visible light range ( 350 nm and 420 nm ) is indicative for the presence of the Fe-S cluster . The elution fractions analysed by SDS-PAGE are indicated . The panels below show Coomassie-stained SDS-gels with the ( heat-stable ) input , flow , and elution fractions . DOI: http://dx . doi . org/10 . 7554/eLife . 08378 . 014 In summary , our results identify the TFEβ FeS domain as essential part of the TFEα/β dimerization interface that comprises also the TFEα WH and tail domains . These are sufficient to form a minimal TFEα/β complex ( Figure 3D ) . Methanocaldococcus jannaschii TFEα binds the RNAP in a bidentate fashion , the WH domain interacts with the tip of the RNAP clamp coiled coil while the ZR domain locates to the base of the clamp and the stalk module ( Grohmann et al . , 2011 ) . In order to characterise the binding characteristics of TFEα/β to Sso RNAP we produced a recombinant RNAP clamp analogously to ( Martinez-Rucobo et al . , 2011 ) . Gel filtration elution profiles show that TFEα forms a stable complex with the recombinant RNAP clamp since both proteins eluted in earlier fractions corresponding to a larger size ( Figure 5 ) . The elution of both proteins was slightly asymmetrical , possibly due to partial dissociation during chromatography . Similarly dimeric TFEα/β forms a stable complex with the clamp since all three proteins co-eluted in a symmetrical fashion . Deletion of the TFEα ZR domain ( TFEα ΔZR , residues 114–147 ) leads to loss of complex formation . In contrast , deleting the TFEβ WH domain ( TFEβ ΔWH , residues 1–73 ) does not impair binding ( Figure 5 ) which suggests that the TFEβ WH domain does not contribute to RNAP binding . 10 . 7554/eLife . 08378 . 015Figure 5 . TFEα/β forms a stable interaction with the RNAP clamp module . 10 μM TFEα/β-His were incubated together with 10 μM recombinant RNAP clamp and the sample was resolved gel filtration . The presence of RNAP clamp and TFEα/β in the fractions was determined by SDS-PAGE and silver staining . The position of peaks for gel filtration marker proteins is indicated on top . For experiments with TFEα/β ΔWH the contrast was enhanced for the lower part of the gel in order to visualize TFEβ ΔWH . DOI: http://dx . doi . org/10 . 7554/eLife . 08378 . 015 In summary , both the TFEα WH and ZR domains are required to anchor TFE to the RNAP clamp . The TFEβ FeS-domains further stabilizes the complex , whereas the TFEβ WH domain appears not to be involved . We developed an electrophoretic mobility shift assay ( EMSA ) to monitor the formation of the preinitiation complex ( PIC ) on the viral SSV1 T6 promoter ( Qureshi et al . , 1997; Werner and Weinzierl , 2005 ) . Due to the intrinsic instability of the closed Sso PIC we used promoter templates that were pre-melted in the region of −4 to −1 relative to the transcription start site ( TSS ) ( Figure 6A ) . PIC formation is strictly dependent on both TBP and TFB ( Figure 6B ) . The addition of TFEα/β to the DNA-TBP-TFB-RNAP complex increases the PIC signal in a concentration-dependent fashion with the PIC now appearing in two different species indicating two different conformations . In contrast , monomeric TFEα had no stimulatory effect on PIC formation ( Figure 6B ) . TFEα/β was not able to bind to the dsDNA template directly ( Figure 6B and data not shown ) . 10 . 7554/eLife . 08378 . 016Figure 6 . TFEα/β increases the stability of the preinitiation complex ( PIC ) and promotes DNA melting . ( A ) NTS sequence of the DNA templates for the promoters tested in electrophoretic mobility shift assay ( EMSA ) and potassium permanganate footprinting assays . The templates incude a 4 nt heteroduplex region ( positions −4 to −1 relative to the TSS , in red ) that was generated by introducing transition mutations into the template strand Asterisks mark permanganate-reactive T residues ( see panel C ) . Additional G residues ( in grey ) were added to stabilize the termini . ( B ) PIC formation on the SSV1-T6 promoter in response to TFEα/β , TFEα and mutant variants using EMSAs . K56E/K57E and K64E denote mutations in the TFEβ WH . ( C ) The non-template strand of the SSV1-T6 and Sso Rpo5 and Sso tRNALeu promoters was probed in potassium permanganate footprinting assays . The position of reactive T residues is indicated on the right . G/A ladder and the DNA sequence are shown on the left . ( D ) Sequence alignment of the WH domains of TFEβ ( gene id: 1455187 ) , WH2 of human C39 ( gene id: 10621 ) , WH2 of yeast C34 ( gene id: 855737 ) , WH2 of human TFIIEβ ( gene id: 2961 ) , and WH2 of yeast Tfa2 ( gene id: 853936 ) . Identical and similar residues are shaded in black and grey , respectively . On top of the alignment the secondary structure of WH2 of human C39 ( PDB id: 2DK5 ) is depicted ( α-helices as barrels and β-strands as arrows ) . Residues K56 , K57 and K64 in the WH domain of TFEβ where mutations were introduced are highlighted in red . DOI: http://dx . doi . org/10 . 7554/eLife . 08378 . 01610 . 7554/eLife . 08378 . 017Figure 6—figure supplement 1 . TFEα/β is part of the PIC . Supershift EMSA with PICs formed in presence or absence of TFEα/β ( 0 . 1 µM ) being exposed to antisera directed against TFEα , TFEβ and TBP . Antibody-PIC complexes migrate slower in the gel compared to PIC alone causing a supershift . DOI: http://dx . doi . org/10 . 7554/eLife . 08378 . 017 Deletion of the TFEα ZR or the TFEβ WH domains reduced the stimulation . The former mutation destabilizes the binding to the RNAP , while the latter variant is not impaired in RNAP binding ( Figure 5 ) , which suggests that the TFEβ WH domain plays a role for the stabilisation of the PIC . EMSA supershift experiments validated the incorporation of TFEβ into the PIC ( Figure 6—figure supplement 1 ) . In order to test the ability of TFEα/β to facilitate DNA melting we carried out permanganate footprinting assays to identify T-residues in single-stranded regions within the NTS . The same −4/−1 pre-melted T6 promoter template was used in EMSA and permanganate assays , which accordingly result in a strong signal at −1 . The ternary complex ( DNA-TBP-TFB ) gives in addition , and to a much lesser extent , a signal at −5 likely due to thermal breathing ( Figure 6C ) . The inclusion of RNAP displayed a similar pattern , while the addition of TFEα/β led to new strong signals at −5 , −7 and −12 , which reflects that the factor stimulates open complex formation ( Figure 6C ) . Deletion of the TFEβ WH or the TFEα ZR domain did not perturb the promoter opening activity of the TFEα/β complex , while TFEα was inactive , congruent with its lack of PIC stabilisation ( Figure 6C ) . Similar experiments with two other −4/−1-premelted endogenous promoters from Sso ( Sso Rpo5 and Sso tRNALeu ) confirmed that TFEα/β strongly enhances open complex formation up to position +4 ( Figure 6C ) . The WH domains of TFEβ , hRPC39/C34 and TFIIEβ/Tfa2 show little conservation , but they commonly carry sets of lysine residues in the wing formed by the two C-terminal β-strands and the loop between them ( Figure 6D ) that may play a role in electrostatic interactions with the phosphate-backbone of the NTS . We generated two charge reversal mutations in the TFEβ WH domain , K56E/K57E and K64E . In EMSA experiments , both mutants showed reduced stimulation in PIC formation when compared to WT TFEα/β ( Figure 6B ) . In permanganate footprinting assays both mutants supported open complex formation similar to WT TFEα/β or TFEα/β ΔWH ( Figure 6C ) . In order to test the influence of TFEα/β on the formation of the first phosphodiester bond we developed a promoter-dependent dinucleotide extension assay . RNAP is able to add one ATP molecule to an ApG dinucleotide to synthesize an ApGpA trinucleotide in a TBP/TFB factor-dependent fashion ( Figure 7A ) . This reaction has a low TFB-independent background , but is strictly dependent on the DNA sequence of the promoter since it relies on ApG and ATP ( Figure 7A ) . TFEα/β ( 1 μM ) stimulates abortive transcription activity on closed promoter templates ( 2 . 3 ± 0 . 5 fold ) as well as pre-opened ( 1 . 6 ± 0 . 1 fold ) templates ( Figure 7A ) while no stimulation was observed with TFEα ( data not shown ) . 10 . 7554/eLife . 08378 . 018Figure 7 . TFEα/β stimulates abortive and productive transcription . ( A ) Abortive transcription assays measuring ApGpA trinucleotide synthesis on −4/−1 heteroduplex and homoduplex SSV1-T6 promoter templates in the presence or absence of 1 µM TFEα/β . ( B ) NTS sequence of the different promoters tested in productive transcription assays . The TSS is boxed and putative TATA-boxes are shown in bold . Residues underlined are C to G mutations in order to construct a C-less cassette . ( C ) Productive transcription assays on four different promoters . Circular relaxed plasmids with different S . solfataricus promoters fused to C-less cassettes were used as templates . The position of the run-off transcripts ( asterisk ) and its expected size are indicated . A recovery marker ( r . m . ) was included . The lower panels show the quantifications of synthesized transcript . ( D ) Effect of the initially melted region on TFEα/β stimulation . Productive transcription assays with hybrid promoters encompassing the TATA-box and surrounding region ( position −46 to −13 ) of the weakly stimulated 16S/23S rRNA promoter with the initially melted and transcribed regions of the stronger stimulated tRNALeu ( −12 to +5 ) or EF-1α promoters ( −12 to +7 ) . ( E ) Effect of TFEα/β mutant variants ( 3 µM ) on productive transcription on the T6 promoter . The mean of three technical replicates is shown . Error bars depict 1× standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 08378 . 01810 . 7554/eLife . 08378 . 019Figure 7—figure supplement 1 . Productive transcription assays using the SSB and Rpo5 promoters . Circular relaxed plasmids with different S . solfataricus promoters fused to C-less cassettes were used as templates . The position of the run-off transcripts ( asterisk ) and its expected size is indicated . A recovery marker ( r . m . ) was included . The lower panels show the quantifications of synthesized transcript . DOI: http://dx . doi . org/10 . 7554/eLife . 08378 . 01910 . 7554/eLife . 08378 . 020Figure 7—figure supplement 2 . Effect of TFEα/β mutant variants on productive transcription on the Rpo5 promoter . Transcription assays and quantification were carried out as described above with 3 µM TFEα/β or variant . The mean of three technical replicates is shown . Error bars depict 1× standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 08378 . 020 In order to ascertain the function of TFEα/β on productive transcription we fused different promoters to a C-less cassette and carried out transcription assays in the presence of GTP , ATP and UTP . We compared the viral T6 and five cellular promoters of protein encoding genes ( Sso EF-1α , Sso SSB , and Sso Rpo5 ) , as well as noncoding RNA genes ( Sso tRNALeu and 16S/23S rRNA ) ( Figure 7B ) . On all tested promoters the addition of TFEα/β stimulated transcription by approximately twofold to fourfold , with the strong T6 and 16S/23S rRNA promoters showing a weaker response ( Figure 7C and Figure 7—figure supplement 1 ) . These results suggest that the stimulation of transcription by TFEα/β is dependent on the sequence of the promoter . Considering that TFEα/β stimulates DNA melting we hypothesised that the initially melted region of the promoter ( −12 to +4 ) determines the amplitude of the stimulation . We generated hybrid promoters encompassing the TATA-box and surrounding region ( position −46 to −13 ) of the weakly stimulated 16S/23S rRNA promoter with the initially melted and transcribed regions of the stronger stimulated Sso tRNALeu ( −12 to +5 ) or EF-1α promoters ( −12 to +7 ) . In absence of TFEα/β the hybrid promoters show reduced activity compared to the wild-type rRNA promoter underlining that the initially melted region contributes to the strength of the ribosomal promoter ( Figure 7D ) . In line with our hypothesis TFEα/β stimulated transcription on the two hybrid promoters to greater extent when compared to the wild-type rRNA promoter , confirming that the sequence of the initially melted region determines the extent of TFEα/β stimulation . Interestingly , both TFEβ WH deletion and charge reversal mutations , and the TFEα ZR deletion mutants were able to stimulate transcription from the T6 and Rpo5 promoters ( Figure 7E and Figure 7—figure supplement 2 ) . In summary , TFEα/β appears as a basal transcription factor that stimulates transcription on mRNA as well as noncoding RNA genes . We have discovered the TFIIEβ homologue in archaea , TFEβ , and characterised its structure and function . The gene encoding TFEβ is essential in S . acidocaldarius and most likely this is true for crenarchaea in general given its strict conservation in the crenarchaeal phylum . On the sequence level TFEβ is homologous to the eukaryotic RNAPIII subunit hRPC39 and we show that both Sso TFEβ and human hRPC39 harbour a cubane FeS cluster at their C-termini . Several proteins involved in nucleic acid metabolism have been reported to harbour FeS clusters including multi-subunit RNAPs and TFIIH subunit XPD/Rad3 ( White and Dillingham , 2012 ) . Notably , the Sso RNAP subunit Rpo3 includes a structural [4Fe-4S] cluster that is highly stable under aerobic conditions ( Hirata et al . , 2008 ) . Although TFEβ could be purified under aerobic conditions with high FeS cluster occupancy , the FeS cluster was lost within a few hours , as apparent from the decolouration of the protein preparation . Furthermore , oxidation with ferricyanide oxidation causes destruction of the TFEβ cluster ( data not shown ) . This difference between the two FeS clusters in Sso RNAP and TFEβ likely reflects different functions . The FeS cluster in the catalytic subunit of yeast DNA Polymerase Pol δ is essential for the interaction with its two auxiliary subunits Pol31 and Pol32 ( Netz et al . , 2012 ) . Similarly , the FeS cluster is required for TFEα/β dimerization and cluster damage results in its dissociation ( Figure 4 ) . Our results demonstrate that steady-state levels of TFEβ are depleted in the stationary phase—the FeS cluster in TFEα/β may provide a handle for fast inactivation of TFEα/β during stress response . In line with this hypothesis we found oxidative stress induced by hydrogen peroxide leads to rapid depletion of Sso TFEβ ( data not shown ) . The minimal domain requirements for the heterodimerisation of TFEα and β encompass the TFEα WH and tail domains , and the TFEβ FeS domain ( Figures 3D , 8A ) . The position of euryarchaeal TFEα on the RNAP clamp and Rpo4/7 stalk has previously been mapped in the context of the complete PIC from the euryarchaeote M . jannaschii ( Grohmann et al . , 2011; Nagy et al . , 2015 ) . Likewise , Sso TFEα/β forms a stable complex with a recombinant RNAP clamp . Our results show that the TFEα ZR makes extensive contacts with the RNAP clamp . The TFEβ FeS domain stabilizes the binding either by directly interacting with the RNAP clamp or indirectly by altering the conformation of TFEα ( Figure 5 ) . In contrast , the TFEβ WH domain is not involved . The overall structural organisation of the crenarchaeal RNAP-TFEα/β ensemble is consistent with PIC models in eukaryotes ( RNAPII , He et al . , 2013 ) as well as euryarchaea ( Grohmann et al . , 2011 ) , and the architecture of RNAPIII ( Vannini et al . , 2010; Wu et al . , 2012 ) . The TFEα WH domain is located on the apex of the RNAP clamp , while its ZR domain projects towards and interacts with the base of the RNAP clamp and stalk module ( Grohmann et al . , 2011; He et al . , 2013 ) ( Figure 8A ) . The TFEβ WH domain is likely to project across the DNA binding channel prone to make contacts with the promoter template . The TFEβ FeS domain may provide additional binding surface to the RNAP clamp . 10 . 7554/eLife . 08378 . 021Figure 8 . Structure and evolution of TFEα/β . ( A ) Model of the dimerization interface of TFEα/β and its interaction with the RNAP clamp based on data presented here and elsewhere ( Grohmann et al . , 2011; Grünberg et al . , 2012; He et al . , 2013 ) . Grey areas indicate domains of TFEα and β that are required for dimerization . Dashed lines indicate interactions between the WH and zinc ribbon domains of TFEα and the RNAP clamp . The FeS domain of TFEβ stabilizes the interaction with the RNAP clamp and might directly bind . ( B ) A scenario for the evolution of TFIIE-like proteins in archaea and eukaryotes under assumption of an archaeal origin of eukaryotes according to the ‘eocyte’ hypothesis ( Guy and Ettema , 2011; Williams et al . , 2013 ) . Our model includes only Euryarchaeota and Crenarchaeota , the two most studied archaeal clades . Five key steps are indicated with red numbers: 1—TFEα and TFEβ WH domains are related to bacterial MarR-type WH domains indicating their early evolutionary origin ( Aravind et al . , 2005; Blombach et al . , 2009 ) . 2—The wide distribution of TFEα and TFEβ encoding genes indicate that both genes date back to the last common archaeal ancestor . 3—In several euryarchaeal species the TFEβ encoding gene was lost ( Blombach et al . , 2009 ) and all biochemically characterised euryarchaeal TFEα appear to function as monomeric factors . Nevertheless , dimeric TFEα/β may exist in other euryarchaeal species . 4—Latest possible emergence of the dimeric TFEα/β-like factor predating the split of Crenarchaeota and the ‘archaeal parent’ of eukaryotes . After gene duplication , this precursor evolved to give rise to TFIIE in the RNAPII transcription machinery and the RNAPIII subcomplex hRPC62/39 . 5—Eukaryotic RNAPI has lost its dependence on TFIIE-like factors . DOI: http://dx . doi . org/10 . 7554/eLife . 08378 . 021 Following RNAP recruitment to the promoter TFEα/β stabilises the PIC ( Figure 6B ) , and this is dependent on the TFEα ZR and the TFEβ WH domain while TFEα alone has no apparent effect on PIC formation ( Figure 6B ) similar to its human homologue TFIIEα ( Peterson et al . , 1991 ) . The proximity of M . jannaschii TFEα WH with the NTS at position −12 ( Grohmann et al . , 2011 ) and the location of the human TFIIEβ WH across the DNA binding channel ( He et al . , 2013 ) suggest that TFEα/β modulates the handling of the DNA strands by RNAP , for example , during DNA melting and open complex formation . Our permanganate footprinting experiments suggest that TFEα/β , and not TFEα , triggers DNA melting creating a bubble ranging in its extremes from position +4 to −12 position relative to the TSS ( Figure 6C ) . TFEα/β , but not TFEα , stimulates the formation of the first phosphodiester bond in an abortive transcription assay as well as productive transcription . The stimulatory effect of TFEα/β on DNA melting and productive transcription appears not to depend on the TFEα ZR and TFEβ WH domains ( Figures 6C , 7E and Figure 7—figure supplement 2 ) . In contrast , in experiments monitoring PIC formation using EMSA that relies to a greater extent on complex stability ( as the complexes have to remain intact during electrophoresis ) , deletion of either TFEα ZR or TFEβ WH strongly reduces PIC stability ( Figure 6B ) . Similarly , stable interaction with the RNAP clamp depends on the TFEα ZR ( Figure 5 ) . This suggests that the core TFEα/β composed of the TFEα WH and the TFEβ FeS domains can trigger DNA melting and thereby stimulate transcription initiation . The TFEα ZR or TFEβ WH domains bring addition stability to the complex that becomes crucial in the cellular context . In line with a stabilizing role for the ZR domain , mutation of the cysteine residues in the yeast Tfa1 ZR in vivo confer a thermosensitive phenotype ( Kuldell and Buratowski , 1997 ) . We assessed the impact of TFEα/β on the transcription directed by a range of protein-encoding and noncoding RNA promoters . While hRPC39 as part of RNAPIII plays a role in the specific recruitment of RNAPIII to its noncoding RNA gene promoters , the hRPC39-like archaeal TFEβ clearly acts as a general transcription initiation factor stimulating transcription from all promoters tested . Our results show that the steady state levels of TFEβ are drastically reduced in stationary phase compared to exponential growth phase , in contrast to other components of the basal Sso transcription apparatus . TFEα/β stimulates transcription of different genes to varying extent , dependent on the sequence of the initiatlly melted region within the promoter . Thus , promoters that are strongly stimulated by TFEα/β could be downregulated in stationary phase , while expression of TFEα/β unresponsive ( or only mildly stimulated ) promoters would be less affected . We do not provide direct evidence for a regulatory role of TFEα/β , however , our results suggest that TFEβ has the potential to reprogram transcription in response to different growth phases . TFEα/β's function as basal transcription factor and potential regulator is not unprecedented; TBP-related factors regulate transcription in metazoans ( Goodrich and Tjian , 2010; Duttke et al . , 2014; Wang et al . , 2014 ) , and multiple TBP and TFB paralogs regulate transcription in a gene-specific fashion in archaea ( Facciotti et al . , 2007 ) . The archaeal TFEα/β factor provides us with a missing link in the evolutionary history of the archaeal and multiple eukaryotic transcription machineries ( Figure 8B ) . In the most parsimonious scenario , the archaeal ‘parent’ of eukaryotes ( Yutin et al . , 2008; Guy and Ettema , 2011 ) included a TFEα/β factor with the domain architecture we have discovered and described here . After the split of the archaeal and eukaryotic lineages about 2 billion years ago , and following gene duplication and speciation of both TFEα and β subunits in eukaryotes , the FeS domain of TFIIEβ was lost in the RNAPII system along with the evolution of a new dimer interface ( Grünberg et al . , 2012 ) . The loss of the FeS cluster in TFIIEβ might have evolved via a route analogous to the apparent loss of the FeS cluster in yeast C34 ( Figure 2G ) . In the RNAPIII system the hRPC62/39 complex with the addition of a third , not universally conserved subunit hRPC32 ( C31 in yeast ) ( Proshkina et al . , 2006 ) became stably incorporated into RNAPIII , most likely via a WH domain duplication turning the single WH domain in TFEα to four WH domains of hRPC62 ( Lefèvre et al . , 2011 ) which provided a drastically enhanced binding surface ( Vannini et al . , 2010 ) . This may have provided a selective advantage considering that transcription by RNAPIII is characterised by high rates of transcription reinitiation and short transcript length . It seems likely that the essential role of the FeS cluster in TFEα/β dimerization is conserved in hRPC62/39 since a truncated form of human hRPC39 encompassing the third WH domain and the FeS domain supports dimerization ( Lefèvre et al . , 2011 ) . Overall the diverse activities of TFEα/β described here are consistent with the activities of TFEα from euryarchaeal transcription systems that do not encompass TFEβ homologues such as M . jannaschii and Pyrococcus furiosus ( Hanzelka et al . , 2001; Werner and Weinzierl , 2005; Naji et al . , 2007; Kostrewa et al . , 2009; Grohmann et al . , 2011 ) . Monomeric Sso TFEα did not stimulate PIC formation or transcription ( Figures 6B , 7C ) . It is worth noting that Sso TFEα binds to the RNAP clamp ( Figure 5 ) and in vitro transcription experiments suggest that Sso TFEα competes with Sso TFEα/β for incorporation into the PIC ( data not shown ) . Hence , Sso TFEα appears to be recruited to the PIC but does not exert any detectable effect on transcriptional activity . Consistent with our results , a previous study suggested that monomeric Sso TFEα does not affect transcription output from the viral T6 promoter when non-limiting TBP concentrations similar to those used in this study were used ( Bell et al . , 2001 ) . Loss of the TFEβ encoding gene in euryarchaeal species might be compensated for by changes in TFEα and RNAP , particularly regarding the mechanisms and dynamics of clamp opening . The structure of the Thermococcus kodakaraensis RNAP revealed the clamp in an open conformation ( Jun et al . , 2014 ) and this species has no TFEβ homologue . The Sso RNAP and yeast RNAPII crystallise with a closed clamp , and both systems utilise TF ( II ) Eβ for initiation ( Bushnell and Kornberg , 2003; Hirata et al . , 2008; Korkhin et al . , 2009; this work ) . We envisage a model where TFE engages with and possibly opens the RNAP clamp , which stimulates open complex formation and transcription . In organisms that lack TFEβ the opening of the clamp occurs more readily , which is reflected in the corresponding RNAP structures; in some species clamp opening occurs entirely factor-indpendent since genes encoding TFEα and TFEβ homologues are missing in the euryarchaeal Thermoplasmata class . What was the selective advantage that led to the emergence of TFEβ ? In budding yeast , the Tfa1 WH domain 1 interacts with the Ssl2 subunit ( termed XPB in human and archaea ) of TFIIH ( Grünberg et al . , 2012 ) which facilitates DNA nucleotide excision repair ( Rouillon and White , 2011 ) and ATP-dependent DNA melting during transcription initiation ( Holstege et al . , 1996; Kim et al . , 2000 ) . In archaea , open complex formation occurs spontaneously due to the torsionally strained topology of the promoter DNA in the PIC ( Nagy et al . , 2015 ) and there is no evidence of any involvement of archaeal XPB in transcription initiation . However , the TFEβ WH domain would be poised to enable the recruitment or integration of XPB-containing complexes into the PIC . The discovery and characterisation of Sso TFEα/β provides an important piece of the evolutionary puzzle of TFIIE-like proteins . It suggests that streamlining occurred in the evolution of the archaeal transcription machinery , that is , the loss of tfeβ in several species ( Blombach et al . , 2009 ) . In eukaryotes on the other hand , evolution led to an increase in complexity chiefly by WH domain duplication events . Cloning via restriction sites was performed with PCR products amplified from S . solfataricus str . P2 genomic DNA . For co-expression of Sso TFEα ( Sso0266 ) and Sso TFEβ-His ( Sso 0944 ) a bicistronic expression construct was produced ( p1076 ) . A corresponding construct for co-expression of hRPC62 and hRPC39 ( p1159 ) was created likewise . To generate a pRSF-1b-based vector for expression of C-terminally His-tagged Sso TFEβ , the gene was amplified from a pET21a+ expression vector including the vector-encoded His-tag and inserted into the pRSF-1b vector that does not encode a His-Tag otherwise . The resulting construct ( p1077 ) was also used as backbone to generate the different Sso TFEβ truncation mutants . All plasmids generated by restriction enzyme-based cloning are listed in Supplementary file 1 . Site-directed mutagenesis of TFEα and TFEβ was performed to yield single or double amino acid substitutions as listed in Supplementary file 2 . For the deletion of nucleotides coding for residues 46–52 in Sso TFEα , a 5′-end phosporylated non-overlapping primer pair was used according to the Phusion Site-directed mutagenesis protocol ( Thermo Scientific/Fisher , Loughborough , United Kingdom ) . All oligonucleotide sequences are listed in Supplementary file 3 . The sequence of all constructs was verified . For isolation of Sso RNAP , pSVA158 was transformed into S . sulfolobus M16 cells resulting in the expression of a C-terminally His10-tagged RNAP subunit Rpo8 and homologous expression was carried out as described previously ( Albers et al . , 2006 ) . Cells were harvested by centrifugation , snap frozen in liquid N2 and stored at −80°C . Cells were resuspended in 30 ml N buffer ( 25 mM Tris/HCl pH 8 . 0 , 10 mM MgCl2 , 100 μM ZnSO4 , 5 mM 2-mercapto-ethanol , 10% glycerol ) with 100 mM NaCl ( N ( 100 ) , salt concentration given in parenthesis ) supplemented with EDTA-free protease inhibitor tablets ( Roche , Burgess Hill , United Kingdom ) and DNase I ( Sigma , Gillingham , United Kingdom ) . After disruption using a French Pressure cell at 16 , 000 psi the lysate was cleared by centrifugation and filtration . The salt concentration was adjusted to 500 mM NaCl and the lysate was loaded onto a 1 ml Histrap ff cartridge ( GE Life Sciences , Uppsala , Sweden ) . The column was washed with 10 ml N ( 500 ) , 20 mM imidazole and RNAP was eluted using a gradient to 250 mM imidazole . Elution fractions were combined , diluted with N ( 0 ) buffer to 100 mM NaCl , and loaded onto a 1 ml HiTrap Heparin HP cartridge ( GE Life Sciences ) . Sso RNAP was eluted with a 10 ml gradient to N ( 1000 ) yielding a single sharp peak for Sso RNAP . The peak fractions were combined and desalted using a PD-10 desalting column ( GE Life Sciences ) to N ( 150 ) . All heterologous expression was performed in E . coli Rosetta 2 ( DE3 ) ( Merck Millipore , Billerica , MA ) or BL21 Star ( DE3 ) ( Life Technologies , Paisley , United Kingdom ) in enriched growth medium at 37°C according to standard procedures . For the expression of Sso TFEβ , Sso TFEα/β and all the mutant versions of these proteins as well as hC62/39 in BL21 Star ( DE3 ) 0 . 5 mM L-cysteine and 0 . 5 mM Ammonium ferric citrate were added after induction in order to maximize FeS cluster occupancy and expression was allowed for 2 . 5 hr . Sso TBP and Sso TFB-His were expressed from plasmids p1121 , 1087 , and p1168 respectively and purified as previously described ( Gietl et al . , 2014 ) . Sso TFEα was expressed in Rosetta 2 ( DE3 ) cells from plasmid p988 . Cells were resuspended in N ( 300 ) and disrupted using a using a French Pressure cell at 16 , 000 psi . The lysate was cleared by centrifugation , incubated for 30 min at 70°C and denatured host proteins were removed by centrifugation . The heat-stable lysate was further purified on a HiPrep 16/60 Sephacryl S-100 HR column ( GE Life Sciences ) . TFEα containing fractions were combined and threefold diluted with N ( 0 ) to 100 mM NaCl . The protein was loaded onto a UnoQ-1 column ( Bio-Rad , Hemel Hempstead , United Kingdom ) and eluted using a 10 ml gradient to N ( 1000 ) . The purity of TFEα was judged by SDS-PAGE and fractions containing TFEα with >95% purity were combined and concentrated by ultrafiltration , snap frozen in liquid nitrogen and stored at −80°C . For the purification of Sso TFEα/β-His and all mutant versions , cells were resuspended in TK buffer ( 20 mM Tris/HCl pH 8 . 0 , 100 μM ZnSO4 , 5 mM DTT ) with 500 mM KCl ( TK ( 500 ) , salt concentration given in parenthesis ) supplemented with 5 mM imidazole and disrupted by sonication . After centrifugation the supernatant was incubated at 65°C for 20 min and centrifuged again . The heat-stable supernatant was purified on a 1 ml Histrap ff cartridge ( GE Life Sciences ) . Sso TFEα/β-His-containing elution fractions were pooled , diluted fivefold with TK ( 0 ) buffer to 100 mM KCl and loaded on a 1 ml HiTrap Heparin HP cartridge ( GE Life Sciences ) and eluted with a 10 ml gradient to TK ( 1000 ) . Fractions were concentrated in AMICON ULTRA 0 . 5 ml MWCO10000 ultrafiltration devices . Aliquots were snap-frozen in liquid nitrogen and stored at −80°C before usage . The whole purification was carried out in a single day to limit oxidation of the Fe-S cluster . For the purification of mutant versions of Sso TFEα/β-His and WT TFEα/β-His serving as control the heparin-affinity chromatography was omitted and instead the protein was buffer changed on a PD-10 column ( GE Life Sciences ) to TK ( 150 ) . Human N-terminally His-tagged C62/C39 was purified as follows: Cells were resuspended in TK ( 150 ) , 20 mM imidazole and disrupted by sonication . After removal of cell debris by centrifugation the supernatant was purified on a 1 ml Histrap ff cartridge ( GE Life Sciences ) . Elution fractions containing protein were combined and diluted with 0 . 5 vol water to 100 mM salt . The protein was further purified and buffer exchanged on a 1 ml HiTrap Heparin HP cartridge ( GE Life Sciences ) equilibrated in 0 . 1 M NH4-Acetate and eluted as a single peak using a 10 ml gradient to 1 M NH4-Acetate in order to have a protein preparation compatible with native MS . A construct for the expression of a recombinant Sso RNAP clamp module ( p1172 ) was designed according to a similar construct for P . furiosus ( Martinez-Rucobo et al . , 2011 ) . The Sso RNAP clamp module was purified as described above for TFEα/β-His . Protein concentrations were determined using the Qubit assay ( Life Technologies ) . Polyclonal rabbit antisera against recombinant Sso TBP , Sso TFEα and Sso TFEβ were raised at Davids Biotechnology ( Regensburg , Germany ) . For the detection of Sso TFB we used antisera raised against S . acidocaldarius TFB ( Gietl et al . , 2014 ) . Rabbit antiserum against S . solfataricus RpoB was obtained from Steve Bell ( Indiana University , USA ) ( Qureshi et al . , 1997 ) . For immunodetection , proteins were generally resolved by 14% SDS-PAGE , transferred to nitrocellulose membranes using a tank or semi-dry blotting system , and immunodetection was performed using PBS buffer with 5% milk powder as blocking reagent . The blots were incubated with the respective rabbit antisera and Dylight 680 conjugated goat anti-rabbit IgG ( Thermo Scientific ) as secondary antibody , scanned on a Typhoon FLA 9500 scanner ( GE Life Sciences ) equipped with a 685 nm laser . For multiplex immunodetection including Alba as loading control , sheep anti-Alba antiserum ( obtained from Malcolm White , University of St . Andrews , UK ) and donkey Dylight 488 conjugated anti-goat IgG were used alongside Dylight 680 conjugated donkey anti-rabbit IgG ( Bethyl Laboratories , Cambridge , United Kingdom ) . We have previously shown that Alba is stably expressed throughout all growth phases ( Blombach et al . , 2014 ) . For homologous expression of Sso0944 the gene was cloned into vector pMJ0503 ( Jonuscheit et al . , 2003 ) . The resulting plasmid p1056 was transformed into S . solfataricus M16 cells and expression was carried out as described previously ( Jonuscheit et al . , 2003 ) . Cells were resuspended in 20 ml TK ( 150 ) supplemented with 2 . 5 mM MgCl2 , DNase I ( Sigma ) , and EDTA-free protease inhibitor ( Roche ) . Cells were disrupted by threefold passage through a French pressure cell ( Thermo Scientific ) at 16 , 000 psi . Cell debris was removed by centrifugation at 30 , 000×g and filtration through a 0 . 22 μm filter . The cleared cell lysate was loaded onto a 1 ml Histrap ff cartridge ( GE Life Sciences ) equilibrated in TK ( 150 ) buffer and the column was washed with 5 ml TK ( 150 ) and 5 ml TK ( 150 ) containing 20 mM imidazole . Protein was eluted with a 5 ml gradient to 250 mM imidazole and 0 . 5 ml fractions were collected . For the more stringent purification presented in Figure 1—figure supplement 1 the second wash step was altered to 20 ml buffer containing 50 mM imidazole . S . solfataricus P2 cells were grown in Brock medium , 0 . 1% NZ-amine , 0 . 2% sucrose to late exponential growth phase ( OD600 = 1 . 1 ) , harvested by centrifugation and stored at −80°C . 5 g cells were resuspended in 20 ml TK ( 150 ) , 10 mM MgCl2 , supplemented with EDTA-free protease inhibitor ( Roche ) and 150 μ DNase I ( Sigma ) and passed thrice through a French Pressure cell at 16 , 000 psi . The lysate was cleared by centrifugation ( 45 , 000×g , 1 hr at 4°C ) and passage through a 0 . 22 μm filter . 250 μl of lysate ( 10 mg/ml protein content ) were fractionated on a Superose 12 10/300 GL column ( GE Life Sciences ) with 0 . 5 ml fraction size . Fractions were analysed by immunodetection . S . solfataricus P2 cells were grown in shake flasks in 1 litre Brock medium ( Zaparty et al . , 2010 ) supplemented with 0 . 1% ( wt/vol ) tryptone and 0 . 2% glucose at 76°C under aerobic conditions in triplicate . After 24 hr the cultures reach O . D . 600 = 0 . 4 ( exponential growth phase ) and 50 ml were withdrawn . After 96 hr at O . D . 600 = 2 . 3 ( stationary growth phase ) 10 ml were withdrawn . All samples were immediately chilled on ice , cells were harvested by centrifugation and stored at −80°C . Cells were then resuspended in 1 ml TK ( 150 ) buffer supplemented with 2 . 5 mM MgCl2 , DNase I ( Sigma ) , and EDTA-free protease inhibitor ( Roche ) . Cells were disrupted using a cup sonicator and lysates were cleared by centrifugation ( 20 min at 20 , 000×g , 4°C ) . Protein concentrations in the cleared lysates were determined using the Qubit assay ( Life Technologies ) . Lysates were diluted to 1 . 2 mg/ml , mixed with SDS loading dye and resolved by 14% SDS-PAGE . Proteins were transferred onto nitrocellulose membranes by semi-dry blotting . To ensure equal loading and transfer , the membranes were stained with Ponceau S before immunodetection performed as described above . In parallel , a twofold dilution series of recombinant untagged TFEα/β or TBP was loaded on to the gels to generate a calibration curve used to allow determination of the expression levels . The procedure to disrupt Saci_1342 basically followed the method described by ( Wagner et al . , 2012 ) using pyrEF as marker gene and the uracil-auxotrophic S . acidocaldarius strain MW001 as parental strain . The upstream and downstream flanks of Saci_1342 ( 751 bp upstream flank including the initial 9 bp of Saci_1342 and 700 bp downstream flank including 26 bp at the 3′ end of Saci_1342 , as this region overlaps with the convergent ORF Saci_1343 ) were PCR-amplified with primer pairs FW282/283 and FW284/FW285 , respectively , fused by overlap extension PCR and inserted into vector pSVA406 via NcoI and BamHI restriction sites yielding plasmid p1058 . After EsaBC41 cytosine-methylation in E . coli ( Grogan , 2003 ) p1058 was transformed into electrocompetent MW001 cells and single-crossover integrants were selected on uracil-deficient plates . Four single-crossover integrants were selected and plated on uracil and 5-FOA containing plates for counterselection against pyrEF triggering excision of p1058 . Strain MW001 Saci1162::Saci1342 was constructed plasmid 1112 similar to ( Meyer and Albers , 2014 ) . Strain MW001 Saci1162::Saci_1342 was then transformed with p1058 as described above . Successful replacement of Saci_1162 and deletion of Saci_1342 was confirmed by DNA sequencing . For co-purification assays , pET-21a ( + ) and pRSF-1b-based vectors for the expression of Sso TFEα and Sso TFEβ or their mutant versions were co-transformed into BL21* ( DE3 ) cells ( Life Technologies ) . Proteins were expressed in enriched growth medium for 2 . 5 hr at 37°C after induction with IPTG . Cells from 400 ml culture were resuspended in 4 ml TK ( 500 ) buffer and disrupted by sonication . The cleared lysate was incubated at 65°C for 20 min . The heat-stable supernatant was mixed with 0 . 2 ml pre-equilibrated His-Select resin ( Sigma–Aldrich ) and incubated for 10 min at 4°C . The resin was washed with 5 ml buffer TK ( 500 ) containing 5 mM imidazole . Bound proteins were eluted by addition of 1 ml TK ( 500 ) containing 250 mM imidazole . Samples were analysed by Tris-Tricine SDS-PAGE . 10 μM Clamp module were mixed with 10 μM of SsoTFEαβ or mutants thereof in TK ( 250 ) buffer , 10% glycerol , 10 mM MgCl2 and incubated at 37°C for 10 min . After brief centrifugation , 250 μl were loaded on a Superose 12 10/300 GL column ( GE Life Sciences ) equilibrated in the same buffer plus 0 . 05% TWEEN 20 . The fraction size was 0 . 5 ml . The elution profile of the respective proteins was visualized on silver-stained SDS-PAGE gels . All samples were measured in an AvaSpec-2048 fiber optic spectrophotometer using a 10 mm pathlength quartz cuvette at a protein concentration of 5 µM and 2 µM for Sso TFEα/β and C62/39 , respectively . For EPR measurements Sso TFEα/β-His and human N-terminally His-tagged C62/C39 samples were aerobically purified and concentrated to 550 and 86 μM respectively , based on A280 absorption and the molar extinction coefficient . As-isolated samples were transferred directly to EPR quartz tubes and flash frozen in liquid nitrogen . Reduced samples were prepared by supplementing a fraction of the same protein stock with 20 mM sodium dithionite by adding 1/100 vol of a 2 M solution to the buffer before freezing . All samples were stored in liquid nitrogen prior to cw-EPR experiments , and showed no change in EPR signal over a period of several months . Cw-EPR measurements were performed on a Bruker EMXplus spectrometer operating at 9 . 4 GHz ( X-band ) equipped with a 4122SHQE resonator , with an Oxford Instruments ESR900 cryostat for measurements at cryogenic temperatures . Typically spectra were acquired in the temperature interval 10–40 K to enable FeS cluster identification . Measurements were performed with a magnetic field sweep from 50 to 600 mT ( to allow the detection of possible high and low spin Fe species ) , a microwave power of 2 mW , modulation amplitude of 0 . 5 mT and a modulation frequency of 100 kHz . Three independent preparations were tested , all of which gave consistent results . Spin quantification was carried out by comparison with a standard solution of Cu ( II ) EDTA according to the method reported in ( chasteen ) . The magnetic field was calibrated with a bismuth doped silicon sample . For nano-electrospray ionization mass spectrometry of TFEα/β-His the heparin-affinity chromatography step was carried out with 0 . 3–1 M NH4-Acetate . Prior to analysis by native mass spectrometry , protein samples were buffer-exchanged into 150 mM ammonium acetate and concentrated to ∼10 μM using Amicon Ultra 0 . 5-ml centrifugal filters ( Millipore ) . Mass spectrometry experiments were carried out on a first-generation Synapt HDMS ( Waters , Manchester , UK ) Quadrupole-TOF , traveling wave ion mobility mass spectrometer ( Pringle et al . , 2007 ) . Samples ( 2- to 3-μl aliquots ) were introduced to the mass spectrometer by means of nanoelectrospray ( nESI ) ionization using gold-coated capillaries that were prepared in-house . Typical instrumental parameters were as follows: source pressure , 5 mbar; capillary voltage , 1 . 0–1 . 3 kV; cone voltage , 40 V; trap energy , 15 V; transfer energy , 10 V; bias , 2 . 0 V . For tandem MS experiments , the backing pressure was reduced to 1 . 4 mbar and the trap and transfer voltages were increased to 60 V and 20 V respectively . MS were smoothed and peak-centered in MassLynx v4 . 1 ( Waters , Elstree , United Kingdom ) . 15 μl samples contained 10 mM MOPS pH 6 . 5 , 11 mM MgCl2 , 150 mM salt ( 115 mM KCl , 27 mM NaCl , 8 mM K-Acetate ) , 10% glycerol , 5 μg/ml heparin , 63 nM TFB , 250 nM TBP , 270 ng RNAP ( 45 nM ) and 125 fmol of 32P 5′-labelled dsDNA templates . Samples were incubated for 5 min at 65°C before loading onto a 5% native Tris-Glycine gels ( 2 . 5% glycerol , 1 mM DTT ) . To generate antibody-supershifts , the Protein A-purified anti-TFEα , anti-TFEβ , or anti-TBP ( 0 . 7 µg ) antiserum was added to the incubated samples and incubation continued for 15 min at room temperature prior to gel loading . Gels were run at 150 V , dried and radiolabelled DNA was detected by phosphorimagery . For potassium permanganate footprinting the samples were as described above but scaled up to 23 μl . TBP and TFB concentrations were raised to 1 μM and 125 nM , respectively , MgCl2 to 26 mM , and DTT was reduced to 0 . 5 mM . Samples were incubated for 5 min at 65°C before the addition of 2 μl 50 mM KMnO4 ( 4 mM final concentration ) . Samples were further incubated at 65°C for 5 min before the addition of 1 . 5 μl 2-mercaptoethanol to stop the reaction . After Proteinase K-treatment and ethanol-precipitation , samples were resuspended in 50 μl 1 M piperidine and incubated at 90°C for 30 min . After chloroform-extraction , samples were ethanol-precipitated and resuspended in formamide loading dye . Samples were resolved on a 10% polyacrylamide , 7 M Urea , 1× TBE sequencing gels . Radiolabelled DNA was detected by phosphor imagery . 15 μl samples for abortive transcription assays contained 500 fmol of dsDNA template pol592/593 ( homoduplex ) or pol592/603 ( −4 to −1 heteroduplex ) , 250 μM ApG dinucleotide , 50 μM ATP ( containing [α-32P]-ATP ) , 125 nM TFB , 1 μM TBP , 270 ng RNAP ( 45 nM ) , and 1 μM TFEα/β . Salt , buffer and heparin concentrations were identical to those used for the mobility shift experiments . Samples were incubated for 10 min at 65°C before addition of 1 vol formamide loading dye . 5 μl of the samples was loaded on a 20% 7 M Urea , 1× TBE PAGE mini-gel and resolved at 300 V for 45 min . Signals were detected by phosphorimagery and quantified using the ImageQuant TL software package ( GE Life Sciences ) . The quantification of TFEα/β stimulation was based on three technical replicates . The signal obtained in absence of TFB was subtracted as background . For promoter-directed in vitro transcription , different promoters fused to a C-less cassette derived from a synthetic 390 nt G-less cassette ( kindly provided by James Goodrich , CO ) ( Sawadogo and Roeder , 1985 ) were cloned into pGEM-T ( Promega ) ( Supplementary file 4 ) . All plasmids were treated converted to relaxed topology with E . coli Topoisomerase I ( NEB ) according to manufacturer's protocol . Transcription reactions were modified as follows: samples contained 500 μM ATP/GTP , 2 . 5 μM UTP ( containing [α-32P]-UTP ) and 200 ng of the respective plasmid template . Reactions were stopped after 10 min by the addition of 9 vol stop mix ( 0 . 3 M Na-acetate pH 5 . 2 , 10 mM EDTA , 0 . 5% SDS , 100 μg/ml glycogen and trace amounts of radiolabelled oligonucleotide serving as recovery marker ) . Samples were purified by phenol/chloroform extraction and ethanol-precipitation and resuspended in formamide-loading dye . Sample were resolved on a 10% polyacrylamide , 7 M Urea , 1× TBE sequencing gel . Transcripts were detected by phosphor imagery and quantification of bands was performed using the ImageQuant TL software ( GE Life Sciences ) . The signals were normalized using the recovery marker . When activities on different promoters were compared , the different specific activities of the transcripts were taken into account .

Life on Earth is often categorized into three domains: the eukaryotes ( which include plants , animals and fungi ) , the bacteria and a group of unusual , single-celled microorganisms called the archaea . But several recent discoveries suggest that the origin of the eukaryotes lies within the archaeal domain . The genetic material of all of these living organisms is made up of DNA , and genes within DNA contain the instructions to make other biological molecules . Making these molecules involves first copying these instructions into a molecule of RNA via a process called transcription . All three domains of life use enzymes called RNA polymerases ( RNAPs ) for transcription , and all RNAPs are thought to have originated from a common ancestor . Archaea and bacteria have a single type of RNAP , whereas all eukaryotes have at least four different kinds of RNAP . The RNAPs found in archaea share many common features with their eukaryotic counterparts . In both cases , the RNAPs do not work alone . Instead , a class of proteins known as transcription factors assist in the first step of the transcription process . One of the eukarotyic RNAPs , termed RNAP II , works with a transcription factor that contains two protein subunits ( called TFIIEα and TFIIEβ ) . While the archaeal counterpart for TFIIEα ( called TFEα ) is known , the counterpart for TFIIEβ is not . Blombach et al . have now identified the archaeal counterpart of TFIIEβ in a species of archaea called Sulfolobus and have renamed it TFEβ . Sulfolobus cells are unable to survive without this protein , which works in a similar way to TFIIEβ in assisting the RNAP to start transcription . Further analyses show that the TFEβ protein is actually related to a protein subunit that is unique to RNAP III , another eukarotyic RNAP . Both of these proteins contain clusters of iron and sulphur . Blombach et al . also found that these iron-sulphur clusters enable TFEβ to bind to its TFEα partner to form a transcription factor that can interact with the RNAP and help it to carry out transcription . These results suggest that the TFEα/β transcription factor found in archaea is likely to resemble the ancestor of the TFIIE transcription factors found in living eukaryotes . This discovery provides new insights in the evolutionary history of both the archaeal and the eukaryotic transcription machineries .

2015

Archaeal TFEα/β is a hybrid of TFIIE and the RNA polymerase III subcomplex hRPC62/39

Complex biological systems rely on cell surface cues that govern cellular self-recognition and selective interactions with appropriate partners . Molecular diversification of cell surface recognition molecules through DNA recombination and complex alternative splicing has emerged as an important principle for encoding such interactions . However , the lack of tools to specifically detect and quantify receptor protein isoforms is a major impediment to functional studies . We here developed a workflow for targeted mass spectrometry by selected reaction monitoring that permits quantitative assessment of highly diversified protein families . We apply this workflow to dissecting the molecular diversity of the neuronal neurexin receptors and uncover an alternative splicing-dependent recognition code for synaptic ligands . The remarkable anatomical and functional complexity of nervous systems relies on molecular programs for cell intrinsic properties and selective cellular interactions . Major advances in transcriptomics have enabled the identification of gene regulatory programs and mRNA targets that underlie specification of neuronal cell types and their plasticity ( Hobert , 2011; Ebert and Greenberg , 2013; Molyneaux et al . , 2015 ) . For example , specific transcriptional programs direct the neurotransmitter phenotypes of neuronal populations , the targeting of axonal projections , or the modification of synapse numbers in response to neuronal activity . In addition , transcript-based studies have uncovered gene families with substantial molecular complexity that may encode neuronal recognition events ( Zipursky and Sanes , 2010; Schreiner et al . , 2014a ) . What remains a major challenge is the exploration of such molecular programs and their function at the protein level . mRNA and protein turnover rates as well as mRNA translation rates exhibit a significant dynamic range . Thus , transcript abundance cannot easily be extrapolated to provide quantitative assessments of the proteome or insights into the stoichiometry of protein complexes ( Helbig et al . , 2011; Schwanhausser et al . , 2011; Vogel and Marcotte , 2012 ) . Notably , such post-transcriptional forms of gene regulation are particularly prevalent in the central nervous system highlighting the need for quantitative approaches that enable targeted dissection of the neuronal proteome . Additionally , many neurons possess long-distance projections . The localization of presynaptic proteins in many cases differs from the anatomical place of their mRNA expression . Thus , the possibility to detect and quantify isoforms at the protein level provides an important advantage in order to understand the functional role of these proteins . New developments in proteomics have driven major advances in understanding the cell biological mechanisms of neuronal development and connectivity ( Bayes and Grant , 2009; Craft et al . , 2013 ) . These led to the delineation of the composition of purified synaptic vesicles , postsynaptic neurotransmitter receptor complexes , or models of presynaptic terminals ( Takamori et al . , 2006; Shanks et al . , 2012; Schwenk et al . , 2014; Wilhelm et al . , 2014 ) . Most of these studies employed semi-quantitative ‘shotgun’ mass spectrometry approaches ( Liu et al . , 2004; Venable et al . , 2004; Neilson et al . , 2011 ) that are based on the random sampling of peptide fragments detected in a sample . In this regard , the complexity of neuronal tissues and the molecular diversity of some neuronal receptor families pose significant limitations . First , many proteins are present in only a fraction of the cells or structures analyzed reducing the chance of detecting peptide levels required for adequate quantification . Second , over the past years alternative splicing has emerged as a key mechanism for the regulation of neuronal recognition ( Aoto et al . , 2013; Takahashi and Craig , 2013; He et al . , 2014; Iijima et al . , 2014; Lah et al . , 2014 ) . Alternative splicing programs can generate families of tens , hundreds , or even thousands of closely related protein isoforms frequently distinguished by only a single peptide . Thus , when applying random sampling ( ‘shotgun’ ) approaches it is challenging to obtain sufficient coverage for isoform detection and quantification . Additionally , the protein inference problem , i . e . , presence of the same peptide sequence in multiple different proteins or protein isoforms limits applicability of ‘shotgun’ approaches for detection and quantification of protein families with high sequence homologies ( Nesvizhskii and Aebersold , 2005 ) . One possibility to circumvent these problems is the application of targeted proteomic approaches , such as selected reaction monitoring ( SRM; also referred to as MRM for ‘multiple reaction monitoring’ ) . While originally developed for characterization of chemical compounds this method has recently emerged as promising technique for the quantitative analysis of protein species in biological samples ( Phanstiel et al . , 2008; Picotti and Aebersold , 2012; Carr et al . , 2014 ) . Instead of sampling a random portion of the proteome , SRM assays use optimized separation and detection parameters for a set of pre-selected peptides ( termed proteotypic peptides or PTPs ) that are specific to a protein or isoform of interest . PTPs are detected based on their chromatographic retention time and mass to charge ratio ( rather than by sequencing as in shotgun proteomics ) . For each PTP , an isotopically labeled reference peptide is added to the sample which then is used as standard for quantification . This strategy greatly reduces experimental variability and significantly increases sensitivity in the detection and quantification of selected peptides ( Bauer et al . , 2014 ) . To date , significant contributions in advancing our knowledge in biology have been made by applying SRM-based approaches , in particular to profile cellular pathways and metabolic states , thereby quantifying low abundance proteins within complex mixtures ( Picotti et al . , 2013; Simicevic et al . , 2013; Kennedy et al . , 2014 ) . SRM assays have also been applied successfully to detect and quantify neuronal proteins ( Zhang et al . , 2012; Craft et al . , 2013 ) . However , the applicability for the dissection of highly diversified protein families in complex tissues remains to be explored . In this study , we developed an array of SRM-based assays for the detection and quantification of highly diversified neuronal receptors . As a case study we focused on neurexins , a class of synaptic adhesion molecules which are widely expressed in the CNS and play an important role in synapse formation and function ( Dean et al . , 2003; Missler et al . , 2003; Graf et al . , 2004; Chih et al . , 2006; Taniguchi et al . , 2007; Uemura et al . , 2010 ) . In mammals , three neurexin genes ( Nrxn1 , 2 , 3 ) are transcribed from two alternative promoters giving rise to long alpha-neurexin and short beta-neurexin transcripts ( Reissner et al . , 2013 ) . Alternative splicing at six alternatively spliced segments ( AS1-6 ) generates >1000 unique isoforms which can be detected in the adult brain at the transcript level ( Baudouin and Scheiffele , 2010; Schreiner et al . , 2014b ) . Importantly , insertion or skipping of alternative exons at two of the alternatively spliced segments ( AS2 and AS4 ) was reported to regulate binding of neurexins to a number of different interaction partners ( Ichtchenko et al . , 1995; Sugita et al . , 2001; Chih et al . , 2006; Koehnke et al . , 2010; Siddiqui et al . , 2010; Uemura et al . , 2010; Reissner et al . , 2013 ) . This raises the possibility that neurexin molecular diversity—and in particular its regulation by alternative splicing—may serve synaptic recognition events that control neuronal wiring and function . The challenge with exploring isoform-specific functions is the lack of suitable tools to detect or quantify endogenous neurexin protein variants . Thus , even fundamental questions have remained unanswered: are neurexins abundant synaptic components ? What is the relative contribution of the individual neurexin isoforms ( NRX1 , 2 , 3 alpha and beta forms ) to the total neurexin repertoire ? Are proteins produced from all of the detected alternative transcripts , in particular transcripts encoding rare splice insertions ? And does alternative splicing at all sites modify receptor–ligand interactions ? Lastly , any studies on selective receptor–ligand interactions have relied on overexpressed or recombinant proteins as no assays are available to specifically probe or detect the endogenous protein isoforms . In this study , we developed , validated , and applied a targeted proteomics workflow for the analysis of neurexin protein variants in the mouse brain using SRM-based assays . We provide a detailed protein expression map of neurexin alternative splice insertions , their absolute quantification , and uncover novel alternative splicing-dependent regulation of neurexin–ligand interactions . This workflow can be highly multiplexed and applied to virtually any protein family and tissue . Thus , it provides important new directions for the dissection of molecular diversity in cellular recognition and pathologies . We established an array of targeted proteomic assays that allow for detection and quantitative determination of neurexin variants in complex samples . The approach includes: ( 1 ) In silico selection of proteotypic peptide candidates , ( 2 ) synthesis and experimental test of candidate-peptide performance in targeted mass spectrometry , and ( 3 ) identification of corresponding endogenous peptides by MS/MS sequencing analysis . We used a customized sequence database containing all known neurexin variants derived from PacBio sequencing data ( Schreiner et al . , 2014b ) . A set of predicted tryptic peptide sequences was selected based on sequence uniqueness , size , and amino acid composition ( Picotti et al . , 2013; Bauer et al . , 2014 ) . Since neurexins are known to be glycosylated we examined peptides for the presence of consensus sequences for carbohydrate modifications and excluded peptides with potential post-translational modification from our assays . To evaluate the suitability of the selected peptides for MS detection , chemically synthesized peptides were analyzed on a triple-quadrupole LC-MS instrument . For each peptide , we determined retention times , fragmentation patterns and optimized collision energies in order to devise highly sensitive and specific SRM assays ( Figure 1—figure supplement 1–7 , see ‘Materials and methods’ for details ) . Peptide identification in SRM is based on the chromatographic properties , mass , and charge of proteotypic peptides but lacks the sequencing capabilities of shotgun approaches ( Carr et al . , 2014 ) . Thus , we separately verified identities of peptides utilized for quantification by MS sequencing using LC-MS of endogenous neurexin proteins from mouse brain and immuno-precipitates obtained with a pan-NRX antibody ( Figure 1B and Figure 1—figure supplement 1–7 ) . In total , we generated and validated 30 SRM assays that enable detection and quantification of 17 specific NRX protein species with multiple PTPs per isoform wherever possible ( Figure 1A ) : ( a ) NRX1 , 2 , and 3 total protein ( pan-peptides ) , ( b ) alpha- and beta-protein variants derived from each Nrxn gene ( alpha and beta-specific peptides ) , ( c ) alternative splice insertions at the alternatively spliced segments 3 , 4 , and 6 . Note that several of the isoform-specific peptides differ in only a single amino acid residue between the NRX1 , NRX2 , and NRX3 isoforms making MS-based approaches the method of choice to specifically detect them . In addition , we established assays for 15 unrelated synaptic proteins to be used for comparison ( the complete list of transitions shown in Supplementary file 1A ) . 10 . 7554/eLife . 07794 . 003Figure 1 . Establishment and validation of SRM-assays . ( A ) Proteotypic peptides for relative quantification of neurexin variants . The position of peptides within the overall domain structure of neurexin proteins is indicated ( laminin-G and EGF-domains are marked , transmembrane domain ( TMD ) shown as box , sites modified by alternative splicing in blue ) . Validated pan-neurexin peptides shared amongst alpha and beta isoforms derived from each Nrxn gene are shown in black , neurexin-alpha specific peptides in orange , neurexin-beta specific peptides in red , and splice isoform-specific peptides in blue . ( B ) Example MS/MS spectra of three endogenous peptides used for detection and quantification of NRX1 splice variants containing insertions at alternatively spliced segments 3 , 4 , and 6 . Letters above the peaks indicate the amino acid sequence of the corresponding peptides ( blue = y-ions; red = b-ions ) . ( C ) Workflow for the quantitative SRM-based protein analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 07794 . 00310 . 7554/eLife . 07794 . 004Figure 1—figure supplement 1 . List of MS/MS spectra of proteotypic endogenous NRX1-pan and NRX1-alpha peptides , retention times and transition patterns of corresponding synthetic heavy peptides used for relative quantification in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 07794 . 00410 . 7554/eLife . 07794 . 005Figure 1—figure supplement 2 . List of MS/MS spectra of proteotypic endogenous NRX1-beta , NRX1-AS3 ( + ) , NRX1-AS6 ( + ) and NRX1-AS4 ( + ) peptides , retention times and transition patterns of corresponding synthetic heavy peptides used for relative quantification in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 07794 . 00510 . 7554/eLife . 07794 . 006Figure 1—figure supplement 3 . List of MS/MS spectra of proteotypic endogenous NRX2-pan peptides , retention times and transition patterns of corresponding synthetic heavy peptides used for relative quantification in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 07794 . 00610 . 7554/eLife . 07794 . 007Figure 1—figure supplement 4 . List of MS/MS spectra of proteotypic endogenous NRX2-alpha , NRX2-beta and NRX2-AS4 ( + ) peptides , retention times and transition patterns of corresponding synthetic heavy peptides used for relative quantification in this study . Note: one NRX2-alpha specific peptide ( VNDGEWCHVDFQR ) could not be identified by shotgun MS/MS . However , read-outs obtained with this peptide were very close to values obtained with two other NRX2-alpha peptides . Thus , results obtained with all three peptides were used for relative quantification . DOI: http://dx . doi . org/10 . 7554/eLife . 07794 . 00710 . 7554/eLife . 07794 . 008Figure 1—figure supplement 5 . List of MS/MS spectra of proteotypic endogenous NRX2-AS3 ( + ) , NRX3-pan and NRX3-alpha peptides , retention times and transition patterns of corresponding synthetic heavy peptides used for relative quantification in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 07794 . 00810 . 7554/eLife . 07794 . 009Figure 1—figure supplement 6 . List of MS/MS spectra of proteotypic endogenous NRX3-alpha , NRX3-beta , NRX3-AS3 ( + ) and NRX3-AS4 ( + ) peptides , retention times and transition patterns of corresponding synthetic heavy peptides used for relative quantification in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 07794 . 00910 . 7554/eLife . 07794 . 010Figure 1—figure supplement 7 . MS/MS spectrum of proteotypic endogenous NRX3-AS6 ( + ) peptide , retention time and transition pattern of corresponding synthetic heavy peptide used for relative quantification in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 07794 . 010 To optimize detection of rare NRX isoforms , we sought a rapid enrichment strategy that would allow for high sample throughput and reliable SRM-based quantification . To this end , we isolated Triton X-100 resistant membrane fractions ( TRM ) from crude synaptosomes . We confirmed enrichment of major synaptic proteins by direct comparison to fractions obtained from conventional synaptosome preparations using shotgun mass spectrometry and quantitative SRM assays ( Figure 2—figure supplement 1 ) . The enrichment of neurexins was comparable with both protocols ( Figure 2—figure supplement 2 ) and no significant bias was detected in the enrichment of alpha- and beta-isoforms ( Figure 2—figure supplement 2 ) . Most importantly , this simplified procedure is fast and yields from 3 mg of starting material ca . 1 . 5 µg of synaptic-enriched proteins sufficient for up to 100 SRM-assays . Thus , we used this simplified protocol for subsequent analyses of relative and absolute NRX protein levels across brain regions . To test whether SRM assays are sufficiently quantitative and reliable to probe region-specific neurexin repertoires in a complex sample we surveyed eight anatomically defined brain regions from mice at postnatal day 30 ( Figure 2A ) . Importantly , all 17 neurexin forms targeted by the SRM assays could be reliably detected across the brain regions interrogated . In independent measurements from 4 animals using all PTPs described in Figure 1A , we observed close agreement in the observed relative isoform distributions confirming the reproducibility of the approach ( Figure 2A ) . 10 . 7554/eLife . 07794 . 011Figure 2 . Quantitative comparison of relative neurexin variant levels across brain areas . ( A , B ) Relative amounts of total neurexin ( NRX-pan ) , NRX-alpha , beta , and splice insertions across P30 mouse brain ( OB = olfactory bulb , CX = cortex , HIP = hippocampus , TH = thalamus , ST = striatum , MB = mid-brain , CB = cerebellum , BS = brain stem ) . Values were normalized to OB . Correlation analysis between expression profiles of pan-neurexins , alpha and beta isoforms was performed using Spearman-correlation test . No significant correlation between expression profiles of alpha and beta isoforms of NRX2 and 3 ( Spearman r = 0 . 6190 , p-value = 0 . 1150 for NRX2 and Spearman r = −0 . 4048 , p-value = 0 . 3268 for NRX3 , respectively ) . Significant correlation ( p-value = 0 . 0046 and Spearman r = 0 . 9048 ) for expression profiles of NRX1-alpha and beta isoforms . Numbers on the bottom of diagrams represent p-values of the t-test analysis of alpha- and beta-NRX in different brain areas ( statistically significant values are marked in red ) . Relative amounts of neurexin AS3 , AS4 , AS6 splice variants across brain regions normalized to respective total NRX protein levels . As the AS3 and AS6 insertions are found exclusively in alpha variants , their measurements were normalized to NRX-alpha protein levels . Means ± SD from 4 biological replicates ( n = 4 , 2 male and 2 female mice , postnatal day 30 ) measured in 2 technical replicates . ( C ) Hierarchical clustering analysis of relative expression of NRX isoforms and splice variants across mouse brain . Hierarchical clustering of ‘protein log2 abundance ratios’ was performed using Ward's algorithm and the correlation distance metric . Subsequently , a heat map was created using the gplots R package . DOI: http://dx . doi . org/10 . 7554/eLife . 07794 . 01110 . 7554/eLife . 07794 . 012Figure 2—figure supplement 1 . Preparation of synaptic proteins enriched Triton-Resistant-Membranes ( TRM ) from mouse brain . ( A ) Workflow of the preparation of Triton X-100 resistant membranes ( TRM ) from brain tissues for enrichment of synaptic proteins . ( B ) Hierarchical clustering of log-transformed protein normalized spectral counts was performed using Ward's algorithm and the Euclidian distance metric . Subsequently , a heat map was created using the gplots R package . To include proteins with zero spectral counts , in one or more conditions , all spectral counts were incremented by one pseudo-count . Heat map diagram and hierarchical clustering of proteins identified in TRM samples and post-synaptic density ( PSD ) samples prepared using a standard protocol . DOI: http://dx . doi . org/10 . 7554/eLife . 07794 . 01210 . 7554/eLife . 07794 . 013Figure 2—figure supplement 2Comparison of enrichments of neurexins and other synaptic proteins in TRM and Postsynaptic-Density ( PSD ) preparations . ( A ) SRM based quantification of the enrichment of neurexin proteins in TRM and PSD samples ( left diagram ) . . In both preparations comparable enrichment ( ∼4 fold ) could be observed . No bias in enrichment of alpha and beta isoforms was observed between two preparations ( right diagram ) . ( B ) Quantitative comparison of the enrichment of core PSD proteins in TRM and purified post-synaptic density ( PSD ) fractions . Relative enrichments compared to input for PSD95 ( Dlg4 ) , Gephyrin , and Bassoon were determined using SRM assays . Purified PSD fraction showed stronger enrichment of core post-synaptic density proteins . ( C ) Ratios of synaptic protein enrichment as determined by SRM assays . For all measured synaptic proteins , higher enrichment could be observed in the purified PSD fraction ( right diagram ) . The enrichment shows significant differences between different synaptic . DOI: http://dx . doi . org/10 . 7554/eLife . 07794 . 013 Interestingly , NRX1 , NRX3pan , and NRX3alpha showed only modest protein level fluctuations across the brain regions examined . By contrast , we discovered highly significant differences in expression for NRX2 and NRX3beta variants as well as usage of specific splice insertion in isoforms derived from all three Nrxn genes . Region-specific alterations in the proteins derived from primary transcripts show up to threefold increased representation of the NRX2beta and 3beta variants in the cerebellum as compared to other brain regions examined ( Figure 2A ) . Notably , this elevation of beta-variants has only a modest impact on the pan-neurexin level in the cerebellum . This suggests that beta-variants make only a small contribution to the total neurexin pool ( see below for absolute quantification of alpha and beta isoform levels ) . Insertions at NRX1 AS3 showed little variation across brain regions , whereas other splice insertions exhibited highly differential expression ( Figure 2B ) . Interestingly , the recently identified AS6 insertion in NRX3alpha ( Treutlein et al . , 2014; Schreiner et al . , 2014b ) is detected in a rostro-caudal gradient across brain regions . Also of note , there is a significant elevation of AS4 insertion-containing NRX1 and NRX2 proteins in the cerebellum , a site that expresses particularly high levels of an AS4-specific ligand ( Uemura et al . , 2010 ) . Hierarchical clustering analysis of the splice variant expression across the mouse brain revealed two major groups with regional co-regulation: one contains variants with AS4 insertion , the other containing variants with AS3 insertion . Remarkably , AS6 containing variants of NRX1 and NRX3 do not segregate into one cluster as this is the case for AS3 and AS4 insertions , indicating independent regulation of AS6 insertion in the Nrxn1 and Nrxn3 genes ( Figure 2C ) . To understand which protein isoforms derived from the three Nrxn genes are most abundant and to dissect stoichiometry of neurexin proteins vis-à-vis other synaptic components , we performed absolute quantification of NRX proteins with isotope-labeled standards . Since approaches based on reference peptides are prone to artifacts resulting from digestion variations ( Carr et al . , 2014 ) we spiked intact heavy protein standards into samples before digestion ( Simicevic et al . , 2013 ) . The standards were in vitro translated as isotope-labeled GFP-fusion proteins ( Figure 3A ) . Together with an accurately quantified amount of an unlabeled GFP protein standard , the heavy protein standards were combined with the brain extracts of interest , proteolytically digested , and PTPs quantified in separate assays by SRM ( an approach that has been referred to as Protein Standard Absolute Quantification ‘PSAQ’ ( Brun et al . , 2007 ) . Absolute amounts of the heavy , in vitro-produced GFP fusion proteins were determined via peptides derived from the light GFP protein standard . The heavy standard in turn served to deduce absolute amounts of the endogenous proteins in the brain extracts using target protein-specific peptides . To assess the sensitivity of these assays , limits of detection were determined for each peptide used for absolute quantification ( Figure 3—figure supplement 1 and Supplementary file 1B ) . To minimize the contribution of any unlabeled protein contaminants present in the in vitro translated protein standards , the ratios of standard-derived to endogenous peptides for each SRM-assay were titrated to 0 . 5 and 2 ( see ‘Materials and methods’ for details ) . 10 . 7554/eLife . 07794 . 014Figure 3 . Absolute quantification of endogenous neurexin isoforms . ( A ) Western blot for recombinant proteins and SRM-based quantification approach using in vitro translated heavy amino acids labeled standards fused to GFP and as heavy standards in SRM assays . For absolute quantification of NRX-beta isoforms full-length NRX-beta fused to GFP was used as heavy standard . For absolute quantification of NRX-alpha proteins , the neurexin-repeat 3 containing laminin-G-like domains 4 and 6 ( LG4-6 ) fused to GFP was used as heavy standards . ( B ) Absolute amounts of NRX1 , 2 , 3 , N-cadherin , and PSD95 in synaptic protein enriched fractions from whole brain of P30 mice . Pan NRX1 , 2 , 3 levels were determined using pan NRX peptides derived from in vitro translated NRX-alpha and beta standards . Shown are means ± SD . NRXs were measured with two independent heavy standards ( alpha and beta ) , in two technical replicates , from 2 animals , n = 2 . N-cad and PSD95 were measured in two technical replicates , from 2 animals , n = 2 . ( C ) Absolute quantification of NRX-alpha and beta in synaptic protein enriched samples from whole brain and cerebellum of adult ( P30 ) mice ( n = 2 mice ) . Quantification was performed with NRX-alpha and beta heavy standards spiked into the brain samples . The amount of NRX-beta in samples measured with alpha-constructs was calculated by subtraction of measured NRX-alpha amount from measured NRX-pan amount . In addition , the amount of NRX-alpha was calculated in samples measured with beta heavy standards . Means ± SD of these independent determinations are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 07794 . 01410 . 7554/eLife . 07794 . 015Figure 3—figure supplement 1 . Examples for the determination of lower limits of detection ( LOD ) and quantification ( LOQ ) for selected NRX1-3 , PSD95 and N-CAD peptides used in this study for absolute quantification . Peptide-specific standard curves were derived by spiking each digested synaptic protein-construct at a concentration range of five values into a digested mixture of mouse whole brain TRM protein extract complemented with GFP protein standard . The measurements were performed in two biological and two technical replicates , including 3 technical replicates of the whole brain protein extract and GFP protein standard mixture alone , here used to assess the background noise level . Lower limits of quantification and lower limits of detection were then established according to the blank and low concentration sample method . DOI: http://dx . doi . org/10 . 7554/eLife . 07794 . 015 Based on these experiments , we found that the amount of total NRX proteins in synapse-enriched fractions from whole brain was approximately 18 fmol/μg protein ( Figure 3B ) . This was similar to the amounts of N-cadherin ( ∼20 fmol/μg ) and approximately threefold lower than amounts of PSD95 ( 54 fmol/μg ) , one of the most abundant proteins in the postsynaptic density of glutamatergic synapses . For all three neurexins , the alpha-isoforms represented the major protein variants , whereas beta variants were 2- to 4-fold less abundant ( NRX1beta: 0 . 87 ± 0 . 36 fmol/μg , NRX1-alpha: 3 . 86 ± 0 . 46 fmol/μg , NRX2-beta: 1 . 39 ± 0 . 49 fmol/μg , NRX2-alpha: 2 . 19 ± 0 . 83 , NRX3-beta: 1 . 97 ± 1 . 6 fmol/μg , NRX3-alpha: 4 . 18 ± 0 . 45 fmol/μg , n = 4 , ±SD ) . Importantly , independent measurements with in vitro translated alpha and beta standard proteins for each primary NRX isoform ( each quantified based on different PTPs ) yielded highly similar results for pan-NRX amounts as well as the contents of alpha and beta isoforms . This further confirms the accuracy of this method . Based on previous estimates for the number of N-cadherin molecules per synapse ( Wilhelm et al . , 2014 ) , we estimate the average number of neurexin molecules to be 7–16 for NRX-beta and 18 to 35 for NRX-alpha per synapse . The ability to detect and accurately quantify individual neurexin isoforms opens the possibility to systematically explore the binding selectivity of endogenous neurexins with synaptic receptors . Thus , we profiled recognition specificity of neurexin isoforms for three different postsynaptic receptors detected at glutamatergic synapses: neuroligin-1 ( NL1B , containing splice insertion B ) , neuroligin-3 ( NL3A2 , containing splice insertion A2 ) , and the leucine-rich transmembrane protein 2 ( LRRTM2 ) ( Chih et al . , 2006; Budreck and Scheiffele , 2007; de Wit et al . , 2009; Siddiqui et al . , 2010 ) . Affinity matrixes containing recombinant postsynaptic receptor proteins were incubated with protein extracts from mouse brain and bound proteins were analyzed by mass spectrometry . Based on the normalized spectral abundance factor ( NSAF ) ( Zybailov et al . , 2006 ) of proteins identified by shotgun analysis , neurexins were amongst the most abundant proteins associated with the receptor proteins . Proteins derived from all three Nrxn genes were recovered on each of the binding partners . Interestingly , LRRTM2 beads led to a more robust isolation of NRX2-derived peptides as compared to NL1B and NL3A2 ( Figure 4A ) . Quantification by spectral counts provides a good estimate of total protein isoforms derived from the individual Nrxn genes but enables only limited detection of isoforms identified by short unique sequence elements , such as beta isoforms or specific splice variants ( e . g . , average spectral counts for NRX-beta isoforms were ≤3 in three independent pull-down experiments ) . By contrast , SRM analysis revealed highly significant differential recovery of specific neurexin isoforms on the postsynaptic ligands . Comparison of neurexin isoforms associated with NL1B and NL3A2 matrices revealed preferential binding of NRX1 and 2 alpha forms to NL3A2 over NL1B . The selectivity of interactions was even more significant when comparing binding to LRRTM2 vs NL1B or NL3A2 . LRRTM2 showed higher recovery of NRX1alpha and 2 alpha isoforms and lower recovery of NRX1beta and 2beta isoforms . Most importantly , NRX1alpha variants containing AS3 and AS6 insertions were significantly de-enriched in LRRTM2 pull-downs as compared to NL1B and NL3A2 . This suggests a major role for neurexin alternative splicing in differential association with these postsynaptic ligands . 10 . 7554/eLife . 07794 . 016Figure 4 . Receptor recognition specificity profiling with splice site-specific SRM assays . ( A ) Ratios of normalized spectral abundance factor ( NSAF ) of NRX1 , 2 , 3-specific peptides recovered from NL1B , NL3A2 , LRRTM2-affinity matrices detected in shotgun experiments . Means of the measurements from three independent experiments are shown . ( B–D ) Pairwise comparison of neurexin protein variant binding to immobilized ligands determined by SRM . Relative peptide recovery normalized to pan-peptides ( for alpha- , beta- , and AS4 ) or alpha-peptides ( for AS3 and AS6 ) is shown as log10 of the ratio of peptide amounts recovered ( NL3A2/NL1B; LRRTM2/NL1B; LRRTM2/NL3A2 ) . n = 6 ( measured in duplicates ) , ±SD . Analysis of significance was performed using nonparametric one-way ANOVA test with Bonferroni's Multiple Comparison Test ( *p ≤ 0 . 05 , **p ≤ 0 . 01 , ***p ≤ 0 . 001 ) . ( E–G ) Quantitative adhesion assays for NRX1alpha splice variant-expressing cells and LRRTM2 or NL3A2-expressing cells . LRRTM2 and NL3A2 cells co-express a surface biotin tag for isolation using streptavidin-magnetic beads . NRX1-alpha variants co-express beta-Galactosidase for quantification . ( n = 4 independent experiments , each with 3 replicates measured in triplicates ) , ±SD . Pairwise comparison was performed using nonparametric one-way ANOVA test with Bonferroni's Multiple Comparison Test ( *p ≤ 0 . 05 , **p ≤ 0 . 01 , ***p ≤ 0 . 001 ) . The expression level of NRX1-alpha constructs was probed by Western blotting . DOI: http://dx . doi . org/10 . 7554/eLife . 07794 . 016 To directly test the impact of AS3 and AS6 splice insertions on NRX1alpha—LRRTM2 interactions we applied a quantitative adhesion assay in heterologous cells ( Schreiner and Weiner , 2010 ) . We compared adhesive interactions between cells expressing NRX1alpha variants containing different combinations of alternative insertions at AS3 and AS6 and cells expressing LRRTM2 or NL3A2 ( Figure 4F ) . These experiments demonstrated a threefold increase in adhesive interactions of NRX1alpha isoforms lacking AS6 insertions as compared to the AS6+ isoforms . Presence or absence of the insertion at AS3 did not affect adhesion to LRRTM2 cells in this assay . By contrast , for NL3A2 presence of the AS6+ insertion in NRX1alpha resulted in slightly increased adhesion and isoforms containing AS6 but lacking AS3 showed the strongest adhesive interactions . These experiments uncover a significant differential regulation of neurexin interactions with two different receptors through a splice code at two alternatively spliced segments that had not previously been implicated in molecular recognition events . We developed an approach for the relative and absolute quantification of protein isoforms derived from highly diversified gene families in complex samples . Recent work uncovered a key function for alternative splicing in the regulation of neuronal recognition events . Remarkably , even small modifications in the amino acid sequence translate into switch-like alterations in cell surface recognition events ( Chih et al . , 2006; Uemura et al . , 2010 ) . Thus , the development of tools that enable a quantitative assessment of closely related protein isoforms in complex samples is of central importance . We validated an array of isoform-specific SRM assays for quantitative detection of neurexin variants resulting from alternative splicing at segments AS3 , 4 , and 6 . The same approach can be extended to the remaining sites modified by alternative splicing ( AS1 , 2 , 5 ) . While we focused in this study on tryptic peptides the use of additional proteases will expand the repertoire of peptides available for analysis , making the quantitative detection of protein isoforms in complex samples even more reliable ( Guo et al . , 2014; Meyer et al . , 2014 ) . The detection of specific splice isoforms at the protein level is particularly important for the recently discovered AS6 variants which are rare ( Schreiner et al . , 2014b ) , and there has been no evidence that such isoforms exist on the protein level and have any functional relevance . In the pull-down experiments with three known ligands we profiled their selectivity for particular endogenous NRX isoforms and uncovered a role for insertions at alternatively spliced segments 3 and 6 ( AS3 and AS6 ) in gating interactions with LRRTM2 vs neuroligin-3 . Notably , crystallographic studies provide a potential structural basis for this regulation ( Chen et al . , 2011; Miller et al . , 2011 ) . The six amino acids incorporated in AS6+ variants are positioned at a hinge region in the L-shaped NRX1alpha protein ( Schreiner et al . , 2014b ) . Thus , the regulation of LRRTM2 binding to laminin-G domain 6 might involve a modification of the flexibility positioning laminin-G domains 1-5 vs the EGF domain 3 and laminin-G domain 6 in the NRX1alpha structure . Moreover , modeling studies predict that NRX1alpha AS3 is positioned in close proximity to an interaction surface with neuroligins ( Chen et al . , 2011; Tanaka et al . , 2012 ) . This might underlie the modulation of NRX1alpha-NL3A2 interactions by AS3 insertions observed in our assays . Based on our absolute quantification experiments , we come to two important conclusions . First , within individual brain regions , neurexin-alpha isoforms are significantly more abundant than beta isoforms . In fact , many previous studies with exogenously expressed neurexins focused on NRX1beta isoforms but our quantitative analysis demonstrates that these forms are the least abundant of all neurexins in vivo . Considering that neurexin-alpha and beta forms differ in their interactions with receptors ( Figure 4 and Boucard et al . , 2005; Chih et al . , 2006; Siddiqui et al . , 2010 ) , this has important functional implications . Second , comparison of absolute levels of neurexins with other synaptic proteins demonstrates that neurexins are rather abundant synaptic proteins . For example , the total amount of all neurexin proteins is similar to that of N-cadherin , a highly abundant synaptic protein and about one third of the amount of PSD95 , one of the most abundant proteins in glutamatergic synapses . During the development of our study , we evaluated the use of AQUA synthetic peptides ( Gerber et al . , 2003 ) for quantification of brain protein repertoires . We observed that several different synthetic AQUA peptides reporting on the same neurexin isoform often yielded divergent results , making a reliable quantification difficult ( data not shown ) . This is most likely due to the variability in tryptic digests of endogenous proteins ( Arsene et al . , 2008 ) . To circumvent these problems , we utilized an in vitro recombinant protein expression system , to use surrogate protein-constructs as standards for quantification . This approach is not only more robust and more accurate , but is also cost-effective , can be multiplexed by using multiple different tags in the isotopically labeled function proteins ( in addition to GFP ) and can be scaled to screen large numbers of different proteins and isoforms . Thus , the experimental pipeline established here is readily transferable to any protein class of interest . Finally , our SRM approach provides an attractive alternative to isotope labeling-based massspectrometry approaches such as SILAC . Notably , SILAC labeling has been most frequently applied to cell culture or microorganisms . This powerful technique has been extended to higher organisms , by using labeled cell lines as an internal standard for quantitation or raising rodents for multiple generations on a diet with heavy isotope-labeled amino acids ( Kruger et al . , 2008 ) . While methods are well suited to assess overall proteome modifications it remains challenging to measure proteins of low abundance or protein isoforms in a consistent manner and across large sample numbers with reasonable throughput , due to fact that neuronal tissues are complex and heterogeneous . In most instances , the SRM-assays developed and validated here for mouse tissues can be directly transferred to assessing protein levels in small amounts of biopsy material from human patients or postmortem brain tissue . This is particularly interesting for the Nrxn gene family as Nrxn1 mutations are associated with autism and schizophrenia in the human population ( Reichelt et al . , 2012 ) . Given the complexity and size of the mammalian neurexin genes it is difficult to predict the impact of such mutations on the expression and function of the neurexin proteins . In the future , the approach established here can be readily applied to iPSC-derived human neurons , biopsy or postmortem tissue and enables exploration of the impact of such mutations on the neurexin isoform repertoires . Brain regions were dissected , weighed , and homogenized in a glass-Teflon homogenizer in 1:20 wt/vol of solution A ( 0 . 32M sucrose 1 mM MgCl2 0 . 5 mM CaCl2 1 mM NaHCO3 complemented with Roche ‘Complete’ protease inhibitors ) . Homogenates were centrifuged at 1 , 400×g at 4°C for 10 min . Supernatants were re-centrifuged ( 14 , 000×g at 4°C for 10 min ) and pellets re-suspended in Triton X-100 buffer ( 12 mM Tris–HCl , pH8 . 1 , 1% Triton X-100 complemented with protease inhibitors ) , rotated at 4°C for 10 min on a rotator . Samples were then centrifuged at 100 , 000×g at 4°C for 1 hr . The pellets representing TRM were dissolved in 50 mM ammonium bicarbonate , 1% sodium deoxycholate ( DOX ) . Protein concentrations were determined using the BCA assay ( Pierce ) . cDNAs encoding the polypeptides of interest were fused in frame with a cDNA encoding GFP in a modified SP6-pF3A WG ( BYDV ) -Flexi vector ( Promega ) . Mouse cDNA constructs encoding the following proteins were used ( always lacking the N-terminal signal sequence ) : NRX1-beta AS4+ ( containing the alternative insertion at AS4 ) , NRX2-beta AS4+ , NRX3-beta AS4- , the third neurexin-repeat ( containing laminin-G-domain 5 , EGF-like domain 3 and laminin-G domain 6 ) of NRX1 , 2 and 3-alpha , each containing insertions at AS4 and AS6 , full-length mouse PSD95 , N-cadherin lacking the pro-domain . Isotopically labeled proteins were expressed using the TnT-SP6 Wheat Germ High Yield Master Mix Minus Amino Acids ( Promega ) . An 18 amino acid mix ( 80 µM each final concentration ) was combined with ( 13C , 15N ) -isotopically labeled Arginine and Lysine ( Cambridge Isotope Laboratories ) to a 1 mM each final concentration and processed according to the manufacturer's instructions . Validation of in vitro-expressed heavy-labeled proteins was performed by Western blot using mouse anti-GFP primary antibodies ( Santa Cruz Biotechnology ) . For proteolytic digest , the in vitro translated proteins were reduced in 5 mM Tris- ( 2-carboxyethyl ) -phosphine hydrochloride ( TCEP ) at 95°C for 10 min , alkylated in 10 mM iodoacetamide at 25°C for 30 min in the dark , and incubated with 12 . 5 mM N-acetyl-cysteine at RT for 10 min . Samples were then incubated with 12 . 0 μg/ml trypsin ( V511C-Promega ) overnight at 37°C . The resulting tryptic peptides were purified by C18 using Macro Spin columns ( The Nest Group ) and re-suspended in 2% acetonitrile 0 . 15% formic acid . Prior to mass spectrometric analyses , digested peptides ( 1 . 5 µg ) were separated by an EASY nano-LC system ( Proxeon Biosystems , Thermo-Scientific ) equipped with a reverse phase HPLC column ( 75 µm × 30 cm ) packed in house with C18 resin ( Reprosil-AQ Pur , 1 . 9 μm , Dr Maisch ) using a linear gradient from 95% solvent A ( 98% water , 2% ACN , 0 . 15% formic acid ) and 5% solvent B ( 98% ACN , 2% water , 0 . 15% formic acid ) to 30% solvent B over 90 min at a flow rate of 0 . 2 μL/min . For LC-MS/MS analysis of separated peptides a LTQ-Orbitrap Velos and Elite mass spectrometer equipped with a nanoelectrospray ion source ( both Thermo Fisher Scientific ) and a custom made column heater set to 60°C . Each MS1 scan was followed by collision-induced-dissociation of the 20 most abundant precursor ions with dynamic exclusion for 60 s . Total cycle time was approximately 2 s . For MS1 , 10E6 ions were accumulated in the Orbitrap cell over a maximum time of 300 ms and scanned at a resolution of 30 , 000 ( 120 , 000 for Elite ) FWHM ( at 400 m/z ) . MS2 scans were acquired at a target setting of 10 , 000 ions , accumulation time of 25 ms . The mass selection window was set to 2 Da and singly charged ions and ions with unassigned charge state were excluded from triggering MS2 events . Besides , the normalized collision energy was set to 35% and one microscan was acquired for each spectrum . The acquired raw-files were converted to the mascot generic file ( mgf ) format using the msconvert tool ( part of ProteoWizard , version 3 . 0 . 4624 ( 2013-6-3 ) ) . Using the MASCOT algorithm ( Matrix Science , Version 2 . 4 . 0 ) , the mgf files were searched against a concatenated target-decoy protein sequence database , comprised of target sequences and decoy entries as follows: ( a ) target sequences/entries: from mouse ( Mus musculus—SwissProt , www . uniprot . org , release date 12/05/2012 , canonical and isoform sequences ) , all NRX alternatively spliced segments generated in silico based on a PacBio sequencing data splicing model ( Schreiner et al . , 2014b ) ; ( b ) a set of common contaminant protein sequences as defined in the MaxQuant software ( Cox and Mann , 2008 ) ; ( c ) decoy entries: the reversed target entries were generated using the SequenceReverser tool from the MaxQuant software ( Version 1 . 0 . 13 . 13 ) . The final database contained 33 , 364 entries in total . The MASCOT search criteria were set as follows: 10 ppm precursor ion mass tolerance , 0 . 6 Da fragment ion mass tolerance , full tryptic specificity was required ( cleavage after lysine or arginine residues unless followed by proline ) , up to 2 missed cleavages were allowed , carbamidomethylation ( C ) , were set as fixed modification and oxidation ( M ) as a variable modification . Next , the database search results were imported to the Scaffold software ( version 4 . 3 . 2 , Proteome Software Inc . , Portland , OR ) and the protein false identification rate was set to 1% based on the number of decoy hits . Specifically , peptide identifications were accepted if they could achieve an FDR less than 1 . 0% by the scaffold local FDR algorithm . Protein identifications were accepted if they could achieve an FDR less than 1 . 0% and contained at least 1 identified peptide . Proteins that contained similar peptides and could not be differentiated based on MS/MS analysis alone were grouped to satisfy the principles of parsimony . Proteins sharing significant peptide evidence were grouped into clusters . All samples were analyzed on a TSQ-Vantage triple-quadrupole mass spectrometer coupled to an Easy-nLC ( Thermo Fisher , Scientific ) equipped with a heated ( 60°C ) reverse phase HPLC column ( 75 µm × 30 cm ) packed in house with C18 resin ( Reprosil-AQ Pur , 3 μm , Dr Maisch ) . In each injection an equivalent of 1 . 5 μg of synaptic protein extract , including purified synaptic heavy-labeled purified protein fusion constructs and the GFP standard were separated using a linear gradient from 95% solvent A ( 98% water , 2% ACN , 0 . 15% formic acid ) and 5% solvent B ( 98% ACN , 2% water , 0 . 15% formic acid ) to 30% solvent B over 90 min at a flow rate of 0 . 2 μl/min . The mass spectrometer was operated in the positive ion mode using ESI with a capillary temperature of 275°C , a spray voltage of +2200 V . All the measurements were performed in an unscheduled mode and a cycle time of 5 s . A 0 . 7 Dalton mass selection window was set for parent- ( Q1 ) and product- ( Q3 ) ion isolation . Fragmentation of parent-ions was performed in Q2 at 1 . 2 mTorr . Each SRM assay was optimized regarding collision energies , parent ion masses and fragment ion selection in pilot experiments using pure heavy peptide reference samples and the Skyline software ( v . 2 . 4 ) ( Carr et al . , 2014 ) . Generally , singly charged peptide fragment ions of the y-ion series with a mass higher than the precursor ion mass to charge ratio were preferably monitored , unless otherwise stated . A complete list of all monitored transitions is provided in Supplementary file 1A . For relative quantification , 5 μg of TRM proteins were digested and a mix of synthetic heavy-labeled peptides ( JPT Peptide Technologies GmbH , Berlin , Germany ) at the final concentration of 50 fmol/µl was spiked in the samples prior to injection . For absolute quantification , 5 μg of TRM proteins were supplemented with an equivalent of 0 . 05 μl of in vitro-expressed heavy-labeled protein construct mixture ( which includes the wheat germ lysate ) along with the GFP protein standard ( Ray Biotechnologies ) at the final concentration of 100–200 fmol/µl , prior to resuming the digestion protocol described above . Calculation of absolute levels of synaptic proteins was performed using a two-step procedure as outlined in Figure 3A . All data analyses were carried out using the Skyline software ( v . 2 . 4 ) . Peptide identification and peak-area integration of the GFP tag peptides and targeted synaptic peptides as well as their transitions were manually verified in Skyline . Individual standard curves were established for all synaptic protein peptides using a concentration range of five values recorded in two biological and two technical replicates ( Figure 3—figure supplement 1 ) . Lower limits of quantification and lower limits of detection were established using the blank and low concentration sample method ( Addona et al . , 2009 ) . For label-free quantification , the generated raw files were imported into the Progenesis LC-MS software ( Nonlinear Dynamics , Version 4 . 0 ) and analyzed using the default parameter settings . MS/MS-data were exported directly from Progenesis in mgf format and analyzed using Mascot , searching the same target-decoy databases as specified above . The search criteria were set as follows: 10 ppm precursor ion mass tolerance , 0 . 6 Da fragment ion mass tolerance , full tryptic specificity required ( cleavage after lysine or arginine residues ) ; maximum 2 missed cleavages; fixed modification: carbamidomethylation ( C ) , variable modification: oxidation ( M ) . Results from the database search were imported into Progenesis . The database search results were filtered , limiting the peptide and protein level FDR to 1% . The Progenesis analysis results were further processed using the SafeQuant R package to obtain protein relative abundances . This analysis included , global data normalization by equalizing the total MS1 peak areas across all channels , summation of MS1 peak areas per protein and LC-MS/MS run , followed by calculation of protein abundance ratios and testing for differential abundance using empirical Bayes method . For the recognition specificity profiling experiments , ectodomains of mouse neuroligin-1 containing insertion B ( NL1B ) , mouse neuroligin-3 containing insertion A2 ( NL3A2 ) , and mouse leucine-rich repeat transmembrane protein 2 ( LRRTM2 ) were expressed as SNAP-tagged proteins using the pDisplay expression vector . For expression of each construct , four 15-cm plates of HEK293T cells were transfected , conditioned media were collected 48 hr post-transfection , and incubated with 100 μl of SNAP-capture agarose resin ( NEB ) overnight . This results in covalent binding of the ‘bait’ proteins to the agarose beads . Beads were washed with pull-down buffer ( 50 mM Tris–HCl , pH7 . 5 , 150 mM NaCl , 10% glycerol ) . For pull-down , one brain of an adult ( P25-P30 ) mouse ( ca . 0 . 4g ) was homogenized in 20 ml using a glass-homogenizer in pull-down buffer completed with protease inhibitors ( Roche ) , 1% Triton X-100 , and 2 mM CaCl2 . Homogenate was centrifuged at 20 , 000×g for 30 min at 4°C . 5 ml of the supernatant was added to each ‘bait’ protein-bead preparation , and incubated overnight . On the next day , beads were washed 4 times with pull-down buffer supplemented with 0 . 1% Triton X-100 , and 2 mM CaCl2 , and bound proteins were eluted from the beads with 50 mM ammonium bicarbonate , 1% sodium deoxycholate and prepared for mass spectrometry as described above . Adhesion assays were performed essentially as described ( Schreiner and Weiner , 2010 ) with some modifications to the protocol . 1 × 106 K562 cells were nucleofected ( Amaxa ) with 5 μg of ‘prey’ and ‘bait’ DNA mixes . ‘Prey’ mixes: NRX1-alpha ( with different combination of insertion at AS3 and AS6 ) or control plasmid 3 μg , beta-Gal 1 . 5 μg , RFP 0 . 5 μg ( transfection control ) . ‘Bait’ mixes: LRRTM2 , NL1 ( B ) , or NL3 ( A2 ) 3 μg plus AP-GFP 1 . 5 μg and BirA-ER 0 . 5 μg . After nucleofection , cells were grown in 2 . 5 ml DMEM in 6 well plates ( one nucleofection per well ) . 24 hr post-nucleofection , 10 μM biotin was added to cells transfected with ‘bait’ mixes . On the next day ‘bait’ cells ( 5 wells ) were pooled , washed 1x with DMEM ( without serum ) , and re-suspend in 10 ml DMEM ( without serum ) in 15-ml tube . 75 μl neutravidin-magnetic beads ( Pierce ) were added and incubated for 15 min ( overhead shaker ) . During pre-incubation of ‘bait’ cells with magnetic beads ‘prey’ cells ( one well for each NRX1-alpha construct or control ) were collected in 15-ml tubes , spun down , and re-suspended in 1 . 5 ml DMEM ( without serum ) . To each tube with ‘prey’ cell suspension , 2 ml of ‘bait’ cells/magnetic beads suspension was added . Suspension mixes were aliquoted ( 1 ml ) in round bottom 2 ml tubes . One 200 μl aliquot was taken ( input ) . Suspensions were incubated for 60 min at RT on an overhead shaker . Magnetic beads were washed once with 1 ml DMEM and bound cells were lysed in 100 μl of 1X passive lysis buffer ( Promega ) complemented with protease inhibitors . Input samples ( 200 μl ) were lysed by adding 100 μl of 3x passive lysis buffer . Beta-gal activity was measured in triplicates: 100 μl of beta-Gal buffer ( 100 mM phosphate buffer , pH7 . 4 , 2 mM MgCl2 , 50 mM beta-mercaptoethanol , 1 mg/ml ONPG ) were pre-pipetted in a 96 well plate and 20 μl of lysates were added per well . Plates were incubated at 37°C . Absorbance was measured at 415 nm and relative binding was calculated from background-subtracted values .

To create a protein , a gene is first copied to form an RNA molecule that contains regions known as introns and exons . Splicing removes the introns and joins the exons together to form a molecule of ‘messenger RNA’ , which is translated into a protein . Over the course of evolution , many groups—or families—of proteins have expanded and diversified their roles . One way in which this can occur is through a process known as alternative splicing , in which different exons can be included or excluded to generate the final messenger RNA . In this way , a single gene can produce a number of different proteins . These closely related proteins are known as isoforms . The brain contains billions of neurons that communicate with one another across connections known as synapses . A family of proteins called neurexins helps neurons to form these synapses . Humans have three neurexin genes , which undergo extensive alternative splicing to produce thousands of protein isoforms . However , it is not known whether all of these isoforms are produced in neurons , as existing experimental techniques were not sensitive enough to easily distinguish one isoform from another . A technique known as ‘selected reaction monitoring’ ( or SRM for short ) has recently emerged as a promising way to identify proteins . This allows proteins containing specific sequences to be separated out for analysis , in contrast to existing techniques that test randomly selected protein samples , which will result in most isoforms being missed . Schreiner , Simicevic et al . have now developed SRM further and show that this technique can detect the identity and amount of the neurexin isoforms present at synapses , including those that are only produced in very small quantities . Using SRM , Schreiner , Simicevic et al . demonstrate that neurexin isoforms differ in how they interact with synaptic receptors . Thus , alternative splicing of neurexins underlies a ‘recognition code’ at neuronal synapses . In the future , this newly developed SRM method could be used to investigate isoforms in other protein families and tissues , and so may prove valuable for understanding how a wide range of cellular recognition processes work .

2015

Quantitative isoform-profiling of highly diversified recognition molecules

During amphibian gastrulation , presumptive endoderm is internalised as part of vegetal rotation , a large-scale movement that encompasses the whole vegetal half of the embryo . It has been considered a gastrulation process unique to amphibians , but we show that at the cell level , endoderm internalisation exhibits characteristics reminiscent of bottle cell formation and ingression , known mechanisms of germ layer internalisation . During ingression proper , cells leave a single-layered epithelium . In vegetal rotation , the process occurs in a multilayered cell mass; we refer to it as ingression-type cell migration . Endoderm cells move by amoeboid shape changes , but in contrast to other instances of amoeboid migration , trailing edge retraction involves ephrinB1-dependent macropinocytosis and trans-endocytosis . Moreover , although cells are separated by wide gaps , they are connected by filiform protrusions , and their migration depends on C-cadherin and the matrix protein fibronectin . Cells move in the same direction but at different velocities , to rearrange by differential migration . The basic body plan of metazoans is established by gastrulation , and at its core is the movement of endoderm and mesoderm from the surface to the interior of the embryo . Among invertebrates , the pre-gastrulation embryo typically consists of a single-layered epithelium , and a common mechanism of germ layer internalisation is invagination , the inward bending of an epithelium at a pre-localised site . A classic example of gastrulation by invagination is the sea urchin embryo ( Kominami and Takata , 2004 ) , and more recently , the invagination of the mesoderm during gastrulation in the fruit fly Drosophila melanogaster has been thoroughly studied ( Rauzi et al . , 2013 ) . Another major internalisation mechanism is ingression , where individual cells leave the epithelial layer to move interiorly . Both modes of internalisation can occur in the same organism . For example , primary mesenchyme ingression precedes invagination in the sea urchin embryo ( Katow and Solursh , 1980; Kominami and Takata , 2004 ) . Within chordates , cephalochordates and tunicates develop from a single-layered blastula . Ingression is not observed in these groups , and internalisation of germ layers occurs by invagination ( Shook and Keller , 2008 ) . Although the blastula wall is single-layered in ascidian tunicates , it is thick relative to the size of the embryo , and the vegetal cells in particular are comparatively large , which gives ascidian invagination a distinctive appearance ( Satoh , 1978; Sherrard et al . , 2010 ) . The transition to the third chordate group , vertebrates , is characterised by a sharp increase in egg size along with the formation of a thick multilayered epithelium that surrounds a blastocoel cavity . Whereas the animal side of the embryo can secondarily become single-layered , the vegetal half always remains as a multilayered cell mass . The corresponding ancestral mode of vertebrate gastrulation , conserved in lampreys , lungfish , and amphibians ( Collazo et al . , 1994; Shook and Keller , 2008 ) , must adapt to this condition . In a second wave of further egg size increase , meroblastic cleavage again requires adaptation of gastrulation movements in various vertebrate groups . For example , germ layer internalisation occurs by ingression at a novel structure , the primitive streak , in birds and mammals ( Arendt and Nübler-Jung , 1999 ) . In the ancestral mode of vertebrate gastrulation , mesoderm is internalised by involution or ingression at the blastopore lip , and the supra-blastoporal endoderm by involution ( Shook and Keller , 2008 ) . The multilayered structure of the sub-blastoporal endoderm of the vegetal cell mass precludes invagination , and ingression of the vegetal surface is also absent . Thus , the question arises of how the vegetal endoderm is internalised . Surprisingly , despite endoderm internalisation being a defining feature of gastrulation , it has scarcely been studied in lower vertebrates . Even in the African clawed frog , Xenopus laevis , the most extensively studied model of vertebrate gastrulation , the inward movement of the sub-blastoporal vegetal endoderm has only been analysed at the tissue level ( Winklbauer and Schürfeld , 1999; Ibrahim and Winklbauer , 2001; Papan et al . , 2007 ) . In X . laevis , the cone-shaped vegetal endoderm is initially narrow inside at the blastocoel floor ( BCF ) , and wide at its outer , epithelial surface . At the equator , it is surrounded by an annulus of mesoderm ( Figure 1A ) ( Keller , 1975; Keller , 1976; Bauer et al . , 1994 ) . At the onset of gastrulation , the vegetal cell mass surges animally into the embryo . It narrows at its vegetal-most part , expands at the BCF , rolls the anterior mesoderm against the ectoderm and displaces the posterior mesoderm in the vegetal direction ( Figure 1A ) ( Winklbauer and Schürfeld , 1999; Ibrahim and Winklbauer , 2001; Papan et al . , 2007; Winklbauer and Damm , 2012 ) . When a mid-sagittal slice of the vegetal half of the gastrula is explanted ( Figure 1B ) , the entire process continues in isolation and appears as rotational movements on the dorsal and the ventral sides , which gave rise to the term vegetal rotation ( Video 1 ) . Further dissection of explants revealed that vegetal rotation is based on active , region-specific tissue deformations within the vegetal cell mass ( Winklbauer and Schürfeld , 1999; Ibrahim and Winklbauer , 2001 ) . However , the cellular mechanisms that drive vegetal rotation are not known , and it is not understood how these processes are related to other modes of gastrulation in chordates . In the present work , we have analysed the cellular mechanism of vegetal rotation . We show that endoderm cells undergo elongation and region-specific re-orientation at the onset of gastrulation and move by amoeboid migration without involving lamellipodial , filopodial or bleb-like protrusions . Spatially graded differences in movement velocity lead to orderly cell rearrangements as cells move over and between each other . Rearrangement by such differential migration narrows the vegetal-most part of the tissue and expands the animal part , which leads to the inward surge of the vegetal mass . During migration , endoderm cells are separated by wide interstitial gaps , which are bridged by dynamic filiform protrusions . Despite these gaps , C-cadherin is required to maintain cell migration , and interaction with the extracellular matrix ( ECM ) protein fibronectin ( FN ) is also necessary . A peculiar mode of ephrinB1-dependent trailing edge retraction by macropinocytosis and trans-endocytosis , combined with the amoeboid characteristics of endoderm cell translocation , suggests that vegetal rotation is a modification of invagination or ingression adapted to the multilayered structure of the vegetal mass , and is based on ingression-type cell migration . To analyse vegetal rotation , we first quantified tissue shape changes performed by vegetal slice explants . Beginning at gastrulation , the upper region of explants expanded , and the BCF widened by 1 . 8-fold in two hours before reaching a plateau . Simultaneously , the equatorial waist of explants narrowed by 0 . 75-fold ( Figure 2A ) . At the cell level , cell rearrangement within the exposed-surface plane of the explant , rearrangement by cells moving in and out of this plane , and cell elongation in region-specific patterns were apparent ( Figure 2B ) . In the absence of strong cell division or cell growth during gastrulation ( Saka and Smith , 2001; Kurth , 2005 ) ; four cells divided in explant shown in Figure 2B ) , changes in apparent cell size must be due to cell shape changes and incomplete intercalation in and out of the plane of view ( Figure 2B ) . Together , the changes in cell shape and position expanded the upper part of the explant at the expense of the lower , narrowing part . Quantitatively , overall cell rearrangement was indicated by changes in cell numbers along landmark lines over time . Average cell number increased from 11 . 7 to 17 . 3 at the BCF , remained unchanged along the animal–vegetal ( A–V ) axis , and decreased from 21 . 3 to 14 . 3 along the waists of explants ( Figure 2C; Figure 2—source data 1 ) . Cells also elongated slightly with the onset of rearrangement in explants ( Figure 2B ) . In embryos , cell elongation was less pronounced , as seen by scanning electron microscopy ( SEM ) ( Figure 2E ) . Cell elongation was accompanied by cell re-orientation and alignment , and elongation was predominantly in the direction of movement ( Figure 2D; Figure 2—figure supplement 1A; Figure 2—figure supplement 1A—source data 1 ) . After explant excision , cells were on average obliquely oriented in all regions . During the next half hour , cells in the top layers turned parallel to the BCF , while those in the middle layers aligned with the A–V axis . Thus , cells near the BCF became perpendicularly oriented relative to cells in the center . Orientation changed similarly in the embryo ( Figure 2D; Figure 2 Data; Figure 2—figure supplement 1B—source data 1 ) . In particular , cells of the expanding BCF flattened in parallel to the tissue surface , perpendicular to cells located farther vegetally ( Figure 2E ) . Cells of the vegetal epithelial layer are known to remain at the surface ( Keller , 1978 ) , but occasionally they showed apical constriction and became wedge-shaped ( Figure 2D ) . Our data support the notion that a combination of cell rearrangement and oriented cell elongation underlies the distinct shape change of vegetal explants . To estimate the relative importance of these processes , we considered in detail a representative explant ( Figure 2B ) . Here , the narrowing of the equator was due to a decrease in cell number , as in other explants ( Figure 2C ) . In part , this was because of a disappearance of cells at the lateral explant margins , which was to some extent offset by the elongation of former sub-marginal cells in parallel to the equator ( Figure 2B ) . We suggest that this ‘edge effect’ was an explant artifact . In the remaining central section of the equator , cell numbers decreased by 0 . 64 ( from 14 to 9 cells ) , matching the 0 . 64-fold decrease in equator length . Apparently , the slight elongation of cells perpendicular to the equator in the center of the explant was offset by a similarly slight net increase in apparent cell size . Cell disappearance was rare in the center of the explant ( Figure 2B ) ; thus , in-plane cell rearrangement constituted the major morphogenetic process to narrow the lower part of the explant . Cells disappeared at the lateral margins both above and below the equator , and to a lesser extent sub-marginally below the equator . However , cells appeared at the explant surface only above the equator , which contributed to expansion of the explant upper region ( Figure 2B ) . As shown below , this type of intercalation was mostly due to superficial cells moving laterally to expose deeper cells . Importantly , cell lengthening parallel to the BCF also contributed to the region’s lateral expansion . Directly at the BCF , net cell number increased 1 . 3-fold ( from 13 to 17 cells ) , and the remainder of the total 1 . 8-fold length increase ( i . e . a 1 . 4-fold contribution ) was due to oriented cell lengthening and some increase in apparent cell size ( Figure 2B ) . As described in the following paragraph , cell lengthening is an integral part of endoderm cell movement; thus , cell rearrangement and its associated cell movements seemed to be the main mechanisms driving vegetal rotation . We next determined the mechanism by which vegetal cells rearrange . Generally , two basic processes of cell neighbour exchange have been identified . An intensively studied paradigm is epithelial cell intercalation by junction remodelling ( Bertet et al . , 2004 ) . For example , in D . melanogaster gastrula ectoderm , a cell–cell boundary constricts and resolves to separate two neighbouring cells while a new , perpendicularly oriented contact is formed between previously non-attached cells . An analogous mechanism was proposed for mesenchymal cell rearrangement in X . laevis mesoderm ( Shindo and Wallingford , 2014 ) . However , mesenchymal rearrangement can also be driven by the migration of cells over each other . A defining feature of migration is that a cell establishes new contacts on a substratum and detaches from previous contacts , thus changing its position . When two cells migrate over each other , one cell serves as substratum at a given instance for the other to translocate across it . For rearrangement by junction remodelling , no such distinction can be made as the common contact areas between two cells shrink or expand together . During vegetal rotation , the endoderm cells rearranged by amoeboid migration ( Figure 3A ) . While cells wedged themselves between neighbours , they underwent cycles of cell body elongation in the direction of movement , expansion of the cell front , narrowing of the cell rear , and retraction of the trailing edge . Cell shapes reminiscent of the amoeboid motility cycle were also seen in the embryo using SEM ( Figure 3B–C ) . Whereas cell tails were often flattened against other cells , leading edges were blunt and locomotory protrusions such as lamellipodia or filopodia were notably absent . To directly show that cells translocated relative to their neighbours , yolk platelets were used as markers of cytoplasmic positions ( Kubota , 1981; Selchow and Winklbauer , 1997 ) . Platelets within an advancing cell maintained their relative positions during migration ( Figure 3A ) , which indicated that the cytoplasm of the cell body advanced as a whole . Cell displacement occurred relative to the yolk platelets of stationary cells on the sides , and contact with these cells was reduced at the rear while new contacts were formed at the front ( Figure 3A ) . The leading edge of the cell remained in contact with the cell ahead of it , and both cells moved in tandem . On other occasions , cells invaded the space evacuated by the retraction of a cell ( Figure 4A ) . At the rear , a lagging cell followed closely , although contact with that cell was gradually reduced ( Figure 3A ) . To confirm this mode of translocation , we followed the lengths of lateral contacts , and the distances of these from leading and trailing edges over a time interval ( Figure 3D ) . While lateral contacts remained stationary , the leading edge advanced and the trailing edge retracted , which led to a net translocation of the cell relative to its neighbours . Again , the trailing edge reduced its contact with the cell behind , but membrane undulations suggesting cell detachment also occurred laterally . Occasionally , vesicles appeared inside the cell . Taken together , endoderm cell displacement shows the hallmarks of migration; that is , contact formation at the cell front and contact resolution at the rear . While lateral contacts remain stationary , the cytoplasmic content of a cell moves forward into the advancing front region . During cell-on-cell migration , cells necessarily move past one another and therefore rearrange locally . To achieve tissue remodelling , rearrangements must be patterned globally such that small local cell displacements translate into large-scale shape change at the tissue level . In the endoderm , the more vegetal part narrows laterally , whereas the animal part expands . In the narrowing region , the elementary rearrangement event consisted of a merging of cell columns . In groups of contiguous cells , animal–vegetal neighbours separated because of the higher velocities of more animally positioned cells , whereas their lateral neighbours converged to fill the spaces left after separation ( Figure 4A; Video 2 ) . Cells also partially or completely disappeared into the deeper layers ( Figures 2B and 4A ) . To a lesser extent , this type of rearrangement also contributed to the narrowing of explants below the equator ( Figure 2B ) . For oriented rearrangements to occur across the whole expanse of the narrowing zone , cell velocity must increase continuously from vegetal to animal . Such a velocity gradient was observed ( Figure 4B ) . Relative to the vegetal surface , cells moved faster when they were located farther animally , up to a zone near the BCF ( Figure 4D; Figure 4—source data 1 ) . Movement was also slightly faster in the center of the explant compared to the periphery ( Figure 4D ) ; that is , cell movements were not restricted to the endoderm periphery as previously suggested ( Winklbauer and Schürfeld , 1999 ) . The timing of explant excision could account for this discrepancy , as vegetal rotation initially spreads from the periphery to the center of the vegetal mass . Rearrangement by the merging of cell columns is easily conceptualised for cells racing side by side on an external surface . However , difficulty arises in a multilayered tissue when a cell is supposed to move consistently slower than the cells directly ahead of it , while it must attach to them to propel itself forward . Using the front cells as substratum would imply that the rear cell moved faster than these cells and would overtake them . As a solution , we found that individual cell movements were intermittent and neighbouring cells usually translocated at different velocities ( Figure 4C ) . Discontinuous movement allowed cells to temporarily use adjacent cells as substratum during bursts of locomotion , and in turn , these cells served as substratum for neighbouring cells in subsequent steps . Nevertheless , average velocity changed gradually in the direction of movement , which promoted cell rearrangement . The migration of single endoderm cells in vitro ( see below ) suggested that their translocation was intrinsically intermittent . In the upper , laterally expanding zone , the A–V velocity gradient was inverted . In explants , cells that were at or near the BCF still moved fast , but changed the direction of their migration ( Video 3 ) from animal to lateral ( Figure 4E , F ) . Consequently , the A–V component of their velocity gradually diminished as they approached the BCF ( Figure 4E ) . This reversal of the velocity gradient changed the direction of intercalation from mediolateral to animal–vegetal; faster cells coming from behind inserted themselves between the slower cells ahead . As cells did not slow down , but changed direction to move sideways , and because cells were elongated in the direction of their active movement , this cell reorientation also contributed to the expansion at the BCF ( see Figure 2B ) . Lastly , as cells migrated away from the central axis , they intercalated not so much by merging of cell rows in the plane of the explant surface , but by exposing cells deeper in the explant; that is , by intercalation perpendicular to the surface plane ( See Figure 2B ) . In the embryo , cells were also obliquely oriented below the BCF and parallel to the BCF at the surface ( Figure 4G , H ) . Together , our data suggest that at the BCF , elongated cells move laterally and cells from deeper layers insert themselves into gaps that open , which expands the BCF . Eventually , the BCF cells will line the remnant of the blastocoel cavity , and as it shrinks and disappears , they will merge with the deep endodermal cell mass ( Ewald et al . , 2004 ) . Endoderm cells moved individually in the direction of their long axes ( Figure 3 ) , and because cells were not oriented strictly in parallel ( Figure 2D ) , one would expect independent trajectories of adjacent cells . However , their mutual attachment integrated their diverging migration trends . As an expression of this effect , net cell movement and cell orientation were not fully aligned in explants . While adjacent cells move in parallel , their long axes diverged such that cells drifted sideways to some extent . Both active movement of the cells in the direction of their long axes and passive drift imposed by the surrounding cells appeared to contribute to their net translocation . When active and passive components are aligned , net movement should be fastest . This behaviour was indeed observed ( Figure 4—figure supplement 1; Figure 4—figure supplement 1—source data 1 ) , which indicates that cells do not migrate individually as within a rigid ECM scaffold , but mutually affect their trajectories . Regional velocity differences were correlated with cell packing densities in the embryo ( Figure 4I , J ) . Endoderm cells were close to each other near the vegetal pole , but became less densely packed toward the BCF . At a given A–V level , packing decreased toward the center of the embryo . The channels that separate cells were narrower near the vegetal base and wider more animally and centrally ( Figure 4K; Figure 4—source data 1 ) . Loose cell packing could reflect lower adhesion , and in calcium-free medium , explants do indeed dissociate faster in the animal region ( Wen and Winklbauer , unpublished ) . In turn , low adhesion could facilitate migration through reduced resistance to movement . In summary , in both zones of the vegetal endoderm , rearrangement was based on gradients of cell velocity in the direction of movement toward the BCF , that is , on the average velocity increasing or decreasing in this direction . We refer to this migration-based intercalation mechanism as differential migration . To investigate how cells could migrate across each other whilst separated by large gaps , we further characterised cell–cell interactions . When gastrulae were examined using SEM , wide interstitial spaces between endoderm cells were observed to be bridged by thin cell processes ( Figure 5A ) . Under transmission electron microscopy ( TEM ) , interstitial spaces 1 μm wide on average and containing ECM material were observed between laterally aligned cells; they widened into large gaps where several cells met ( Figure 5B ) . Cells were directly attached to each other over short stretches only . In these close contact regions , membranes approached to within 30 nm ( Figure 5B–C ) , a distance compatible with cadherin adhesion . Moreover , thin cytoplasmic protrusions formed stitch-like contacts between cells ( Figure 5B ) . Thus , cells were mostly surrounded by an ECM-filled interstitial space , but were in direct contact in small areas , and through thin cell processes . Gaps between cells were also seen in live explants ( Figure 5D ) . In medium containing AvidinFITC , cells were separated by fluorescent spaces about 1 µm wide on average , similar to the gaps observed using SEM and TEM ( Figure 5E; Figure 5—source data 1 ) . In TEM , densely stained globular material on the surface of cells likely represented collapsed ECM material ( Figure 5F ) . In live explants , long ( 1–10 µm ) , thin ( 0 . 5–1 . 0 µm ) , F-actin-filled protrusions were seen to connect cells for short time intervals while rapidly extending and retracting ( Figure 5D , G ) . We observed that a 5 µm long protrusion could undergo a complete cycle of extension , attachment and retraction in under 3 min to provide highly dynamic , reversible cell contact . Amoeboid migration is commonly noted for its independence from specific substratum adhesion , but in the section below , we provide evidence to show that endoderm cells required defined molecular interactions with the ECM and with each other for proper cell migration . A putative substratum molecule in the ECM is FN . Endoderm cells adhere strongly to the RGD cell binding site of FN in vitro ( Winklbauer , 1988 ) , and they secrete and accumulate FN on their surface ( Winklbauer , 1998 ) . However , in contrast to their in situ behaviour , isolated endoderm cells on FN adsorbed to tissue culture plastic are multipolar , extend lamellipodia , and yet do not migrate ( Wacker et al . , 1998; Luu et al . , 2008 ) . We confirmed that endoderm cells on FN/plastic were multipolar and extended numerous protrusions ( Figure 6A ) . On gelatin-coated dishes , cells were non-adherent , but on a substratum of FN adsorbed to gelatin , cells adopted in vivo-like features ( Figure 6A–B ) . They attached and elongated weakly , and established polarity through the development of discernable trailing ends , whereas lamellipodia were absent ( Figure 6A ) , as in the embryo ( Figure 6B ) . We then examined cell migration on the different substrates ( Figure 6C–D; Video 4 ) . On FN/plastic , cells spread but did not translocate ( Figure 6C , E ) . On gelatin alone , cells also remained stationary ( Figure 6E; Figure 6—source data 1 ) , but on FN/gelatin cells translocated , mimicking their migration in the tissue context ( Figure 6D ) . Cells moved intermittently , at an average velocity of 4 µm/min , which was faster than in the intact tissue ( Figure 6E ) . Thus , interaction with appropriately supplied FN is sufficient to support endoderm amoeboid migration . On FN/plastic or FN/gelatin , the onset of substratum-specific motile behaviour occurred within minutes , which indicates that the modulation of cell behaviour was unlikely due to changes in gene expression , but constituted a direct adaptation to the cell environment . Migration ‘plasticity’ in response to substratum properties is a well-documented phenomenon in many cell types ( Paluch et al . , 2016; Te Boekhorst et al . , 2016 ) . To assess the requirement for FN in situ , we blocked cell–FN interaction using an Arg–Gly–Asp ( RGD ) -containing peptide that competitively inhibits integrin binding to FN . When injected into blastula stage embryos , RGD peptide caused cell rounding and detachment by the gastrula stage . Embryos injected with RGE control peptide appeared normal ( Figure 6F ) . Thus , FN was required to maintain endoderm cell adhesion in the embryo . Exposure of vegetal slice explants to RGD peptide also perturbed cell morphology ( Figure 6G , H; Figure 6—source data 1 ) . Compared with control RGE peptide treatment , RGD-treated cells became significantly more rotund . However , because cells can deviate randomly from a spherical shape , we determined the component of elongation that paralleled the A–V axis , that is the approximate axis of migration ( Figure 6—figure supplement 1 ) . This ‘elongation congruity’ was unity in RGD-treated cells , which indicated that elongation was indeed random and not aligned with migration , in contrast to control peptide treated explants ( Figure 6I; Figure 6—source data 1 ) . Importantly , cells became nearly non-migratory upon RGD treatment , whereas RGE-treated cells moved at normal velocities ( Figure 6J; Figure 6—source data 1 ) . Our results show that endoderm cell interaction with FN is sufficient for translocation in vitro , and necessary for cell movement within the tissue . The sites of close ( 20–30 nm ) cell–cell contacts in the endoderm are compatible with cadherin binding . C-cadherin is the main isoform in the early gastrula ( Kühl and Wedlich , 1996 ) . Its knockdown by a well characterised , C-cadherin mRNA rescuable morpholino antisense oligonucleotide ( CcadMO; Ninomiya et al . , 2012 ) caused rounding of cells in vegetal slice explants ( Figure 7A–D; Figure 7—source data 1 ) . This shape change was unexpected given that endoderm cell–cell contact is largely mediated through short-lived , lateral filiform protrusions . However , the behaviour of α-catenin indicated that C-cadherin was indeed functional in these transient contacts . In the X . laevis gastrula , α-catenin is associated with the cadherin–β-catenin complex at sites of adhesion ( Kurth et al . , 1999 ) . We found that α-catenin was indeed recruited upon contact , but not only to the tips of filiform protrusions . It also accumulated at their bases , outside of the direct contact area , and it dispersed after protrusion retraction ( Figure 7A–B; Figure 7—figure supplement 1 ) . This response suggests that cadherins of adjacent cells transiently engaged in binding . Strikingly , CcadMO-injected cells possessed fewer lateral protrusions ( Figure 7A–B , E; Figure 7—source data 1 ) . Those that formed were as stable as in controls , but protrusion initiation itself was reduced upon C-cadherin knockdown ( Figure 7F , G; Figure 7—source data 1 ) . In turn , less α-catenin accumulated laterally in cells as fewer protrusions were present at any point in time in morphant cells ( Figure 7A–B ) . A feedback between cadherin-dependent α-catenin recruitment and filiform protrusion formation appeared to be involved in endoderm cell interaction . Ultimately , this process amounted to strengthening of cell–cell adhesion . CcadMO-injection reduced tissue surface tension , a measure of tissue cohesion ( Winklbauer , 2015 ) in the endoderm ( Figure 7H; Figure 7—source data 1 ) . Thus , although contacts compatible with cadherin adhesion were infrequent and minute , C-cadherin contributed significantly to the mutual attachment of vegetal endoderm cells . C-cadherin-based interaction is essential for endoderm cell migration . In vegetal slice explants , cell trajectories were reduced by CcadMO to the level of the most vegetal cells in normal explants ( Figure 7I–K; Figure 7—source data 1 ) . Paradoxically , it was exactly in the regions where packing was less dense that C-cadherin was required for cells to elongate and to attain their full migration velocity . Our results indicated that in these regions , interactions of cells with the ECM component FN and through the cell adhesion receptor C-cadherin were both required for amoeboid differential migration . We noted a unique mechanism of tail retraction in endoderm cells . The process appeared as a simple narrowing of the cell–cell contact area at the rear end ( Figure 3A , D ) when viewed in a plane outside the actual retracting rim ( Figure 8A , B ) . If the complete tail was exposed , however , it resembled a lamelliform protrusion ( Figure 8A , B ) that was actin-rich and actively protruded and retracted while attached to the surface of an adjacent cell ( Figure 8C , D; Videos 5 and 6 ) . However , the protrusion pointed opposite to the direction of cell movement , and was dragged behind the advancing cell body while accumulating localised membrane clusters ( Figure 8D; Video 5 ) . Increased membrane undulations were also seen more laterally at the trailing edge , and FITC-conjugated dextran was taken up in vesicles that initiated from membrane pits ( Figure 8E ) . When viewed under TEM , vesicles were found in cell tails in densely packed clusters ( Figure 8F ) , and structures consistent with different stages of endocytosis were discernable at the membrane ( Figure 8G ) . Trailing edge vesicles were also seen in the SEM where the cell surface was incidentally broken , whereas the intact surface showed pits of a similar size ( Figure 8—figure supplement 1 ) . Vesicle sizes varied from 0 . 05 to 3 μm ( Figure 8H; Figure 8—source data 1 ) , which well exceeded the 0 . 1–0 . 2 μm range for clathrin-mediated endocytosis ( McMahon and Boucrot , 2011 ) , but were consistent with formation by macropinocytosis . Rab5 is associated with early stages of macropinocytosis ( Lanzetti et al . , 2004 ) , and it was enriched in the trailing domain . Rab5-CFP puncta accumulated and resolved on the scale of minutes alongside membrane clusters that entered the cytoplasm ( Figure 8I; Video 7 ) . The association of vesicle internalisation with high protrusive activity , the large and heterogeneous size of vesicles , and their interaction with Rab5 strongly suggest that endocytosis at the trailing edge was based on macropinocytosis . Thus , translocation of the endoderm cell rear occurred by the forward movement of the yolk-rich content of the cell body , which left behind a lamelliform , highly protrusive tail that was , at least partially , resorbed by macropinocytosis . EphrinB , a transmembrane protein signalling ligand , has been implicated in macropinocytosis ( Bochenek et al . , 2010 ) , and X . laevis endoderm expresses a full complement of EphB and ephrinB isoforms ( Rohani et al . , 2011 ) . Because we had noted an effect of ephrinB1 on vegetal rotation ( Figure 9—figure supplement 1 ) , we examined its localisation in migrating cells using a fluorophore-fused construct , ephrinB1-mCherry . Immediately after explantation , ephrinB1-mCherry was evenly expressed at the cell membrane , but as cells attained an elongated morphology , it progressively accumulated at the rear membrane ( Figure 9A; Video 8 ) . Enrichment at the trailing edge was maintained during migration ( Figure 9B , C; Figure 9—figure supplement 2; Figure 9—source data 1 ) . In ephrinB1-MO morphant cells , endosome number in the trailing domain was decreased ( Figure 9E; Figure 9—source data 1 ) . Overexpression of ephrinB1 increased endosome numbers ( Figure 9E ) and induced ectopic vesicle internalisation throughout the cell surface , leading to cell rounding and detachment ( Figure 9D , H ) , although cells still showed a ‘kneading’ motion of their surface . In summary , we concluded that ephrinB1 regulated endocytosis in X . laevis endoderm cells . Despite being largely separated from cells laterally ( Figures 5B and 8F ) , the elongated shape of cell tails indicated attachment at some point . The detachment required for eventual tail retraction could also be mediated by ephrinB1 , and a double role was in fact indicated by the presence of two types of vesicles . At interstitial spaces , large vesicles pinching off from the cell surface co-localised with ephrinB1-mCherry ( Figure 10A ) , and consistent with the cytoplasmic attachment of the mCherry tag , vesicles were labelled on their cytoplasmic surfaces ( Figure 10B ) . However , some vesicles showed more intense membrane labelling and had the mCherry label on both the inner and outer surfaces ( Figure 10C ) . This finding suggested that the vesicles were generated by trans-endocytosis , a mechanism whereby an ephrin/Eph receptor-associated cell contact is resolved; a cell endocytoses its own membrane together with that of the adjacent cell to which it is linked through the ephrin/Eph interaction ( Gaitanos et al . , 2016 ) . Indeed , we observed in the TEM , in addition to numerous simple vesicles , a smaller number of double-layered ones , that is , cytoplasm-filled inner vesicles within outer vesicles ( Figure 10E ) , as expected from trans-endocytosis ( Figure 10C ) . Whereas macropinocytosis removes free cell surfaces , trans-endocytosis can break cell contacts at the rear , and remove surfaces in contact with adjacent cells . Together , ephrinB1-dependent macropinocytosis and trans-endocytosis could permit tail resorption and retraction in endoderm cells ( Figure 10D ) . In fact , endocytosis at the cell rear was correlated with trailing edge retraction . Normally , the width of the trailing edge decreased continually , whereas in ephrinB1-MO morphant cells , it remained unchanged ( Figure 9F , G; Figure 9—source data 1 ) . Consistent with impaired tail retraction , ephrinB1-MO morphant cells became elongated ( Figure 9H ) , and vegetal rotation was halted ( Figure 9—figure supplement 1 ) . We propose that resorption of the tail by endocytosis is an essential component in the retraction of the trailing edge of vegetal endoderm cells , which in turn is necessary for their migration . However , we cannot exclude the possibility that ephrinB1 also functions outside the trailing edge to modulate cell contact behaviours required for translocation . In invertebrates , internalisation of mesoderm and endoderm starts from a single-layered cell array , typically an epithelium . In the sister group to vertebrates , the urochordates , a single layer of endoderm cells invaginate by constricting apically and laterally , thus inverting its wedge shape ( Figure 11A ) ( Satoh , 1978; Sherrard et al . , 2010 ) . In X . laevis , the corresponding region has increased dramatically in size . The large vegetal blastomeres also initially form a single-layered array , but soon become subdivided into a multilayered vegetal cell mass by periclinal and anticlinal cleavage divisions ( Figure 11B ) . Nevertheless , the vegetal endoderm as a whole performs a similar inversion of its wedge-shaped configuration to narrow at the external surface and to widen inside the embryo ( Bauer et al . , 1994 ) ( Figure 11B ) . This tissue deformation is based on deep cell rearrangement . The outer , epithelial layer does not invaginate except in a small zone at its margin where bottle cells form ( Figure 11B ) . Otherwise , the epithelium constricts moderately to accommodate the narrowing of the vegetal mass ( Keller , 1978 ) , and later bends upward at its periphery during archenteron formation ( Evren et al . , 2014 ) ( Figure 11B ) . However , the epithelial layer retains its integrity , and cells do not leave the surface by ingression ( Keller , 1978 ) . Deep vegetal endoderm cell rearrangement is due to the migration of cells over each other without the use of lamellipodial , filopodial or bleb-like protrusions; thus , we have classified this mode of migration as amoeboid ( Figure 11E ) . Amoeboid migration is typically seen when single cells move through three-dimensional ECM or between stationary cells , as documented for human leukocytes , zebrafish primordial germ cells , and cancer cells ( Mandeville et al . , 1997; Sahai and Marshall , 2003; Wolf et al . , 2003; Blaser et al . , 2006 ) . X . laevis vegetal endoderm provides an example of amoeboid translocation in the context of collective cell movement ( Figure 11D ) . Previously , Holtfreter ( 1944 ) noted that in the salamander Ambystoma , prospective endoderm cells take on a ‘sausage shape’ when isolated , and engage in an ‘obscure gliding movement’ . Amoeboid creeping of elongated endoderm cells on agarose was also observed for the newt Cynops pyrrhogaster ( Kubota , 1981 ) , and ‘sausage-like’ cells were isolated from the endoderm of the frog Rana pipiens ( LeBlanc et al . , 1981 ) . These findings suggest that amoeboid migration is a widespread mechanism for amphibian endoderm internalisation . Several sub-types of amoeboid migration can be distinguished ( Paluch et al . , 2016 ) , but the specific mechanism of amphibian endoderm cell translocation is not known . Amoeboid translocation is typically associated with low substratum adhesion and movement in confined spaces where traction is generated by repeated cell shape changes ( Paluch et al . , 2016; Te Boekhorst et al . , 2016 ) . Vegetal endoderm is indeed the least cohesive tissue in the gastrula ( David et al . , 2014 ) , and cells cycle through elongation of the cell body , bulging of the cell front , and narrowing and retraction of the trailing edge ( Figure 11E ) . Ambystoma endoderm cells also employ shape changes in vitro ( Holtfreter , 1944 ) , whereas C . pyrrhogaster cells glide continuously ( Kubota , 1981 ) . However , all of these cells can move in isolation while lightly attached to a surface , without the need for lateral confinement , which excludes mechanisms such as ‘chimneying’ ( Yip et al . , 2015 ) . Moreover , cells move their yolk-filled contents forward relative to sites of substratum attachment , but no uropod pushes the cytoplasm by myosin II-powered contractions , and the yolk platelets show no signs of flowing within cells in a fountain movement; instead , they rigidly maintain their relative positions . In C . pyrrhogaster endoderm cells , membrane flows continuously from the front to the rear end , relative to a yolk-filled interior that behaves as a coherent body ( Kubota , 1981 ) . Further study will be necessary to identify the mechanism that moves the cell content forward relative to the substratum-attached membrane . The most striking difference from other instances of amoeboid migration is seen at the cell’s rear end . Instead of being pulled forward by a uropod ( Vicente-Manzanares et al . , 2007; Lämmermann et al . , 2008 ) , the trailing edge of X . laevis endoderm cells is at least in part resorbed by massive macropinocytosis . Likewise , in C . pyrrhogaster endoderm cells , large vesicles accumulate and membrane protrusions are present at the rear end ( Kubota , 1981 ) . The combination of locomotion by amoeboid leading-edge behaviour and macropinocytotic tail retraction links endoderm locomotion more closely to bottle cell invagination and ingression than to other examples of amoeboid migration ( Figure 11C , D ) . Invagination by bottle cell formation , and gastrula-stage ingression in the sea urchin or at the primitive streak of birds and mammals , start similarly . The basal end of cells bulges interiorly while the apical portion narrows ( Figure 11C ) . This is the end state in bottle cell invagination . Ingressing cells continue to move to the inside , across like cells from which they are separated by ECM-filled gaps , without using locomotory protrusions , until they detach from the surface , retract their necks , and leave the epithelial layer ( Figure 11C ) ( Balinsky and Walther , 1961; Perry et al . , 1966; Granholm and Baker , 1970; Wakely and England , 1977; Batten and Haar , 1979; Katow and Solursh , 1980; Komazaki , 1995; Viebahn et al . , 1995; Lee and Harland , 2010; Williams et al . , 2012 ) . During this process , the apical cell surface forms actin-rich protrusions and internalises large vesicles , which often become packed into clusters . This process resembles macropinocytosis , which can occur at the apical membrane of epithelial cells ( Mettlen et al . , 2006 ) . In addition , cytoplasm-filled vesicles-within-vesicles occur in bottle cells of X . laevis and the newt Triturus ( Perry et al . , 1966; Lee and Harland , 2010 ) , and the pinching-off of such a vesicle was captured in a mouse primitive streak cell ( Batten and Haar , 1979 ) . Based on their co-occurrence with labelled ephrinB1-containing vesicles , we tentatively identified similar structures in X . laevis endoderm as the products of trans-endocytosis . This mechanism drives cell–cell separation upon Eph receptor–ephrin interaction – these molecules actually form strong bonds – by one cell engulfing the interaction site including the membrane of the adjacent cell ( Gaitanos et al . , 2016 ) . The high protrusive activity associated with intense vesicle formation at the trailing edge of endoderm cells , combined with the amoeboid translocation of the cell body , suggests that vegetal cell migration may be derived from an invagination or ingression process ( Figure 11C , D ) . A further commonality between endoderm cell migration and ingression is that cells use each other as substratum for translocation in a collective cell movement . Whereas amoeboid migration in general is considered to be independent of specific cell adhesion mechanisms , X . laevis endoderm cells require interaction with FN and cadherin for proper migration . Bottle cells form transiently or permanently during ingression or invagination , respectively , and both processes are mechanistically related . For example , whereas mesoderm invaginates in D . melanogaster , it ingresses in the midge Chironomus riparius , and a change in the expression of one or two effector genes of Snail is sufficient to switch between the two mechanisms ( Urbansky et al . , 2016 ) . In vegetal endoderm cells , the stereotypical motile behaviour characteristic of ingression is executed repeatedly , constituting a special case of amoeboid translocation , ingression-type migration . Ingression proper is absent from the vegetal epithelial layer ( Keller , 1978 ) , and we propose that vegetal rotation is derived from invagination . Apical constriction of vegetal epithelial layer cells was occasionally observed in the SEM . Because deep vegetal cells do not possess an apical domain through which they would be linked , they simply move past each other when executing their program of bottle cell formation ( Figure 11C , D ) . Endoderm migration is directional , with cells initially polarised in the A–V direction , which suggests that during the cleavage divisions that generate the multilayered structure of the vegetal mass , the primary apical–basal polarity of the vegetal blastomeres is transmitted to all deep cells by an unknown mechanism . Animal–vegetal polarisation of the vegetal mass is apparent in the graded cell packing density along this axis . Moreover , numerous genes show A–V differences in expression at the initial gastrula stage ( Taverner et al . , 2005 ) . For example , a hyaluronic acid synthase is expressed in the animal half of the vegetal cell mass , consistent with hyaluronan supporting the extended interstitial space in this region . Likewise , the Wnt/planar cell polarity components Frizzled-7 and Prickle are enriched in the animal part of the region . Whether cells are also individually polarised or gain polarity and orientation cues during migration from these spatial inhomogeneities remains unclear . Endoderm cells migrating on FN-gelatin are polarised , but we do not know whether this polarity is identical to that within the tissue , or generated by spontaneous symmetry breaking in vitro . The cues that re-orient endoderm cells as they approach the BCF likewise remain to be identified . Cell rearrangement by junction remodelling in epithelial monolayers has been extensively studied , but less detailed information is available for the corresponding processes in multilayered tissues . Differential migration may be an archetypical mode of rearrangement in such compact tissues . Mesoderm involution ( Evren et al . , 2014 ) and the radial intercalation of prechordal mesoderm ( Damm and Winklbauer , 2011 ) are examples of this mechanism in the X . laevis embryo . Likewise , the dorsal migration of lateral mesoderm in zebrafish involves differential migration ( Roszko et al . , 2009; Yin et al . , 2009 ) , and fin or limb bud morphogenesis in fish and mice , respectively , depend on three-dimensional patterns of cell migration ( Ede et al . , 1974; Wyngaarden et al . , 2010; Mao et al . , 2015 ) . Vegetal rotation is another example of cell rearrangement by differential migration . Appropriately graded velocity differences have been observed in explants , and time-lapse X-ray microtomography revealed similar velocity gradients in the embryo ( Moosmann et al . , 2013 ) . Like ingressing cells in the sea urchin , chicken or mouse ( Balinsky and Walther , 1961; Granholm and Baker , 1970; Wakely and England , 1977; Batten and Haar , 1979; Katow and Solursh , 1980; Komazaki , 1995; Viebahn et al . , 1995; Lee and Harland , 2010; Williams et al . , 2012 ) , translocating endoderm cells in X . laevis and other amphibians ( e . g . Nakatsuj and Nakatsuj , 1975 ) are separated by prominent interstitial gaps . ECM material is present in the gaps , but a self-supporting , trans-cellular ECM structure is unlikely to permeate the endoderm for several reasons . First , when fixed for TEM , the ECM collapses locally into densely stained , isolated globules at the surface of cells , and when endoderm is dissociated in calcium-free medium , no conspicuous , stable ECM scaffold remains . Secondly , vegetal explants deform rapidly beyond their initial contours by cell migration , a behaviour that is difficult to reconcile with a stable , pre-formed matrix scaffold as an external substratum . In addition , the movement of individual endoderm cells is the resultant of active and passive components , which is inconsistent with individual cells migrating through a stable , external ECM scaffold . Lastly , proteins that form covalently cross-linked ECM structures , such as collagens , laminin or fibrillin , are not expressed in the early gastrula ( Winklbauer and Ettensohn , 2004 ) . Thus , the ECM in the vegetal cell mass is best viewed as a cell surface coat that contributes to weak and flexible adhesion between cells . A highly hydrated , space-filling constituent of the ECM in the X . laevis gastrula is hyaluronic acid ( Müllegger and Lepperdinger , 2002 ) , which may act as a spacer between cells , whereas FN contributes to cell adhesion . FN and its receptor α5β1 integrin are expressed in gastrula endoderm ( Winklbauer and Ettensohn , 2004 ) , and the knockdown of FN diminishes cell–cell contacts in the endoderm ( Barua et al . , 2017 ) . Likewise , inhibition of FN–integrin interaction with RGD peptide leads to vegetal cell detachment . FN-mediated cell–cell adhesion has been demonstrated for other cell types ( Robinson et al . , 2003 , 2004 ) . FN inhibition impedes endoderm cell migration and affects internalisation of the vegetal endoderm ( Ramos and DeSimone , 1996; Winklbauer and Keller , 1996 ) , which may be a consequence of diminished adhesion , or due to reduced integrin signalling . Vegetal endoderm cells are also connected via small direct membrane contacts and filiform lateral protrusions . Despite the infrequent occurrence of cell–cell contact , C-cadherin is required for proper endoderm cell migration . Cadherins couple cells mechanically , but also act as signalling molecules that control the actin cytoskeleton ( Priya and Yap , 2015 ) . During cell rearrangement by junction remodelling , both roles are important to drive neighbour exchanges in cells closely attached to each other through adherens junctions . In the loosely packed endoderm , knockdown of C-cadherin impedes lateral cell–cell contacts through attenuation of filopodia formation . Cells round up and fail to translocate , but tissue cohesion is still retained; we speculate that here C-cadherin mainly assumes a signalling role . Cadherin-based contact could control the lateral cell cortex through the recruitment of α-catenin ( Drees et al . , 2005; Amack and Manning , 2012; Winklbauer , 2015 ) , and permit for example the lengthening of the cell body as an essential step in translocation . Altogether , amoeboid migration in the vegetal endoderm is not independent of specific adhesion mechanisms . This difference from other instances of amoeboid translocation may reflect the fact that endoderm cells move not single , but as a coherent mass to rearrange by differential migration . Interstitial space between migrating amoeboid cells is common in vertebrate embryos . In sections , it often appears as channels between cells that run in parallel to the longitudinal axes of the cell bodies , are of even width , and follow the contours of neighbouring cells , consistent with cell–cell attachment across the gaps ( e . g . Granholm and Baker , 1970; Viebahn et al . , 1995 ) . This pattern is also seen in adult tissues of basal metazoans , such as in poriferan archaeocytes ( Weissenfels , 1982 ) , where cell adhesion is mediated by large , space-filling carbohydrate–protein complexes ( Fernàndez-Busquets and Burger , 2003 ) . Movement of a space-filling ECM , together with the migrating cells that it surrounds , has been observed , for example , in the chicken gastrula , when internalised mesoderm moves away from the primitive streak ( Vanroelen et al . , 1980; Van Hoof and Harrisson , 1986 ) . The amoeboid migration of cells in contact with each other through ECM coats may be an ancestral mechanism of cell rearrangement and morphogenesis in metazoans . Adult X . laevis were housed in aquaria , water temperature 19–20°C . Research animals were used in accordance with guidelines approved by the University Animal Care Committee ( Protocol no . 20011765 , University of Toronto , Canada ) . X . laevis embryos were obtained by in vitro fertilisation ( Winklbauer , 1986 ) . Briefly , eggs were obtained and fertilised with macerated testes . To remove the jelly coat , embryos were incubated in a solution of 2% cysteine ( w/v , pH 8 . 0 ) in 0 . 1X Modified Barth's Saline ( MBS; 88 mM NaCl , 1 mM KCl , 2 . 4 mM NaHCO3 , 0 . 82 mM MgSO4 , 0 . 33 mM Ca ( NO3 ) 2 , 0 . 41 mM CaCl2 , 10 mM Hepes ( +NaOH ) , 1% Streptomycin , 1% Penicillin , pH 7 . 4 ) for 5 min and subsequently cultured in 0 . 1X MBS until the desired stage for experimentation . Embryos were staged according to Nieuwkoop and Faber , 1967 . Embryos were microinjected in a 3% Ficoll ( w/v; Sigma-Aldrich , St . Louis , MO ) solution in 1X MBS on plasticine-coated injection dishes . All embryos were seeded into Ficoll dishes 15 min before injection and allowed to heal for 1 hr in Ficoll after injection . All microinjections were performed at the four-cell stage . Microsurgical manipulations were carried out on sterilised plasticine-coated petri dishes in 1X MBS . Tissues were excised with eyebrow knife and hair-loop tools then transferred onto glass-bottom dishes ( MatTek , Ashland , MA ) or tissue-culture dishes ( Cellstar , Germany ) rendered non-adhesive by pre-coating with 1% bovine serum albumin ( w/v , BSA ) solution in 1X MBS . Dissections were carried out under a Zeiss Stemi SV 11 microscope , observed with a Zeiss AxioCam MRc digital camera using AxioVision 4 . 8 software . Vegetal explants were excised from stage 10 gastrulae . A mid-sagittal vegetal slice ( without the blastocoel roof ) about 5 cell layers thick was secured onto a glass bottom dish with a cover glass , altogether held in place by silicon grease as described ( Winklbauer , 1998 ) . All live recordings were captured in an ambient temperature of 21–23°C . Bright-field and epi-fluorescence time-lapse videos were recorded on a Zeiss Axiovert 200 M inverted microscope with Zeiss AxioCam MRm digital camera using AxioVision 4 . 8 software . High resolution images were captured with a Leica TCS SP8 confocal laser scanning microscope equipped with HCX-PL-APO-CS 10x/NA0 . 40 , HC-PL-APO-CS 20x/NA0 . 75 , 40x/NA1 . 30 , 63x/NA1 . 20 , HC-PL-APO-CS2 100x/NA1 . 4 oil-immersion objectives , and HCX-IRAPO-L 25x/NA0 . 95 , HC-PL-APO-CS2 63x/NA1 . 20 water-immersion objectives and resonant scanning system using Leica LAS AF 3 . 2 software . Embryos were fixed in 2 . 5% glutaraldehyde , and 2% paraformaldehyde in 50 mM HEPES buffer ( pH 7 . 4 ) overnight at 4°C . Embryos were rinsed in 0 . 1 M sodium cacodylate ( pH 7 . 0 ) and bisected mid-sagittally using a microsurgical knife . Post-fixation was performed by incubating embryos with 1% osmium tetroxide in 0 . 1 M sodium cacodylate overnight at 4°C . Samples were then dehydrated through a graded ethanol series ( 30% , 50% , 70% , 2 × 100% ) , dried overnight , and sputter-coated with gold–palladium . Images were obtained using the Hitachi S2500 scanning electron microscope . X . laevis gastrulae with the vitelline membrane removed were gently punctured in the blastocoel roof with a tungsten needle to facilitate stain infusion . Perforated gastrulas were fixed for one week at 4°C in 3% glutaraldehyde ( Fisher Scientific , Pittsburgh , PA ) , 2% paraformaldehyde ( Fisher Scientific ) , and 1% alcian blue ( Sigma-Aldrich ) in 0 . 05 M sodium cacodylate ( Sigma-Aldrich , pH 7 . 0 ) . Embryos were rinsed in 0 . 1 M sodium cacodylate ( pH 7 . 0 ) , bisected mid-sagittally and embedded in 3% low-melting point agarose ( Sigma-Aldrich ) , then fixed overnight at 4°C in 1% osmium tetroxide ( Electron Microscopy Sciences , Hatfield , PA ) , and 1% lanthanum nitrate ( Sigma-Aldrich ) in 0 . 1 M sodium cacodylate ( pH 7 . 0 ) . Embryos were then rinsed in 0 . 1 M sodium cacodylate ( pH 7 . 0 ) and dehydrated in a graded series of ethanol solutions ( 30% , 50% , 70% , 90% , 2 × 100% ) then gradually infiltrated with Spurr’s resin overnight at room temperature . Embryos were then cured at 65°C overnight . Semi-thin ( 1–1 . 5 µm ) sections were made using a Leica RM2235 microtome , and ultrathin ( 90–100 nm ) sections were prepared using a Leica EM UC6 microtome . Semi-thin sections were stained with 1% toluidine blue for sample inspection under bright-field microscopy , while ultrathin sections were stained using 3% uranyl acetate and Reynold’s lead citrate to provide contrast for imaging using the Hitachi HT7700 transmission electron microscope operated at 80 kV . Tissue aggregates round up into drop-shapes in vitro . The drop-shape of an aggregate at equilibrium represents a balance of forces between tissue surface tension ( which acts to minimise the tissue into a sphere ) and gravity ( which acts to flatten the tissue ) . Given that gravity is known , tissue surface tension can be deduced by the Laplace equation using the curvature radii of the aggregate profile ( David et al . , 2009 ) . Since surface tension is numerically equal to surface energy , which is the energy required to expand the surface of a droplet by a unit of area which is also equal to half of the energy required to split a droplet into two equal parts ( i . e . creating two new surfaces ) ; tissue surface tension is an effective measure of tissue cohesion . Tissue excised from the early gastrula ( at NF stage 10 unless otherwise specified ) were placed in a well for 30 min to facilitate rounding . Tissue aggregates were transferred onto BSA-coated tissue culture dishes to equilibrate for 2 hr before aggregate curvature profiles were imaged using an inclined mirror calibrated at a 45° angle to the substrate surface . Tissue surface tension was quantified using a modified Axisymmetric Drop-Shape Analysis ( ADSA; Del Río and Neumann , 1997 ) adapted for use with tissue explants ( David et al . , 2009; Luu et al . , 2011; David et al . , 2014 ) . Endoderm cells were isolated using tissue dissected from a central column of the vegetal cell mass between the vegetal pole and the blastocoel floor and incubated in cell dissociation buffer ( 88 mM NaCl , 1 mM KCl , 2 . 4 mM NaHCO3 , 10 mM Hepes ( +NaOH ) , pH 7 . 4 ) to separate cells . Culture dishes were coated with 10% ( w/v ) fish skin gelatin ( Sigma-Aldrich ) over-night then surface coated with bovine plasma FN ( 5 µg/cm2; Sigma-Aldrich ) for 2 hr at room-temperature prior to seeding . Cells were then seeded onto FN–gelatin dishes and incubated in modified Holtfreter’s Solution ( 59 mM NaCl , 0 . 67 mM KCl , and 2 . 4 mM NaHCO3 ) . The cell body length–width ratio ( LWR ) provides a measure of cell elongation parallel to the long axis of the cell by dividing the length of the cell by its width perpendicular to the long axis . LWR was measured using cell outlines captured from time-lapse videos of membrane-labelled explants and scanning electron micrographs of mid-sagittally fractured embryos in various gastrulation stages as indicated . To assess whether cell elongation was congruent with the direction of cell migration , we analysed cell congruity ratio with respect to the animal-vegetal trajectory . Cell body orientation was measured by determining the angle of the long-axis of cells relative to the vertical axis . To determine the long-axis of individual endoderm cells , best-fit ellipses were matched to traced cell outlines ( Blanchard et al . , 2009 ) . Individual fittings were manually checked to contain the same area , orientation and centroid as the original cell . The long-axis of the ellipse was then used as to represent the long-axis of cells . Cell movement trajectories were tracked using MtrackJ ( Meijering et al . , 2012 ) . Cell displacement and velocity measurements were made by connecting the centroid of individual cells from lapsed time-points . Cell path alignment was measured by determining the angle of cell displacement path relative to the long-axis of cells . All experiments were replicated at least three times , and representative images are shown . Statistical testing was conducted using two-tailed unpaired Student’s t-tests to compare different populations . Significant findings are shown by asterisks indicating p-values<0 . 05 ( * ) , <0 . 01 ( ** ) , and <0 . 001 ( *** ) . Standard deviation ( S . D . ) or standard error of the mean ( S . E . ) are indicated . Figures were cropped and presented after intensity adjustment using Photoshop ( Adobe Systems , San Jose , CA ) . All adjustments were performed equally within each experiment . Figures were composed using Illustrator ( Adobe Systems ) .

In most animals , the early embryo consists of a single layer of cells that forms a hollow sphere . This simple structure first gains complexity by organising into multiple layers that are fated to become specialised tissues in the adult , such as muscle or skin . To form the primitive gut , a group of cells known as the endoderm must move from the surface to the interior in a process called gastrulation . Since the early 1900s , the sea urchin and the frog have been the standard species used to study gastrulation . In sea urchin embryos , gastrulation entails bending the sheet of cells that form the surface of the embryo inward at a predetermined site to generate a pocket that will become the digestive system . By contrast , frog embryos begin gastrulation as multilayered structures , and the embryo’s surface does not bend . Furthermore , classic studies of frog gastrulation have found that cells do not leave the surface of the embryo to enter its interior . So despite generations of students having learned about how gastrulation occurs in backboned animals from studying frogs , the cell behaviours that internalise the endoderm are still not understood . Wen and Winklbauer now show that endoderm cells in the frog move using the same set of behaviours that cells in other organisms use to break loose from or bend sheets of cells . Individual cells move by simultaneously pushing their front end forward while retracting their rear in a peculiar manner , by engulfing their own cell surface at a large scale . In the frog embryo , this movement is coordinated into an organised pattern where cells use the surfaces of their slower or stationary neighbours to propel past each other , and then slow down to return the favour . This constitutes a newly defined type of movement referred to as ingression-type migration . Frog embryos are remarkably large because each of their cells is packed with yolk to support development until the animals are able to feed . As an adaptation to this large size , some frog gastrulation movements appear unusual . However , Wen and Winklbauer show that the cell behaviours that drive these movements are similar to the behaviours of cells in single-layered embryos , and indeed the behaviour of single-celled organisms such as amoebae . Further research is now needed to investigate how these cells find their way straight to the interior of the embryo .

2017

Ingression-type cell migration drives vegetal endoderm internalisation in the Xenopus gastrula

Animals vary considerably in size , shape , and physiological features across individuals , but yet achieve remarkably similar behavioral performances . We examined how animals compensate for morphophysiological variation by measuring the system dynamics of individual knifefish ( Eigenmannia virescens ) in a refuge tracking task . Kinematic measurements of Eigenmannia were used to generate individualized estimates of each fish’s locomotor plant and controller , revealing substantial variability between fish . To test the impact of this variability on behavioral performance , these models were used to perform simulated ‘brain transplants’—computationally swapping controllers and plants between individuals . We found that simulated closed-loop performance was robust to mismatch between plant and controller . This suggests that animals rely on feedback rather than precisely tuned neural controllers to compensate for morphophysiological variability . Animals routinely exhibit dramatic variations in morphophysiology between individuals but can nevertheless achieve similar performance in sensorimotor tasks ( Sponberg et al . , 2015; Bullimore and Burn , 2006 ) . Further , individual animals can experience rapid changes in their own morphophysiological features , such as extreme weight changes that occur during and between bouts of feeding . For example , mosquitoes can consume more than their body weight ( Van Handel , 1965 ) and hummingbirds can consume up to 20% of their body weight ( Hou et al . , 2015 ) in a single feeding . How neural control systems accommodate these changes is not known . The behavioral performance of an individual animal is determined via an interplay between its ‘controller’ and ‘plant’ ( Kiemel et al . , 2011; van der Kooij and Peterka , 2011; Cowan et al . , 2014; Dickinson et al . , 2000; Hedrick et al . , 2009 ) . The plant typically includes musculoskeletal components that interact with the environment to generate movement ( Hedrick and Robinson , 2010; Sefati et al . , 2013; Maladen et al . , 2009 ) . The controller typically includes sensory systems and neural circuits used to process information to generate motor commands ( Cowan and Fortune , 2007; Kiemel et al . , 2011; Lockhart and Ting , 2007; Roth et al . , 2014 ) . From the perspective of control theory , one might expect the dynamics of the controller to be precisely tuned to the dynamics of the plant , resulting in an optimal control law that reduces variability in task performance ( Todorov , 2004; Franklin and Wolpert , 2011; Bays and Wolpert , 2007 ) . Were this the case , variations across individuals in morphophysiological features of their plants should manifest in commensurate differences in each animal’s controller . Alternatively , the central nervous system ( CNS ) may be implementing robust feedback control that attenuates morphophysiological variability at the behavioral level without the need for precise tuning . Investigating these relationships requires separate estimates for plants and controllers . However , the classical input–output system identification of behavioral tasks—using only the sensory input and the behavioral output—is limited to generating closed-loop control models of behavioral responses . Data-driven system identification of the underlying neural controllers or locomotor plants requires additional observations such as a measurement of the control output . Electromyograms ( EMGs ) are the most commonly used proxy for the output of the neural controller . EMGs allow separate data-driven estimates of the controller and plant ( Kiemel et al . , 2011; van der Kooij and Peterka , 2011 ) but require understanding the coordination strategy across multiple groups of muscles that interact in fundamentally nonlinear ways ( Ting and Macpherson , 2005 ) . We studied refuge tracking in a species of weakly electric fish Eigenmannia virescens ( Figure 1A ) , a system that permits identification of input–output dynamics as well as the locomotor plant via behavioral observations alone . Like an ‘aquatic hummingbird’ , Eigenmannia precisely hover in place , making rapid and nuanced adjustments to its position in response to the movement of the refuge in which it is hiding ( Rose and Canfield , 1993; Roth et al . , 2011; Uyanik et al . , 2019b; Figure 1—video 1 ) . During refuge tracking , individual Eigenmannia generate fore-aft thrust forces using undulatory motions of their ventral ribbon fin . Undulations are initiated at the rostral and caudal ends of the fin resulting in counter propagating waves that travel towards each other ( Sefati et al . , 2013; Ruiz-Torres et al . , 2013 ) . The two traveling waves meet at a position along the ribbon fin known as the nodal point ( Figure 1—video 2 ) . In a task in which the fish maintains position in a stationary refuge , Eigenmannia shift the rostrocaudal position of the nodal point as a function of steady-state swimming speed ( Sefati et al . , 2013; Figure 1—figure supplement 1 ) , providing a behavioral proxy for the controller’s output , without reliance on EMGs . We measured tracking performance and moment-to-moment position of the nodal point in three fish during a refuge tracking task . Despite the fact that tracking performance of the three fish were similar , there were nevertheless large variations in the movement of the nodal point , reflecting morphophysiological differences between individuals . We used computational techniques , specifically data-driven system identification of feedback control models , to explore how neural control systems cope with individual variability in locomotor dynamics . We estimated a data-driven model for the plant dynamics P⁢ ( s ) of each fish . For the purpose of visualizing the plant models graphically , it is useful to treat the plant , P⁢ ( s ) , as a filter through which motor commands are processed . At a given frequency , ω , the filter’s behavior can be represented as a complex number , called a phasor , that is obtained by evaluating the transfer function P⁢ ( s ) at that frequency , namely P⁢ ( j⁢ω ) , where j=-1 . The locus of phasors as a function of frequency is called the frequency response function . We estimated the frequency responses of the locomotor dynamics of each fish using the position of the fish y⁢ ( t ) and its nodal point u⁢ ( t ) . To visualize the variability across plants , we used a Nyquist plot . On such a plot , the gain and phase of the response of the plant at a given frequency are represented in polar form , where the x axis is the real part and the y axis is the imaginary part—both dimensionless . The gain is the distance from the origin 0+j0 , and the phase is the angle measured counterclockwise from the real axis . Nyquist plots of each individual's estimated plant revealed substantial differences between locomotor plants of individual fish ( Figure 2A ) . Despite these differences , the frequency responses shared a common structure: the locomotor dynamics of each fish acted as a low-pass filter on the movements of the nodal point . This common structure facilitated the application of parametric modeling , reducing the complexity of analysis while enabling computational manipulations of the model system . We used the physics-based parametric model of locomotor dynamics of Eigenmannia described by Sefati et al . ( 2013 ) for the plant ( see Materials and methods for derivation ) : ( 1 ) P⁢ ( s ) =km⁢s2+b⁢s Here , m , k , and b represent mass , gain , and damping , respectively , and s is complex frequency in Laplace domain ( see , e . g . Roth et al . , 2014 ) . We estimated the parameters in the parametric plant model for each fish based on their individualized frequency responses via numerical optimization ( see Materials and methods ) ( Figure 2B ) . Frequency responses for the estimated parametric models are illustrated in Figure 2C ( see black lines in Figure 2A for corresponding Nyquist plots ) . Finally , we estimated the plant for a ‘merged’ fish , in which the data from the three fish were concatenated as a single fish . The differences in the frequency responses between individuals resulted in substantial differences ( about twofold ) in estimated model parameters ( Table 1 ) . Moreover , the merged fish has plant dynamics that differ from each of the individual fish ( Figure 2A , bottom ) , highlighting the need to use individualized plants for the analysis of the control system of each fish . Despite the differences in plant dynamics , fish produced remarkably similar tracking performance , consistent with previously published reports ( Cowan and Fortune , 2007; Roth et al . , 2011 ) . This behavioral robustness could be achieved via precise tuning between the controller and plant dynamics of each fish . Alternatively , the central nervous system ( CNS ) may be implementing robust feedback control without the need for precise tuning . To test these hypotheses , we built feedback control models that permit computational manipulation of the relationships between controller and plant . Specifically , we swapped the controllers and plants between fish using these computational models ( Figure 3 ) . If each fish required precise tuning for consistent behavior , we would expect to see increased variability for the swapped models . Alternatively , a robust feedback controller might be insensitive to mismatch between C⁢ ( s ) and P⁢ ( s ) pairs . We used the second order model proposed by Cowan and Fortune ( 2007 ) to represent the input–output behavioral response of the fish: ( 2 ) G⁢ ( s ) =A⁢ωn2s2+2⁢ζ⁢ωn⁢s+ωn2 We estimated the model parameters for each fish using the position of the fish y⁢ ( t ) and the refuge r⁢ ( t ) . In other words , we generated individualized parametric transfer functions that capture the input–output behavioral performance of each fish . Parameters varied by about 15–20% ( Table 2 ) . We investigated how the variability in plant dynamics ( parameters varied by about twofold; Table 1 ) is mitigated at the level of behavior ( parameters varied by 15–20%; Table 2 ) . Specifically , we inferred a controller for each fish using models of their respective plant dynamics and input–output behavioral responses . Given the plant P⁢ ( s ) and behavioral performance G⁢ ( s ) of each individual fish , we can infer the controller for each fish using the following equation ( see Materials and methods and Cowan and Fortune , 2007 ) : ( 3 ) C⁢ ( s ) =G⁢ ( s ) ( 1-G⁢ ( s ) ) ⁢P⁢ ( s ) Estimates of controllers and plants for individual fish allowed us to implement computational manipulations in the system models . Given a model of each fish’s plant and controller , we computed the transfer function of the closed-loop system using each fish’s own controller when matched with its own plant ( ‘matched’ , Figure 4A ) , via the equation ( see Equation 11 , Materials and methods ) : ( 4 ) Gmatched , i⁢ ( s ) =Pi⁢ ( s ) ⁢Ci⁢ ( s ) 1+Pi⁢ ( s ) ⁢Ci⁢ ( s ) , for i=1 , 2 , 3 . Then , to test the hypothesis that the animals rely on precise tuning between their plants and controllers , we substituted the controller of each fish with the plant dynamics of other fish ( see ‘swapped’ in Figure 4A ) , that is a simulated brain transplant , namely ( 5 ) Gswapped , i , j⁢ ( s ) =Pi⁢ ( s ) ⁢Cj⁢ ( s ) 1+Pi⁢ ( s ) ⁢Cj⁢ ( s ) whereby each controller j=1 , 2 , 3 is paired with another fish’s plant i≠j , for a total of 6 ‘swapped’ cases . If the controllers and plants need to be co-tuned , then we would expect a significant increase in variability in the swapped models . To quantify such changes in variability and to evaluate the fitness of a given computational model for explaining biological data , we defined two metrics termed ‘model variability’ and ‘behavioral variability’ . Model variability quantifies the variability of the complex-valued frequency responses of matched or swapped models across a range of frequencies . Behavioral variability , on the other hand , provides a conservative estimate of the variability observed across trials in the real fish . See Model variability and Bootstrap estimate of behavioral variability in Materials and methods . Using these metrics , we computed the variability of the matched models and compared against the behavioral variability observed in real fish . Unsurprisingly , variability of the matched models ( orange region in Figure 4B ) remained well below the behavioral variability across the entire frequency range of interest ( dashed line in Figure 4B ) . Surprisingly , however , the model variability remained below the behavioral variability even for swapped models ( Figure 4B , gray region versus dashed line ) . These results highlight the fact that sensory feedback can attenuate the variability of closed-loop models despite mismatch between the controller and plant pairs . In other words , feedback models do not require precise tuning between the controller and plant to achieve the low variability we observed in the behavioral performance of the animals . Having established that variability of the closed-loop models is robust to the relations between the plant and controller in a feedback system , we examined the role of feedback . This was achieved by examining the loop gain , that is the dynamic amplification of signals that occurs in feedback systems ( Figure 4C ) . The only difference between the loop gain model and the closed-loop model is the absence and presence of feedback , respectively . For our model , this was calculated as the product of the plant and controller in both matched and swapped cases . This removal of feedback revealed dramatic variability in the loop gain at frequencies below about 1 Hz ( Figure 4D ) . This variability was well above the behavioral variability observed in fish . In contrast , at frequencies above 1 Hz , the model variability was slightly reduced . These results indicate that sensory feedback attenuates behavioral variability in the biologically relevant range of tracking frequencies at a cost of slightly increased variability at high frequencies . As predicted by Cowan and Fortune ( 2007 ) , each of the feedback controllers obtained above for the averaged fish responses had high-pass filtering characteristics despite the differences in their dynamics ( Figure 5A ) . What is the range of neural controllers that , when used in this feedback control topology , leads to behavior that is indistinguishable from the real fish ? In other words , how well tuned to the plant does the neural controller need to be to achieve these behavioral performances ? We used the Youla-Kučera parametrization to obtain a range of controllers that generate similar behavioral responses ( Kučera , 2011 ) . Specifically , this parametrization provided a parametric transfer function describing all stabilizing controllers for the merged plant: ( 6 ) C∅⁢ ( s ) =Q⁢ ( s ) 1-Pm⁢ ( s ) ⁢Q⁢ ( s ) Here , Pm⁢ ( s ) is the transfer function of the merged plant and Q⁢ ( s ) is any stable and proper function . Equation 6 parametrizes all stabilizing controllers for the merged plant Pm⁢ ( s ) . However , we were interested in finding the subset of controllers that yields indistinguishable behavioral performances from real fish . To achieve this , we computed the range for the input–output system dynamics , G⁢ ( s ) , of real fish response . Specifically , we calculated the bounds for the gain and phase responses of the 1000 input–output transfer function models estimated while computing the behavioral variability ( see Materials and methods ) . The gray-shaded areas in Figure 5B serve as the range of frequency responses that are consistent with behavioral variability of the real fish . For each of these 1000 transfer functions G⁢ ( s ) that were consistent with the behavioral variability of the fish , we selected Q⁢ ( s ) =G⁢ ( s ) /Pm⁢ ( s ) to generate 1000 corresponding controllers using Equation 6 . The bounds for gain and phase responses of these 1000 controllers ( the gray shaded regions ) show the breadth of controllers that , when implemented within the feedback control topology , produce behavioral outputs consistent with the performance of the real fish ( see Figure 5A ) . Note that the controllers calculated in Equation 3 also satisfy the structure of Equation 6 when Q⁢ ( s ) =G⁢ ( s ) /P⁢ ( s ) for the associated plant dynamics P⁢ ( s ) of each fish; in other words , the controller estimates are guaranteed to be stabilizing for the associated plants . These results indicate that the neural controllers need not be precisely tuned to their associated plant dynamics . We found a wide range of controllers that the fish could implement to generate consistent behavioral performance . We note that each of these controllers had high-pass filtering characteristics . A key contribution of this work is the identification of a data-driven plant model for the locomotor dynamics of a freely behaving animal based on behavioral observations only . To achieve this , we adopted a grey-box system identification approach that seeks to reconcile a physics-based parametric transfer function model with a non-parametric data-driven model ( i . e . , the frequency-response function ) . Developing a model from first principles , for example Newton’s laws , is sometimes an effective modeling approach for describing the dynamics of a physical system . For instance , a widely used model in legged locomotion is the spring-loaded inverted pendulum ( SLIP ) model for describing running dynamics in the sagittal plane ( Blickhan and Full , 1993; Full and Koditschek , 1999 ) . While physics-based models have proven to be successful in modeling the dynamics of biological and mechanical movements , there are limitations . Physics-based approaches for modeling behaviors at lower levels ( e . g . , the spiking activity of all motor neurons ) may lead to a very complex model that does not accurately capture high-level behavior . Data-driven system identification approaches are used to directly identify a dynamical model based on empirical data ( Ljung , 1998; Kiemel et al . , 2011; Uyanik et al . , 2019a ) . In general , data-driven system identification may take a black-box approach in which only a general model structure is assumed ( say , an ODE or frequency response function ) . However , data-driven techniques typically generate numerical transfer function estimates to represent animal behavior . Alternatively , the so-called grey-box approach that we adopt in this paper integrates the structure of a specific physics-based model but leaves its parameters free , relying on data-driven system identification to fit those parameters . In this case , prior knowledge about the underlying dynamical model informs and constrains data-driven system identification . Grey-box identification can provide a bridge between top-down , data-driven modeling and bottom-up , physics-based modeling . We utilized the parametric dynamical model of Sefati et al . ( 2013 ) for the plant but estimated the model parameters using data-driven system identification techniques . Our results show that the data-driven estimates for the plant dynamics match the structure of this model ( Figure 2A ) . Our results reveal two complementary perspectives on variability in plant dynamics . On the one hand , estimates of the closed-loop controllers were highly sensitive to the dynamics of the plant of individual fish . This was an inevitable consequence of our strategy for estimating the controllers—inferring the controllers from the plant and closed-loop dynamics . On the other hand , our closed-loop control models were robust to variability of either the plant or controller , indicating that precise tuning is not needed for this behavior . A control-theoretic sensitivity analysis demonstrates that these results are not unique to this example but rather are a general property of feedback control systems , see Csete and Doyle ( 2002 ) for a review . Specifically , consider the frequency dependent sensitivity function of the feedback controller C⁢ ( s ) with respect to plant P⁢ ( s ) , in the closed-loop topology: ( 7 ) S ( C←P ) ( jω ) =∂C∂P=[G ( jω ) G ( jω]−1]1P2 ( jω ) The sensitivity of the controller to the plant dynamics is a frequency dependent function . It depends on the gain and phase of both the measured closed-loop transfer function G⁢ ( j⁢ω ) as well as the plant model P⁢ ( j⁢ω ) . At low frequencies , fish track extremely well and thus G⁢ ( j⁢ω ) -1≈0 . At high frequencies , the low-pass plant P⁢ ( j⁢ω ) is small . Combining these factors , we expect the sensitivity |S⁢ ( j⁢ω ) | to be large across frequencies . In other words , there is an inescapable sensitivity to plant dynamics when the controllers are estimated using this computational strategy . We conducted a complementary analysis to compute the sensitivity of the closed-loop tracking response G to perturbations in the combined controller and plant dynamics P⁢C . We treated the controller–plant pair P⁢C as a single variable and obtained the frequency-dependent sensitivity function as ( 8 ) S ( G←PC ) ⁢ ( j⁢ω ) =∂⁡G∂⁡P⁢C=1 ( 1+P⁢C⁢ ( j⁢ω ) ) 2 . At low frequencies , G⁢ ( j⁢ω ) has nearly unity gain and thus P⁢C⁢ ( j⁢ω ) goes to ∞ . As a result , sensitivity S ( G←PC ) approaches zero . At high frequencies , P⁢C⁢ ( j⁢ω ) goes to zero and thus sensitivity is bounded around 1 . Thus , despite the fact that the controller estimates are sensitive to plant variations , the closed-loop transfer function ( in the presence of sensory feedback ) is robust against variability of controller–plant pairs . These findings suggest that a fish could implement a range of controllers in its nervous system for refuge tracking . These controllers must have high-pass filtering characteristics , but their details may be inconsequential . This has two implications for neurophysiological analysis of neural control systems . First , neurophysiological activity within control circuits in open-loop experiments ( e . g . playback and anesthetized/immobilized animals ) need not appear to be well tuned for the control of a behavioral performance . This poor tuning , which may manifest in variability that appears across levels of functional organization—from variability in neural activity within neurons , variability in tuning across neurons , and variability across individuals—is refined via feedback during behaviors in which the feedback loops are intact . Second , there must be mechanisms by which the controllers are slowly ( at time constants greater than that necessary for the moment-to-moment control of the behavior ) tuned to the dynamics of the animal’s locomotor plant . For instance , adaptation of cerebellar activity in relation to mismatch between intended versus actual motor performances contribute to the retuning of neural controllers ( Morton and Bastian , 2006; Bell et al . , 1997; Pisotta and Molinari , 2014 ) . Feedback is mediated both through the effects of behavior on sensory receptors and via descending pathways in the brain . Behavior generally results in concomitant activation of receptor types across the animal , which can include , for example , simultaneous stimulation of stretch receptors embedded in muscles and visual receptors in eyes . Correlations in feedback-related activity across sensory modalities likely contribute to robust control ( Roth et al . , 2016 ) . Internal feedback pathways , interestingly , have been recently shown to synthesize sensory filtering properties of behaviorally-relevant stimuli . Descending neural feedback is used to dynamically synthesize responses to movement ( Metzen et al . , 2018; Huang et al . , 2018; Clarke and Maler , 2017 ) . How do other animal systems manage variability , broadly speaking , to achieve consistent output ? In pyloric neural circuits of crustaceans , the oscillatory output of the system is consistent despite dramatic variations in the dynamics of cellular and membrane properties of neurons within these circuits ( Goaillard et al . , 2009; Marder and Taylor , 2011 ) . How these circuits maintain consistent output despite underlying variability remains an open question ( Hamood and Marder , 2014 ) but likely relies on feedback regulation that is intrinsic to the neural network itself . The mechanisms by which systems maintain consistent output can be assessed through behavioral analysis ( Krakauer et al . , 2017 ) of responses to systematic perturbations ( Cowan et al . , 2014 ) . For example , perturbations have been used in many species including flying insects ( Bender and Dickinson , 2006; Matthews and Sponberg , 2018 ) , walking sticks ( Dallmann et al . , 2019; Diederich et al . , 2002 ) , and humans ( Lee and Perreault , 2019 ) to reveal how control systems manage mechanical and sensory variation . These analyses and others show that animals rely on sensory feedback to modulate moment-to-moment movement to maintain consistent task performances . Ultimately , understanding how robustness emerges in closed loop requires investigating the interplay between plants and controllers , which are inextricably linked ( Cowan and Fortune , 2007; Tytell et al . , 2011; Tytell et al . , 2018 ) . A commonly implemented strategy , for example , is the use of low-pass plant dynamics . This strategy can avoid instabilities that arise from long-latency feedback , an inescapable feature of biological control systems ( Sponberg et al . , 2015; Madhav and Cowan , 2020 ) . Specifically , delay introduces phase lag that increases with frequency . As the phase angle of the loop gain P⁢C approaches −180° , the likelihood of instability increases . A low-pass plant can mitigate this instability by ensuring that the gain of |P⁢C|≪1 near this ‘phase crossover frequency . ’ In turn , the animal exhibits low behavioral sensitivity , namely S ( G←PC ) ≲1 in Equation 8 . In short , a well-tuned neuromechanical plant can simplify feedback control by rendering the dynamics passively self-stabilizing ( Hedrick et al . , 2009; Sefati et al . , 2013 ) , while nevertheless maintaining behavioral flexibility ( Cowan et al . , 2014 ) . The experimental apparatus is similar to that used in previous studies ( Stamper et al . , 2012; Biswas et al . , 2018; Uyanik et al . , 2019b ) . A refuge machined from a PVC pipe with a length of 15 cm and 5 . 08 cm diameter was placed in the experimental tank with the fish . The bottom face of the refuge was removed to allow video recording from below . Six windows , 0 . 625 cm in width and spaced within 2 . 5 cm intervals , were machined onto each side to provide visual and electrosensory cues . During experiments , we actuated the refuge using a linear stepper motor with 0 . 94 μm resolution ( IntelLiDrives , Inc Philadelphia , PA , USA ) driven via a Stepper motor controller ( Copley Controls , Canton , MA , USA ) . MATLAB ( MathWorks , Natick , MA , USA ) scripts were used to control the movement of the refuge and to capture video . Video data were captured using a pco . 1200s high speed camera ( Cooke Corp . , Romulus , MI , USA ) with a Micro-Nikkor 60 mm f/2 . 8D lens ( Nikon Inc , Melville , NY , USA ) . All videos used for data analysis were shot at 30 frames per second with 1280 × 1024 pixel resolution . Some videos of ribbon fin motion were shot at 100 frames per second . Refuge movement consisted of single sine waves of amplitude 0 . 1 cm and of frequencies 0 . 55 , 0 . 95 , and 2 . 05 Hz . The amplitude of refuge movements was chosen because fish rely on counter propagating waves for tracking in this regime ( Roth et al . , 2011 ) . At higher amplitudes , fish often will use a uni-directional wave in the ribbon fin for locomotion . The frequencies were selected to be within the normal tracking regime as determined in previous studies ( Stamper et al . , 2012; Biswas et al . , 2018; Uyanik et al . , 2019b ) . Trials were randomized with respect to frequency . Each trial lasted 60 seconds . The stimulus amplitude was linearly ramped up over the first ten seconds to prevent startle responses from the fish . During the experimental phase , the stimulus frequency and amplitude were maintained for 40 seconds . Finally , the stimulus amplitude was ramped down during the final ten seconds . Trials were separated by a minimum break of 2 minutes . Sefati et al . ( 2013 ) developed and tested a second-order , linear , ordinary differential equation that describes how changes in fore-aft position of the animal , y⁢ ( t ) , relate to changes in position of the nodal point , u⁢ ( t ) : ( 9 ) m⁢d2⁢yd⁢t2+b⁢d⁢yd⁢t=k⁢u⁢ ( t ) Here , m , k , and b represent mass , gain , and damping , respectively . This equation follows from Equation S13 in the Supporting Information of Sefati et al . ( 2013 ) . Note that the present paper uses slightly different nomenclature; in particular u⁢ ( t ) , y⁢ ( t ) , and b in the present paper correspond to Δ⁢L , x⁢ ( t ) , and β , respectively , in Sefati et al . ( 2013 ) . The Laplace transform provides a computationally convenient means to represent dynamics of linear , time-invariant systems such as the one in Equation 9 ( Roth et al . , 2014 ) . Taking the Laplace transform of both sides of Equation 9 , neglecting initial conditions , and algebraically simplifying , we arrive at the plant model in Equation 1: ( 10 ) Y ( s ) U ( s ) =kms2+bs⏞P ( s ) Given the feedback control topology in Figure 1 , the closed-loop dynamics relating the movement of the refuge to the movement of the fish are given in the Laplace domain by the following equation: ( 11 ) G⁢ ( s ) =P⁢ ( s ) ⁢C⁢ ( s ) 1+P⁢ ( s ) ⁢C⁢ ( s ) This equation is also shown in Equation 7 of Cowan and Fortune ( 2007 ) with a slightly different nomenclature; in particular P⁢ ( s ) and G⁢ ( s ) in the present paper correspond to G⁢ ( s ) and H⁢ ( s ) , respectively , in Cowan and Fortune ( 2007 ) . Given G⁢ ( s ) and P⁢ ( s ) , one can compute the complementary controller C⁢ ( s ) using Equation 11 as ( 12 ) C⁢ ( s ) =G⁢ ( s ) ( 1-G⁢ ( s ) ) ⁢P⁢ ( s ) . The position of the refuge and fish were tracked for each video using custom software ( Hedrick , 2008 ) . The videos were analyzed to extract 3 to 10 seconds segments , where the fish used counter propagating waves for refuge tracking . Then , the nodal point was hand clicked in these video segments: 18 , 000 nodal point measurements were made over a total of 106 segments of data . The physics-based plant model in Sefati et al . ( 2013 ) was previously validated with quasi-static open-loop experiments . Here we reconciled the physics-based plant model from Sefati et al . ( 2013 ) ( Equation 1 ) with the data that were collected in tracking experiments . For each frequency of refuge movement , M segments of nodal point data were extracted . Each segment of data consists of the following measurements: nodal point shift {u1m , u2m , … , unm} and fish position {y1m , y2m , … , ynm} as a function of time {t1m , t2m , … , tnm} , where n is the number of samples and m={1 , 2 , … , M} . We estimated the magnitude and phase of the plant model for each frequency of refuge movement . The average value of nodal point shift and fish position were computed from M data segments per fish for each frequency of refuge movement . We aligned each data segment based on the phase of refuge signals . The segments are not completely overlapping: we selected the largest time window with at least 50 percent overlap of data segments . A sine wave function of the following form was fitted to the average nodal point data , ua⁢v⁢g⁢ ( t ) , and average fish position data , ya⁢v⁢g⁢ ( t ) , as ( 13 ) uavg ( t ) =Ausin⁡ ( 2πfit+ϕu ) +Bu , ( 14 ) yavg ( t ) =Aysin⁡ ( 2πfit+ϕy ) +By , where input–output pairs ( Au , Ay ) , ( ϕu , ϕy ) and ( Bu , By ) correspond to magnitudes , phases and DC offsets in polar coordinates , respectively . Note that this fitting was done separately for each refuge frequency , fi={0 . 55 , 0 . 95 , 2 . 05} Hz . After computing the magnitude and phase for both the average nodal shift and fish position , we estimated the magnitude and phase for the plant transfer function at ωi=2⁢π⁢fi as ( 15 ) |P^ ( jωi ) |=AyAu , ( 16 ) ∠P^ ( jωi ) =ϕy−ϕu , We obtained a non-parametric estimate of the plant transfer function for each frequency ωi , that is P^⁢ ( j⁢ωi ) by estimating magnitude and phases . We used P^⁢ ( j⁢ωi ) to estimate the parameters of the transfer function model given in Equation 1 . In this model , there are three unknown parameters , namely m , k , and b . However , for the fitting purposes we reduced the number of unknown parameters to two by normalizing the ‘gain’ ( k ) and ‘damping’ ( b ) by the ‘mass’ ( m ) . The normalized plant transfer function in Fourier domain takes the form ( 17 ) P⁢ ( j⁢ω ) =k/m ( j⁢ω ) 2+ ( b/m ) ⁢ ( j⁢ω ) =k/m-ω2+ ( b⁢j⁢ω/m ) . where j=-1 . For an ideal deterministic system , for each frequency P⁢ ( j⁢ωi ) =P^⁢ ( j⁢ωi ) , where P^⁢ ( j⁢ωi ) corresponds to the non-parametrically computed frequency response function . For this reason , estimates of transfer function parameters were made by minimizing a cost function using gradient descent method: ( 18 ) J⁢ ( k/m , b/m ) =∑i=13|P⁢ ( j⁢ωi ) -P^⁢ ( j⁢ωi ) |2 . We estimated behavioral variability using bootstrap estimates derived from individual experimental trials at the three test frequencies . Across all three fish , we made 37 observations of the frequency response at fi=0 . 55 Hz , 35 observations at fi=0 . 95 Hz , and 34 observations at fi=2 . 05 Hz , namely: ( 19 ) {G^1 ( j2π0 . 55 ) , G^2 ( j2π0 . 55 ) , … , G^37 ( j2π0 . 55 ) }{G^1 ( j2π0 . 95 ) , G^2 ( j2π0 . 95 ) , … , G^35 ( j2π0 . 95 ) }{G^1 ( j2π2 . 05 ) , G^2 ( j2π2 . 05 ) , … , G^34 ( j2π2 . 05 ) } To estimate the behavioral variability at frequencies that were not explicitly tested , we used a parametric approach . Specifically , we constructed N=1000 triplets by randomly selecting one frequency response function from each of the test frequencies in Equation 19 . For each of the 1000 triplets , we estimated a transfer function , Gbootstrap , i⁢ ( s ) , i=1 , … , N of the form in Equation 2 using Matlab’s transfer function estimation method ‘tfest’ . Let xi and yi be real and imaginary parts of the complex-valued frequency response function , namely Gbootstrap , i⁢ ( j⁢ω0 ) =xi+j⁢yi , where ω0 is a frequency in the range 0 . 2 Hz to 2 . 05 Hz . The covariance matrix for the estimated frequency response function in the complex domain was calculated as ( 20 ) Covω0=[σx⁢xω0σx⁢yω0σy⁢xω0σy⁢yω0 , ]where ( 21 ) σxxω0=1N−1∑i=1N ( xi−μx ) 2 , ( 22 ) σyyω0=1N−1∑i=1N ( yi−μy ) 2 , ( 23 ) σxyω0=σyxω0=1N−1∑i=1N ( xi−μx ) ( yi−μy ) . Here , μx and μy are mean values of xi and yi , ∀i∈{1 , 2 , … , N} , respectively . The final bootstrap estimate of behavioral variability was calculated as the largest singular value of the central covariance matrix , Covω0 . In addition , the range of the gain and phase of these 1000 transfer function models was plotted in Figure 5B . The variability across ‘matched’ and ‘swapped’ models was calculated for both the closed-loop transfer function G⁢ ( s ) ( Figure 4B ) and the loop gain P⁢ ( s ) ⁢C⁢ ( s ) ( Figure 4D ) . We evaluated each of the three fish-specific controllers and plant transfer functions at frequencies between 0 . 2 Hz and 2 . 05 Hz; for each frequency ω0 in this range , we have C1⁢ ( j⁢ω0 ) , C2⁢ ( j⁢ω0 ) , and C3⁢ ( j⁢ω0 ) and P1⁢ ( j⁢ω0 ) , P2⁢ ( j⁢ω0 ) , and P3⁢ ( j⁢ω0 ) . To calculate the matched closed-loop variability , we first calculated the real ( xi ) and imaginary ( yi ) parts of Ci⁢ ( j⁢ω0 ) ⁢Pi⁢ ( j⁢ω0 ) / ( 1+Ci⁢ ( j⁢ω0 ) ⁢Pi⁢ ( j⁢ω0 ) ) . Using these values , the matched closed-loop variability was calculated as the largest singular value of the central covariance matrix of these ordered pairs . The matched loop-gain variability was calculated similarly , using the real and imaginary parts of Ci⁢ ( j⁢ω0 ) ⁢Pi⁢ ( j⁢ω0 ) . For each of these calculations , N=3 , because there are three sets of matched models . The closed-loop and loop-gain swapped variability was calculated identically , except using the N=6 swapped permutations of control–plant pairs . Figure 4B , D illustrates the variability of the matched and swapped models both for closed-loop control and loop gain .

People come in different shapes and sizes , but most will perform similarly well if asked to complete a task requiring fine manual dexterity – such as holding a pen or picking up a single grape . How can different individuals , with different sized hands and muscles , produce such similar movements ? One explanation is that an individual’s brain and nervous system become precisely tuned to mechanics of the body’s muscles and skeleton . An alternative explanation is that brain and nervous system use a more “robust” control policy that can compensate for differences in the body by relying on feedback from the senses to guide the movements . To distinguish between these two explanations , Uyanik et al . turned to weakly electric freshwater fish known as glass knifefish . These fish seek refuge within root systems , reed grass and among other objects in the water . They swim backwards and forwards to stay hidden despite constantly changing currents . Each fish shuttles back and forth by moving a long ribbon-like fin on the underside of its body . Uyanik et al . measured the movements of the ribbon fin under controlled conditions in the laboratory , and then used the data to create computer models of the brain and body of each fish . The models of each fish’s brain and body were quite different . To study how the brain interacts with the body , Uyanik et al . then conducted experiments reminiscent of those described in the story of Frankenstein and transplanted the brain from each computer model into the body of different model fish . These “brain swaps” had almost no effect on the model’s simulated swimming behavior . Instead , these “Frankenfish” used sensory feedback to compensate for any mismatch between their brain and body . This suggests that , for some behaviors , an animal’s brain does not need to be precisely tuned to the specific characteristics of its body . Instead , robust control of movement relies on many seemingly redundant systems that provide sensory feedback . This has implications for the field of robotics . It further suggests that when designing robots , engineers should prioritize enabling the robots to use sensory feedback to cope with unexpected events , a well-known idea in control engineering .

2020

Variability in locomotor dynamics reveals the critical role of feedback in task control

We investigated the structural development of superficial-layers of medial entorhinal cortex and parasubiculum in rats . The grid-layout and cholinergic-innervation of calbindin-positive pyramidal-cells in layer-2 emerged around birth while reelin-positive stellate-cells were scattered throughout development . Layer-3 and parasubiculum neurons had a transient calbindin-expression , which declined with age . Early postnatally , layer-2 pyramidal but not stellate-cells co-localized with doublecortin – a marker of immature neurons – suggesting delayed functional-maturation of pyramidal-cells . Three observations indicated a dorsal-to-ventral maturation of entorhinal cortex and parasubiculum: ( i ) calbindin-expression in layer-3 neurons decreased progressively from dorsal-to-ventral , ( ii ) doublecortin in layer-2 calbindin-positive-patches disappeared dorsally before ventrally , and ( iii ) wolframin-expression emerged earlier in dorsal than ventral parasubiculum . The early appearance of calbindin-pyramidal-grid-organization in layer-2 suggests that this pattern is instructed by genetic information rather than experience . Superficial-layer-microcircuits mature earlier in dorsal entorhinal cortex , where small spatial-scales are represented . Maturation of ventral-entorhinal-microcircuits – representing larger spatial-scales – follows later around the onset of exploratory behavior . The representation of space in the rodent brain has been investigated in detail . The functional development of spatial response properties has also been investigated in the cortico-hippocampal system ( Ainge and Langston , 2012; Wills et al . , 2014 ) , with studies suggesting the early emergence of head-directional selectivity ( Tan et al . , 2015; Bjerknes et al . , 2015 ) , border representation ( Bjerknes et al . , 2014 ) and place cell firing , but a delayed maturation of grid cell discharges ( Wills et al . , 2010; Langston et al . , 2010 ) . Even though there is information on the emergence of functional spatial properties in the hippocampal formation , remarkably little is known about the structural development of the microcircuits which bring about these properties . To understand this , we investigated the development of the architecture of the medial entorhinal cortex ( MEC ) and parasubiculum ( PaS ) , two key structures in the cortico-hippocampal system . In adult animals , layer 2 of MEC contains two types of principal cells , stellate and pyramidal cells ( Alonso and Klink , 1993; Germroth et al . , 1989 ) . Stellate and pyramidal neurons are distinct in their intrinsic conductance ( Alonso and Llinás , 1989; Klink and Alonso , 1997 ) , immunoreactivity ( Varga et al . , 2010 ) , projections ( Lingenhöhl and Finch , 1991; Canto and Witter , 2012 ) and inhibitory inputs ( Varga et al . , 2010 ) . Pyramidal neurons in layer 2 of MEC can be identified by calbindin-immuno-reactivity ( Varga et al . , 2010 ) and are clustered in patches across various mammalian species ( Fujimaru and Kosaka , 1996; Ray et al . , 2014; Naumann et al . , 2016 ) , while stellate cells can be identified by reelin-immuno-reactivity ( Varga et al . , 2010 ) and a lack of structural periodicity ( Ray et al . , 2014 ) . In rodents , the grid-like arrangement of pyramidal cell patches is aligned to cholinergic inputs ( Ray et al . , 2014; Naumann et al . , 2016 ) . Functionally , about a third of all cells in layer 2 exhibit spatial tuning with grid , border , irregular and head-directional discharges being present ( Tang et al . , 2014 ) . Neurons in layer 3 of MEC are characterized by rather homogenous in vitro intrinsic and in vivo spatiotemporal properties ( Tang et al . , 2015 ) . A majority of cells exhibit a lack of spatial modulation , and the remaining are mainly dominated by irregular spatial responses ( Tang et al . , 2015 ) with a fraction also exhibiting grid , border and head-directional responses ( Boccara et al . , 2010 ) . The parasubiculum is a long and narrow structure flanking the dorsal and medial extremities of MEC ( Video 1 ) . The superficial parasubiculum , corresponding to layer 1 of MEC is divided into large clusters , while the deeper part , corresponding to layers 2 and 3 of MEC , is rather homogenous ( Tang et al . , 2016 ) . In terms of functional tuning of cells , a majority of the cells of PaS show spatially tuned responses , and include grid , border , head-directional and irregular spatial cells ( Boccara et al . , 2010; Tang et al . , 2016 ) . 10 . 7554/eLife . 13343 . 003Video 1 . Medial entorhinal cortex and parasubiculum in the rat brain . The medial entorhinal cortex and parasubiculum are situated at the posterior extremity of the rat neocortex . This schematic video illustrates the location of the medial entorhinal cortex and parasubiculum in situ , the tangential sectioning process and the layout of parasubicular patches and calbindin-patches in the medial entorhinal cortex . DOI: http://dx . doi . org/10 . 7554/eLife . 13343 . 003 Here we investigate the emergence of the periodic pyramidal-cell patch pattern in layer 2 of MEC , as well as the development of cellular markers that characterize the architecture of adult MEC and PaS . The results indicate an early emergence of pyramidal cell organization , a delayed maturation of pyramidal but not stellate cells and a dorsal-to-ventral maturation of MEC circuits . We first investigated development of brain size and thickness of layers of the MEC ( Figure 1 ) by observing rats at E18 , P0 , P4 , P8 , P12 , P16 , P20 , P24 and adults ( >P42 ) . The majority of the brain development takes place within the first few weeks postnatally ( Figure 1a ) , with the brain size increasing 1000% from 0 . 12 ± 0 . 00 g at E18 ( mean ± SD; n=3 ) to 1 . 23 ± 0 . 07 g at P12 ( n=5 ) . Subsequently , the growth plateaus to ~25% with the brain weighing 1 . 71 ± 0 . 08 g at P24 ( n=6 ) and having a weight of 2 . 11 ± 0 . 14 g in adults ( n=9 ) ( Figure 1b ) . The superficial layers ( layers 1–3 ) of the MEC ( Figure 1c ) double in thickness during this early postnatal period from 243 ± 35 μm at P0 ( mean ± SD; n=21 , 4 rats ) to 652 ± 50 μm at P12 ( n=24 , 4 rats ) . A similar increase is also observed in the deeper layers ( layers 4–6 ) from 167 ± 21 μm at P0 ( n=21 , 4 rats ) to 329 ± 54 μm at P12 ( n=24 , 4 rats ) . The overall thickness plateaus around this point to 981 ± 81 μm at P12 ( n=24 , 4 rats ) and remains at 882 ± 78 μm in adults ( n=24 , 4 rats ) ( Figure 1d ) . Proportionally , the thickness of the layers remains similar during development , with layer 2 accounting for ~20% and layers 3 and 5/6 each accounting for ~30% of the MEC . Layers 1 and 4 are the thinnest at about 10% and 5% of the total thickness respectively ( Figure 1d ) . 10 . 7554/eLife . 13343 . 004Figure 1 . Rat brain and medial entorhinal cortex laminar development . ( a ) Growth in rat brain size from E18 , P0 , P4 , P8 , P12 , P16 , P20 to adult . Brains are overlaid on a 1 cm x 1 cm grid . ( b ) Mean weight ( in grams ) of E18 ( n=3 ) , P0 ( n=6 ) , P4 ( n=5 ) , P8 ( n=5 ) , P12 ( n=5 ) , P16 ( n=5 ) , P20 ( n=5 ) , P24 ( n=6 ) and in adult ( n=9 ) rat brains . Error bars indicate SD . ( c ) Parasagittal section double stained for calbindin-immunoreactivity ( green ) and Purkinje cell protein 4 immunoreactivity ( pcp4; red ) , illustrating the superficial layers of the medial entorhinal cortex and parasubiculum . Calbindin+ neurons ( green ) are in layer 2 , pcp4+ neurons ( red ) are in layer 3 MEC . ( d ) Development of mean layer width ( in μm ) of layer 1 ( light-blue ) , layer 2 ( green ) , layer 3 ( red ) , layer 4 ( gray-blue ) and layer 5/6 ( purple ) from P0 to P24 and in adult rat medial entorhinal cortex . Scale bars 250 µm . PaS- Parasubiculum; L1- Layer 1; L2- Layer 2; L3- Layer 3; D- Dorsal; V-Ventral . DOI: http://dx . doi . org/10 . 7554/eLife . 13343 . 00410 . 7554/eLife . 13343 . 005Figure 1—source data 1 . Laminar widths ( in μm ) of the medial entorhinal cortex for P0 , P4 , P8 , P12 , P16 , P20 , P24 and adult rats . DOI: http://dx . doi . org/10 . 7554/eLife . 13343 . 005 We next investigated the microcircuit organization of superficial layers of MEC . Calbindin , a calcium binding protein , is selectively expressed in layer 2 pyramidal cells ( Varga et al . , 2010; Fujimaru and Kosaka , 1996 ) , which form a grid-like arrangement in adult animals ( Ray et al . , 2014 ) . Concurrently , reelin , an extracellular matrix protein , is selectively expressed in stellate cells in layer 2 of MEC , which are scattered throughout ( Ray et al . , 2014 ) layer 2 . To visualize the development of entorhinal microcircuits we first prepared tangential sections ( see our video animation on preparing tangential sections , Video 1 ) through layer 2 of medial entorhinal cortex and stained for calbindin-immunoreactivity . From the earliest postnatal stages , calbindin+ neurons in the MEC exhibited clustering , forming patches at P0 ( Figure 2a ) . The calbindin+ patches at P0 exhibited a grid-like ( Figure 2a , b ) regular arrangement ( Figure 2c ) , determined by spatial autocorrelation analysis and grid scores , similar to that observed in adults ( Ray et al . , 2014; Naumann et al . , 2016 , Figure 2d–f ) . Similar preparations for visualizing stellate cells by reelin-immunoreactivity ( Figure 2—figure supplement 1 ) , exhibited the presence of stellate cells in early postnatal stages ( Figure 2—figure supplement 1a , b ) and a lack of periodicity ( Figure 2—figure supplement 1c ) , similar to observations made in adults ( Ray et al . , 2014 , Figure 2—figure supplement 1d–f ) . Calbindin+ pyramidal neurons in the MEC ( Figure 2g ) also received preferential cholinergic innervation early postnatally ( Figure 2h–i ) , similar to adults ( Ray et al . , 2014; Naumann et al . , 2016 , Figure 2j–l ) . 10 . 7554/eLife . 13343 . 006Figure 2 . Adult-like grid layout and cholinergic innervation of calbindin+ pyramidal neurons in layer 2 of MEC at early postnatal stages . ( a ) Tangential sections of the MEC processed for calbindin-immunoreactivity ( green ) . Patches of calbindin+ neurons are evident already in the MEC , while the parasubicular patches at the right extremity also show calbindin-immunoreactivity in P0 rats . ( b ) Inset from ( a ) , rotated 90 degrees clockwise , for presentation . ( c ) Two-dimensional spatial autocorrelation of the MEC region shown in ( b ) showing a periodic spatial organization of calbindin+ patches . The grid score is 0 . 59 . ( d ) as ( a ) for adult animals . ( e ) Inset from ( d ) . ( f ) Two-dimensional spatial autocorrelation of the MEC region shown in ( e ) showing a periodic spatial organization of calbindin+ patches . The grid score is 1 . 18 . ( g ) Tangential section in a P4 animal processed for calbindin-immunoreactivity ( green ) . Also note the calbindin-immunoreactive parasubicular patches present in a P4 rat . ( h ) Section from ( g ) co-stained for acetylcholinesterase activity ( brown ) . ( i ) Overlay of inset regions from ( g ) and ( h ) shows overlap between calbindin and acetylcholinesterase in MEC in P4 rats . ( j–l ) as ( g–i ) for adult animals . ( d–f , j–l ) modified from Ray et al . ( 2014 ) . Colour scale of spatial autocorrelation , -0 . 5 ( blue ) through 0 ( green ) to 0 . 5 ( red ) . Scale bars 250 µm . D- Dorsal; V- Ventral; M- Medial; L- Lateral . Orientation in ( d ) applies to all sections apart from ( b ) , where it’s rotated 90 degrees clockwise . DOI: http://dx . doi . org/10 . 7554/eLife . 13343 . 006© 2014 The American Association for the Advancement of Science2014The American Association for the Advancement of ScienceFigure 2 , panels d-f and j-l are adapted from Ray S , Naumann R , Burgalossi A , Tang Q , Schmidt H , Brecht M . 2014 . Grid-layout and theta-modulation of layer 2 pyramidal neurons in medial entorhinal cortex . Science 343:891–896 . doi:10 . 1126/science . 1243028 . Reprinted with permission from AAAS . 10 . 7554/eLife . 13343 . 007Figure 2—figure supplement 1 . Adult-like scattered distribution of reelin+ stellate cells in early postnatal stages . ( a ) Tangential sections of the MEC processed for reelin-immunoreactivity ( red ) in a P4 rat . ( b ) Inset from ( a ) . ( c ) Two-dimensional spatial autocorrelation of the MEC region shown in ( b ) showing a lack of periodicity of reelin+ neurons . The grid score is -0 . 09 . ( d–f ) as ( a–c ) for adult animals . The grid score in ( f ) is 0 . 03 . Scale bars 250 µm . D- Dorsal; V- Ventral; M- Medial; L- Lateral . Orientation in ( d ) applies to all sections . DOI: http://dx . doi . org/10 . 7554/eLife . 13343 . 007 In the parasubiculum , a transient presence of calbindin was observed with ~15 clusters of calbindin+ neurons at P0 ( Figure 2a ) and P4 ( Figure 2g ) . This expression was curtailed in older animals , with only a calbindin+ stripe persisting in adults ( Figure 2d ) . To visualize the laminar development of MEC , we stained parasagittal sections for calbindin ( Figure 3 ) and reelin ( Figure 4 ) immunoreactivity . Indications of calbindin+ neuronal clusters were visible prenatally at E18 ( Figure 3a ) . However , the calbindin+ patches in the MEC did not exhibit clustering of their dendrites , as previously described in adults ( Ray et al . , 2014 ) at E18 and P0 ( Figure 3a , b ) . Some dendritic clustering could be observed at P4 ( Figure 3c ) , while from P8 ( Figure 3d–h ) the dendritic clustering of calbindin+ pyramidal neurons was similar to that in adults . In layer 3 of the MEC , we observed a transient presence of calbindin expression . The number of calbindin+ neurons in layer 3 declined progressively from prenatal stages to P20 ( Figure 3a–g ) , where it attained adult-like levels with rarely any calbindin+ neurons in layer 3 ( Figure 3h ) . Quantitatively , calbindin+ neuronal density ( calbindin+ neurons per mm2 ) decreased from 955 ± 315 ( mean ± SD; count refers to n=3776 neurons in 8 rats ) in P4-P8 rats to 333 ± 99 ( n=2104 neurons , 8 rats ) in P12-P16 rats to 141 ± 56 ( n=828 neurons , 7 rats ) in adults ( Figure 3i ) . 10 . 7554/eLife . 13343 . 008Figure 3 . Transient presence of calbindin+ neurons in layer 3 of MEC in early postnatal stages reduces progressively to adult-like state by third postnatal week . Parasaggital sections of the MEC processed for calbindin-immunoreactivity ( green ) . The sections show clustering of calbindin+ pyramidal cells in layer 2 and a transient presence of calbindin+ neurons in layer 3 , which decrease with age in ( a ) E18 rat . ( b ) P0 rat . ( c ) P4 rat . ( d ) P8 rat . ( e ) P12 rat . ( f ) P16 rat . ( g ) P20 rat . ( h ) Adult rat . ( i ) Decreasing density of calbindin+ neurons in layer 3 of MEC from P4-P8 ( n=3776 neurons , 8 rats ) ; to P12-P16 ( n=2104 neurons , 8 rats ) to adults ( n=828 neurons , 7 rats ) . Error bars denote SD . Scale bars 250 µm . D- Dorsal; V- Ventral . Orientation in ( h ) applies to all sections . DOI: http://dx . doi . org/10 . 7554/eLife . 13343 . 00810 . 7554/eLife . 13343 . 009Figure 3—source data 1 . Calbindin+ neurons counted and areas ( in µm2 ) in layer 3 for determining calbindin+ neuronal density in layer 3 in P4-P8 , P12-P16 and adult rats . DOI: http://dx . doi . org/10 . 7554/eLife . 13343 . 00910 . 7554/eLife . 13343 . 010Figure 4 . Increase of reelin expression in layer 3 neurons of MEC through development . Parasaggital sections of the MEC processed for reelin-immunoreactivity ( red ) . The sections show reelin+ stellate cells in layer 2 and an increasing reelin expression in layer 3 neurons with development in ( a ) P4 rat . ( b ) P8 rat . ( c ) P12 rat . ( d ) P16 rat . ( e ) P20 rat . ( f ) Adult rat . ( g ) Increasing density of reelin+ neurons in layer 3 of MEC from P4-P8 ( n=1405 neurons , 4 rats ) ; to P12-P16 ( n=3309 neurons , 3 rats ) to adults ( n=5039 neurons , 3 rats ) . Error bars denote SD . Scale bars 250 µm . D- Dorsal; V- Ventral . Orientation in ( f ) applies to all sections . DOI: http://dx . doi . org/10 . 7554/eLife . 13343 . 01010 . 7554/eLife . 13343 . 011Figure 4—source data 1 . Reelin+ neurons counted and areas ( in µm2 ) in layer 3for determining reelin+ neuronal density in layer 3 in P4-P8 , P12-P16 and adult rats . DOI: http://dx . doi . org/10 . 7554/eLife . 13343 . 011 Reelin was present in layer 2 from early postnatal stages ( Figure 4a; Figure 2—figure supplement 1a , b ) , though the most prominent reelin-immunoreactive cells in the first two postnatal weeks were present in layer 1 ( Figure 4a–c ) . Reelin expression increased in layer 3 of the MEC from early postnatal stages to P20 ( Figure 4a–e ) , where it attained adult-like levels ( Figure 4f ) . Quantitatively , reelin+ neuronal density in layer 3 increased from 729 ± 435 ( n=1405 neurons , 4 rats ) in P4-P8 rats to 1549 ± 115 ( n=3309 neurons , 3 rats ) in P12-P16 rats to 1996 ± 208 ( n=5039 neurons , 3 rats ) in adults . Three observations indicated a dorsal-ventral developmental gradient in the superficial layers of medial entorhinal cortex and parasubiculum: First , the transient calbindin expression in layer 3 disappeared from dorsal to ventral . Thus , most of layer 3 had calbindin+ neurons at P8 ( Figure 5a ) , only the ventral half of layer 3 showed calbindin expression at P16 ( Figure 5b ) , and in adults calbindin expression was largely absent from layer 3 of MEC ( Figure 5c ) . This transient expression of calbindin in layer 3 followed a dorso-ventral developmental profile ( Figure 5d ) . Early postnatally , in P4-P8 rats , we observed equitable densities of calbindin+ cells in dorsal , intermediate and ventral levels of MEC ( n=3776 neurons , 8 rats ) . In contrast , around the end of the second postnatal week , in P12-P16 rats , we observed significantly lower densities ( p=0 . 010 , Mann-Whitney two tailed ) in the dorsal ( 225 ± 96 cells / mm2 ) , as opposed to the ventral ( 449 ± 161 cells / mm2 ) MEC ( n=2104 neurons , 8 rats ) . In adults ( n=828 neurons , 7 rats ) , calbindin+ neurons were largely absent in layer 3 , but among the remaining population the density waxed from dorsal to intermediate and ventral MEC . The development of reelin expression in layer 3 neurons on the other hand ( Figure 5—figure supplement 1a–c ) occurred in equitable proportions in dorsal , intermediate and ventral levels of MEC ( Figure 5—figure supplement 1d ) with increasing age . 10 . 7554/eLife . 13343 . 012Figure 5 . Dorsal-to-ventral disappearance of layer 3 calbindin expression . Parasaggital sections showing superficial layers of the MEC processed for calbindin-immunoreactivity ( green ) . ( a ) Calbindin expression is seen throughout layer 3 in P8 rats . ( b ) Calbindin expression is seen only in ventral half of layer 3 in P16 rats . ( c ) Calbindin expression is largely absent in layer 3 in adult rats . ( d ) Proportion of layer 3 calbindin+ neurons in dorsal ( white ) , intermediate ( gray ) and ventral ( black ) MEC in P4-P8 ( n=3776 neurons , 8 rats ) ; P12-P16 ( n =2014 neurons , 8 rats ) ; and adult ( n=828 neurons , 7 rats ) rats . The numbers represent layer 3 calbindin+ neuronal density and decay in a dorsal to ventral gradient with age as evident with the reduced proportions of the white ( dorsal MEC ) and gray ( intermediate MEC ) sections of the columns with increasing age . Scale bars 250 µm . L1- Layer 1; L2- Layer 2; L3- Layer 3; D- Dorsal; V-Ventral . Orientation in ( c ) applies to all sections . DOI: http://dx . doi . org/10 . 7554/eLife . 13343 . 01210 . 7554/eLife . 13343 . 013Figure 5—figure supplement 1 . Dorsal- ventral distribution of layer 3 reelin expression . Parasaggital sections showing superficial layers of the MEC processed for reelin-immunoreactivity ( red ) . ( a ) Reelin expression is sporadic throughout layer 3 in P8 rats . ( b ) Reelin expression equitably increases in layer 3 in P16 rats . ( c ) Reelin expression is present throughout layer 3 in adult rats . ( d ) Proportion of layer 3 reelin+ neurons in dorsal ( white ) , intermediate ( gray ) and ventral ( black ) MEC in P4-P8 ( n=1405 neurons , 4 rats ) ; P12-P16 ( n =3309 neurons , 3 rats ) ; and adult ( n=5039 neurons , 3 rats ) rats . The numbers represent layer 3 reelin+ neuronal density and increase equitably with age as evident with the similar proportions of the white ( dorsal MEC ) , gray ( intermediate MEC ) and black ( ventral MEC ) sections of the columns with increasing age . Scale bars 250 µm . L1- Layer 1; L2- Layer 2; L3- Layer 3; D- Dorsal; V-Ventral . Orientation in ( c ) applies to all sections . DOI: http://dx . doi . org/10 . 7554/eLife . 13343 . 01310 . 7554/eLife . 13343 . 014Figure 5—figure supplement 1—source data 1 . Calbindin+ neurons ( Figure 5 ) and reelin+ neurons ( Figure 5—figure supplement 1 ) counted and areas ( in μm2 ) in dorsal , intermediate and ventral parts of layer 3 for determining calbindin+ and reelin+ neuronal densities respectively in P4-P8 , P12-P16 and adult rats . DOI: http://dx . doi . org/10 . 7554/eLife . 13343 . 014 Second , layer 2 calbindin+ patches in the MEC also exhibited a dorsal-to-ventral maturation profile . The calbindin+ patches ( Figure 6a ) co-localized with doublecortin ( Figure 6b ) , a well-established marker for immature neurons ( Brown et al . , 2003 ) throughout layer 2 at P8 ( Figure 6c , d ) . At P16 , the dorsal calbindin+ patches ( Figure 6e , g ) did not express doublecortin ( Figure 6f , g ) , while ventral calbindin+ patches still co-localized with doublecortin ( Figure 6h ) . In adults , calbindin+ patches ( Figure 6i ) did not exhibit doublecortin ( Figure 6j ) in either dorsal ( Figure 6k ) or ventral ( Figure 6l ) parts . A similar dorsal-to-ventral development gradient was evident in the PaS , with doublecortin being present throughout the PaS in P8 ( Figure 6b ) , only in the ventral part in P16 ( Figure 6f ) and not present in adults ( Figure 6j ) . To quantify the overlap between calbindin and doublecortin we performed spatial cross-correlations ( Figure 6m ) . P8-P12 rats exhibited a high degree of overlap between calbindin and doublecortin in both dorsal ( 0 . 74 ± 0 . 05; mean ± SD , Pearson’s cross-correlation coefficient ) and ventral ( 0 . 61 ± 0 . 10 ) parts ( n=9 regions , 5 rats ) . In P16-P20 rats ( n=16 regions , 8 rats ) , the dorsal regions showed low correlations ( 0 . 14 ± 0 . 17 ) , while the ventral regions still showed significantly higher overlap ( 0 . 60 ± 0 . 07; p=0 . 0008 , Mann-Whitney two tailed ) . In adults ( n=7 regions , 4 rats ) , both dorsal ( 0 . 19 ± 0 . 07 ) and ventral ( 0 . 20 ± 0 . 07 ) regions had low overlap . The difference in the Pearson’s cross correlation coefficient between overlapping regions ( dorsal and ventral in P8-P12; ventral in P16-P20 ) and non-overlapping regions ( dorsal in P16-P20; dorsal and ventral in adults ) was significant at p=0 . 000001 ( Mann-Whitney two tailed ) . 10 . 7554/eLife . 13343 . 015Figure 6 . Dorsal-to-ventral maturation of layer 2 calbindin+ patches and parasubiculum . Tangential sections of the MEC double-stained for calbindin immunoreactivity ( green ) and doublecortin immunoreactivity ( red ) . Doublecortin is a marker for immature neurons and disappears in a dorsal-ventral gradient . ( a ) Calbindin-expression ( green ) in P8 rats . ( b ) Doublecortin-expression ( red ) in P8 rats . Note the presence of doublecortin throughout the dorso-ventral extent of MEC and parasubiculum . ( c ) Overlay of the dorsal inset region ( dashed ) in ( a ) and ( b ) , showing overlap of calbindin and doublecortin ( hence the yellowish color ) . ( d ) Overlay of the ventral inset region ( dashed ) in ( a ) and ( b ) , showing overlap of calbindin and doublecortin . ( e–h ) as ( a–d ) for P16 rats , respectively . However , note that dorsal inset region lacks doublecortin ( g ) while ventral inset region shows overlap of calbindin and doublecortin ( h ) . Also , note the absence of doublecortin in the dorsal but not the ventral parasubiculum ( f ) . ( i–k ) as ( a–d ) for adult rats . No doublecortin is present in either dorsal ( k ) or ventral ( l ) regions . ( m ) Spatial cross-correlations of calbindin and doublecortin in MEC showing high overlap in both dorsal and ventral regions in P8-P12 rats ( dark green; n=9 regions , 5 rats ) ; low correlation in dorsal but high overlap in ventral in P16-P20 rats ( green; n=16 regions , 8 rats ) and low correlations in both dorsal and ventral in adult rats ( light green; n=7 regions , 4 rats ) . The Pearson’s cross-correlation coefficient can vary from -1 ( anti-correlated ) through 0 ( un-correlated ) to 1 ( correlated ) . Scale bars 250 µm . D- Dorsal; V- Ventral; M- Medial; L- Lateral . Orientation in ( i ) applies to all sections . DOI: http://dx . doi . org/10 . 7554/eLife . 13343 . 015 A closer analysis of the co-localization of the immature neuronal marker doublecortin with calbindin+ pyramidal cells and reelin+ stellate cells ( Figure 7a–c ) revealed doublecortin to be mostly co-localized with calbindin+ rather than reelin+ neurons ( Figure 7d ) . Spatial cross-correlations between doublecortin and either calbindin or reelin ( Figure 7e; n=8 rats from ages P8 - P20 ) from triple-immunostained calbindin , reelin and doublecortin regions of layer 2 of the MEC revealed a greater overlap of doublecortin with calbindin ( 0 . 54 ± 0 . 10 ) than with reelin ( 0 . 08 ± 0 . 13 ) . This difference in the Pearson’s cross correlation coefficient was significant at p=0 . 0009 ( Mann-Whitney two tailed ) . 10 . 7554/eLife . 13343 . 016Figure 7 . Higher co-localization of doublecortin with calbindin+ pyramidal than reelin+ stellate cells in the developing medial entorhinal cortex . Tangential sections of the MEC layer 2 triple-stained for calbindin immunoreactivity ( CB; blue ) , doublecortin immunoreactivity ( DCX; red ) and reelin immunoreactivity ( RL; green ) . Pyramidal but not stellate cells are structurally immature during early postnatal stages . ( a ) Calbindin-expression ( blue ) in layer 2 of MEC . ( b ) Doublecortin-expression ( red ) in layer 2 of MEC . ( c ) Reelin-expression ( green ) in layer 2 of MEC . ( d ) Overlay of the inset region ( dashed ) in ( a ) , ( b ) and ( c ) , showing a higher co-localization of doublecortin ( red ) with calbindin ( blue ) , than reelin ( green ) . ( e ) Spatial cross-correlations of doublecortin with calbindin and reelin showing high overlap of doublecortin with calbindin but not reelin ( n=8 regions , 8 rats ) . Scale bars ( a–c ) 250 µm; ( d ) 100 µm . D- Dorsal; V- Ventral; M- Medial; L- Lateral . Orientation in ( c ) applies to all sections . DOI: http://dx . doi . org/10 . 7554/eLife . 13343 . 016 Third , wolframin expression , a marker which co-localizes with calbindin+ pyramidal neurons in layer 2 of MEC in adult rodents ( Kitamura et al . , 2014 ) , develops from dorsal to ventral in layer 2 medial entorhinal cortex and parasubiculum ( Figure 8 ) . Specifically , wolframin expression starts to appear in the dorsal MEC and the dorsal PaS shortly after birth ( Figure 8a ) and is present only in the dorsal ~10% of the PaS . It extends progressively more ventrally ( Figure 8b ) and covers ~40% at P8 and ~75% at P12 of PaS . At P20 it is expressed throughout the full extent of medial entorhinal cortex and the parasubiculum ( Figure 8c ) . 10 . 7554/eLife . 13343 . 017Figure 8 . Dorsal-to-ventral maturation of wolframin expression in the medial entorhinal cortex and parasubiculum . ( a ) Tangential sections of the MEC and PaS ( outlines dashed ) double-stained for calbindin-immunoreactivity ( green ) and wolframin immunoreactivity ( red ) in a P4 rat . Shown is an overlay of red and green fluorescence . ( b ) as ( a ) for a P8 rat . ( c ) as ( a ) for a P20 rat . Wolframin is present in the dorsal ~10% of the parasubiculum at P4 , ~40% at P8 and 100% at P20 . Note that wolframin expression co-localizes with calbindin-expression in the MEC ( hence the yellowish color ) and increases from dorsal to ventral with age . Scale bars 250 µm . D- Dorsal; V- Ventral; M- Medial; L- Lateral . Orientation in ( a ) applies to all sections . DOI: http://dx . doi . org/10 . 7554/eLife . 13343 . 017 Neurogenesis in the medial entorhinal cortex is completed prior to E18 ( Bayer , 1980a; 1980b ) , and at this time the basic laminar organization of medial entorhinal cortex is already evident . While the basic structure of medial entorhinal cortex appears early , we observe massive developmental changes in the cortical structure , including a doubling of the thickness of the superficial layers during the first postnatal week . The clustering of layer 2 MEC calbindin+ neurons into patches is also an early developmental event , and key aspects of the grid-layout of calbindin+ neurons are already present at birth . This observation indicates that the periodic structure of patches is a result of genetic signaling rather than spatial experience . Periodic patterns are ubiquitous in nature , and several chemical patterning systems have been explained on the basis of interaction between dynamical systems ( Turing , 1952 ) . Since it has been suggested that the grid layout of calbindin+ neurons is functionally relevant for grid cell activity ( Brecht , 2014 ) , it would be interesting to investigate , whether genetic manipulations would result in changes of layout periodicity and have functional effects . The dendritic clustering of calbindin+ pyramidal neurons is similar to dendritic development in the neocortex ( Petit et al . , 1988 ) and is established by the end of the first postnatal week . The cholinergic innervation of the calbindin+ patches was present by P4 in line with other long-range connectivity patterns in the MEC ( O’Reilly et al . , 2015 ) , which are also established early in development . Reelin is an important protein in cortical layer development ( D'Arcangelo et al . , 1995 ) and in the early stages of postnatal development we see the strongest reelin expression in layer 1 , where reelin secreting Cajal-Retzius cells are involved in radial neuronal migration ( Pesold et al . , 1998 ) . Stellate cells in layer 2 of MEC , which can be visualized by reelin-immunoreactivity ( Varga et al . , 2010 ) , were scattered ( Ray et al . , 2014 ) throughout postnatal development . Layer 3 of the MEC features a complementary transition of calbindin+ and reelin+ neurons during the first couple of postnatal weeks . While the density of reelin+ neurons increases , there is a concurrent decline in calbindin+ neuronal density in layer 3 of MEC , though part of the calbindin+ neuronal density decline can be attributed to the increasing brain size . Taken together with the presence of radial neuronal migration promoting Cajal-Retzius cells in layer 1 during this period , it would be interesting to investigate whether the transient calbindin+ neurons are migrating to layer 2 or changing their phenotype to reelin+ neurons , and what layer and cell-type specific functional differences are observed in this early postnatal development stage . An interesting observation is the presence of clusters of neurons in the parasubiculum , which transiently express calbindin in early postnatal stages , and subsequently express wolframin . Transient expression of calbindin has been observed in early postnatal development in the neocortex ( Hogan and Berman , 1993 ) and midbrain regions ( Liu and Graybiel , 1992 ) , but its functional significance remains largely unknown . Our data show , however , that at early developmental stages the parasubiculum and medial entorhinal cortex share a similar organization in calbindin+ patches . Additionally , the expression of wolframin in the parasubiculum persists in adults , while calbindin+ neurons in MEC layer 2 also exhibit wolframin ( Kitamura et al . , 2014 ) from the end of the first postnatal week . Current studies generally focus on cell-type specific investigations using proteins expressed by these cells . However , investigations to study the specific roles of these proteins ( Li et al . , 1995 ) might provide interesting insights towards understanding the finer differences in the functionalities exhibited by these cells . For instance , calbindin is a calcium buffer , and reduces the concentration of intracellular calcium ( Mattson et al . , 1991 ) , while wolframin is implicated in increasing intracellular calcium levels ( Osman et al . , 2003 ) . With the medial entorhinal cortex and parasubiculum having many similarities in their spatial discharge properties ( Tang et al . , 2014; Boccara et al . , 2010; Tang et al . , 2016 ) , a structure-function comparison of the wolframin+/transiently-calbindin+ neurons in the parasubiculum and the wolframin+/ permanently-calbindin+ neurons in the medial entorhinal cortex would be worthwhile . A dorsal-to-ventral development profile was observed in the superficial layers of the MEC and parasubiculum . This conclusion was suggested by the progressive disappearance of the calbindin expression in layer 3 from dorsal to ventral; the progressive disappearance of doublecortin expression in layer 2 and parasubiculum from dorsal to ventral; and the progressive appearance of the wolframin expression in superficial layer 2 of MEC and parasubiculum from dorsal to ventral . Homing behavior in rats , as well as spontaneous exploratory behavior develops around the end of second postnatal week ( Wills et al . , 2014; Bulut and Altman , 1974 ) while spontaneous exploration of larger environments outside the nest emerge towards the end of the third postnatal week ( Wills et al . , 2014 ) . This is coincident with the timeline of maturation of calbindin+ patches in the dorsal and ventral MEC respectively . Since the dorsal MEC represents smaller spatial scales and the ventral MEC progressively larger scales ( Hafting et al . , 2005; Stensola et al . , 2012 ) , these data may indicate that the rat’s navigational system matures from small to large scales . Early eyelid opening experiments have indicated an accelerated development of spatial exploratory behaviour ( Kenny and Turkewitz , 1986; Foreman and Altaha , 1991 ) , and similar experiments might provide insights into whether early behavioral development is accompanied by an accelerated development of the microcircuit underlying spatial navigation . The higher co-localization of doublecortin with calbindin+ pyramidal cells than reelin+ stellate cells , supports further the dichotomy of structure-function relationships exhibited by these two cell types ( Ray et al . , 2014; Tang et al . , 2014 ) . Grid and border cells have been implicated to be largely specific to pyramidal and stellate cells respectively ( Tang et al . , 2014 ) and the delayed structural maturation of pyramidal cells might reflect the delayed functional maturation of grid cells ( Wills et al . , 2010; Langston et al . , 2010 ) , with the converse being applicable to stellate and border cells ( Bjerknes et al . , 2014 ) . The divergent projection patterns of pyramidal and stellate cells , with the former projecting to CA1 ( Kitamura et al . , 2014 ) and contralateral MEC ( Varga et al . , 2010 ) and the latter to dentate gyrus ( Varga et al . , 2010; Ray et al . , 2014 ) and deep layers of MEC ( Sürmeli et al . , 2015 ) , have differing theoretical interpretations in spatial information processing . The same sets of neurons , which correspond to grid and border cells ( Tang et al . , 2014 ) , have also been implicated to be differentially involved in temporal association memory ( Kitamura et al . , 2014 ) and contextual memory ( Kitamura et al . , 2015 ) respectively . An underlying differential structural maturation timeline of the microcircuit governing these processes may also translate into a differential functional maturation profile of these memories . We conclude that the structural maturation of medial entorhinal cortex can be coarsely divided into an early appearance of the calbindin+ neuron patches and a progressive cell-type specific refinement of the cellular structure , which proceeds along the dorsal to ventral axis . Male and female Wistar rats ( n=83 ) from E18 to P24 and adults ( >P42 ) were used in the study . The ages were accurate to ± 1 day . Animals were anaesthetized by isoflurane , and then euthanized by an intraperitoneal injection of 20% urethane . They were then perfused transcardially with first 0 . 9% phosphate buffered saline solution , followed by 4% formaldehyde , from paraformaldehyde , in 0 . 1 M phosphate buffer ( PFA ) . For prenatal animals , pregnant rats at E18 were perfused in the aforesaid manner and the E18 animals were then extracted from the uterus . Subsequently , brains were removed from the skull and postfixed in PFA overnight . Brains were then transferred to 10% sucrose solution for one night and subsequently immersed in 30% sucrose solution for at least one night for cryoprotection . The brains were embedded in Jung Tissue Freezing Medium ( Leica Microsystems Nussloch , Germany ) , and subsequently mounted on the freezing microtome ( Leica 2035 Biocut ) to obtain 60 μm thick sagittal sections or tangential sections parallel to the pia . Tangential sections of the medial entorhinal cortex were obtained by separating the entorhinal cortex from the remaining hemisphere by a cut parallel to the surface of the medial entorhinal cortex ( Video 1 ) . For subsequent sectioning the surface of the entorhinal cortex was attached to the block face of the microtome . Acetylcholinesterase ( AChE ) activity was visualized according to previously published procedures ( Ichinohe et al . , 2008; Tsuji , 1998 ) . After washing brain sections in a solution containing 1 ml of 0 . 1 M citrate buffer ( pH 6 . 2 ) and 9 ml 0 . 9% NaCl saline solution ( CS ) , sections were incubated with CS containing 3 mM CuSO4 , 0 . 5 mMK3Fe ( CN ) 6 , and 1 . 8 mM acetylthiocholine iodide for 30 min . After rinsing in PB , reaction products were visualized by incubating the sections in PB containing 0 . 05% 3 , 3’- Diaminobenzidine ( DAB ) and 0 . 03% nickel ammonium sulfate . Immunohistochemical stainings were performed according to standard procedures . Briefly , brain sections were pre-incubated in a blocking solution containing 0 . 1 M PBS , 2% Bovine Serum Albumin ( BSA ) and 0 . 5% Triton X-100 ( PBS-X ) for an hour at room temperature ( RT ) . Following this , primary antibodies were diluted in a solution containing PBS-X and 1% BSA . Primary antibodies against the calcium binding protein Calbindin ( Swant: CB300 , CB 38; 1:5000 ) , the extracellular matrix protein Reelin ( Millipore: MAB5364; 1:1000 ) , the transmembrane protein Wolframin ( Proteintech: 11558-1-AP; 1:200 ) , the microtubule associated protein Doublecortin ( Santa Cruz Biotechnology: sc-8086; 1:200 ) and the calmodulin binding protein Purkinje cell protein 4 ( Sigma: HPA005792; 1:200 ) were used . Incubations with primary antibodies were allowed to proceed for at least 24 hr under mild shaking at 4°C in free-floating sections . Incubations with primary antibodies were followed by detection with secondary antibodies coupled to different fluorophores ( Alexa 488 , 546 and 633; Invitrogen ) . Secondary antibodies were diluted ( 1:500 ) in PBS-X and the reaction allowed to proceed for two hours in the dark at RT . For multiple antibody labeling , antibodies raised in different host species were used . For visualizing cell nuclei , sections were counterstained with DAPI ( Molecular Probes: R37606 ) . After the staining procedure , sections were mounted on gelatin coated glass slides with Vectashield mounting medium ( Vectorlabs: H-1000 ) . An Olympus BX51 microscope ( Olympus , Shinjuku Tokyo , Japan ) equipped with a motorized stage ( LUDL Electronics , Hawthorne NY ) and a z-encoder ( Heidenhain , Shaumburg IL , USA ) , was used for bright field microscopy . Images were captured using a MBF CX9000 ( Optronics , Goleta CA ) camera using Neurolucida or StereoInvestigator ( MBF Bioscience , Williston VT , USA ) . A Leica DM5500B epifluorescence microscope with a Leica DFC345 FX camera ( Leica Microsystems , Mannheim , Germany ) was used to image the immunofluorescent sections . Alexa fluorophores were excited using the appropriate filters ( Alexa 350 – A4 , Alexa 488 – L5 , Alexa 546 – N3 , Alexa 633 – Y5 ) . Fluorescent images were acquired in monochrome , and color maps were applied to the images post acquisition . Post hoc linear brightness and contrast adjustment were applied uniformly to the image under analysis . To determine the width of different layers of the medial entorhinal cortex , we prepared parasagittal sections and stained them for calbindin-immunoreactivity , Purkinje cell protein-immunoreactivity and DAPI . Measurements were taken from dorsal , medial and ventral parts of each section analyzed using Leica Application Suite AF ( Leica Microsystems , Mannheim , Germany ) . To determine the spatial periodicity of calbindin+ patches , we determined spatial autocorrelations . The spatial autocorrelogram was based on Pearson’s product moment correlation coefficient ( Sargolini et al . , 2006 ) . rτx , τy=n∑​f ( x , y ) fx−τx , y−τy−∑​fx , y∑​fx−τx , y−τyn∑​f ( x , y ) 2−∑​f ( x , y ) 2n∑​f ( x−τx , y−τy ) 2−∑​f ( x−τx , y−τy ) 2 where , r ( τx , τy ) is the autocorrelation between pixels or bins with spatial offset τx and τy . f is the monochromatic image without smoothing , n is the number of overlapping pixels . Autocorrelations were not estimated for lags of τx and τy , where n<20 . Grid scores were calculated , as previously described ( Ray et al . , 2014 ) , and can vary from −2 to 2 . To determine the degree of overlap between doublecortin and calbindin or reelin , we determined spatial crosscorrelations . Spatial crosscorrelations were determined based on Pearson’s product moment correlation coefficient . r=n∑​f1 ( x , y ) f2 ( x , y ) −∑​f1x , y∑​f2 ( x , y ) n∑​f1 ( x , y ) 2−∑​f1 ( x , y ) 2n∑​f2 ( x , y ) 2−∑​f2 ( x , y ) 2 where , r is the cross-correlation between the monochromatic images f1 and f2 without smoothing . n is the number of pixels in the image . The Pearson’s cross-correlation coefficient can vary from -1 ( anti-correlated ) through 0 ( un-correlated ) to 1 ( correlated ) . For analysis of dorso-ventral variation in overlap between doublecortin with calbindin , two regions of the same size were selected from a section double-stained for calbindin and doublecortin . One region was selected from the dorsal half of the section and another from the ventral half and the regions were represented as pairs . Where , due to section damage , it was not possible to obtain regions from both dorsal and ventral parts , the data was presented as unpaired . For analysis of variation in overlap between doublecortin and calbindin/reelin , comparisons were performed between the same regions from a section triple stained for calbindin , reelin and doublecortin .

Many animals , from rats to humans , need to navigate their environments to find food or shelter . This ability relies on a kind of memory known as spatial memory , which provides a map of the outside world within the animal’s brain . Specifically , cells in a part of the brain called the medial entorhinal cortex act like the grids present on a map , and are known as grid cells . Other cells in this region represent boundaries in the environment and are known as border cells . These cells and other cells connect to each other to make the spatial memory circuit . Previous research had reported that the grid cells were not present in the very early stages of an animal’s life . It was also not clear how the different cell types involved in spatial memory develop after birth . Ray and Brecht have now studied rats and found that certain characteristic structures in the circuit are present at birth . For example , cells that were most likely to become grid cells , were already laid out in a grid , indicating that this layout is instructed by genetic information rather than experience . Ray and Brecht also found that the cells that most likely become grid cells matured later than the cells that most likely become border cells . Further analysis then revealed that the circuits in the top part of the medial entorhinal cortex , which represents nearby areas , matured earlier than those in the bottom part of this region , which represent farther areas . These findings could therefore explain why rats explore nearby areas earlier in life before going on to explore further away areas at later stages . More work is needed to characterize other components of the neural circuits involved in spatial memory to provide a complete understanding of how these memories are formed . Future experiments could also ask if encouraging young rats to explore a wider area can cause the circuits to mature more quickly .

2016

Structural development and dorsoventral maturation of the medial entorhinal cortex

Phosphorylation of eIF2α controls translation initiation by restricting the levels of active eIF2-GTP/Met-tRNAi ternary complexes ( TC ) . This modulates the expression of all eukaryotic mRNAs and contributes to the cellular integrated stress response . Key to controlling the activity of eIF2 are translation factors eIF2B and eIF5 , thought to primarily function with eIF2-GDP and TC respectively . Using a steady-state kinetics approach with purified proteins we demonstrate that eIF2B binds to eIF2 with equal affinity irrespective of the presence or absence of competing guanine nucleotides . We show that eIF2B can compete with Met-tRNAi for eIF2-GTP and can destabilize TC . When TC is formed with unphosphorylated eIF2 , eIF5 can out-compete eIF2B to stabilize TC/eIF5 complexes . However when TC/eIF5 is formed with phosphorylated eIF2 , eIF2B outcompetes eIF5 and destabilizes TC . These data uncover competition between eIF2B and eIF5 for TC and identify that phosphorylated eIF2-GTP translation initiation intermediate complexes can be inhibited by eIF2B . In eukaryotic translation initiation , initiator tRNA ( Met-tRNAi ) recognises AUG start codons in mRNA as part of a larger complex bound to the 40 S ribosome . Eukaryotic translation initiation factor 2 ( eIF2 ) is the key factor that delivers Met-tRNAi to ribosomes ( Hinnebusch , 2014; Dever et al . , 2016 ) . eIF2 is a G-protein and binds GTP and Met-tRNAi forming a ternary complex ( TC ) ( Schmitt et al . , 2010 ) . Together with other initiation factors the TC-bound ribosomal preinitiation complex ( PIC ) binds near the mRNA 5’ cap and scans the mRNA usually to the first AUG codon where Met-tRNAi anticodon/mRNA codon interactions help facilitate translation initiation . Hence TC is critical for delivering tRNAiMet to 40 S ribosomes , for scanning the 5’UTR and AUG codon recognition ( Hinnebusch , 2014; Dever et al . , 2016; Llácer et al . , 2015 ) . Importantly , in all eukaryotes studied , translation initiation is controlled by regulating the activity of eIF2 , see below . Met-tRNAi has a ~10 fold greater affinity for eIF2-GTP over eIF2-GDP ( Kapp and Lorsch , 2004 ) . Hence hydrolysis of eIF2-bound GTP and Pi release during AUG codon recognition , facilitates loss of eIF2-GDP from initiating ribosomes ( Algire et al . , 2005 ) . This eIF2-GDP must be converted to an active GTP-bound form for continued active translation initiation . Reactivation of eIF2 relies on eIF2B , a multifunctional guanine nucleotide exchange factor ( GEF ) ( Pavitt , 2005 ) . eIF2B GEF activity facilitates GDP release from eIF2-GDP complexes ( Panniers et al . , 1988; Pavitt et al . , 1998 ) enabling GTP and Met-tRNAi binding to eIF2 . However , the precise mechanism of eIF2B-mediated GEF action is not yet understood ( Mohammad-Qureshi et al . , 2008 ) , see Figure 1A . One highly conserved form of translational control involves signalling to and activation of eIF2 kinases that each phosphorylate serine 51 of the eIF2α subunit , for example during periods of cell stress ( Pavitt , 2005 ) and known widely as the integrated stress response ( ISR ) . In the ISR , the phosphorylated form of eIF2 [eIF2 ( αP ) ] binds eIF2B unproductively forming a GEF-inhibited complex that restricts TC levels ( Rowlands et al . , 1988; Pavitt et al . , 1998 ) ( Figure 1A ) . This causes a general reduction in protein synthesis initiation rates while at the same time activating translation of ISR-responsive mRNAs ( Young and Wek , 2016 ) . Genetic and biochemical evidence shows that phosphorylated eIF2α binds to a regulatory site on eIF2B formed by the eIF2Bαβδ subcomplex ( Pavitt et al . , 1997; Krishnamoorthy et al . , 2001; Kashiwagi et al . , 2016b ) . In contrast GEF function is provided by eIF2Bε , which interacts with eIF2β and the GDP/GTP-binding eIF2γ subunit ( Gomez and Pavitt , 2000; Asano et al . , 1999; Alone and Dever , 2006 ) . The γ subunit of eIF2B binds to , and stimulates the GEF action of , eIF2Bε ( Pavitt et al . , 1998; Jennings et al . , 2013 ) . 10 . 7554/eLife . 24542 . 003Figure 1 . eIF2 affinity for eIF2B is unaffected by guanine nucleotides . ( A ) Current model for eIF2 activation and inhibition by phosphorylated eIF2 . Interactions and activities are explained in the introduction . ( B ) . Overview of eIF2/eIF2B equilibrium binding assay . ( C–E ) . Affinity determined by mixing 2 nM eIF2 with increasing concentrations of eIF2B immobilised on anti-Flag resin . Flag resin was pelleted and the eIF2 remaining in each supernatant fraction was resolved by SDS-PAGE and immunoblotted with two eIF2 subunit antibodies ( inset , eIF2α and eIF2γ ) . Fraction bound at equilibrium was determined by quantification: total ( lane 0 nM eIF2B ) minus fraction remaining in the supernatant ( graphs ) and used calculate the dissociation constants ( nM ± standard error ( SE ) ) indicated . Assays were done either without nucleotide ( C ) or in the presence of either 1 mM GDP ( D ) or 1 mM GTP ( E ) . ( F ) . Cartoon of figure conclusion . DOI: http://dx . doi . org/10 . 7554/eLife . 24542 . 003 A second eIF2 regulatory factor is eIF5 . eIF5 primarily functions as a GTPase activating protein ( GAP ) for eIF2-GTP within the 48S pre-initiation complex ( PIC ) ( Paulin et al . , 2001; Algire et al . , 2005 ) . GTP hydrolysis within eIF2-GTP and subsequent inorganic phosphate release are key events that signal AUG start codon recognition by initiating ribosomes ( Algire et al . , 2005 ) . These events facilitate release of eIF2-GDP/eIF5 complexes from the initiation complex , enabling subsequent 60S joining . eIF5 binds inactive eIF2-GDP and active TC with similar affinity ( Kd = 23 nM; [Algire et al . , 2005] ) and we found that eIF5 has a second activity with eIF2-GDP that impairs spontaneous release of GDP from eIF2 . This GDP-dissociation inhibitor ( GDI ) function is important during the ISR when eIF2α is phosphorylated and translation is attenuated as GDI prevents eIF2B independent release of GDP and ensures tight translational control ( Jennings and Pavitt , 2010a , 2010b; Jennings et al . , 2016 ) . Under optimal cell growth conditions , when free eIF2B is available , we found that eIF2B ( specifically eIF2Bγε ) actively displaces eIF5 from eIF2-GDP prior to its action as a GEF ( Jennings et al . , 2013 ) . Thus the combined actions of eIF2B and eIF5 regulate eIF2 nucleotide status ( Figure 1A ) . GEFs typically function to activate their cognate G protein partners by destabilizing the binding of Mg2+ and GDP prompting GDP release to a nucleotide free intermediate that allows GTP to bind . To promote nucleotide exchange GEFs have a higher affinity for nucleotide-free forms of their G protein partners than to the nucleotide bound forms ( Bos et al . , 2007 ) . For example , translation elongation factor eEF1A/eEF1B affinity is 0 . 125 µM , while affinity of eEF1B for the nucleotide-bound forms of eEF1A are >10 fold less stable ( Kd = 1 . 5 µM ) ( Gromadski et al . , 2007 ) . For the prokaryotic EF-Tu and EF-Ts , the nucleotide free complex is 1000-fold more stable ( Kd = 3 nM ) than nucleotide bound forms ( Kd = 2–6 µM ) ( Gromadski et al . , 2002 ) . eIF2γ is structurally conserved with the tRNA-binding elongation factors eEF1A and EF-Tu as well as aIF2 found in archaea and all bind to both GDP/GTP and to tRNAs ( Schmitt et al . , 2010 ) . However , unlike the translation elongation factors , eIF2/aIF2 each have three subunits ( α-γ ) and analyses indicate that both the eIF2α and β subunits also make important contacts with Met-tRNAi within the TC and modulate nucleotide binding to eIF2γ ( Huang et al . , 1997; Naveau et al . , 2013; Llácer et al . , 2015 ) . eIF2αβ also influence interactions with eIF2B and eIF5 , as indicated above . The additional eIF2 and eIF2B subunit complexity over other translation G-proteins and GEFs suggested to us that eIF2B might not behave as a typical GEF . In addition a recent studies examining the EF-Tu interactions with its GEF , EF-Ts , suggest that EF-Ts can stimulate EF-Tu TC formation and can form an EF-Tu-GTP/tRNA/EF-Ts quaternary complex ( Burnett et al . , 2013 , 2014 ) . These findings prompted us to investigate the interplay between eIF2B and eIF5 with eIF2-GTP and Met-tRNAi during eIF2 activation and its control by eIF2 ( αP ) . We report the unexpected finding that eIF2B binding affinity for eIF2 at steady state is not influenced by guanine nucleotides . We also report an antagonistic ability of eIF2B , where eIF2B can destabilise TC and compete with tRNAi for eIF2 . We show that these new antagonistic roles for eIF2B are counteracted by eIF5 , wherein eIF5 stabilises TC preventing the destabilizing effects of eIF2B . Hence eIF5 is required for robust rates of TC formation and transition initiation . Together these findings suggest that the competition between eIF2B and eIF5 for eIF2 continues even after TC formation and that TC/eIF5 complexes represent a robust product of nucleotide exchange . Finally we demonstrate that eIF2B can better compete with eIF5 for binding to TC formed with eIF2 ( αP ) , leading to increased TC destabilization . We propose these new interactions of eIF2B with TC reveal a second step of translation initiation regulated by eIF2 ( αP ) and that these events ensure tight control of translation initiation . GTP-binding protein interactions with their GEFs are typically modulated by the nucleotide bound to the G-protein , wherein the affinity for the nucleotide-free exchange intermediate is greatest and contributes to the forward direction of the G protein activation process ( Bos et al . , 2007 ) . This preference has been found for translation elongation factors eEF1A and EF-Tu , that , like eIF2 , also bind tRNAs during protein synthesis ( Gromadski et al . , 2002 , 2007 ) . We investigated eIF2/eIF2B complex formation and in contrast to our expectations we found that , at steady state , eIF2 bound eIF2B with equal affinity independent of its nucleotide status . Our assay used Flag affinity resin to immobilize purified Flag tagged eIF2B over a wide range of concentrations ( 0–100 nM ) incubated with limiting apo-eIF2 ( 2 nM eIF2 incubated to remove any co-purifying nucleotide , see Materials and methods ) in the presence of 1 mM GDP , GTP or nucleotide free ( Figure 1B ) . The fraction of eIF2 remaining unbound was quantified and used to calculate equilibrium dissociation constants ( Kds ) . Kds showed minimal variations according to nucleotide status 26 . 6 nM ( +GDP ) to 32 . 2 nM ( apo-eIF2 ) ( Figure 1C–E ) . eIF2B has recently been shown to be a dimer ( Gordiyenko et al . , 2014; Kashiwagi et al . , 2016a ) and so capable of binding two eIF2 molecules per dimer . Our analyses provided no evidence for co-operative binding and the data fit well to a model where a 5-subunit eIF2B monomer binds one eIF2 heterotrimer ( see Materials and methods ) . So in these and all following experiments our eIF2B concentrations assume a 5-subumit monomer forming a 1:1 complex with eIF2 and differ by two-fold from values that would be obtained by an eIF2B dimer . The measured eIF2/eIF2B Kds are similar to the measured affinity between eIF5 and eIF2-GDP and between eIF5 and TC ( both ~23 nM ) ( Algire et al . , 2005 ) . Hence neither regulator of eIF2 shows a preference for eIF2 nucleotide status . Both of these eIF2 regulatory proteins bind via interactions with both eIF2β and γ ( Asano et al . , 1999; Mohammad-Qureshi et al . , 2007a ) . As eIF5 has opposing GDI and GAP functions with different forms of eIF2 , these observations open up the possibility that eIF2B has additional roles regulating eIF2-GTP . As eIF2B does not have enhanced affinity for nucleotide-free eIF2 ( Figure 1F ) or a reduced affinity for GTP bound eIF2 , the GEF step alone cannot provide a driving force for TC formation to promote rounds of protein synthesis initiation . We hypothesized that eIF2B binding to eIF2-GTP might impact on the ability of eIF2 to form TC with Met-tRNAi and therefore promote eIF2-TC formation . Indeed recent studies with bacterial EF-Tu showed that its GEF , EF-Ts , could accelerate both the formation and decay of EF-Tu TC via a transient quaternary complex ( Burnett et al . , 2014 ) . We used BOP-N-Met-tRNAi ( tRNA probes ) in a fluorescence spectroscopy assay where increasing concentrations of eIF2 and excess nucleotides enabled us to monitor TC formation in solution . Consistent with recent studies ( Jennings et al . , 2016 ) , eIF2-GTP TC formed readily ( Kd tRNAi=1 . 13 nM ) , while eIF2-GDP significantly reduced Met-tRNAi binding by ~50 fold ( Kd tRNAi=55 . 5 nM; Figure 2A ) . Surprisingly , we found that the addition of eIF2B significantly inhibited TC formation in a concentration dependent manner ( Figure 2B ) . Our assays contained 20 nM eIF2 and even a ten fold lower amount of eIF2B antagonised TC formation . Importantly , this assay was performed without GDP and with an excess of GTP making eIF2B GDP/GTP exchange redundant for TC formation . This experiment reveals that eIF2B competes with Met-tRNAi for binding to eIF2-GTP . The inhibitory effect of eIF2B on TC formation observed here is greater than predicted by the individual binding affinities of Met-tRNAi and eIF2B for eIF2-GTP and a standard competitive inhibition model ( Schön et al . , 2011 ) . The reasons for this are not yet clear . One idea is that despite GTP presence in vast excess , eIF2B binding to eIF2 may promote GTP release , as in the absence of bound Met-tRNAi , the rate of spontaneous GTP release from eIF2 is high ( Figure 2—figure supplement 1 ) . Hence , as GTP loss impairs Met-tRNAi affinity significantly ( Figure 2A ) , this could reduce the apparent affinity for Met-tRNAi in the presence of eIF2B . 10 . 7554/eLife . 24542 . 004Figure 2 . eIF2B competes with Met-tRNAi for binding to eIF2-GTP . ( A ) Binding curves titrating 20 nM BOP-N-Met-tRNAi with eIF2 in the presence of 1 mM GTP or GDP . Dissociation constants ( nM ) ± SE are indicated ( inset ) . ***p<0 . 001 two-tailed T-test . ( B ) . As in A only in the presence of 1 mM GTP ± increasing concentrations of eIF2B ( 0–200 nM ) . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 two-tailed T-test C ) . Supernatant depletion affinity capture as done in Figure 1 with 1 mM GTP +2 nM Met-tRNAi with calculated affinity constants ± SE D ) . Dissociation curves for 20 nM BOP-N-Met-tRNAi pre-bound to eIF2 ( 20 nM ) and saturating GTP ( 1 mM ) upon titration of eIF2B ( 0–300 nM ) ( blue triangles ) . eIF5 ( 20 nM ) was added to some reactions ( red circles ) . Calculated IC50 values ± SE for dissociation of TC are shown in inset box ( nM ) . *p<0 . 05 two-tailed T-test . ( E ) . Model for eIF2B competition with tRNA . DOI: http://dx . doi . org/10 . 7554/eLife . 24542 . 00410 . 7554/eLife . 24542 . 005Figure 2—figure supplement 1 . Guanine nucleotide release from eIF2 . Fluorescent-BODIPY labelled GDP or GTP release from eIF2 ( 20 nM ) in the presence of 1 mM ‘dark’ nucleotide ± eIF2B ( 5 nM ) . Calculated Koff ± SE is shown in the box . ***p<0 . 001 , two-tailed T-test . DOI: http://dx . doi . org/10 . 7554/eLife . 24542 . 00510 . 7554/eLife . 24542 . 006Figure 2—figure supplement 2 . eIF2B does not bind Met-tRNAi . Fluorescence binding assay in the absence of eIF2 , showing eIF2B does not bind to or alter fluorescence of BOP-N-Met-tRNAi . DOI: http://dx . doi . org/10 . 7554/eLife . 24542 . 006 In contrast to the large negative effect of eIF2B on TC formation , monitoring eIF2/eIF2B interactions in the presence of GTP and Met-tRNAi revealed that Met-tRNAi only minimally weakened the eIF2-GTP/eIF2B interaction ( Kd eIF2=45 . 2 nM; Figure 2C ) compared with eIF2-GTP/eIF2B affinity ( 28 . 3 nM; Figure 1C ) . Importantly Met-tRNAi remained in the supernatant in our binding assay ( Figure 2C ) and excess eIF2B did not alter BOP-N-Met-tRNAi fluorescence ( Figure 2—figure supplement 2 ) , indicating that eIF2B does not independently bind Met-tRNAi . To assess if eIF2B could disrupt pre-formed TC , we altered our assay set-up by pre-forming eIF2-TC to assess the ability of eIF2B to release BOP-N-Met-tRNAi . In the absence of added eIF2B , TC was stable during the course of our assays ( 0 nM eIF2B in Figure 2D ) , however the addition of eIF2B disrupted TC , altering tRNA fluorescence with an IC50 = 53 . 4 nM ( blue triangles , Figure 2D ) . Thus at steady-state , eIF2B impairs eIF2 TC formation rates and can disrupt pre-formed TC complexes . Hence eIF2B is a competitive inhibitor of TC formation . By analogy with EF-Ts ( Burnett et al . , 2014 ) , one idea is that competition can be effected by formation of an unstable eIF2-TC/eIF2B intermediate ( Figure 2E ) . These data suggest that eIF2 TC is not a final stable product of the eIF2B catalysed nucleotide exchange reaction and that an additional step is required to stabilize eIF2 TC . In models of translation initiation eIF2 TC interacts with eIF5 and with eIF1 , −1A and −3 to stimulate TC binding to 40 S ribosomes ( Hinnebusch , 2014 ) . Hence , unlike the translation elongation factor G protein TCs that bind directly to translating 80 S ribosome A-sites , additional protein factors modulate eIF2 TC downstream functions in translation initiation . We hypothesized that eIF5 may be a good candidate factor to modulate the apparent negative/competitive interfering role of eIF2B in destabilizing eIF2 TC ( Figure 2 ) because eIF5 binds TC ( Algire et al . , 2005 ) to promote its interaction with eIF3 ( Singh et al . , 2005 ) and eIF1 and eIF4G ( Yamamoto et al . , 2005; Luna et al . , 2012 ) and within the 48 S PIC eIF5 promotes start codon recognition during scanning and eIF2-GTP hydrolysis steps ( Huang et al . , 1997; Luna et al . , 2012 ) . Potentially at odds with this idea , we previously showed that eIF2B could readily displace eIF5 from eIF2-GDP binary complexes via its GDI-displacement factor ( GDF ) function ( IC50 eIF2B=15 . 1 nM ) ( Jennings et al . , 2013 ) ( Figure 1A ) . How Met-tRNAi binding to eIF2/eIF5 would impact eIF2B GDF was not known . We therefore examined the impact of eIF5 on the ability of eIF2B to destabilize eIF2-GTP and eIF2-TC formation . Pre-binding eIF2-GTP to eIF5 , using 20 nM eIF5 ( a 10-fold excess over eIF2 concentration in our assay ) had only minimal impact on the eIF2-GTP/eIF2B interaction ( Kd = 37 nM , Figure 3A ) compared to assays without eIF5 ( Kd = 28 nM; Figure 1C ) . This result is consistent with our previous report characterising the eIF2B GDF function that disrupts eIF2/eIF5 complexes ( Jennings et al . , 2013 ) . In contrast , addition of eIF5 and Met-tRNAi significantly impaired the ability of eIF2B to bind eIF2-GTP ( Kd = 223 nM; Figure 3B ) . Similarly , in assays monitoring the binding of Met-tRNAi to eIF2-GTP , whereas eIF2B reduced the affinity for BOP-N-Met-tRNAi affinity for eIF2 ( Kd increased from 1 . 13 to 14 . 5 nM , Figure 3C ) , eIF2B and eIF5 together did not affect the Met-tRNAi affinity for eIF2 ( Kd = 1 . 43 ) . Similarly , eIF5 alone had no impact ( Kd = 1 . 04 ) . Finally , pre-incubation of TC with eIF5 effectively prevented TC destabilization by eIF2B ( IC50 = 215 nM; Figure 2D ) . Together these data indicate that eIF2B can disrupt eIF2-GTP/eIF5 and TC ( eIF2-GTP/Met-tRNAi ) but not the quaternary complex of ( eIF2-GTP/Met-tRNAi/eIF5 ) . They suggest a model for eIF2 recycling and TC formation that requires an additional step over the currently accepted pathway: the formation of TC/eIF5 complex that enables Stabilization of TC ( STC ) , Figure 3D . These experiments define STC as a role for eIF5 to promote directionality to eIF2 recycling . By preventing pathway reversal by eIF2B , eIF5 STC function should help ensure eIF2 recycling proceeds in a forward direction . 10 . 7554/eLife . 24542 . 007Figure 3 . eIF5 stabilizes Met-tRNAi binding to TC . ( A ) eIF2/eIF2B equilibrium binding assay in the presence of eIF5 as described in legend to Figure 1 . eIF2 ( 2 nM ) was pre-bound ( prior to mixing with eIF2B ) with eIF5 ( 20 nM ) and GTP ( 1 mM ) then mixed with increasing concentrations of eIF2B immobilised on anti-Flag resin . ( B ) . As in A , but with Met-tRNAi ( 20 nM ) also added prior to eIF2B . ( C ) . Binding curves titrating BOP-N-Met-tRNAi ( 20 nM ) with eIF2 in the presence of GTP ( 1 mM ) as in Figure 2A , but with eIF5 ( 20 nM ) , eIF2B ( 20 nM ) or both eIF5 and eIF2B ( 20 nM each ) . Dissociation constants ( nM ) are indicated . ***p<0 . 001 , two-tailed T-test . ( D ) . Model for eIF5 stabilization of TC . DOI: http://dx . doi . org/10 . 7554/eLife . 24542 . 007 Next we wished to define the eIF5 domain requirements for stabilization of eIF2-TC . eIF5 is a single polypeptide with separate functional domains . Its amino terminal domain ( NTD; residues 1–152 ) is required for GAP activity , while its carboxyl terminal domain ( CTD; residues 241–405 ) is responsible for promoting interactions with eIF2 and other factors within the PIC ( Yamamoto et al . , 2005; Luna et al . , 2012 ) . The eIF5 NTD and CTD are joined by a linker region ( LR; residues 153–240 ) . The LR and CTD together are required for GDI activity ( Jennings and Pavitt , 2010a ) . We assessed the ability of eIF5 domains to exhibit STC function in our TC formation assay . The eIF5-CTD alone was sufficient to prevent eIF2B destabilizing eIF2-TC equivalent to full-length eIF5 ( Figure 4A and Figure 4—source data 1A , left panel ) . In contrast , the NTD alone was not able to stop eIF2B destabilizing eIF2-TC , as expected because the N-terminal domain has poor affinity for eIF2 ( Jennings and Pavitt , 2010a ) . This result provides a clear distinction between STC function that requires only the eIF5 CTD and the previously described eIF5 GDI function that requires the LR in addition to the CTD . 10 . 7554/eLife . 24542 . 008Figure 4 . eIF2Bε antagonises eIF5 STC function . ( A ) Kd measurements from eIF2-TC formation assays shown in Figure 4—source data 1 . Experiments were performed as in Figure 2A . ±20 nM eIF2B ±20 nM eIF5 , eIF5-NTD or eIF5-CTD . ( B ) . As panel A except ±5 nM or 20 nM of full eIF2B complex , eIF2Bγε subcomplexes or eIF2B epsilon alone . ( C ) . Left , Serial dilution growth assay of gcn2∆ yeast cells bearing multi-copy plasmids overexpressing the indicated combination of tRNAi ( IMT4 ) and eIF2Bε ( GCD6 ) , eIF2Bγ ( GCD1 ) , eIF5 ( TIF5 ) or eIF2Bεγ ( GCD6 + GCD1 ) grown on minimal and 3AT medium . Right , Western blot of strains used confirming overexpression of indicated proteins . ( D ) . Model showing eIF2Bε antagonism of TC/eIF5 and TC . **p<0 . 01 , ***p<0 . 001 , NS non-significant ( p≥0 . 05 ) , two-tailed T-test . DOI: http://dx . doi . org/10 . 7554/eLife . 24542 . 008 10 . 7554/eLife . 24542 . 009Figure 4—source data 1 . Destabilisation of TC requires eIF2Bε and stabilisation of TC requires eIF5-CTD . ( A ) Source data for Figure 4A , Top: cartoon summary of assay . Left panel: eIF2B competition with eIF2-TC is overcome by eIF5-CTD . Right panel: eIF2Bε antagonises eIF5 stabilization of eIF2-TC . Binding curves titrating 20 nM BOP-N-Met-tRNAi with eIF2 in the presence of 1 mM GTP ±20 nM eIF2B ( left ) or indicated eIF2B subunit combinations ( right ) . Dissociation constants ( nM ) ± SE are shown in insert boxes . ( B ) . Source data for Figure 4B . Top: cartoon summary of assay . Destabilization of eIF2-TC requires eIF2Bε . Binding curves as in panel A ± 5 nM or 20 nM of eIF2B subunit combinations shown . DOI: http://dx . doi . org/10 . 7554/eLife . 24542 . 009 As indicated in the introduction , eIF2B is a large multi-subunit complex encoded by five different genes . We wished to determine which was responsible for destabilization of eIF2 TC . eIF2B GEF activity requires eIF2Bε and is boosted by complex formation with eIF2Bγ . Similarly both subunits are needed for GDF ( Jennings et al . , 2013 ) . We found that both eIF2Bε alone and eIF2Bγε subcomplexes were as or more effective than full eIF2B complexes in competing with BOP-N-Met-tRNAi binding to eIF2-GTP ( Figure 4B and Figure 4—source data 1B ) . For example 20 nM eIF2B complexes reduced Met-tRNAi affinity by ~10 fold from 1 . 13 to 14 . 5 nM , while 20 nM eIF2Bε alone reduced Met-tRNAi affinity by ~20 fold ( to 27 nM; Figure 4B ) . To further explore eIF5 and eIF2B antagonism we repeated these TC formation assays with 20 nM eIF2B subcomplexes and 20 nM eIF5 . Here , eIF2Bε alone and eIF2Bγε subcomplexes behaved very differently to the presence of eIF5 . eIF2Bγε subcomplexes were out-competed by eIF5 STC function and behaved like full eIF2B complexes . In contrast eIF2Bε was unaffected by eIF5 and was able to impair TC formation ( Figure 4A and Figure 4—source data 1A , right panel ) . This result suggests that an unexpected and novel function of eIF2Bγ is to prevent eIF2Bε antagonising TC/eIF5 complexes and thereby facilitate eIF5 STC activity . It was shown previously that overexpression of eIF2Bε in yeast cells caused an unexpected phenotype . Excess eIF2Bε gene dosage promotes Gcn2-independent activation of the general amino acid control response ( GAAC ) , the yeast counterpart to the mammalian ISR ( Richardson et al . , 2004 ) . Activation of GAAC under these conditions implies cells have reduced active eIF2-TC levels that in turn stimulate GCN4 translation to levels that can overcome an imposed amino acid limitation ( Hinnebusch , 2005; Dever et al . , 2016 ) . This result is not expected from eIF2Bε’s known functions in eIF2 recycling , but is consistent with the results described above and shown in Figure 4A . Our data predict that by increasing Met-tRNAi gene dosage we might suppress the aberrant GAAC response by mass action . We therefore tested this idea , monitoring the growth of gcn2∆ cells on 3-aminotriazole containing medium ( SD + 3AT ) to assess GAAC activation . In accord with our prediction , increasing Met-tRNAi gene dosage suppressed the aberrant Gcn2-independent GAAC response associated with excess eIF2Bε gene dosage ( Figure 4C , Compare rows 3 and 4 ) . We also found that co-overexpressing eIF2Bγ with ε resulted in the same Gcn2-independent growth and suppression by excess Met-tRNAi ( rows 5 and 6 ) . As expected expressing eIF2Bγ alone did not confer the GAAC phenotype , ( Figure 4C , rows 7 and 8 ) , while excess eIF5 did ( row 9 ) . Excess eIF5 was previously shown to antagonise eIF2Bε ( Singh et al . , 2006 ) and in line with expectations this phenotype was not suppressed by excess Met-tRNAi gene dosage ( row 10 ) . Western Blotting confirmed that analysis excess Met-tRNAi did not alter protein overexpression levels ( Figure 4C , right panels ) and quantification indicated that eIF2Bε was overexpressed ~24 fold over wild-type , while eIF2Bγ was overexpressed ~12 fold over wild-type levels . Hence the cells shown in rows 5 and 6 ( Figure 4C ) likely contain a mix of excess eIF2Bγε complexes and excess free eIF2Bε that both contribute to the observed phenotype . Together with the biochemical analyses , these data show that eIF2Bε alone can act as a rogue factor to antagonise TC and TC/eIF5 complex formation ( Figure 4D ) . These results are consistent with the idea that eIF2B subunit complexity contributes to effective eIF2 recycling by reducing the ability of eIF2Bε to antagonise the recruitment of eIF2 TC by eIF5 and hence promote translation initiation . Because phosphorylation of eIF2α at serine 51 by eIF2α kinases is a universal and potent inhibitor of protein synthesis , we investigated whether this would have any impact on these new activities of eIF2B and eIF5 . As phosphorylated eIF2 [eIF2 ( αP ) ] can form a tight complex with eIF2B via binding to its regulatory αβδ subunits ( Pavitt et al . , 1998; Krishnamoorthy et al . , 2001 ) , we hypothesized that this may enable eIF2B to further antagonise eIF5/TC complexes . We used purified PKR to phosphorylate eIF2 and employed Phos-tag SDS-PAGE gel immunoblots to demonstrate that eIF2α was phosphorylated to at least 80% ( Figure 5—figure supplement 1A ) . To assess the impact of phosphorylated α on eIF2 kinetic parameters we repeated a large selection of our established assays with this new substrate . Importantly , eIF2 ( αP ) did not alter the intrinsic Koff GDP observed in our assays , but did prevent eIF2B stimulation of BODIPY-GDP release in line with previous findings ( Figure 5—figure supplement 1B ) . In addition , eIF2 ( αP ) exhibited a 8–10-fold enhanced affinity for eIF2B over non-phosphorylated eIF2 , irrespective of eIF2 nucleotide status ( Figure 5A and Figure 5—source data 1 panels A-C ) . Phosphorylation did not impact the affinity of eIF2 for BOP-N-Met-tRNAi , either in the presence of GTP ( Kd tRNA1 . 35 nM ) or GDP ( Kd tRNA49 . 3 nM ) ( Figure 5—figure supplement 2 ) . The lack of impact of phosphorylation on Met-tRNAi affinity eIF2 is in line with expectations given that structural analysis of eIF2-tRNA interactions indicates that there is no direct contact between the serine 51 of eIF2α and Met-tRNAi ( Llácer et al . , 2015 ) and because in yeast mutations in eIF2 or eIF2B subunits permit cells to grow at normal rates with very high eIF2αP levels ( Pavitt et al . , 1997; Vazquez de Aldana and Hinnebusch , 1994; Vazquez de Aldana et al . , 1993 ) : experiments implying that eIF2αP inhibits eIF2B function only . 10 . 7554/eLife . 24542 . 010Figure 5 . eIF2B antagonizes eIF2 ( αP ) -TC/eIF5 complexes . ( A ) Kd measurements ± SE from eIF2 ( αP ) /eIF2B complex formation assays shown in Figure 5—source data 1 and compared with measurements made with non-phosphorylated eIF2 shown in previous figures . ( B ) . Dissociation curves for 20 nM BOP-N-Met-tRNAi pre-bound to eIF2 ( αP ) ( 20 nM ) and saturating GTP ( 1 mM ) upon titration of eIF2B ( 0–300 nM ) ( red circles ) . eIF5 ( 20 nM ) was added to some reactions ( open circles ) . Data shown in Figure 2D is reproduced for comparison in gray symbols . Calculated IC50 values ± SE are shown in the box . ( C ) . Model for eIF2B inhibition of eIF2 ( αP ) -TC/eIF5 complexes . *p<0 . 05 , ***p<0 . 001 , two-tailed T-test . DOI: http://dx . doi . org/10 . 7554/eLife . 24542 . 010 10 . 7554/eLife . 24542 . 011Figure 5—source data 1 . Binding between eIF2 ( αP ) and eIF2B in the presence of different ligands . Top cartoon depicting assay . ( A–F ) . eIF2/eIF2B interaction assay equivalent to that shown in Figures 1C–E , 2C , 3A and B , only with eIF2 phosphorylated by PKR prior to assay . Calculated Kds ± SE and example western blot images are shown as insets . Kds ± SE are plotted in the histogram in Figure 5A . DOI: http://dx . doi . org/10 . 7554/eLife . 24542 . 011 10 . 7554/eLife . 24542 . 012Figure 5—figure supplement 1 . Phosphorylation of eIF2α and its impact on eIF2B activity . ( A ) Phos-Tag gel western blots of eIF2 ± PKR with total eIF2α ( left ) and ser-51 phospho-specific ( right ) antibodies . ( B ) . Fluorescent-BODIPY labelled GDP release from eIF2 or eIF2 ( αP ) ( 20 nM ) in the presence of 1 mM ‘dark’ nucleotide ±5 nM eIF2B . Calculated Koff ± SE shown . ***p<0 . 001 , NS non-significant ( p≥0 . 05 ) , two-tailed T-test . DOI: http://dx . doi . org/10 . 7554/eLife . 24542 . 01210 . 7554/eLife . 24542 . 013Figure 5—figure supplement 2 . eIF2 ( αP ) affinity for Met-tRNAi . Top , Cartoon descripting assay . Bottom: Fluorescence of 20 nM BOP-N-Met-tRNAi binding to 2 nM eIF2 ( αP ) in the presence of 1 mM GDP or GTP . Calculated Kds ± SE shown in inset box . ***p<0 . 001 , two-tailed T-test . DOI: http://dx . doi . org/10 . 7554/eLife . 24542 . 013 To examine completion between eIF2B , eIF5 and Met-tRNAi for eIF2 ( αP ) we repeated our previous interaction assays . eIF2B had only a modest 1 . 2–1 . 4-fold enhanced affinity for eIF2 ( αP ) -GTP over eIF2-GTP in the presence of excess Met-tRNAi or eIF5 alone ( Figure 5A and Figure 5—source data 1 , panels D-E ) . In contrast when 10-fold molar excess of both Met-tRNAi and eIF5 over eIF2 ( αP ) were both included in the binding assay , eIF2 ( αP ) /eIF2B complexes formed with 4-fold enhanced affinity over non-phosphorylated eIF2 ( Kd eIF2 =57 nM vs 223 nM for unphosphorylated eIF2/eIF2B ( Figure 5A and Figure 5—source data 1 , panel F ) . These experiments suggest that eIF2B is better equipped to disrupt TC/eIF5 complexes when eIF2 is phosphorylated , than when non-phosphorylated . To quantify this we preformed TC with eIF2 ( αP ) ± eIF5 and asked if eIF2B could disrupt Met-tRNAi binding to eIF2 . As shown in Figure 5B , eIF2B is able to displace Met-tRNAi from eIF2 ( αP ) -TC /eIF5 complexes ( red open circles ) ~4x as well as it can with non-phosphorylated TC / eIF5 ( gray open circles ) . Indeed Met-tRNAi displacement from eIF2 ( αP ) -TC/eIF5 complexes by eIF2B proceeded as effectively as Met-tRNAi displacement from TC complexes lacking eIF5 ( gray filled circles in Figure 5B ) . Together these data are consistent with the idea that in addition to regulation of eIF2B GEF activity , eIF2B can also further inhibit translation initiation by disrupting eIF2 ( αP ) -TC/eIF5 complexes and forming eIF2 ( αP ) -GTP/eIF2B complexes ( Figure 5C ) . Unlike bound GDP , GTP readily dissociates from eIF2 ( αP ) ( Figure 5—figure supplement 1B ) which should further impede Met-tRNAi re-acquisition . We propose that these inhibitory interactions with eIF2 ( αP ) -GTP act as secondary ‘fail-safe’ mechanisms to ensure tight-regulation of translation by phosphorylation of eIF2 . We have evaluated interactions between eIF2B and eIF2 , Met-tRNAi and eIF5 in the formation and regulation of eIF2-GTP/Met-tRNAi ternary complexes . Our studies reveal five key novel findings about the mechanisms driving translation initiation and its regulation by eIF2αP , a key element of the ISR . Firstly we show that eIF2B affinity for eIF2 is not governed by the nucleotide bound ( Figure 1 ) . This is highly unusual as GEFs typically bind their G proteins with higher affinity for a nucleotide-free form . Such a binding mode helps drive the extraction of the inactive GDP and promote binding of GTP ( Bos et al . , 2007 ) . This lack of specificity of eIF2B for nucleotide-free eIF2 is likely because eIF2 has additional subunits over other translational G proteins . It is known that eIF2β is important for interacting with both regulatory factors ( Asano et al . , 1999 ) , hence eIF2 regulator interactions are not dependent solely on the nucleotide-binding eIF2γ subunit . A likely consequence of a lack of preference for the nucleotide free form is that eIF2B-promoted GDP release and subsequent binding of GTP does not by itself provide a strong forward momentum to promote eIF2B release from eIF2-GTP and drive TC formation . When investigating the influence of eIF2B on eIF2-TC formation we found that eIF2B and Met-tRNAi compete for eIF2-GTP . eIF2B therefore paradoxically promotes translational activation via GEF action and also acts as a competitive inhibitor impeding TC formation . We assume that because eIF2B concentrations in vivo are lower than eIF2 and Met-tRNAi ( Singh et al . , 2007; von der Haar and McCarthy , 2002 ) , that tRNAi abundance helps favour the forward reaction during times of optimal growth . The TC destabilizing activity of eIF2B appears important when translational down-regulation by eIF2 phosphorylation is required , ( see later discussion ) . Hence the negative impact of eIF2B on TC formation is likely the price cells pay to permit tight regulation of TC levels by eIF2B . eIF2B and Met-tRNAi competition for eIF2 only requires eIF2Bε ( Figure 4B ) . This is consistent with the idea that Met-tRNAi and eIF2Bε compete for an overlapping or shared binding site on eIF2 and that binding is mutually exclusive . Such competition between eIF2B and Met-tRNAi helps to explain why excess eIF2Bε aberrantly activates GCN4 translation . Although excess eIF2Bε enhances eIF2B GEF activity ( Pavitt et al . , 1998 ) , this is not reflected by GCN4 activity in vivo which is elevated in cells lacking gcn2 but with excess eIF2Bε . Gcn2-independent growth on 3AT medium is termed a Gcd- phenotype , which signals limiting active TC levels . Genetic suppression of the Gcd- phenotype by excess tRNAi gene dosage ( Figure 4C ) is in accord with a model where eIF2Bε and Met-tRNAi can compete for eIF2-GTP . eIF5 is known to promote eIF2-TC recruitment to other initiation factors ( Asano et al . , 2000; Yamamoto et al . , 2005; Luna et al . , 2012 ) . Here we found that eIF5 stabilizes eIF2-TC , effectively preventing eIF2B releasing eIF2 from TC/eIF5 complexes ( Figure 3 ) . These data are consistent with free eIF5 providing a driving force stabilizing TC in preparation for translation initiation and preventing eIF2B antagonizing eIF2-TC . One interpretation of this data is that eIF2-TC/eIF5 complexes represent the final product of nucleotide exchange rather than TC itself . A further implication of our findings is that by down-regulating eIF5 levels or activity it may be possible to influence the availability of TC/eIF5 complexes for translation . eIF5 levels are regulated through uORFs in mammalian cells ( Loughran et al . , 2012 ) , and eIF5 mimic proteins have been described that can antagonise eIF2/eIF5 functions , suggesting additional regulatory inputs could potentially control TC activity ( Singh et al . , 2011; Kozel et al . , 2016 ) . Only the eIF5-CTD was required to stabilize TC ( Figure 4A ) . This is consistent with prior results that eIF5-CTD is critical for assembling the eIF2 multifactor complex and 43S PIC ( Asano et al . , 2000; Yamamoto et al . , 2005 ) and shows that TC stabilization does not require the eIF5 linker-region , which is necessary for eIF5 GDI function with eIF2-GDP ( Jennings and Pavitt , 2010b ) . We found recently that the eIF2Bγε subcomplex is necessary for eIF2B to actively dissociate eIF2 from eIF2/eIF5 complexes ( Jennings et al . , 2013 ) . Neither eIF2Bε or eIF2Bγ alone could stimulate eIF5 release . eIF2Bγε sub-complexes also have enhanced GEF activity compared with eIF2Bε alone , activity equivalent to eIF2B full complexes ( Pavitt et al . , 1998; Jennings et al . , 2013 ) . Here we show that eIF2Bγ has a third important role , as it prevents eIF2Bε destabilizing TC/eIF5 complexes ( Figure 4A ) . We interpret these observations as indicating that eIF2Bγ plays multiple critical roles ensuring efficient eIF2 recycling . Firstly it facilitates release of eIF2-GDP from eIF2-GDP/eIF5 complexes ( GDF ) , next it boosts GDP release ( GEF ) and finally it prevents eIF2Bε destabilizing TC/eIF5 complexes . Together these data show eIF2Bγ is critical for efficient eIF2 recycling through multiple steps under conditions of active growth . eIF2 ( αP ) inhibits eIF2B GEF activity ( Pavitt et al . , 1998; Jennings et al . , 2013 ) by forming a tight complex between eIF2 ( αP ) -GDP and eIF2B via binding eIF2α to its regulatory αβδ subunits ( Pavitt et al . , 1998; Krishnamoorthy et al . , 2001; Kashiwagi et al . , 2016b ) . However our experiments provide evidence that eIF2B can also disrupt eIF2 ( αP ) -GTP functions independently of blocking eIF2B nucleotide exchange . eIF2B can destabilize eIF2 ( αP ) -TC/eIF5 complexes releasing both eIF5 ( Figure 5—source data 1 , panel F ) and Met-tRNAi ( Figure 5B ) . We interpret these findings as evidence of a secondary control system , or ‘failsafe’ mechanism to enable free eIF2B to mop up any eIF2 ( αP ) in its vicinity , even when the latter was bound to Met-tRNAi and eIF5 . As shown in cartoon form in Figure 6 , these activities likely allow eIF2B to wind-back initiation complex intermediates to boost eIF2 ( αP ) /eIF2B complexes . Unlike its tight interaction with GDP , GTP-bound eIF2 , is very unstable ( Figure 2—figure supplement 1 ) . Release of GTP from inhibited complexes would further ensure tight control of translation . We envisage this mechanism as providing additional safeguard to rapidly limit protein synthesis upon the onset of eIF2 ( αP ) , rather than restricting protein synthesis inhibition to eIF2 ( αP ) -GDP complexes . 10 . 7554/eLife . 24542 . 014Figure 6 . Model summary of new activities of eIF2 and eIF5 . A summary model of interactions between eIF2 , eIF2B , eIF5 and Met-tRNAi to generate eIF2-TC/eIF5 complexes and their inhibition by eIF2 ( αP ) . See text for details . DOI: http://dx . doi . org/10 . 7554/eLife . 24542 . 014 How effective this additional mechanism of control is in vivo is not easy to evaluate , as there is no tool currently available to decouple regulation of GTP complexes from GDP complexes . eIF2B is generally thought present in limiting concentrations compared with eIF2 , eIF5 and Met-tRNAi , and this is consistent both with maintaining active translation and limiting the ability of eIF2B to compete with Met-tRNAi when optimal growth and translation are required . Although it has been observed that eIF2B can be concentrated within cellular granules or bodies whose role is not clear , but which likely alter local relative eIF2:eIF2B ratios . For example eIF2B is found in diffusible cytoplasmic foci termed eIF2B bodies or filaments ( Campbell et al . , 2005; Taylor et al . , 2010; Noree et al . , 2010 ) , the abundance of which has been shown to be regulated by some cell stresses ( Petrovska et al . , 2014 ) . The eIF2 kinase Perk is an ER-membrane tethered kinase thought to preferentially regulate the translation of ER-localized protein synthesis during the unfolded protein response . In addition stress granules form upon a wide variety of cellular stresses in a variety of cell types including stresses that generate phosphorylated eIF2 ( Buchan and Parker , 2009; Kedersha et al . , 2013 ) . The ratios of translation factors in these various granules likely differ from their overall relative abundances in the cytoplasm . Such variations may permit lower affinity interactions to be biologically meaningful . Together , these findings provide a functional framework to our understanding of eIF2B complexity and its multiple roles in eIF2 recycling and suggest additional eIF2 ( αP ) controlled steps operate within the translation initiation pathway . Strain GP7124 ( MATα ura3-52 leu2-3 leu2-112 ino1 gcn2∆ sui3∆ trp1∆::hisG ura3-52::PHIS4-LacZ pAV2443 [Flag-SUI3 TRP1 CEN] ) was grown in yeast extract peptone dextrose medium as described ( Amberg et al . , 2005 ) and transformed using lithium acetate method ( Gietz and Woods , 2006 ) with plasmids pAV2330 [IMT4 URA3 2 µm] , pAV1754 [GCD6 LEU2 2 µm] , pAV1413 [GCD6 GCD1 LEU2 2 µm] , pAV1875 [TIF5-FLAG LEU2 2 µm] , pAV1162 [GCD1 LEU2 2 µm] or empty vector plasmid controls . Transformed strains were 10-fold serial diluted and spotted onto synthetic dextrose ( SD ) medium ±10 mM 1 , 2 , 3-aminotriazole and grown at 30°C . For protein expression analysis , strains were grown in selective media to an A600 of 0 . 6 then harvested by centrifugation at 5000 x g . Cells were resuspended in lysis buffer ( 30 mM HEPES pH 7 . 4 , 100 mM KCl , 0 . 1 mM MgCl2 , 10% glycerol + EDTA-free protease inhibitors ( Thermo Fisher Scientific , Loughborough , United Kingdom ) . 200 μl of acid washed glass beads ( Sigma-Aldrich , Poole , United Kingdom were added and cells were lysed by using a FastPrep ( MP Bio , Santa Ana , CA ) for 3 × 20 s at 6 . 5 ms−1 with cooling on iced water for 5 min between cycles . Cell debris was removed from the cell extract by centrifugation at 10 , 000 × g for 15 min at 4°C . SDS-PAGE and immuno-blotting were performed as previously described ( Jennings and Pavitt , 2010a ) using specific antibodies for eIF2α , eIF2γ , eIF2Bε , eIF5 , eIF2Bγ , Flag-M2 ( Sigma-Aldrich , Poole , United Kingdom ) , PAB1 ( Encor Biotechnology , Gainesville , FL ) , ser-51 phosphorylated eIF2α ( Abcam , Cambridge , UK ) , and eIF5 . Secondary antibody probing and quantification was performed using IRDye 800CW goat anti-rabbit IgG with an Odyssey Fc imaging system ( Li-Cor , Cambridge , United Kingdom ) . eIF2 was purified from yeast strain GP3511 as previously described ( Pavitt et al . , 1998 ) . To obtain apo-eIF2 free from nucleotide , eIF2 was dialysed with EDTA ( 30 mM HEPES , 100 mM KCl , 1 mM DTT , 1 mM EDTA , 10% glycerol , pH 7 . 4 ) then with magnesium ( 30 mM HEPES , 100 mM KCl , 1 mM DTT , 0 . 1 mM MgCl2 , 10% glycerol , pH 7 . 4 ) . GST-eIF5 was purified from Escherichia coli as described ( Jennings and Pavitt , 2010a ) . Flag-eIF2B complexes , subunits and sub-complexes were purified from yeast , as described ( Mohammad-Qureshi et al . , 2007b ) and ( Jennings et al . , 2013 ) . Flag-PKR was also purified from yeast as previously described ( Jennings et al . , 2013 ) Purified eIF2 was phosphorylated using purified PKR . Typically , 5 μg of eIF2 was incubated with 0 . 3 μL PKR , 0 . 1 mM ATP , and 5 mM NaF for 15 min at room temperature . To assess the level of phosphorylation , samples were resolved by SuperSep Phos-tag ( Wako , Osako , Japan ) SDS-PAGE to separate phosphorylated and nonphosphoylated eIF2α prior to immunoblotting with eIF2α-specific antisera . To monitor Met–tRNAi binding to eIF2 , 20 nM BOP-N-Met-tRNAi ( tRNA Probes , College Station , TX ) in 180 μl of assay buffer ( 30 mM HEPES , 100 mM KCl , 10 mM MgCl2 , pH7 . 4 ) was measuring using a Fluoromax-4 spectrophotometer ( Horiba , Stanmore , United Kingdom ) ( 490 nm excitation , 509 nm emission ) . Change in fluorescence intensity was measured upon addition of increasing amounts of apo-eIF2 , incubating for 5 min at room temperature each time . Each measurement was blanked against a control without nucleotide to account for any affect of eIF2 and data were corrected for dilution effects caused by volume addition and normalised to starting values before being fitted to a single site binding model: y = 1 + [ ( ∆Fmax - 1 ) * ( x/ ( x + Kd ) ) ] to obtain the dissociation constant ( Kd ± SE ) . To monitor dissociation of TC , 20 nM BOP-N-Met-tRNAi was premixed with eIF2 +/- eIF5 in 180 μl of assay buffer before monitoring fluorescence intensity . Change in intensity was then monitored upon addition of increasing amounts of eIF2B , incubating for 5 min at room temperature each time . Each measurement was blanked and volume corrected . Data from 7–13 individual experiments was fitted to an exponential curve to calculate IC50 values ± SE . To monitor the interaction between eIF2 and eIF2B , increasing amounts of eIF2B was pre-bound to 50 µl of anti-Flag M2 resin ( Sigma-Aldrich , Irvine , United Kingdom ) . 2 nM of apo eIF2 was added to the Flag resin ( ±1 mM nucleotides , 20 nM eIF5 , 2 nM met-tRNAi ) in a total volume of 1 ml and bound for 1 hr at 4°C . Flag resin was pelleted and the eIF2 remaining in each supernatant fraction was concentrated ( SpeedVac concentrator , Thermo Fisher Scientific , Loughborough , United Kingdom ) then resolved by SDS-PAGE and immunoblotted with two eIF2 subunit antibodies ( inset , eIF2α and eIF2γ ) . Fraction bound at equilibrium was determined by quantification and then subtracting the amount remaining upon addition of eIF2B from the total using a control where no eIF2B was added . Data from seven individual experiments were fitted to [ ( A*x ) / ( Kd+x ) ] to calculate the equilibrium dissociation constant ( Kd ) ± SE . eIF5 in the supernatant was also probed using a specific antibody . Met-tRNAi in the supernatant was monitored by semi-quantitative RT-PCR using the oligonucleotide primers IMTF ( AGCGCCGTGGCGCAGTGGAAGCGCGCA ) and IMTR ( TAGCGCCGCTCGGTTTCGATCCGAG ) , Onestep RT-PCR Kit ( Qiagen Ltd , Manchester , United Kingdom ) . Fluorescent eIF2•BODIPY-GDP binary complex was formed by incubating apo-eIF2 with a 2x excess of BODIPY-FL-GDP ( Thermo Fisher Scientific , Loughborough , United Kingdom ) for 20 min at room temperature . Excess nucleotide was removed by passing through a G-50 Sephadex column ( GE Healthcare , Little Chalfont , United Kingdom ) . Labelling efficiency was calculated to exceed 90% . To measure GDP release , 20 nM eIF2•BODIPY-GDP was quickly mixed with 1 mM of unlabelled GDP ( ± eIF2B and ± GST-eIF5 ) in 180 µl of assay buffer ( 30 mM HEPES , 100 mM KCl , 10 mM MgCl2 , pH 7 . 4 ) and fluorescence intensity was continuously measured using a Fluoromax-4 spectrophotometer ( Horiba , Stanmore , United Kingdom ) ( 490 nm excitation , 509 nm emission , 0 . 1 s integration time ) . 5 nM of eIF2B was added to stimulate nucleotide exchange . Experimental data were fitted to exponential dissociation curves to determine the rate constants ( Koff ) ± SE . To determine statistical significance , standard errors ( SE ) reported from nonlinear curve regression were compared . Degrees of freedom were calculated as the sum of the data points in each fit minus the sum of the variables fit ( two variables per fit , totaling four in all cases ) . T Scores were calculated as T= ( Fit1-Fit2 ) /√ ( SE1+SE2 ) . P values were then calculated based on a two-tailed T-test .

All cells sense and react to changes in the world around them . One way that cells react to threats to their health is by switching off genes required for their normal activity and diverting this energy to switching on genes that deal with the specific stress . A protein called eIF2 controls this general response , which is known as the “integrated stress response” . The eIF2 protein is switched on when it binds to a molecule of GTP and switched off when it binds to a molecule of GDP; these two molecules are swapped by another protein called eIF2B . Previous research revealed that the eIF2B protein attaches to eIF2 that is carrying GDP ( also known as eIF2-GDP ) and decides whether to switch it to eIF2-GTP or not depending on signals from the integrated stress response . However it was not known if eIF2B could also attach to eIF2-GTP or how this might affect the eIF2 protein , the activity of genes and the integrated stress response . Now , using proteins extracted from baker’s yeast as a model , Jennings et al . studied the interactions between eIF2 and eIF2B . The experiments revealed that , under stressful conditions , eIF2B not only triggers the integrated stress response through eIF2-GDP but also halts any active eIF2-GTP not under the control of this response . Jennings et al . suggest that these two processes represent a fail-safe switch that ensures that the integrated stress responses occurs rapidly whenever the cell is stressed . It is not clear how important this proposed fail-safe switch is for different types of cells , and so further studies will explore this question . In particular , people with mutations in the five genes that encode eIF2B develop a fatal brain disease and have an impaired integrated stress response , and so studies might check to see if some of these mutations affect the fail-safe switch .

2017

Fail-safe control of translation initiation by dissociation of eIF2α phosphorylated ternary complexes

Cerebellar plasticity underlies motor learning . However , how the cerebellum operates to enable learned changes in motor output is largely unknown . We developed a sensory-driven adaptation protocol for reflexive whisker protraction and recorded Purkinje cell activity from crus 1 and 2 of awake mice . Before training , simple spikes of individual Purkinje cells correlated during reflexive protraction with the whisker position without lead or lag . After training , simple spikes and whisker protractions were both enhanced with the spiking activity now leading behavioral responses . Neuronal and behavioral changes did not occur in two cell-specific mouse models with impaired long-term potentiation at their parallel fiber to Purkinje cell synapses . Consistent with cerebellar plasticity rules , increased simple spike activity was prominent in cells with low complex spike response probability . Thus , potentiation at parallel fiber to Purkinje cell synapses may contribute to reflex adaptation and enable expression of cerebellar learning through increases in simple spike activity . Active touch is important for exploring our environment , allowing us to assess the shape , substance and movements of objects and organisms around us ( Prescott et al . , 2011 ) . Throughout the animal kingdom , various systems have evolved for this purpose; these include for example the antennae of insects , the fingertips of primates and the well-developed whisker systems of rodents and sea mammals ( Ahl , 1986; Dehnhardt et al . , 2001; Staudacher et al . , 2005; Dere et al . , 2007; Anjum and Brecht , 2012 ) . Activation of these sensory organs can provoke reactive movements , often occurring as a reflex ( Nguyen and Kleinfeld , 2005; Bellavance et al . , 2017; Brown and Raman , 2018; Staudacher et al . , 2005 ) . For survival , it is important to maintain optimal control of such reflexes in daily life and to be able to adapt these movements ( Voigts et al . , 2015; Anjum and Brecht , 2012; Arkley et al . , 2017 ) . Given the impact of cerebellar plasticity on a wide variety of motor learning tasks ( Herzfeld et al . , 2015; Herzfeld et al . , 2018; Medina and Lisberger , 2008; ten Brinke et al . , 2015; Thier et al . , 2002; Voges et al . , 2017; Yang and Lisberger , 2017 ) , it can be anticipated that adaptation of reflexive whisker movements is also partly controlled by plastic processes in the cerebellum . Historically , most studies on cerebellar learning have suggested that long-term depression ( LTD ) at the parallel fiber to Purkinje cell ( PC ) synapse may act as the main cellular mechanism underlying induction of cerebellar motor learning ( Albus , 1971; Konnerth et al . , 1992; Ito , 1989; Koekkoek et al . , 2003; Medina and Lisberger , 2008; Boele et al . , 2018; Narain et al . , 2018 ) . However , parallel fiber LTD is unlikely to be the sole cellular mechanism underlying cerebellar learning ( Gao et al . , 2012; Hansel et al . , 2001; D'Angelo et al . , 2016 ) . Short-term forms of plasticity probably also contribute , as some forms of behavioral adaptation can be linked to changes in PC activity during the previous trial ( Yang and Lisberger , 2014; Herzfeld et al . , 2018 ) . Moreover , long-term potentiation ( LTP ) of parallel fiber to PC synapses may also be relevant , as various PC-specific mutants with impaired LTP show deficits in cerebellar learning ( Schonewille et al . , 2010; Schonewille et al . , 2011; Rahmati et al . , 2014; Gutierrez-Castellanos et al . , 2017 ) . Possibly , different cerebellar cellular mechanisms dominate the induction of different forms of learning , dependent on the requirements of the downstream circuitries involved ( De Zeeuw and Ten Brinke , 2015; Suvrathan et al . , 2016 ) . While many studies have focused on the synaptic mechanism ( s ) that may induce cerebellar motor learning , the spiking mechanisms that are responsible for the expression thereof remain relatively unexplored . To date , whereas evidence is emerging that the expression of conditioned eyeblink responses is mediated by a long-lasting suppression of simple spikes of PCs in the deep fissure of lobule simplex ( Heiney et al . , 2014; Halverson et al . , 2015; ten Brinke et al . , 2015 ) , it is unclear to what extent enduring increases in simple spike activity can also contribute to the expression of cerebellar learning , and if so for what forms of learning . Here , we developed a novel whisker training paradigm that is likely to generate plasticity in the cerebellar cortex and to produce increases in simple spike activity at the PC level following induction of LTP at the parallel fiber to PC synapse ( D'Angelo et al . , 2001; Lev-Ram et al . , 2002; Lev-Ram et al . , 2003; Coesmans et al . , 2004; Ramakrishnan et al . , 2016; van Beugen et al . , 2013 ) . Indeed , we show that a brief period of 4 Hz air-puff stimulation of the whiskers can enhance touch-induced whisker protraction as well as PC simple spike firing for tens of minutes . Moreover , these behavioral and neuronal changes are both absent in two independent mouse mutant lines deficient for parallel fiber to PC LTP , bridging the putative mechanism of memory expression with that of memory induction . The large facial whiskers are a prime source of sensory information for many mammals , in particular for rodents that can make elaborate movements with their large facial whiskers ( Arkley et al . , 2017; Brecht , 2007; Welker , 1964; Bosman et al . , 2011; Vincent , 1913 ) . It has been noted that passive touch can trigger active whisker movements in mice ( Bellavance et al . , 2017; Brown and Raman , 2018; Nguyen and Kleinfeld , 2005; Ferezou et al . , 2007 ) , but this behavior has not been described in great detail yet . Here , we studied whisker movements following rostro-caudal air-puff stimulation of the whisker pad in 16 awake , head-restrained mice ( Figure 1A–C ) . The air-puffer was placed in such a way that most , if not all , large mystacial whiskers were affected by the air flow from the front . The mice made active whisker protractions following the retractions induced by the air flow in the large majority ( 82% ) of stimulus trials ( Figure 1D–E; Figure 1—figure supplement 1C ) . Because of the systematic full-field air-flow from the front , the touch-induced protraction was typically performed by all whiskers simultaneously ( data not shown ) , which is in line with the presumed reflexive nature of this movement ( Bellavance et al . , 2017; Brown and Raman , 2018; Nguyen and Kleinfeld , 2005 ) . Moreover , as reported previously ( Ferezou et al . , 2007 ) , the touch-induced whisker protraction was followed in about half the trials ( 51% ) by extended periods of active whisker movements during the subsequent 200 ms interval ( Figure 1D; Figure 1—figure supplement 1A–C ) . However , under our experimental conditions with a 2 s inter-trial interval , spontaneous whisking in between the stimuli was relatively rare . Across all 16 mice measured , we found spontaneous movements ( with an amplitude exceeding 10° ) only in 12% of the 100 trials per mouse during the 200 ms interval prior to stimulus onset . To find out whether touch-induced whisker protraction can indeed be described as a reflex ( Bellavance et al . , 2017; Brown and Raman , 2018; Nguyen and Kleinfeld , 2005 ) , we wanted to know to what extent the movements also show signs characteristic of startle responses or voluntary events , which have a different identity . A startle response would be expected to be not only highly stereotypic , but to also show relatively little direction-specificity , and to reveal signs of pre-pulse inhibition ( Gogan , 1970; Swerdlow et al . , 1992; Moreno-Paublete et al . , 2017 ) . Instead , if the air-puff triggered a conscious , explorative movement , the animal would most likely make spontaneous movements towards the source of the air-puff , dependent on its specific position . To explore these possibilities , we placed a second air-puffer at the caudal side of the whisker field and a third air-puffer at the front of the contralateral whisker field , and we provided air-puffs from the three different orientations , intermingling trials with and without brief pre-pulses in random order ( Figure 1—figure supplement 2A ) . An air-puff from the front on the ipsilateral side induced a retraction prior to the active protraction . Such a retraction was mostly absent when stimulating from the back . Contralateral stimulation also evoked a slight retraction , followed by a much larger forward sweep ( Figure 1—figure supplement 2B–E; Supplementary file 1A ) . Thus , applying the air-puff from different angles produced different retractions and different subsequent protractions , arguing against a stereotypical startle behavior that occurs independent from the stimulus conditions . Moreover , we did not observe a diminishing effect of the weaker pre-pulse on the reaction to the stronger pulse ( p=0 . 268; Dunn’s pairwise post-hoc test after Friedman’s ANOVA; p=0 . 003; Fr = 13 . 933; df = 3 ) . Finally , we also did not observe distinct explorative movements linked to the puff sources , which might have suggested dominant voluntary components ( Figure 1—figure supplement 2B–C ) . Altogether , the reactive nature of the touch-induced whisker movements in the absence of characteristic signs of startle or voluntary responses indicates that the air-puff induced protraction is indeed a reflexive movement . In line with the fact that PCs receive sensory whisker input not only directly from the brainstem but also indirectly from thalamo-cortical pathways ( Figure 2F ) ( Kleinfeld et al . , 1999; McElvain et al . , 2018; Bosman et al . , 2011; Brown and Raman , 2018; Kubo et al . , 2018 ) , the dynamics of their responses upon whisker stimulation are heterogeneous ( Brown and Bower , 2001; Loewenstein et al . , 2005; Bosman et al . , 2010; Chu et al . , 2011 ) . To study the anatomical distribution of these responses within cerebellar lobules crus 1 and crus 2 , we mapped the complex spike and simple spike firing of their PCs following ipsilateral whisker pad stimulation with air-puffs in awake mice . Of the 132 single-unit PCs from which we recorded , 118 ( 89% ) showed significant complex spike responses , albeit with large variations in latency and amplitude ( Figure 2A–C , Figure 2—figure supplement 1A–B ) . We considered a response to be significant if it passed the threshold of 3 s . d . above the average of the pre-stimulus interval . Cluster analysis revealed that in terms of complex spike modulation PCs can better be considered as two separate clusters rather than a continuous spectrum ( indicated by the lowest absolute BIC value for two components ( 437 , compared to 490 and 442 for one and three components , respectively; Figure 2—figure supplement 1D ) . We refer to the cells of the cluster with the higher complex spike response probability as ‘strong’ ( 34% , with a peak response above 1 . 98%; see Materials and methods ) and the other as ‘weak’ ( 66% ) responders ( Figure 2—figure supplement 1D–F ) . Similarly , of the 132 recorded PCs 127 ( 96% ) showed a significant simple spike response . Simple spike responses were often bi-phasic , consisting of a period of inhibition followed by one of excitation , or vice versa ( Figure 2D–E , Figure 2—figure supplement 1C ) . The trough of the simple spike responses typically correlated in a reciprocal fashion with the peak of the complex spike responses ( Figure 2A–E; Figure 2—figure supplement 1A–C ) ( De Zeeuw et al . , 2011; Zhou et al . , 2014; Badura et al . , 2013 ) . Only 2 PCs , out of the 132 , did not show any significant modulation ( i . e . for neither complex spikes nor simple spikes ) . To chart the spatial distribution of the PCs with different response kinetics upon whisker stimulation we first combined electrolytic lesions ( Figure 2G ) with reconstructions of the electrode entry points , generating a map of the locations of the PCs from which we recorded with the quartz/platinum electrodes ( n = 132 ) . Complex spike responses to whisker stimulation were found to be especially strong in parts of crus 1 overlapping with large areas of the C2 , D1 and D2 zones ( Figure 2H ) , whereas the primary simple spike responses were predominantly facilitating in adjacent areas in the medial and lateral parts of crus 1 and crus 2 , as predicted by the overall tendency for reciprocity ( Figure 2H–I; Figure 2—figure supplement 1G–H ) . This distribution was verified using double-barrel glass pipettes with which we injected the neural tracer , BDA 3000 , at the recording spot after recording complex spike responses . Following identification of the source of the climbing fibers in the inferior olive and the projection area in the cerebellar nuclei ( Figure 2—figure supplement 2A–C ) , we defined the cerebellar area in which the recorded PC was located ( Apps and Hawkes , 2009; Voogd and Glickstein , 1998 ) . These experiments confirmed that the PCs with strong complex spike responses were situated most prominently in centro-lateral parts of crus 1 , whereas the PCs with weak complex spike responses were predominant in adjacent areas in crus 1 and crus 2 ( Figure 2—figure supplement 2D ) . As complex spikes have been reported to be able to encode , at the start of a movement , the destination of arm movements ( Kitazawa et al . , 1998 ) , we wondered whether a similar association could be found for whisker movements . Therefore , we asked whether trials that started with a complex spike involved larger or smaller protractions . To this end , we separated all trials of a session based upon the presence or absence of a complex spike during the first 100 ms after stimulus onset in a single PC . It turned out that during the trials with a complex spike , the protraction was significantly larger ( see Figure 3A for a single PC; Figure 3B for the population of 55 PCs of which we had electrophysiological recordings during whisker tracking and that responded to air-puff stimulation ) . A direct comparison between the timing of the complex spike response and the difference in whisker position between trials with and without a complex spike revealed that the peak in complex spike activity preceded the moment of maximal difference in position by 63 ± 4 ms ( mean ±SEM; n = 55; Figure 3C–D ) . The maximal difference in protraction in trials with a complex spike equaled 0 . 80° ( median , with IQR of 2 . 80°; p<0 . 001 ) , whereas this was only 0 . 28° ( 0 . 92° ) for retraction ( p=0 . 002; Wilcoxon matched pairs tests , significant after Bonferroni correction for multiple comparisons: α = 0 . 05/3 = 0 . 017 ) ( Figure 3E ) . These findings imply that trials that started with a complex spike showed bigger whisker protractions than those without a complex spike . We next questioned whether there was a correlation between the strength of the complex spike response and the difference in maximal protraction . This did not seem to be the case ( R = 0 . 119; p=0 . 386; Pearson correlation; Figure 3—figure supplement 1A ) . Thus , in general , the complex spike of any PC showing whisker-related complex spike activity could have a similar predictive power for the amplitude of the subsequent protraction . In line with this , a map showing the distribution of the PCs based upon the correlation of their complex spikes with whisker protraction was fairly homogeneous . Only in an area overlapping with the rostral part of crus 1 , a small cluster of PCs was observed whose complex spikes correlated with an unusually large difference in protraction ( Figure 3—figure supplement 1B ) . However , since sensory-induced complex spikes were typically more frequent in lateral crus 1 , PCs in this area appeared to have overall a stronger correlation with increased touch-induced whisker protraction than the PCs in the surrounding areas ( Figure 3—figure supplement 1C ) . Previous studies showed that motor control can be related to the coherence of complex spike firing of adjacent PCs ( Mukamel et al . , 2009; Hoogland et al . , 2015 ) . We therefore expected to observe also increased coherence at the trial onsets in our experiments . To test this , we performed two-photon Ca2+ imaging to study the behavior of adjacent groups of PCs in crus 1 around the moment of whisker pad air-puff stimulation in awake mice . After injection of the Ca2+-sensitive dye Cal-520 we could recognize the dendrites of PCs as parasagittal stripes , each of which showed fluorescent transients at irregular intervals ( Figure 3—figure supplement 2A–B ) . Previous studies identified these transients as the result of PC complex spike firing ( Ozden et al . , 2008; Tsutsumi et al . , 2015; Schultz et al . , 2009; De Gruijl et al . , 2014 ) . Occasionally , signals could be found that were shared by many PCs , even in the absence of sensory stimulation ( Figure 3—figure supplement 2B ) in line with earlier reports ( Ozden et al . , 2009; De Gruijl et al . , 2014; Mukamel et al . , 2009; Schultz et al . , 2009 ) . Upon whisker pad stimulation , however , complex spike firing occurred much more often collectively in multiple PCs ( Figure 3—figure supplement 2C ) . To quantify this form of coherent firing , we counted the number of complex spikes fired per frame ( of 40 ms ) and determined the level of coherence using cross-correlation analyses ( Figure 3—figure supplement 2D ) ( see also Ju et al . , 2018 ) . The levels of coherence increased to such strength that they were extremely unlikely to have occurred by the increase in firing frequency alone ( compared to a re-distribution of all events based on a Poisson distribution; Figure 3—figure supplement 2E ) . In other words , firing of a single or a few PCs was the dominant mode of activity in the absence of stimulation , and this changed toward the involvement of multiple PCs upon stimulation , firing coherently as can be seen in the change in distribution of coherently active PCs ( Figure 3—figure supplement 2F–G ) . We conclude that groups of adjacent PCs respond to whisker pad stimulation by increased complex spike firing with an enhanced level of coherence , which is likely to further facilitate the occurrence of bigger whisker reflexes ( see above ) . The firing rate of simple spikes has been shown to correlate with whisker position: in the large majority of PCs , simple spike firing is correlated with protraction and in a minority it correlates with retraction ( Brown and Raman , 2018; Chen et al . , 2016 ) . This led us to study the correlation in simple spike firing during touch-induced whisker protraction . At first sight , variation in simple spike firing roughly correlated to periods with whisker movement ( Figure 4A–B ) . To study this in more detail , we made use of the inter-trial variations in simple spike rate and whisker position , allowing us to make a correlation matrix between these two variables on a trial-by-trial basis ( see ten Brinke et al . , 2015 ) . In a representative example ( Figure 4C ) , the whisker protraction and peak in simple spike firing were roughly simultaneous . In the correlation matrix , this is visualized by the yellow color along the 45° line . This turned out to be the general pattern in 25 of the 56 PCs ( 45% ) of which we had electrophysiological recordings during whisker tracking ( Figure 4D ) . In all these 25 PCs , there was a positive correlation between instantaneous simple spike firing and whisker protraction that occurred relatively late during the movement , in particular between 80 and 200 ms after the start of the stimulus ( Figure 4C–D; Figure 4—figure supplement 1A–C ) , thus well after the complex spike responses occurred ( Figure 3C ) . In the 31 remaining PCs , that is the ones that did not display a significant correlation when evaluated at the level of individual cells , we still observed a slight , yet significant , correlation at the population level . Remarkably , this correlation was slightly negative , that is possibly reflecting a correlation between simple spike firing and retraction ( Figure 4—figure supplement 1 ) . We conclude that during the touch-induced whisker reflex simple spikes predominantly correlate with whisker protraction and that this correlation is maximal without a clear time lead or lag , unlike the complex spikes , the occurrence of which tended to precede the reflexive protraction . Next , we investigated whether sensory experience could modulate the touch-induced whisker protraction , the frequency of simple spike firing and the relation between them . We hypothesized that whisker movements might be enhanced following air-puff stimulation at 4 Hz , as this frequency has been shown to be particularly effective in inducing potentiation at the parallel fiber-to-PC synapse ( Coesmans et al . , 2004; D'Angelo et al . , 2001; Lev-Ram et al . , 2002; Ramakrishnan et al . , 2016 ) . Indeed , application of this 4 Hz air-puff stimulation to the whisker pad for only 20 s was sufficient to induce an increase in the maximal protraction ( average increase 17 . 9 ± 3 . 9%; mean ±SEM; p<0 . 001; Wilcoxon-matched pairs test; n = 16 mice ) ( Figure 5A–B; Supplementary file 1B ) . This change in the amplitude of the touch-induced whisker protraction was not accompanied by any substantial change in the complex spike response to whisker pad stimulation ( p=0 . 163; Wilcoxon matched pairs test; n = 55 PCs ) ( Figure 5C; Supplementary file 1B ) . However , the rate of simple spike firing upon air-puff stimulation was markedly increased after 20 s of 4 Hz air-puff stimulation . This was especially clear during the first 60 ms after the air-puff ( p=0 . 003; Wilcoxon matched pairs test; n = 55 PCs ) ( Figure 5D; Supplementary file 1B ) . Overlaying the averaged whisker traces and PC activity profiles highlighted the earlier occurrence of facilitation in simple spike firing after the 4 Hz air-puff stimulation protocol ( Figure 5E ) . To study this timing effect in more detail , we repeated the trial-based correlation analysis ( cf . Figure 4C–D ) . The short period of 4 Hz air-puff stimulation caused an anticipation of the moment of maximal correlation between simple spike firing and whisker position . Along the 45° line – thus regarding only the zero-lag correlation between simple spike firing and whisker position – this changed from 152 . 1 ± 18 . 1 ms to 90 . 7 ± 9 . 4 ms ( means ± SEM ) ; p=0 . 020; t = 2 . 664; df = 13; paired t test; n = 14 PCs ) ( Figure 5—figure supplement 1A–B ) . The slope of the correlation between the instantaneous simple spike frequency and the whisker position remained unaltered ( p=0 . 197 , t = 1 . 360 , df = 13 , n = 14 , paired t test ) ( Figure 5—figure supplement 1C–D ) . However , the point of maximal correlation was no longer with a zero-lag , but after induction the simple spikes led the whisker position ( pre-induction: Δtime = 0 ± 10 ms; post-induction: Δtime = 20 ± 30 ms; medians ± IQR; n = 14; p=0 . 001; Wilcoxon matched-pairs test ) ( Figure 5F ) . Thus , not only the simple spike rate increased , but also its relative timing to the touch-induced whisker protraction changed , now preceding the likewise increased touch-induced whisker protraction . During the entrainment itself ( i . e . during the 20 s period with 4 Hz air-puff stimulation ) , the whisker responses as well as the complex spike and the simple spike responses to each air-puff were weakened compared to the pre-induction period during which we used 0 . 5 Hz stimulation . More specifically , the touch-induced whisker protraction decreased by 62 . 2% ( median; IQR = 37 . 5% ) . The maximum response of the complex spikes significantly decreased from a median of 1 . 27% ( with an IQR of 1 . 89% ) during pre-induction to 0 . 52% ( with an IQR of 0 . 43% ) during induction ( p<0 . 001; Wilcoxon matched pairs tests , n = 55 PCs ) , and the average modulation of the simple spikes in the first 200 ms after the puff decreased from a median of 5 . 9% ( with an IQR of 18 . 2% ) during pre-induction to -0 . 3% ( IQR = 3 . 13 ) during induction ( p=0 . 039 , Wilcoxon matched-pairs test ) ( Figure 5—figure supplement 2 ) . Thus , during the 4 Hz training stage , all responses – both at the behavioral and neuronal level – diminished compared to the preceding 0 . 5 Hz stimulation stage . Given the correlation between instantaneous simple spike rate and whisker position described above , one would expect that contralateral air-puff stimulation – which triggers a stronger protraction ( Figure 1—figure supplement 2 ) – also triggers a stronger simple spike response . To test this hypothesis , we recorded PC activity while stimulating the ipsi- and contralateral whiskers in a random sequence ( Figure 5—figure supplement 3A–B ) . The change in maximal protraction was considerable ( difference in maximal protraction: 7 . 30 ± 1 . 24° ( mean ±SEM ) ; n = 9 mice ) ( Figure 5—figure supplement 3C; cf . Figure 1—figure supplement 2E ) . Possibly , the absence of the direct mechanical retraction on the ipsilateral side during contralateral air-puff stimulation can explain part of this difference , which is also in line with the earlier onset of the protraction during contralateral stimulation ( Figure 5—figure supplement 3C ) . However , in addition a change in simple spikes may contribute to this difference as well , as the simple spikes increased significantly more during contralateral stimulation ( increase during first 60 ms after air-puff onset for contra- vs . ipsilateral stimulation: 13 . 7 ± 5 . 3%; mean ±SEM; p=0 . 023; t = 2 . 413; df = 26; paired t test; n = 27 PCs ) ( Figure 5—figure supplement 3E ) . Such a contribution is compatible with the fact that most mossy fiber pathways related to whisker movement are bilateral with a contralateral preponderance ( Bosman et al . , 2011 ) . Instead , the complex spikes were less activated during contralateral stimulation ( complex spike peak response: ipsilateral: 1 . 40% ( 1 . 25% ) ; contralateral: 0 . 71% ( 0 . 81% ) ; medians ( IQR ) ; p<0 . 001; Wilcoxon matched-pairs test; n = 27 PCs ) ( Figure 5—figure supplement 3D ) . This response is in line with a bilateral component of the projection from the trigeminal nucleus to the olive ( De Zeeuw et al . , 1996 ) . To establish a causal link between increases in simple spike firing and whisker protraction , artificial PC stimulation would also have to affect whisker movement . Previously , it has been shown that simple spikes modulate ongoing whisker movements rather than initiate them ( Brown and Raman , 2018; Chen et al . , 2016; Proville et al . , 2014 ) . To find out whether simple spike firing could modulate touch-induced whisker protraction under our recording conditions , we investigated the impact of activation of PCs by optogenetic stimulation . To this end we used Pcp2-Ai27 mice , which express channelrhodopsin-2 exclusively in their PCs and which respond with a strong increase in their simple spike firing upon stimulation with blue light ( Witter et al . , 2013 ) . We placed an optic fiber with a diameter of 400 µm over the border between crus 1 and crus 2 and compared air-puff induced whisker movements among randomly intermingled trials with and without optogenetic PC stimulation . The period of optogenetic stimulation ( i . e . 100 ms ) was chosen to mimic preparatory activity of PCs and thus corresponded well to the period during which we observed increased simple spike firing after 4 Hz air-puff stimulation ( Figure 5D ) . As expected , the whisker protraction was substantially bigger during the period of optogenetic stimulation ( p<0 . 001; t = 4 . 411; df = 12; paired t test; n = 13 mice; Figure 5—figure supplement 4 ) . Thus , even though optogenetic stimulation of PCs can also trigger secondary feedback mechanisms that may influence the outcome ( Witter et al . , 2013; Chaumont et al . , 2013 ) , we conclude that increases in simple spike firing can cause stronger whisker protraction . As cerebellar plasticity is bi-directional and under control of climbing fiber activity ( Ohtsuki et al . , 2009; Lev-Ram et al . , 2003; Coesmans et al . , 2004 ) , we wanted to find out to what extent plastic changes in simple spike activity can be related to the strength of the complex spike response generated by climbing fibers . To this end we compared for each PC the strengths of the complex spike and simple spike responses before , during and after the 4 Hz air-puff stimulation . As expected , we found a significant negative correlation between the strength of the complex spike response , as measured by the peak of the PSTH before the 4 Hz air-puff stimulation , and the change in simple spike response following this 4 Hz stimulation ( R = 0 . 311; p=0 . 021; Pearson correlation; n = 55 PCs ) ( Figure 6A ) . We further substantiated these findings by looking separately at the average complex spike firing frequency of the strong and weak responders ( cf . Figure 2—figure supplement 1E ) . The correlation found between the frequency of complex spike firing and the change in simple spike activity after 4 Hz air-puff stimulation proved to be present only in the weak responders , taking the firing rate during the pre-induction and induction period into account ( Figure 6—figure supplement 1 ) . This is again in line with the notion that parallel fiber activity in the absence of climbing fiber activity promotes parallel fiber to PC LTP ( Coesmans et al . , 2004; Lev-Ram et al . , 2002; Ramakrishnan et al . , 2016 ) . The PCs with the strongest effect of 4 Hz air-puff stimulation on simple spike firing were mainly located in the lateral part of crus 2 ( Figure 6B ) , posterior to the crus 1 area with the strongest complex spike responses ( Figure 2H ) . We compared the location of this lateral crus 2 area to that of the PCs with the strongest correlations between simple spike firing and whisker protraction and we found these two crus 2 locations to match well ( Figure 6B–C ) . The impact of 4 Hz air-puff stimulation on the simple spike activity of PCs with a weak complex spike response lasted as long as our recordings lasted ( i . e . at least 30 min ) , whereas that on PCs with a strong complex spike response was not detectable during this period ( Figure 6D–F ) . Indeed , the weak responders differed significantly from the strong responders in this respect ( weak vs . strong responders: p=0 . 005; F = 3 . 961; df = 4 . 424; two-way repeated measures ANOVA with Greenhouse-Geyser correction; n = 8 weak and n = 6 strong responders; Figure 6F ) . Likewise , the impact of the 4 Hz air-puff stimulation on touch-induced whisker protraction also lasted throughout the recording in that the protraction sustained ( Figure 6—figure supplement 2 ) . Thus , both simple spikes and whisker muscles remained affected by the 4 Hz air-puff stimulation for as long as our recordings lasted . In reduced preparations , 4 Hz stimulation of the parallel fiber inputs leads to LTP of parallel fiber to PC synapses ( Coesmans et al . , 2004; Lev-Ram et al . , 2002; Ramakrishnan et al . , 2016 ) . At the same time , parallel fiber LTP is inhibited by climbing fiber activity ( Coesmans et al . , 2004; Lev-Ram et al . , 2003; Ohtsuki et al . , 2009 ) . Hence , our data appear in line with a role for parallel fiber LTP as a potential mechanism underlying the observed increase in simple spike firing upon a brief period of 4 Hz stimulation . To further test a potential role for LTP , we repeated our 4 Hz air-puff stimulation experiments in Pcp2-Ppp3r1 mice , which lack the PP2B protein specifically in their PCs , rendering them deficient of parallel fiber-to-PC LTP ( Schonewille et al . , 2010 ) ( Figure 7A ) . The impact of 4 Hz air-puff stimulation on the maximal protraction was significantly less in the Pcp2-Ppp3r1 mutant mice compared to wild types ( p=0 . 044 , t = 2 . 162 , df = 19 , t test; Figure 7B–D ) . Accordingly , in contrast to those in their wild-type ( WT ) littermates ( p<0 . 001 , t = 4 . 122 , df = 15 , t test ) , the maximal touch-induced whisker protraction before and after induction was not significantly different in Pcp2-Ppp3r1 mice ( p=0 . 647 , t = 0 . 470 , df = 12 , t test; Figure 7E ) . Thus , Pcp2-Ppp3r1 mice do not show increased touch-induced whisker protraction after 4 Hz air-puff stimulation . In line with the absence of increased touch-induced whisker protraction , also the increase in simple spike firing observed in wild type mice was absent in Pcp2-Ppp3r1 mice . As the strong complex spike responders in WTs did not show changes in simple spike activity ( cf . Figure 6 ) , we compared weak complex spike responders of both genotypes . Simple spike responses were stably increased in WT PCs with a weak complex spike response following 4 Hz air-puff stimulation ( as shown in Figure 6F ) , but not in those of Pcp2-Ppp3r1 mice ( effect of genotype: p=0 . 003 , F = 4 . 361 , df = 4 . 137 , two-way repeated measures ANOVA with Greenhouse-Geyser correction; n = 8 WT and n = 9 Pcp2-Ppp3r1 PCs ) ( Figure 7F ) . Despite the lack of potentiation , we found that the Pcp2-Ppp3r1 mice still had a significant correlation between the complex spike frequency during the induction block and changes in simple spike activity ( R = 0 . 489 , p=0 . 013 , Pearson correlation; Figure 7—figure supplement 1A ) ; this correlation may result from other forms of plasticity that are still intact in Pcp2-Ppp3r1 mice ( Schonewille et al . , 2010 ) . Yet , in line with the absence of increased simple spike responsiveness , the correlation between changes in simple spike firing during the induction block and the impact of 4 Hz air-puff stimulation , as present in the WT PCs , was absent in the Pcp2-Ppp3r1 mice ( Figure 7—figure supplement 1B ) . Thus , in the absence of the PP2B protein in PCs , the impact of 4 Hz air-puff stimulation on touch-induced whisker protraction as well as on the simple spike responsiveness was not detectable . These correlations between complex spike and simple spike firing on the one hand and modification of the simple spike response to whisker pad stimulation on the other hand further strengthen our hypothesis that parallel fiber to PC LTP is one of the main mechanisms that underlies the long-term changes that can be observed at both the level of simple spike activity and whisker protraction after 4 Hz air-puff stimulation . To control for compensatory mechanisms specific for Pcp2-Ppp3r1 mice we used a second , independent , yet also PC-specific , mutant mouse line deficient in parallel fiber LTP . In these mice ( Pcp2-Gria3 ) , PCs lack the AMPA receptor GluA3 subunit ( Gutierrez-Castellanos et al . , 2017 ) . As in the Pcp2-Ppp3r1 mice , we did not find evidence for increased whisker protraction after 4 Hz air-puff stimulation ( e . g . , change in whisker angle during the first 120 ms after air-puff onset: WT vs . Pcp2-Gria3 mice: p=0 . 007 , Tukey’s post-hoc test after ANOVA ( p=0 . 001 , F = 9 . 111 , df = 2 ) , n = 16 WT and n = 6 Gria3 deficient mice ) ( Figure 8A–C ) . Moreover , as in the Pcp2-Ppp3r1 mice , also the increase in simple spike responsiveness after 4 Hz stimulation was absent in Pcp2-Gria3 mice ( difference in simple spike count between WT and Pcp2-Gria3 PCs during the first 60 ms after air-puff onset: p=0 . 004; Tukey’s post-hoc test after ANOVA ( p=0 . 002 , F = 6 . 681 , df = 2 ) , n = 35 WT PCs and n = 13 Gria3 KO PCs , next to n = 23 Pcp2-Ppp3r1 KO PCs , all with weak complex spike responses ) ( Figure 8D–F ) . Thus an independent line of evidence supports the findings made in the Pcp2-Ppp3r1 mice . For control we compared the basic electrophysiological profiles of PCs in the three genotypes used in this study . When averaged over the entire period with episodes of stimulation , the overall complex spike rate , simple spike rate and simple spike CV2 value ( i . e . parameter for level of irregularity ) of PCs in the Pcp2-Ppp3r1 KO mice were moderately , but significantly , reduced compared to those in WTs ( Figure 8—figure supplement 1A–D; Supplementary file 1C ) . However , as the Pcp2-Gria3 mice did not show any significant deviations in these overall firing properties ( Figure 8—figure supplement 1A–D; Supplementary file 1C ) , it is unlikely that the aberrant firing properties of Pcp2-Ppp3r1 mice could explain the lack of adaptation at both the behavioral and electrophysiological level . Comparing the response probabilities to whisker pad stimulation we found that both the number of complex spikes and simple spikes after the air-puff were reduced in Pcp2-Ppp3r1 mice ( Figure 8—figure supplement 1E–J; Supplementary file 1C ) . The predominantly suppressive simple spike responses were not found in Pcp2-Gria3 mice , but the latter also had a reduced complex spike response to air-puff stimulation . Since a reduced complex spike response acts permissive for the adaptive increase in the simple spike response , it is unlikely that the observed reduction in complex spike firing would be the cause of the observed lack of simple spike enhancement in both mutant mouse lines . Moreover , the amplitudes of the touch-induced whisker protraction as measured before the induction phase were similar between the WT and the mutant mice ( Figure 8—figure supplement 2 ) . We therefore conclude that the absence of simple spike potentiation and the concomitant increase in touch-induced whisker protraction is likely due to the absence of parallel fiber LTP caused by the genetic mutations rather than to altered firing patterns of the PCs involved . Although most mammals have whiskers , only few species use their whiskers to actively explore their environment by making fast , rhythmic whisker movements ( Vincent , 1913; Ahl , 1986; Welker , 1964; Woolsey et al . , 1975 ) . In ‘whisking’ animals , such as mice and rats , both whisker protraction and retraction are under direct muscle control , while especially whisker retraction can additionally reflect a passive process involving skin elasticity ( Berg and Kleinfeld , 2003; Simony et al . , 2010; Haidarliu et al . , 2015; Moore et al . , 2013; Deschênes et al . , 2016 ) . Animals can modify the pattern of whisker movements upon sensory feedback during natural behavior , as has been demonstrated for example during gap crossing and prey capture ( Anjum and Brecht , 2012; Voigts et al . , 2015 ) . The neural control of adaptation of reflexive whisker movements is still largely unknown . Given the widespread networks in the brain controlling whisker movements ( Bosman et al . , 2011; Kleinfeld et al . , 1999 ) , it is likely that multiple brain regions contribute . We show here , at least for a specific reproducible form of whisker adaptation , that parallel fiber to PC LTP and enhancement in PC simple spike activity may contribute to the induction and expression of this form of motor learning , respectively . Our electrophysiological recordings indicate that the simple spike activity correlates well with whisker protraction , especially in PCs located in crus 2 , and that this relation is context-dependent . Under the baseline condition of our paradigm , during the 0 . 5 Hz whisker pad stimulation , simple spikes correlate positively with the position of the whiskers during protraction on a single-trial basis . The correlation between the rate of simple spikes and that of protraction was also found when comparing the impact of contralateral vs . ipsilateral whisker pad stimulation . The absence of a clear time lag or lead between simple spike activity and whisker movements under this condition suggests that during normal motor performance without sensorimotor mismatch signaling the simple spikes predominantly represent ongoing movement . Our data under baseline conditions are compatible with those obtained by the labs of Chadderton and Léna ( Chen et al . , 2016; Proville et al . , 2014 ) . In their studies on online motor performance , the simple spike activity of most PCs in the lateral crus 1 and/or crus 2 regions correlated best with protraction of the set point , defined as the slowly varying midpoint between maximal protraction and maximal retraction . During and after training with 4 Hz air-puff stimulation the temporal dynamics of the simple spikes shifted in that the simple spikes were found to precede the whisker movement and to predict the magnitude of the protraction , suggesting the emergence of an instructive motor signal . Optogenetic stimulation experiments confirmed that increased simple spike firing during the early phase of touch-induced whisker protraction can promote whisker protraction . Thus , the current dataset confirms and expands on previous studies , highlighting a role of the cerebellar PCs injecting additional accelerating and amplifying signals into the cerebellar nuclei during entrainment ( De Zeeuw et al . , 1995 ) . Synaptic plasticity in the cerebellar cortex has , next to that in the cerebellar and vestibular nuclei ( Lisberger and Miles , 1980; Lisberger , 1998; Zhang and Linden , 2006; McElvain et al . , 2010 ) , generally been recognized as one of the major mechanisms underlying motor learning ( Ito , 2001; Ito , 1989 ) . For forms of motor learning that require a decrease in simple spike activity for expression of the memory , such as eyeblink conditioning ( Halverson et al . , 2015; ten Brinke et al . , 2015; Jirenhed et al . , 2007 ) , long-term depression ( LTD ) of the parallel fiber to PC synapse may play a role during the initial induction stage ( Ito , 1989; Koekkoek et al . , 2003 ) . In LTD-deficient mouse models the potential contribution of LTD is most apparent when compensatory mechanisms that involve activation of the molecular layer interneurons are blocked ( Boele et al . , 2018 ) . However , for forms of motor learning that require an increase in simple spike activity for expression of the procedural memory it is less clear which forms of cerebellar cortical plasticity may contribute . Here , we show that increasing whisker protraction by repetitive sensory stimulation requires an increase in simple spike activity and that blocking induction of parallel fiber to PC LTP prevents changes in both spiking and motor activity following the same training paradigm . Possibly , adaptation of the vestibulo-ocular reflex ( VOR ) follows partly similar learning rules in that various genetic mouse models with impaired induction of parallel fiber to PC LTP show reduced VOR learning ( Gutierrez-Castellanos et al . , 2017; Rahmati et al . , 2014; Schonewille et al . , 2010; Ly et al . , 2013; Peter et al . , 2016 ) and that optogenetic stimulation of PCs in the flocculus of the vestibulocerebellum can increase VOR gain ( Voges et al . , 2017 ) . In this respect , it will be interesting to find out to what extent increases in simple spike activity in the flocculus can also be correlated with an entrained increase in VOR gain on a trial-by-trial basis , as we show here for whisker learning . The differential learning rules highlighted above indicate that different forms of cerebellar plasticity may dominate the induction of different forms of learning ( Hansel et al . , 2001; Gao et al . , 2012; D'Angelo et al . , 2016; De Zeeuw and Ten Brinke , 2015 ) . The engagement of these rules may depend on the requirements of the downstream circuitries involved ( Suvrathan et al . , 2016; De Zeeuw and Ten Brinke , 2015 ) . Indeed , whereas the eyeblink circuitry downstream of the cerebellar nuclei comprises purely excitatory connections and hence requires a simple spike suppression of the inhibitory PCs to mediate closure of the eyelids , the VOR circuitry comprises an additional inhibitory connection and hence requires a simple spike enhancement so as to increase the compensatory eye movement ( De Zeeuw and Ten Brinke , 2015; Voges et al . , 2017 ) . The circuitry downstream of the cerebellum that mediates control of whisker movements is complex ( Bosman et al . , 2011 ) . Possibly , the cerebellar nuclei may modulate the trigemino-facial feedback loop in the brainstem that controls the touch-induced whisker protraction ( Bellavance et al . , 2017 ) . This could be done via the intermediate reticular formation , which receives a direct input from the cerebellar nuclei ( Teune et al . , 2000 ) and projects to the facial nucleus where the whisker motor neurons reside ( Zerari-Mailly et al . , 2001; Herfst and Brecht , 2008 ) . As the latter projection is inhibitory ( Deschênes et al . , 2016 ) , the same configuration may hold as described for the VOR pathways ( De Zeeuw and Ten Brinke , 2015 ) in that adaptive enhancement of the whisker reflex may require induction of parallel fiber to PC LTP and increases in simple spike activity . Thus , given the current findings and the known neuro-anatomical connections in the brainstem , the picture emerges that cerebellar control of whisker movements follows the general pattern which suggests that the predominant forms of PC plasticity and concomitant changes in simple spike activity align with the requirements downstream in the cerebellar circuitry ( De Zeeuw and Ten Brinke , 2015 ) . A minority of the PCs we recorded had a high complex spike response probability upon air-puff stimulation of the whisker pad . These PCs were predominantly located in the centro-lateral part of crus 1 . Most of the other PCs , in particular those in the medial part of crus 1 and in crus 2 , showed a low , yet significant , complex spike response probability to sensory whisker stimulation . In these cells the absence of a strong complex spike response to air-puff stimulation probably acted as a permissive gate to increase the simple spike response following training , which is in line with current theories on cerebellar plasticity ( Coesmans et al . , 2004; Ito , 2001; Lev-Ram et al . , 2002; Ohtsuki et al . , 2009 ) . The PCs with a relatively high complex spike response probability were not prone for increases in simple spike activity following our training protocol . Instead , they may dynamically enhance reflexive whisker protraction through increases in their coherent complex spike firing , likely engaging ensemble encoding ( Hoogland et al . , 2015; Mukamel et al . , 2009; Ozden et al . , 2009; Schultz et al . , 2009 ) . This enhancement does not require a repetitive training protocol and also occurs during single trial stimulation . Indeed , these complex spike responses , which tended to precede the active whisker movement , could be correlated to the strength of the touch-induced whisker protraction under baseline conditions . This is in line with previous studies showing that complex spikes can facilitate the initiation of movements and define their amplitude ( Hoogland et al . , 2015; Kitazawa et al . , 1998; Welsh et al . , 1995 ) . Thus , PCs with strong complex spike responses to whisker stimulation – especially those located in the D2 zone of crus 1 – show poor simple spike enhancement to mediate whisker adaptation , but they might facilitate execution of touch-induced whisker protraction under baseline conditions by relaying coherent patterns of complex spikes onto the cerebellar nuclei neurons . Based on a known form of reflexive whisker movements , we introduced a novel adaptation paradigm and investigated the underlying cerebellar plasticity mechanism and spiking learning rules . A brief period of increased sensory input appeared to be sufficient to induce a lasting impact on touch-induced whisker protraction: the whisker reflex started earlier and had a bigger amplitude . This motor adaptation probably requires induction of parallel fiber LTP in PCs that can be identified by their weak but present complex spike response to sensory stimulation . The resultant increased simple spike firing of these PCs may affect the brainstem loop controlling touch-induced whisker protraction via the reticular formation in the brainstem , in line with optogenetic stimulation experiments . Thus , our study proposes induction of parallel fiber to PC LTP as a cellular mechanism for enhancing PC simple spike responsiveness that facilitates the expression of the entrained whisker protraction . For most of the experiments in this study , we used two different mutant mouse lines on a C57BL/6J background . Comparisons of electrophysiological parameters were always made between the mutant mice and their respective wild-type ( WT ) littermates , although for easier visualization the WTs were sometimes grouped as indicated in the figure legends . Both mouse lines had been used before and details on their generation have been published . Briefly , Pcp2-Ppp3r1 mice ( Tg ( Pcp2-cre ) 2Mpin;Ppp3r1tm1Stl ) lacked functional phosphatase 2B ( PP2B ) specifically in their PCs . They were created by crossing mice in which the gene for the regulatory subunit ( Cnbl1 ) of PP2B was flanked by loxP sites ( Zeng et al . , 2001 ) with transgenic mice expressing Cre-recombinase under control of the Pcp2 ( L7 ) promoter ( Barski et al . , 2000 ) as described in Schonewille et al . , 2010 . Pcp2-cre+/--Ppp3r1f/f mice ( ‘Pcp2-Ppp3r1 mice’ ) were compared with Pcp2-cre-/--Ppp3r1f/f littermate controls . We used 35 WT mice ( 17 males and 18 females of 21 ± 9 weeks of age ( average ±s . d . ) ) and 22 Pcp2-Ppp3r1 mice ( 6 males and 16 females of 18 ± 10 weeks of age ( average ±s . d . ) ) . Pcp2-Gria3 mice ( Tg ( Pcp2-cre ) 2Mpin;Gria3tm2Rsp ) lacked the AMPA receptor GluA3 subunit specifically in their PCs . They were created by crossing mice in which the Gria3 gene was flanked by loxP sites ( Sanchis-Segura et al . , 2006 ) with transgenic mice expressing Cre-recombinase under control of the Pcp2 promoter ( Barski et al . , 2000 ) as described in Gutierrez-Castellanos et al . , 2017 . We used Pcp2-cre+/--Gria3f/f mice ( ‘Pcp2-Gria3 mice’ ) and Pcp2-cre-/--Gria3f/f as littermate controls . We used 5 WT male mice ( 25 ± 3 weeks of age ( average ±s . d . ) ) and 9 Pcp2-Gria3 mice ( 6 males and 3 females of 26 ± 4 weeks of age ( average ±s . d . ) ) . Mutants and wild-types were measured in random sequence . For the two-photon Ca2+ imaging experiments , we used six male C57BL/6J mice ( Charles Rivers , Leiden , the Netherlands ) of 4–12 weeks of age . The photostimulation experiments were performed on seven mice ( 3 males and 4 females of 25 ± 1 weeks of age ( average ±s . d . ) ) expressing Channelrhodopsin-2 exclusively in their PCs ( Tg ( Pcp2-cre ) 2Mpin;Gt ( ROSA ) 26Sortm27 . 1 ( CAG-COP4*H134R/tdTomato ) Hze ) as described previously ( Witter et al . , 2013 ) . All mice were socially housed until surgery and single-housed afterwards . The mice were kept at a 12/12 hr light/dark cycle and had not been used for any invasive procedure ( except genotyping shortly after birth ) before the start of the experiment . All mice used were specific-pathogen free ( SPF ) . All experimental procedures were approved a priori by an independent animal ethical committee ( DEC-Consult , Soest , The Netherlands ) as required by Dutch law and conform the relevant institutional regulations of the Erasmus MC and Dutch legislation on animal experimentation . Permissions were obtained under the following license numbers: EMC2656 , EMC2933 , EMC2998 , EMC3001 , EMC3168 and AVD101002015273 . All mice that were used for electrophysiology received a magnetic pedestal that was attached to the skull above bregma using Optibond adhesive ( Kerr Corporation , Orange , CA ) and a craniotomy was made on top of crus 1 and crus 2 . The surgical procedures were performed under isoflurane anesthesia ( 2–4% V/V in O2 ) . Post-surgical pain was treated with 5 mg/kg carprofen ( ‘Rimadyl’ , Pfizer , New York , NY , USA ) , 1 µg lidocaine ( Braun , Meisingen , Germany ) , 1 µg bupivacaine ( Actavis , Parsippany-Troy Hills , NJ ) and 50 µg/kg buprenorphine ( ‘Temgesic’ , Indivior , Richmond , VA ) . After 3 days of recovery , mice were habituated to the recording setup during at least two daily sessions of approximately 45 min each . In the recording setup , they were head-fixed using the magnetic pedestal . The mice used for two-photon imaging received a head plate with a sparing on the location of the craniotomy instead of a pedestal . The head plate was attached to the skull with dental cement ( Superbond C and B , Sun Medical Co . , Moriyama City , Japan ) . To prevent the growth of scar tissue , which could affect image quality , two-photon recordings were made on the day of the surgery ( recording started at least 1 hr after the termination of anesthesia ) . Air-puff stimulation to the whisker pad was applied with a frequency of 0 . 5 Hz s at a distance of approximately 3 mm at an angle of approximately 35° ( relative to the body axis ) . The puffs were delivered using a tube with a diameter of approximately 1 mm with a pressure of ~2 bar and a duration of 30 ms . During the induction period , the stimulation frequency was increased to 4 Hz and 80 puffs were given . In a subset of experiments , a 2 ms air-puff ( pre-pulse ) was delivered 100 ms prior to the 30 ms puff . Videos of the whiskers were made from above using a bright LED panel as backlight ( λ = 640 nm ) at a frame rate of 1 , 000 Hz ( 480 × 500 pixels using an A504k camera from Basler Vision Technologies , Ahrensburg , Germany ) . The whiskers were not trimmed or cut . Electrophysiological recordings were performed in awake mice using either glass pipettes ( 3–6 MΩ ) or quartz-coated platinum/tungsten electrodes ( 2–5 MΩ , outer diameter = 80 µm , Thomas Recording , Giessen , Germany ) . Unless specified otherwise , recordings were made in Pcp2-Ppp3r1 WT mice . The latter electrodes were placed in an 8 × 4 matrix ( Thomas Recording ) , with an inter-electrode distance of 305 µm . Prior to the recordings , the mice were lightly anesthetized with isoflurane to remove the dura , bring them in the setup and adjust all manipulators . Recordings started at least 60 min after termination of anesthesia and were made in crus 1 and crus 2 ipsilateral to the side of the whisker pad stimulation at a minimal depth of 500 µm . The electrophysiological signal was digitized at 25 kHz , using a 1–6000 Hz band-pass filter , 22x pre-amplified and stored using a RZ2 multi-channel workstation ( Tucker-Davis Technologies , Alachua , FL ) . For the neural tracing experiments , we used glass electrodes filled with 2 M NaCl for juxtacellular recordings . After a successful recording of a PC , neural tracer was pressure injected ( 3 × 10 ms with a pressure of 0 . 7 bar ) either from the same pipette re-inserted at the same location or from the second barrel or a double barrel pipette . We used a gold-lectin conjugate has described previously ( Ruigrok et al . , 1995 ) ( n = 3 ) or biotinylated dextran amine ( BDA ) 3000 ( 10 mg/ml in 0 . 9% NaCl; ThermoFisher Scientific , Waltham , MA ) ( n = 7 ) . Five days after the tracer injection , the mice were anesthetized with pentobarbital ( 80 mg/kg intraperitoneal ) and fixated by transcardial perfusion with 4% paraformaldehyde . The brains were removed and sliced ( 40 µm thick ) . The slices were processed by Nissl staining . Experiments were included in the analysis if the electrophysiology fulfilled the requirements mentioned below with a recording duration of at least 50 s and if the tracer was clearly visible . For BDA 3000 , this implied that it was taken up by the PCs at the injection spot and transported to the axonal boutons a single subgroup in the cerebellar nuclei . BDA 3000 was also found in the inferior olive . For the gold-lectin conjugate the subnucleus of the inferior olive was considered . Based upon the subnuclei of the cerebellar nuclei and/or the inferior olive , the sagittal zone of the recording site was identified according to the scheme published in Apps and Hawkes , 2009 . After the recordings made with the quartz/platinum electrodes , electrolytic lesions were applied to selected electrodes in order to retrieve the recording locations . To this end , we applied a DC current of 20 µA for 20 s . This typically resulted in a lesion that could be visualized after Nissl staining of 40 µm thick slices made of perfused brains . We accepted a spot as a true lesion if it was visible in at least two consecutive slices at the same location . In total , we could retrieve 16 successful lesions . Recording locations were approximated using pictures of the entry points of the electrodes in combination with the locations of the lesions . After the surgery ( see above ) with the dura mater intact , the surface of the cerebellar cortex was rinsed with extracellular solution composed of ( in mM ) 150 NaCl , 2 . 5 KCl , 2 CaCl2 , 1 MgCl2 and 10 HEPES ( pH 7 . 4 , adjusted with NaOH ) . After a 30 min recovery period from anesthesia , animals were head-fixed in the recording setup and received a bolus-loading of the cell-permeant fluorescent Ca2+ indicator Cal-520 AM ( 0 . 2 mM; AAT Bioquest , Sunnyvale , CA , USA ) . The dye was first dissolved with 10% w/V Pluronic F-127 in DMSO ( Invitrogen ) and diluted 20x in the extracellular solution . The dye solution was pressure injected into the molecular layer ( 50–80 μm below the surface ) at 0 . 35 bar for 5 min . After dye loading , the brain surface was covered with 2% agarose dissolved in saline ( 0 . 9% NaCl ) in order to reduce motion artefacts and prevent dehydration . Starting at least 30 min after dye injection , in vivo two-photon Ca2+ imaging was performed of the molecular layer using a setup consisting of a titanium sapphire laser ( Chameleon Ultra , Coherent , Santa Clara , CA ) , a TriM Scope II system ( LaVisionBioTec , Bielefeld , Germany ) mounted on a BX51 microscope with a 20 × 1 . 0 NA water immersion objective ( Olympus , Tokyo , Japan ) and GaAsP photomultiplier detectors ( Hamamatsu , Iwata City , Japan ) . A typical recording sampled 40 × 200 µm with a frame rate of approximately 25 Hz . We included all mice measured during this study , with the exception of one mouse where video-analysis revealed that the air-puff was delivered more to the nose than to the whisker pad . Single-unit data was included if the recording was of sufficient quality and reflected the activity of a single PC according to the rules defined below ( see section Electrophysiological analysis ) . Whisker movements were tracked offline as described previously ( Rahmati et al . , 2014 ) using a method based on the BIOTACT Whisker Tracking Tool ( Perkon et al . , 2011 ) . We used the average angle of all trackable large facial whiskers for further quantification of whisker behavior . The impact of 4 Hz air-puff stimulation on air-puff-triggered whisker movement was quantified using a bootstrap method . First , we took the last 100 trials before induction and divided these randomly in two series of 50 . We calculated the differences in whisker position between these two series , and repeated this 1000 times . From this distribution we derived the expected variation after whisker pad air-puff stimulation . We took the 99% confidence interval as the threshold to which we compared the difference between 50 randomly chosen trials after and 50 randomly chosen trials before induction . Spikes were detected offline using SpikeTrain ( Neurasmus , Rotterdam , The Netherlands ) . A recording was considered to originate from a single PC when it contained both complex spikes ( identified by the presence of stereotypic spikelets ) and simple spikes , when the minimal inter-spike interval of simple spikes was 3 ms and when each complex spike was followed by a pause in simple spike firing of at least 8 ms . The regularity of simple spike firing was expressed as the local variation ( CV2 ) and calculated as 2|ISIn+1-ISIn|/ ( ISIn+1+ISIn ) with ISI = inter simple spike interval ( Shin et al . , 2007 ) . Only single-unit recordings of PCs with a minimum recording duration of 200 s were selected for further analysis . However , for the neural tracing experiments ( see above ) , on which no quantitative analysis was performed , we accepted a minimum recording duration of 50 s . Image analysis was performed offline using custom made software as described and validated previously ( Ozden et al . , 2008; Ozden et al . , 2012; De Gruijl et al . , 2014 ) . In short , we performed independent component analysis to define the areas of individual PC dendrites ( Figure 3—figure supplement 2A ) . The fluorescent values of all pixels in each region of interest were averaged per frame . These averages were plotted over time using a high-pass filter . A 8% rolling baseline was subtracted with a time window of 0 . 5 ms ( Ozden et al . , 2012 ) . Ca2+ transients were detected using template matching . For the aggregate peri-stimulus time histograms ( PSTHs ) , we calculated per individual frame the number of complex spikes detected and made a PSTH color coding the number of simultaneously detected complex spikes . Based on the total number of complex spikes and dendrites per recording , we calculated the expected number of simultaneous complex spikes per individual frame based upon a Poisson distribution . The actual number of simultaneous complex spikes was compared to this calculated distribution and a p value was derived for each number based upon the Poisson distribution . For each PC recording , we constructed PSTHs of complex spikes and simple spikes separately using a bin size of 10 ms for display purposes . For further quantitative analyses of the PSTHs , we used a bin size of 1 ms and convolved them with a 21 ms wide Gaussian kernel . Complex spike responses were characterized by their peak amplitude , defined as the maximum of the convolved PSTH and expressed in percentage of trials in which a complex spike occurred within a 1 ms bin . Latencies were taken as the time between stimulus onset and the time of the response peak , as determined from the convolved PSTH . For some analyses , we discriminated between the sensory response period ( 0–60 ms after stimulus onset ) and inter-trial interval ( 500 to 200 ms before stimulus onset ) . We considered a PC responsive for sensory stimulation if the peak or trough in the PSTH in the 60 ms after the stimulus onset exceeded the threshold of 3 s . d . above or below the average of the pre-stimulus interval ( 1 ms bins convolved with a 21 ms Gaussian kernel , pre-stimulus interval 200 ms before stimulus onset ) . Long-term stability of electrophysiological recordings was verified by heat maps of time-shifted PSTHs . The time-shifted PSTH was processed by calculating the simple spike PSTH for 20 air-puffs per row , which were shifted by five air-puffs between neighboring rows . The simple spike rates per row are calculated at 1 ms resolution and convolved with a 21 ms Gaussian kernel and color-coded relative to baseline firing rate ( −1000 to −200 ms relative to air-puff time ) . A principal component analysis showed that the heterogeneity among the sensory complex spike responses was driven almost exclusively by one parameter , the maximum amplitude peak of the convolved complex spike PSTH . We performed a univariate Gaussian mixture model using only that variable . The Bayesian information criterion ( BIC ) indicated that the model with two components with unequal variances yielded the best approximation of the data . Then we applied the function Mclust ( data ) in R ( R Foundation , Vienna , Austria ) which use the expectation-maximization algorithm in order to assert the main parameters of the resulting models ( probability , mean and variance of each population ) . Trial-by-trial correlation between instantaneous simple spike firing rate and whisker position was performed as described before ( ten Brinke et al . , 2015 ) . In short: spike density functions were computed for all trials by convolving spike occurrences across 1 ms bins with an 8 ms Gaussian kernel . Both spike and whisker data were aligned to the 200 ms baseline . For cell groups , data was standardized for each cell for each correlation , and then pooled . The spike-whisker Pearson correlation coefficient R was calculated in bins of 10 ms , resulting in a 40 × 40 R-value matrix showing correlations for −100 to 300 ms around the air-puff presentation . Group sizes of the blindly acquired data sets were not defined a priori as the effect size and variation were not known beforehand . A post-hoc power calculation based upon the results of the potentiation of the PC responses to whisker pad stimulation of the ‘weak complex spike responders’ indicated a minimum group size of 12 PCs ( α = 5% , β = 20% , Δ = 9 . 65% , s . d . = 10 . 59% , paired t test ) . This number was obtained for the ‘weak complex spike responders’ in WT ( n = 35 ) , Pcp2-Ppp3r1 ( n = 21 ) and Pcp2-Gria3 PCs ( n = 13 ) , as well as for the relatively rare ‘strong complex spike responders’ in WT mice ( n = 20 ) . This was further substantiated by other independent analyses , including ANOVA and linear regression , as described in the Results section . Variations in success rate , especially considering recordings of longer duration in combination with video tracking , explain why some groups are larger than others . Data was excluded only in case of a signal to noise ratio that was insufficient to warrant reliable analysis . For data visualization and statistical analysis , we counted the number of PCs as the number of replicates for the spike-based analyses and the number of mice for the behavior-based analyses . We tested whether the observed increase in coherence after sensory stimulation ( Figure 3—figure supplement 2D–G ) was more than expected from the increased firing rate induced by the stimulation . The expected coherence based on the firing rate was calculated from 1000 bootstrapped traces from the inhomogeneous Poisson spike trains made for each neuron . The resultant distribution was compared to the measured distribution using a two-sample Kolmogorov-Smirnov test . Stacked line plots were generated by cumulating the values of all subjects per time point . Thus , the first line ( darkest color ) represents the first subject , the second line the sum of the first two , the third line the sum of the first three , etcetera . The data were divided by the number of subjects , so that the last line ( brightest color ) represents , next to the increase from the one but last value , also the population average . Sample size and measures for mean and variation are specified throughout the text and figure legends . For normally distributed data ( as evaluated using the Kolmogorov-Smirnov test ) parametric tests were used . Comparisons were always made with 2-sided tests when applicable . Unpaired t test were always made with Welch correction for possible differences in s . d . . The data for the box plots is available as Source Data Files . Custom written Matlab code to complement the whisker tracking analysis by the BIOTACT Whisker Tracking Tool was used as described previously ( Rahmati et al . , 2014 ) and is available via GitHub ( Spanke and Negrello , 2018; copy archived at https://github . com/elifesciences-publications/BWTT_PP ) .

Rodents use their whiskers to explore the world around them . When the whiskers touch an object , it triggers involuntary movements of the whiskers called whisker reflexes . Experiencing the same sensory stimulus multiple times enables rodents to fine-tune these reflexes , e . g . , by making their movements larger or smaller . This type of learning is often referred to as motor learning . A part of the brain called cerebellum controls motor learning . It contains some of the largest neurons in the nervous system , the Purkinje cells . Each Purkinje cell receives input from thousands of extensions of small neurons , known as parallel fibers . It is thought that decreasing the strength of the connections between parallel fibers and Purkinje cells can help mammals learn new movements . This is the case in a type of learning called Pavlovian conditioning . It takes its name from the Russian scientist , Pavlov , who showed that dogs can learn to salivate in response to a bell signaling food . Pavlovian conditioning enables animals to optimize their responses to sensory stimuli . But Romano et al . now show that increasing the strength of connections between parallel fibers and Purkinje cells can also support learning . To trigger reflexive whisker movements , a machine blew puffs of air onto the whiskers of awake mice . After repeated exposure to the air puffs , the mice increased the size of their whisker reflexes . At the same time , their Purkinje cells became more active and the connections between Purkinje cells and parallel fibers grew stronger . Artificially increasing Purkinje cell activity triggered the same changes in whisker reflexes as the air puffs themselves . Textbooks still report that only weakening of connections within the cerebellum enables animals to learn and modify movements . The data obtained by Romano al . thus paint a new picture of how the cerebellum works in the context of whisker learning . They show that strengthening these connections can also support movement-related learning .

2018

Potentiation of cerebellar Purkinje cells facilitates whisker reflex adaptation through increased simple spike activity

Insulin secretion from β-cells is reduced at the onset of type-1 and during type-2 diabetes . Although inflammation and metabolic dysfunction of β-cells elicit secretory defects associated with type-1 or type-2 diabetes , accompanying changes to insulin granules have not been established . To address this , we performed detailed functional analyses of insulin granules purified from cells subjected to model treatments that mimic type-1 and type-2 diabetic conditions and discovered striking shifts in calcium affinities and fusion characteristics . We show that this behavior is correlated with two subpopulations of insulin granules whose relative abundance is differentially shifted depending on diabetic model condition . The two types of granules have different release characteristics , distinct lipid and protein compositions , and package different secretory contents alongside insulin . This complexity of β-cell secretory physiology establishes a direct link between granule subpopulation and type of diabetes and leads to a revised model of secretory changes in the diabetogenic process . Blood glucose levels are maintained in a narrow range by secretion of the hormone insulin from pancreatic β-cells . Insufficient insulin secretion leads to elevated levels of blood glucose resulting in diabetes ( Rorsman and Ashcroft , 2018 ) . Type 1 diabetes ( T1D ) results primarily from immune-mediated killing of pancreatic β-cells . In contrast , type 2 diabetes ( T2D ) arises when peripheral resistance to insulin signaling causes persistent demand for insulin secretion and eventual β-cell exhaustion ( Guthrie and Guthrie , 2004; Rorsman and Ashcroft , 2018 ) . Understanding molecular deficiencies of the secretory pathway in diabetes is essential for identifying the underlying causes of the disease . Secretory defects have been observed in diabetic human islet cells . Treatments that mimic these defects have been identified and used to model diabetes in healthy human cells , rodent islets , and immortalized cell lines ( Aslamy et al . , 2018a; Aslamy et al . , 2018b; Gandasi and Barg , 2014; Hoppa et al . , 2009; Olofsson et al . , 2007 ) . Onset of T1D is modeled by treatment of insulin-secreting cells with proinflammatory cytokines TNF-α , INF-γ , and IL-1β , which leads to decreased insulin release and a loss of Doc2B , an established biomarker for the disease ( Aslamy et al . , 2018a; Aslamy et al . , 2018b ) . T2D is associated with extended exposure to elevated free fatty acids ( FFA ) caused by high fat diets ( Grill and Qvigstad , 2000 ) . The lipotoxicity component of T2D is frequently modeled by sustained treatment of insulin-secreting cells with palmitate , resulting in reduced glucose-stimulated insulin secretion ( Hoppa et al . , 2009; Sako and Grill , 1990 ) . Stimulated secretion of insulin normally occurs in two sequential phases , an acute first phase and a sustained second phase , which are differently affected in the T1D and T2D models . Modeling the inflammation of T1D by cytokine treatment causes a defect in insulin secretion attributed to the loss of Doc2B ( Aslamy et al . , 2018a; Aslamy et al . , 2018b ) , which primarily suppresses the second phase of secretion ( Ramalingam et al . , 2012 ) . Conversely , modeling T2D by palmitate treatment strongly decreases the first phase ( Rorsman and Ashcroft , 2018 ) . These observations show distinct insulin secretion profiles in the two diseases , and understanding the molecular and cellular mechanisms of how these different outcomes arise would provide a fundamentally new molecular view on the pathology of diabetes . Although the secretory defects leading to T1D and T2D are well established , the cell and molecular biology that underlies the phenotypes of T1D and T2D model treatments is not known . To address this fundamental gap in knowledge , we performed large-scale purification of secretory granules on either untreated , palmitate-treated , or cytokine-treated insulin-secreting cells . These granules were characterized comparatively for SNARE-mediated fusion properties in single particle fusion assays with SNARE-containing reconstituted target membranes . This novel cellular dissection and reconstitution approach , which was corroborated in intact insulin-secreting cells undergoing the same treatments , revealed a previously unknown heterogeneity among insulin granules with different subpopulations of secretory vesicles being lost following one or the other of the two diabetes-mimicking treatments . Further study showed remarkable differences in size , composition , and fusion characteristics of these subpopulations . As a result , we were able to identify a strong correlation between the distinct contents of the two subpopulations and known changes in β-cell output of signaling molecules by pancreatic islets during the onset of each type of diabetes . Taken together , these studies lead to a revised model about how the organization and regulation of two insulin secretory pathways potentially impact physiological intra-islet communication and pathological changes that accompany the diabetogenic process . While secretion of insulin is known to be compromised in diabetes , we have been interested in exploring whether the diabetogenic process might entail detectable mechanistic changes in how insulin granules fuse during exocytosis . To address this question , insulin granules were purified by iso-osmotic density centrifugation ( Kreutzberger et al . , 2019 ) from INS 1 cells that stably express human proinsulin with GFP inserted into the C-peptide domain and that are referred to as GRINCH cells ( Haataja et al . , 2013 ) and used for biochemical characterization and reconstituted fusion assays . Binding and fusion of purified granules with planar supported membranes containing the SNARE proteins syntaxin-1a and SNAP-25 and lipids that mimic the plasma membrane in the presence of Munc18 and complexin were recorded by total internal reflection fluorescence ( TIRF ) microscopy in a reconstituted single particle fusion assay by monitoring the release of C-peptide-GFP ( Figure 1A , Figure 1—figure supplement 1A and D ) . In this assay ( Kreutzberger et al . , 2017a ) , full-length syntaxin 1a and SNAP-25 are reconstituted into planar supported membranes and preincubated with Munc18 and complexin . Purified insulin granules are docked in the absence of calcium in a SNARE-specific manner ( Figure 1—figure supplement 1B ) . In the absence of calcium , fusion is largely suppressed in the presence of Munc18 and complexin ( Figure 1—figure supplement 1C ) . However , when calcium is injected , fusion occurs as monitored by a spike and decay in C-peptide-GFP TIRF fluorescence ( Figure 1—figure supplement 1D ) . When triggered with 100 μM calcium , approximately 60% of bound granules fuse after a short time delay ( Figure 1—figure supplement 1D and E ) and the response is sensitive to the calcium concentration ( Figure 1—figure supplement 1F ) . Quantification of fusion as a function of calcium concentration revealed a biphasic response with granules purified from untreated cells ( Figure 1B , green , control ) . Granules purified from cells following a 72 hr treatment with palmitate ( delivered as a BSA complex and used to model elevated free fatty acid [FFA] observed during T2D Rorsman and Ashcroft , 2018 ) exhibited a monophasic high-affinity calcium response ( K1/2 = 12 ± 2 µM , Figure 1B , blue ) . In contrast , granules purified from cells following a 20 hr treatment with cytokines TNF-α , INF-γ , and IL-1β that model the inflammation associated with the onset of T1D ( Aslamy et al . , 2018a; Aslamy et al . , 2018b ) showed a shift to a low-affinity monophasic response ( K1/2 = 41 ± 4 µM , Figure 1B , red ) . Overexpression of Doc2B prior to cytokine treatment , which had previously been shown to reverse cytokine effects on β-cells ( Aslamy et al . , 2018b ) , prevented the shift of the calcium dose-response ( Figure 1C , purple ) . Fusion events of granules labeled with a soluble fluorescent content marker have a distinctive shape ( Figure 1D , Figure 1—figure supplement 1D , Figure 1—figure supplement 2 and see Kreutzberger et al . , 2017a; Kreutzberger et al . , 2017b; Kreutzberger et al . , 2019 ) . This is characterized by a decrease in fluorescence at the onset of fusion , a rise in fluorescence as the marker is pulled forward into the TIRF field as the granule membrane collapses into the supported membrane , and a continued decrease in fluorescence as the content probe diffuses away from the site of fusion . A two-step mathematical model can be used to describe the events in which the intensity change ( ΔIC ) indicates the amount of content released when the vesicle collapses into the supported membrane ( Figure 1—figure supplement 2 ) and ( Kreutzberger et al . , 2017a; Kreutzberger et al . , 2017b; Kreutzberger et al . , 2019 ) . Averaging the fluorescent signals from multiple fusion events revealed that insulin granules had distinct slow and fast release characteristics for C-peptide-GFP release ( Figure 1D ) . Quantifying individual granule fusion events from untreated cells showed slow release modes ( smaller ΔIC ) at all calcium concentrations ( blue ) , while fast release modes appeared at higher calcium concentrations ( Figure 1E ) . Granules purified following diabetic model treatments exhibited only slow release modes from palmitate-treated cells and only fast release modes from cytokine-treated cells when fusion was assayed at 100 μM calcium ( Figure 1F ) . Overexpression of Doc2B restored the slow release mode of granules purified from cytokine-treated cells ( Figure 1F ) . Corresponding fast and slow release events were observed when analyzing secretion of C-peptide-GFP in intact GRINCH cells ( Figure 2 and Figure 2—figure supplement 1 ) . Opening of voltage-dependent calcium channels increases intracellular calcium levels , which trigger granule exocytosis ( Rorsman and Ashcroft , 2018 ) . Increased depolarization , which increased calcium influx ( Figure 2—figure supplement 1A ) , decreased the duration of C-peptide release events from intact cells ( Figure 2B and C and Figure 2—figure supplement 1B and C ) corresponding to fast and slow fusion events observed in the reconstitution assay ( Figure 1D ) . At both high and low stimulation strengths , palmitate treatment increased release duration whereas cytokine treatment decreased release duration ( Figure 2C and Figure 2—figure supplement 1C ) corresponding to the shifts in release modes observed in the reconstitution assay ( Figure 1F ) . Purified insulin granules were further examined for treatment-induced morphology changes using cryo-EM ( Figure 1G ) . Granules had a diameter of 195 ± 61 nm when purified from untreated cells , 173 ± 43 nm when purified from palmitate-treated cells , and 222 ± 53 nm when purified from cytokine-treated cells . Overexpression of Doc2B shifted the size distribution of purified granules back to 191 ± 37 nm . The shifts in granule size were confirmed using fluorescence correlation spectroscopy ( FCS ) to measure the hydrodynamic radii of the C-peptide-GFP containing granules with diameters of 194 ± 11 nm , 183 ± 4 nm , 220 ± 2 nm , and 187 ± 3 nm for granules from untreated , palmitate-treated , cytokine-treated , and Doc2B-protected cytokine-treated cells , respectively ( Figure 1H and Figure 1—figure supplement 3 ) . In combination , the data on calcium affinity , release modes , and granule size are consistent with the presence of two distinct subpopulations of insulin granules: larger fast releasing granules that are susceptible to palmitate T2D-mimicking treatment and smaller slow releasing granules that are susceptible to cytokine T1D-mimicking treatment . The calcium sensor for exocytosis is synaptotagmin ( syt ) , a granule membrane-associated protein with two C2 domains ( Brose et al . , 1992 ) that promotes assembly of the SNARE complex and catalyzes fusion between the granule and plasma membranes ( Jahn and Fasshauer , 2012 ) , leading to the release of insulin from the cell . Syt7 and syt9 are the two predominant isoforms relevant for insulin secretion ( Gustavsson et al . , 2008; Iezzi et al . , 2005 ) , with syt7 and syt9 known to have a high and low affinity for calcium , respectively ( Zhang et al . , 2011 ) . The levels of both of these isoforms were unaffected in cell lysates after palmitate or cytokine treatment ( Figure 3A and B ) . However , fractionation of cell lysates by density gradient centrifugation revealed that treatments of cells with T1D-mimicking cytokines or T2D-mimicking palmitate caused distinct and different shifts in syt isoform distribution ( Figure 3A and C ) . Granules from palmitate-treated cells show a loss of the lower affinity syt9 while still containing the high-affinity calcium sensor syt7 . Conversely , granules from cytokine-treated cells show a loss of syt7 while still containing syt9 ( Figure 3C ) . Doc2B protected the redistribution of syt7 from fraction 9 of cytokine-treated cells ( Figure 3C ) . Immunodepletion of intact granules from untreated cells using antibodies against either syt7 or syt9 revealed that these two isoforms reside on separate subpopulations of insulin granules ( Figure 3D ) . The presence of distinct subpopulations is further supported by cryo-EM and FCS measurements on the immunodepleted supernatants . Residual syt7 granules ( after syt9 depletion ) had diameters of 168 ± 36 nm ( cryo-EM ) and 185 ± 4 nm ( FCS ) , whereas residual syt9 granules ( after syt7 immunodepletion ) were 211 ± 46 nm ( cryo-EM ) and 197 ± 7 nm ( FCS ) ( Figure 3E and Figure 1—figure supplement 3 ) . We note that our iodixanol purified fraction 9 ( Figure 3C ) may contain a minor contamination from other organelles , but that this would not affect our results because we follow C-peptide-GFP , which is an authentic insulin granule marker . Immunodepleted insulin granules were also examined by single particle fusion to reconstituted planar target membranes . The dose response to calcium was monophasic with a K1/2 = 10 ± 1 µM for syt7 granules ( depleted of syt9 ) and monophasic with a K1/2 = 48 ± 3 µM for syt9 granules ( depleted for syt7 , Figure 3F ) . These results closely correlate with granules that were purified from PC12 cells engineered to express only the single respective syt isoforms ( Figure 3—figure supplement 1 ) and with the observation that syt isoforms give rise to different release rates in chromaffin cells ( Rao et al . , 2014; Rao et al . , 2017 ) . Henceforth , insulin granule subpopulations will be referred to by the remaining syt isoform after depletion . Analysis of the modes of C-peptide-GFP release obtained from individual fusion events in the planar membrane fusion assay showed that the syt isoform also defines the release mode , with the syt7 granules releasing contents more slowly than the syt9 granules ( Figure 3G and Figure 3—figure supplement 2 ) . The similarity in shifts of calcium affinity , release mode , and size distribution lead us to conclude that T1D-mimicking cytokine treatment results in loss of syt7 granules , whereas T2D-mimicking palmitate treatment results in loss of syt9 granules . Formation of distinct insulin granule subpopulations strongly implicates a membrane sorting mechanism that begins in the trans-Golgi network ( TGN ) ( Simons and Ikonen , 1997 ) . Segregation potentially entails generation of granules with distinct lipid compositions analogous to those observed for other TGN-derived pathways ( Deng et al . , 2016 ) . This was explored first by extracting lipids from all isolated granules ( control ) , as well as from syt7 and syt9 subpopulations , and examining the relative amounts of cholesterol and sphingomyelin using colorimetric assays . Strikingly , syt7 granules were substantially enriched in both cholesterol and sphingomyelin compared to unseparated control granules , while syt9 granules contained much less of both lipid species ( Figure 4A and Figure 4—figure supplement 1 ) . The marked difference in sphingomyelin partitioning between the two granule types was validated in both purified granule fractions and intact cells using a secretory pathway fluorescent biosensor , EQ-SM-Kate ( Deng et al . , 2016 ) . EQ-SM-Kate consists of a mutated and nontoxic sphingomyelin-specific biosensor that is coupled to the fluorescent protein mKate and a signal sequence that directs entry into the secretory pathway . It was used in combination with a non-binding control EQ-sol-GFP encoding the same toxin further mutated so it does not bind membranes ( Deng et al . , 2016 ) . mKate and GFP fluorescence of purified granules from INS 832/13 cells that had been co-transfected with both biosensors was measured before and after immunodepletion with antibodies against either syt7 or syt9 ( Figure 4B ) . The GFP signal decreased in roughly equal amounts when immunodepleted of either syt isoform . In contrast , the SM-Kate signal only decreased when syt7 granules were depleted ( Figure 4B and C ) . As further confirmation of differential labeling by EQ-SM-Kate , the distributions of EQ-SM-Kate and EQ-sol-GFP in transfected INS 832/13 cells were compared by confocal fluorescence microscopy . Punctate fluorescence signals were observed for both constructs ( Figure 4D ) . The EQ-SM-Kate signal predominantly colocalized with the GFP signal while there were numerous GFP-positive puncta that lacked the EQ-SM-Kate signal ( Figure 4D and E ) . This indicates that EQ-sol-GFP broadly labels vesicles in the secretory pathway while the SM-Kate sensor selectively labels organelles/granules enriched in sphingomyelin . To further investigate other possible compositional differences of lipid species between granule populations , lipidomic analysis by mass spectroscopy of the phospholipids was performed . Purified granules divided into equal volumes were immunodepleted by syt isoform or left untouched ( as a control ) . Then granule lipids were extracted and mass spectroscopy was performed . The relative abundance ratios of all detected lipid species in syt9 over syt7 granules were determined along with their significance values ( Figure 4—source data 2 ) and plotted in a log-log volcano plot ( Figure 4F ) . No detectable significant phospholipid species differences between the two granules were revealed with the exception of an enrichment of 42:1 sphingomyelin in syt7 granules with a significance value p<0 . 05 ( Figure 4F ) . When sphingomyelin 42:1 , which likely contains the two predominant sphingomyelin tails , 18:0 in the sphingosine and 24:1 in the acyl chain positions ( O'Brien and Rouser , 1964 ) , is visualized in a bar graph and compared to all other pooled lipid species in each headgroup class , it is apparent that sphingomyelin 42:1 is 3-fold enriched in syt7 compared to syt9 granules ( Figure 4G ) . None of the other lipid classes and not even other sphingomyelin species show any significant enrichment in either type of granule . Previously , posttranslational palmitoylation of syt has been suggested to facilitate sorting to sphingomyelin- and cholesterol-enriched organelles . This has been shown for syt1 in neurons ( Kang et al . , 2004 ) and syt7 in immune cells ( Flannery et al . , 2010 ) . To probe for this possibility in insulin secreting cells , palmitate containing a click-iT group was incubated with INS 832/13 cells to allow incorporation into endogenous proteins by posttranslational modification . Granules were purified from these cells , solubilized in detergent , and labeled with a corresponding click chemistry Alexa647 fluorescent dye . Subsequent immunoprecipitation with anti-syt antibodies revealed that syt7 but not syt9 was palmitoylated ( Figure 5A ) . This likely explains the selective association of syt7 with sphingomyelin- and cholesterol-rich granules . β-Cells package proinsulin into nascent granules that bud from the TGN . Proinsulin is processed into insulin and C-peptide by granule-associated prohormone convertases PC1/3 and PC2 . Insulin is then complexed with zinc and condensed into a crystalline core within mature granules . Stored granules are released by a signaling cascade , in which glucose uptake and subsequent metabolism increase the cellular ATP/ADP ratio , causing closure of plasma membrane ATP-dependent potassium channels and thereby depolarizing the cell . The realization that there exist two subpopulations of insulin granules with distinct syt isoforms , fusion characteristics , and lipid compositions led us to examine other potential differences in content and membrane protein composition . Quantitative western blotting of immunodepleted granules was used to measure distributions between granule subpopulations of proteins that have functions in β-cell secretion or islet signaling and have established relationships to diabetes . The proteins considered include the membrane-anchored vesicular R-SNARE proteins , soluble SNARE-interacting proteins , non-insulin peptide hormones , their processing enzymes , and membrane transporters that concentrate classical transmitters within granules ( Figure 5B and Figure 5—figure supplement 1 ) . The vesicular R-SNARE proteins , VAMP2 , VAMP3 , VAMP4 , VAMP7 , and VAMP8 are implicated in fusion of multiple post-Golgi trafficking pathways ( Dingjan et al . , 2018 ) . VAMP2 , VAMP3 , VAMP4 , and VAMP8 were found in roughly equal amounts in both syt7 and syt9 granules , whereas VAMP7 was significantly enriched in syt7 granules ( Figure 5B ) . Of note , VAMP7 has been implicated in crinophagy and macro-autophagy ( Csizmadia et al . , 2018 ) , processes that are involved in insulin granule turnover ( Marsh et al . , 2007 ) and activated by the cytokine IL-1β ( Sandberg and Borg , 2006 ) . Among upstream ( of fusion ) regulatory proteins calcium activated protein for secretion 2 ( CAPS2 ) and Doc2B have been found to co-purify with the insulin granules ( Kreutzberger et al . , 2019 ) . They each contain C2 calcium binding domains that interact with PI ( 4 , 5 ) P2 and the SNARE fusion machinery ( Pinheiro et al . , 2016 ) . CAPS2 functions in both granule maturation and priming for exocytosis ( Kreutzberger et al . , 2017a; Speidel et al . , 2008 ) , while Doc2B functions in priming likely through its recruitment of Munc13 ( Groffen et al . , 2004 ) , a priming protein that does not co-purify with secretory granules ( Kreutzberger et al . , 2019 ) . CAPS2 was found to be enriched on syt-9 granules , while Doc2B was enriched on syt7 granules ( Figure 5B ) , indicating that the fusion regulatory machinery differs in part between the granule subpopulations . Moreover , there is a striking correlation between the loss of syt7 granules and the loss of Doc2B with the cytokine-induced type 1 diabetic phenotype ( Figure 1B and Aslamy et al . , 2018a; Aslamy et al . , 2018b ) . The preservation of the syt7 granules by overexpression of Doc2B suggests a role for Doc2B in granule formation or in protecting these granules during cytokine treatment . Other β-cell secretory products implicated in the progression of diabetic phenotypes include neuropeptide Y ( NPY ) , macrophage migration inhibitory factor ( MIF ) , insulin-like growth factor 2 ( IGF2 ) , and islet amyloid polypeptide ( IAPP ) . These secretory products enable local signaling in islets and distal regulation of other tissues . NPY promotes β-cell proliferation by inhibiting glucose-induced insulin secretion ( Rodnoi et al . , 2017 ) . IGF2 promotes β-cell proliferation , and its autocrine action via the IGF1 receptor on β-cells ( Cornu et al . , 2009 ) has been shown to be elevated in T2D ( Casellas et al . , 2015 ) . IAPP has been shown to inhibit insulin secretion and forms amyloid aggregates associated with T2D ( Westermark et al . , 2011 ) . MIF increases β-cell insulin secretion via autocrine stimulation ( Waeber et al . , 1997 ) and suppresses islet resident macrophages ( Stojanovic et al . , 2012 ) , an important signal for β-cell-to-immune system communication . Notably , IGF2 , IAPP aggregates , and MIF have been shown to be elevated in patients with T2D , while MIF has been shown to be suppressed in T1D . These secreted factors revealed a remarkable degree of segregation between the granule subpopulations . MIF , IGF2 , and IAPP are detected in syt7 granules while NPY is predominantly detected in syt9 granules ( Figure 5B and Figure 5—figure supplement 1 ) . These findings are easily explained with the loss of syt7 granules in response to cytokines and their presence following extended palmitate treatment ( Figure 1B ) . Evidently , our observed susceptibilities of distinct insulin granule subpopulations to the diabetes-mimicking treatments correlate well with previous clinical observations showing that the same respective granule specific markers are co-released with insulin in T1D and T2D patients ( Herder et al . , 2006 ) . Overall , these results strongly support the notion that syt7 granules are lost selectively during early stages of T1D while syt9 granules are lost selectively in T2D . The proprotein convertases PC1/3 and PC2 also exhibited distinct distributions . PC1/3 was equally distributed between the two granule subpopulations , while PC2 was markedly enriched in syt7 granules ( Figure 5B and Figure 5—figure supplement 1 ) . These differences suggest that processing specificity as well as processing kinetics may differ for the precursor polypeptides including insulin within the two granule types . Secretion of small molecules including glutamate , GABA , ATP , and dopamine from β-cells plays a role in auto- and paracrine regulation of insulin secretion and in communication with pancreatic resident immune cells ( Bai et al . , 2003; Gammelsaeter et al . , 2004; Garcia Barrado et al . , 2015; Geisler et al . , 2013; Weitz et al . , 2018 ) . These molecules are taken up and concentrated in insulin granules by specific transporters that use the electrochemical gradient generated by the VATPase . The vesicular transporters known to function in β-cells include VGLUT1 ( glutamate uptake ) , VGAT ( GABA uptake ) , VNUT ( ATP uptake ) , and VMAT2 ( dopamine uptake ) ( Anne and Gasnier , 2014 ) . VMAT2 is equally distributed between the two subpopulations of granules . However , VGLUT1 is present only in syt9 granules , whereas VNUT and VGAT are present only in syt7 granules ( Figure 5B ) . Thus , different neurotransmitters communicating with the immune system segregate into syt7 and syt9 granules . The distinct protein compositions of syt7 and syt9 granules are summarized in Figure 5C . In order to verify the partitioning of different secretory products to granules with different syt isoforms and therefore different calcium affinities for granule exocytosis , secretion of three different products ( C-peptide-GFP , used here as a surrogate for insulin and measured in a bulk secretion assay; ATP , measured by a luciferase assay of the cell media; and glutamate , measured in cell media by the iGluSnFr fluorescent biosensor ) were examined using mild ( 25 mM KCl ) and strong ( 90 mM KCl ) depolarization eliciting weaker and stronger calcium responses , respectively ( Figure 2—figure supplement 1A ) . Untreated GRINCH cells secreted C-peptide-GFP in increasing amounts as the depolarization strength was increased , consistent with its presence in both granule subpopulations ( Figure 6A , top panel ) . ATP secretion measured with INS 832/13 cells was maximal at 25 mM KCl , consistent with its selective concentration in syt7 granules . In contrast , glutamate secretion from INS 832/13 cells was only detected at 90 mM KCl , consistent with its selective concentration in syt9 granules ( also see Figure 6—figure supplement 1 , for calibrations and absolute amounts of secretion under different conditions ) . For palmitate-treated cells , C-peptide-GFP secretion was stimulated at 25 mM KCl and the response was not further increased at 90 mM KCl . Palmitate also had no observable effect on the secretion of ATP , but drastically reduced the secretion of glutamate ( Figure 6A , middle panel ) . By contrast , the cytokine treatment reduced the secretion of C-peptide-GFP ( especially upon mild stimulation ) and significantly decreased the secretion of ATP , while secretion of glutamate was not much altered as compared to untreated cells ( Figure 6A , bottom panel ) . Secretion of C-peptide , ATP , and glutamate from cells stimulated by strong ( 90 mM KCl ) depolarization corroborates the calcium dose response for fusion of purified insulin granules ( Figure 1 ) and the secretion by stimulated cells ( Figure 2 ) . The differential effects of T1D and T2D model treatments on C-peptide , ATP , and glutamate show a striking relationship between differential β-cell secretion and the pathology of diabetes . The differing secretory effects of the diabetes-mimicking treatments potentially shed new insight regarding the onset and contrasting secretory changes for T1D and T2D . As summarized in Figure 6B , two subpopulations of granules containing either syt7 or syt9 are found in untreated cells . They have distinct affinities for calcium , exhibit distinct modes of content release , and they have distinct lipid and protein compositions . Induction of a T1D model by treatment with cytokines results in loss of syt7 granules that normally secrete MIF , IGF2 , IAPP , ATP , and GABA alongside insulin . On the other hand , induction of lipotoxicity in T2D by treatment with palmitate causes selective loss of syt9 granules , which normally release NPY and glutamate alongside insulin . Our results reveal that insulin-secreting cells harbor distinct subpopulations of insulin granules that are selectively activated for secretion to release different secretory molecules . Therefore , insulin granules can no longer be considered generic storage units for releasing the full range of secretory molecules without regard to timing and signal strength . Regulated exocytosis provides a complex network of β-cell derived signals that serve as autocrine and paracrine governors of the secretion of insulin and other islet-derived hormones . The current work shows that these secretory regulators are parsed out selectively among the two insulin granule subpopulations that we have characterized in great molecular detail . The biphasic secretory pattern of insulin release in response to glucose has been previously described by its correlation to electrical activity in stimulated islet β-cells ( Rorsman and Ashcroft , 2018 ) . Because the granule subpopulations with secretory cargo compositions have different calcium affinities furnished by different synaptotagmin isoforms ( Figure 3 ) , we propose that these phases can be interpreted as a staged release of selected peptides and transmitters over time and levels of intracellular calcium . The detailed characteristics of the granule subpopulations and their correlations to T1D and T2D are summarized in Table 1 . In our new model linking the progression of T1D and T2D to different subpopulations of insulin secretory granules in the same cells , we posit that the syt9 granules release their content primarily in the first phase of secretion during the initial spike in intracellular calcium supported by their close proximity to calcium channels ( Hoppa et al . , 2009 ) . Granules harboring the high-affinity calcium sensor syt7 would release their content during both phases . Indeed , both first- and second-phase secretion are decreased in human β-cells upon syt7 knockout ( Dolai et al . , 2016 ) and upon Doc2b knockout in mice ( Ramalingam et al . , 2012 ) , while treatment of human β-cells with free fatty acids inhibits the peak during the initial release phase ( Hoppa et al . , 2009 ) . It will be interesting to see if physiologic responses can be correlated to time-dependent release of selected cargoes that have local and systemic regulatory roles and if there is further heterogeneity within the discovered subpopulations . Our studies further illustrate the very important contribution of granule membrane protein and lipid composition in supporting the complex signaling capability of β-cell exocytosis . The striking difference in lipid composition between the two granule subpopulations , especially with regard to sphingomyelin and cholesterol content ( Figure 4 ) , almost certainly contributes to the specific sorting of the fusion regulatory proteins and transporters ( Figure 5 ) . These sorting events enable granule subpopulations to be mobilized according to signaling strength and thereby export different regulatory signals during different phases of T1D and T2D development . Going forward , it will be interesting to define these sorting processes in more detail , determine how they interface with other post-Golgi pathways , and thereby contribute to the pathology of diabetes . A central and striking new finding of the present study is the discovery that the two markedly different insulin granule subpopulations are each selectively sensitive to T1D and T2D model treatments . This brings new insight to the differing secretory pathologies that arise in β-cells during the two types of diabetes . Further , the differential loss of specific granule subpopulations can explain previously observed changes in insulin release characteristics such as more restricted fusion pores after palmitate treatment ( Hoppa et al . , 2009 ) , which would be consistent with slower content release that we observed after this treatment . Because the levels of neither syt7 nor syt9 are decreased in lysates of cells following diabetic model treatments ( Figure 3B ) , it appears that shifts in granule subpopulations induced by the treatments are not explainable solely by amplified autophagic degradation of a particular type of granule , although the process of granule turnover by crinophagy in β-cells ( Marsh et al . , 2007 ) may be membrane-conservative and thus not lead to synaptotagmin degradation . In the case of palmitate treatment , however , it will be necessary to evaluate this deduction in relation to the recently reported crinophagy of newly formed insulin granules under experimental conditions that differ from ours ( Pasquier et al . , 2019 ) . Follow-up studies may also be needed to examine whether diabetogenic signaling is targeted to granule formation where a role for CAPS has already been defined ( Speidel et al . , 2008 ) or to enhanced stimulated secretion ( Olofsson et al . , 2007 ) . As well , it will be interesting to learn whether perturbed clustering of L-type calcium channels by palmitate ( Gandasi and Barg , 2014 ) and perturbed function of syntaxin-4 caused by proinflammatory cytokines ( Oh et al . , 2012; Wiseman et al . , 2011 ) interface with the signaling that is targeted to the insulin granule subpopulations . Regardless , the loss of a whole granule subpopulation and with it the loss of a variety of signaling molecules during the onset of T1D or throughout T2D likely explains multiple previously unknown effects on the progression of the disease . For example , losing granules secreting ATP and MIF ( Figure 6A ) during the onset of T1D might compromise both local signaling and islet protection from immune cells ( Stojanovic et al . , 2012; Weitz et al . , 2018 ) . Although there were prior hints that high and low sensitivity calcium pools co-exist in insulin secreting cells ( Wan et al . , 2004; Yang and Gillis , 2004 ) , the identities of different granule subpopulations have not been previously described and mapping of the complex arrays of secretory products within these subpopulations has not been achieved . Also , the discovery that these subpopulations and their secretory products are linked to distinct diabetic states of insulin producing cells is novel and provides new mechanistic insight to established defects in diabetic ( T1D and T2D ) secretory signals . Our granule purification and functional reconstitution approach was ideally suited to characterize heterogeneity of granules subpopulations in extensive detail , which has not been previously achieved using traditional cellular approaches . A recent study demonstrates that the described granule heterogeneity likely extends to human and rodent pancreatic β-cells with differential modulation of calcium sensitivity of exocytosis conferred by synaptotagmin isoforms during pancreatic development and maturation ( Huang et al . , 2018 ) . Therefore , our current work provides a road map for examining changes of secretory signals within diabetes in authentic islets . The new insights gained here will likely inform the design of future studies to explore granule heterogeneity in human and other mammalian islet secretion of specific products stimulated by appropriate signals under different diabetic states . The presence of functionally and compositionally distinct secretory vesicles revealed here for insulin secreting cells ( Table 1 ) is not unique to pancreatic β-cells and derivatives , but is likely a general property of a variety of regulated secretory cells . Secretory vesicle heterogeneity has been shown to have consequences on synaptic secretion and likely affects secretion from most endocrine cell types as well as from immune cells functioning in host defense . In neurons , the partitioning of VGAT and VGLUT between different vesicles has been observed ( Farsi et al . , 2016 ) but no mechanism of how a cell facilitates sorting of these transporters into distinct synaptic vesicles has been explored . The tools developed and presented in this work should open new avenues for future studies of the characteristics and mechanisms of sorting processes in these other cell types and thus help with a better molecular understanding of the diseases that are associated with them . The following materials were purchased and used without further purification: porcine brain L-α-phosphatidylcholine ( bPC ) , porcine brain L-α-phosphatidylethanolamine ( bPE ) , porcine brain L-α-phosphatidylserine ( bPS ) , and L-α-phosphatidylinositol ( liver , bovine ) ( PI ) , and porcine brain phosphatidylinositol 4 , 5-bisphosphate ( bPIP2 ) were from Avanti Polar Lipids; cholesterol , sodium cholate , EDTA , calcium , Opti-Prep Density Gradient Medium , sucrose , MOPS , glutamic acid potassium salt monohydrate , potassium acetate , adenosine 5'-triphosphate ( ATP ) magnesium ( Mg2+ ) salt and glycerol were from Sigma; CHAPS and DPC were from Anatrace; HEPES was from Research Products International; Click iT palmitic acid azide was from Molecular Probes; Sodium palmitate , Alexa Fluor 647 alkyne triethylammonium salt , chloroform , ethanol , Contrad detergent , all inorganic acids and bases , and hydrogen peroxide were from Fisher Scientific; fetal bovine serum ( FBS ) from Atlanta Biological . Water was purified first with deionizing and organic-free three filters ( Virginia Water Systems ) and then with a NANOpure system from Barnstead to achieve a resistivity of 18 . 2 MΩ/cm . Antibodies for syt1 ( mouse monoclonal cat . 105 011 ) , syt7 ( rabbit polyclonal cat . 105 173 ) , syt9 ( rabbit polyclonal cat . 105 053 ) , Doc2B ( rabbit polyclonal cat . 174 103 ) , VGAT ( mouse monoclonal cat . 131 011 ) , VATPase ( rabbit polyclonal cat . 109 002 ) , VMAT2 ( rabbit polyclonal cat . 138 302 ) , VGLUT1 ( mouse monoclonal cat . 135 011 ) , CAPS2 ( rabbit polyclonal cat . 262 103 ) , synaptobrevin-2/VAMP2 ( mouse monoclonal cat 104 211 ) , and VAMP8 ( rabbit polyclonal cat . 104 302 ) were from Synaptic Systems . The calnexin antibody ( rabbit polyclonal cat . ADI-SPA-860 ) was from Enzo Life Sciences , the secretogranin II antibody ( mouse monoclonal cat . LS-C335666 ) was from Biodesign International , the GFP antibody ( mouse monoclonal cat . sc-9996 ) was from Santa Cruz , the succinate ubiquinone oxidoreductase antibody ( mouse monoclonal cat . 459200 ) was from Molecular Probes , PC1/3 ( rabbit polyclonal cat . Ab220363 ) , PC2 ( rabbit polyclonal cat . Ab3533 ) , VAMP7 ( rabbit polyclonal cat . Ab224535 ) , NPY ( rabbit polyclonal cat . Ab30914 ) , IGF2 ( rabbit polyclonal cat . Ab9574 ) , and MIF ( rabbit polyclonal cat . Ab7207 ) were from Abcam , IAPP ( rabbit polyclonal cat . MBS8501010 ) was from MyBioSource , VNUT ( rabbit polyclonal cat . Abn84 ) was from EMD Millipore , VAMP4 ( rabbit polyclonal cat . PA1-768 ) was from Thermo Fisher , and VAMP3 ( rabbit antiserum ) was a gift from Pietro DeCamilli ( Yale School of Medicine ) . INS 1 cell-derived 832/13 cells were obtained originally from Christopher Newgard , Duke University School of Medicine ( Hohmeier et al . , 2000 ) . Cells were cultured on 10 cm plates in RPMI 1640 medium ( Gibco ) , 10 mM HEPES , 1 mM sodium pyruvate , 50 µM β-mercaptoethanol , 1x pen/strep , and 10% fetal bovine serum at 37°C and 5% CO2 as described in Hussain et al . , 2018 . INS 832/13 cell-derived GRINCH cells stably expressing human proinsulin tagged with GFP in the C-peptide region , cultured with 20 µg/mL of G418 antibiotic , were originally described in Haataja et al . , 2013 and were a gift from Peter Arvan , University of Michigan Medical School . Cells were authenticated by immunofluorescent staining for endogenous insulin and the fact that a majority of our studies involved evaluating the insulin granule marker C-peptide-GFP . Potential contamination by mycoplasma was controlled by ciprofloxacin when needed . Free fatty acid ( FFA ) treatment with palmitate ( sodium palmitate from Fisher , cat . P000725G ) was performed as described by Hoppa et al . , 2009 . This was done by dissolving palmitate in 95% ethanol adding stoichiometric amounts of NaOH . The solution was dried with N2 gas , water was added , and the sample was heated creating a hot soap . The solution was stirred and BSA was added to a final concentration of 10% w/v , creating a 10x stock solution . The pH was set to 7 . 4 . Cells were cultured in media containing 0 . 5 mM palmitate in the presence of 1% BSA for 72 hr . This was estimated to give a FFA concentration of 26 nM ( Hoppa et al . , 2009 ) . Cytokine treatments described by others ( Ahn et al . , 2016; Aslamy et al . , 2018b ) were performed by incubating 10 ng/mL TNF-α , 100 ng/mL of IFN-γ , and 5 ng/mL of IL-1β in cell culture media for 24 hr prior to experiments . Prior to cytokine treatment cells were transfected with ( Myc-DDK-tagged ) -human Doc2B ( described in Aslamy et al . , 2018b and purchased from OriGene , cat . RC218949 ) . Transfection was performed by electroporation using a ECM 830 Electro Square Porator ( BTX ) . After harvesting and sedimentation , cells were suspended in electroporation buffer 120 mM KCl , 10 mM KH2PO4 , 0 . 15 mM CaCl2 , 2 mM EGTA , 25 mM Hepes-KOH , 5 mM MgCl2 , 2 mM adenosine triphosphate , and 5 mM glutathione ( pH 7 . 6 ) ( van den Hoff et al . , 1992 ) and then counted and diluted . Cell suspension of ~10×106 cells in 700 µL and 30 µg of DNA were placed in an electroporator cuvette with a 4 mm gap , and two 255 V , 8 ms electroporation pulses were applied . Cells were transferred to a 10 cm cell culture dish with 10 mL of normal growth medium . Cytokine treatment was then applied 3 days after transfection . Pheochromocytoma cells ( PC12 ) were cultured on 10 cm plates in DMEM high-glucose media ( Gibco ) supplemented with 5% horse serum ( CellGro ) , and 5% calf serum ( HyClone ) , and 1% pen/strep . Stable knockdowns of endogenous synaptotagmins ( isoforms 1 and 9 ) were generated ( first described in Kreutzberger et al . , 2017a ) individual isoforms were over expressed with shRNA resistant plasmids for synaptotagmin-1 , –7 , or −9 ( Bendahmane et al . , 2020; Kreutzberger et al . , 2017a; Kreutzberger et al . , 2019 ) . Granules were purified using a previously described method ( Kreutzberger et al . , 2017a; Kreutzberger et al . , 2019 ) . Cells were scraped ( ~20–30 10 cm plates ) into PBS . Cells were pelleted by centrifugation and then suspended and washed in homogenization medium ( 0 . 26 M sucrose , 5 mM MOPS , 0 . 2 mM EDTA ) by pelleting and resuspending . Following resuspension in 3 mL of medium containing protease inhibitor ( Roche Diagnostics ) , the cells were cracked open using a ball bearing homogenizer with a 0 . 2507 inch bore and 0 . 2496 diameter ball . The homogenate was spun at 1500 x g for 10 min at 4o C in fixed-angle microcentrifuge to pellet nuclei and larger debris . The post-nuclear supernatant was collected and spun at 5900 x g for 15 min at 4°C to pellet mitochondria . The post-mitochondrial supernatant was then collected , adjusted to 5 mM EDTA , and incubated 10 min on ice . A working solution of 50% Optiprep ( iodixanol ) ( 5 vol 60% Optiprep: 1 vol 0 . 26M sucrose , 30 mM MOPS , 1 mM EDTA ) and homogenization medium was used to prepare solutions for discontinuous gradients in Beckman SW55 tubes: 0 . 5 mL of 30% iodixanol on the bottom and 3 . 8 mL of 14 . 5% iodixanol , above which 1 . 2 ml EDTA-adjusted supernatant was layered . Samples were spun at 190 , 000 x g for 5 hr . A clear white band at the interface between the 30% iodixanol and the 14 . 5% iodixanol was collected as the dense core granule sample . Gradient fractions were prepared for western blotting by diluting individual fractions with PBS , followed by pelleting the membranes with a high-speed spin and resuspending the membranes in SDS loading buffer . The dense core granule sample used for functional experiments was extensively dialyzed in a cassette with 10 , 000 kD molecular weight cutoff ( 24–48 hr , 3 × 5L ) into the fusion assay buffer ( 120 mM potassium glutamate , 20 mM potassium acetate , 20 mM Hepes , pH 7 . 4 ) . Samples were suspended in 1X SDS buffer . Samples were loaded and run on 4–20% TGX protein gels ( Bio-Rad 4561094 ) . Gels were then transferred onto a PVDF membrane with Bio-Rad Trans-Blot Turbo Transfer system ( Bio-Rad 1704150 ) ; dried membranes were activated with methanol and washed in TBST ( 0 . 5% Tween-20 in Tris-buffered saline ) before use . Membranes were blocked in 5% milk in TBST for 1 hr , washed , and incubated in primary antibody overnight . All primary and secondary antibodies for immunoblotting were diluted in Millipore signal boost immunoreaction enhancer solutions ( Millipore 407207 ) . The following day , membranes were washed and probed with HRP conjugated secondary antibody diluted 1:10 , 000 ( Jackson Immunoresearch cat . rabbit: 211032171 , cat . mouse: 115035174 ) . Membranes were washed and incubated in Pico PLUS chemiluminescent substrate ( Thermofisher 34580 ) before imaging on Fujifilm LAS-3000 luminescent analyzer . 100 µL of Protein A magnetic SureBeads ( Bio-Rad ) were washed three times with PBS with 1% Tween , three times with PBS , and then washed five times with fusion buffer ( 120 mM potassium glutamate , 20 mM potassium acetate , 20 mM HEPES , pH 7 . 4 ) . The beads were then incubated with 5 µg of the indicated synaptotagmin isoform antibody in 100 µL of fusion buffer for 20–30 min . Beads were then washed three times with fusion buffer . Antibody-bound beads were then incubated with 0 . 5–1 mL of purified insulin granules ( sufficient to adsorb all granules containing an individual synaptotagmin isoform; Figure 3D ) . Supernatant was used for fusion experiments or was pelleted and resuspended in SDS loading buffer for western blots . Western blots of immuno-depleted granules ( Figure 5—figure supplement 1 ) were quantified with the amount of protein in depleted samples being determined relative to the amount of an equal volume of granules that were not depleted ( Figure 5B and Figure 5—figure supplement 1 ) . Imaging was performed on an Olympus cellTIRF-4Line microscope ( Olympus , USA ) equipped with a TIRF oil immersion objective ( NA 1 . 49 ) and an additional 2x lens in the emission path between the microscope and the cooled electron-multiplying charge-coupled device Camera ( iXon 897 , Andor Technology ) . The final pixel size was 80 nm . Series of images were acquired at ~20 Hz using CellSense software with an exposure time of 30 ms and an EM gain of 200 . C-peptide-GFP was excited using a 488 nm laser . The cells were imaged in buffer ( 145 mM NaCl , 5 . 6 mM KCl , 2 mM CaCl2 , 0 . 5 MgCl2 , 3 mM glucose , and 15 mM HEPES , pH 7 . 4 ) . Cells were individually stimulated using a needle ( 100 µm in diameter ) connected to a perfusion system under positive pressure ALA-VM4 ( ALA Scientific Instruments , Westerbury , NY ) . To trigger exocytosis , cells were first perfused with buffer for 5–10 s and then stimulated by increasing KCl to 25 mM or 90 mM with a subsequent decrease in NaCl by equal amounts . Images were analyzed using a homemade program written in LabVIEW ( National Instruments ) . Stacks of images were filtered by a moving average filter . The maximum intensity for each pixel over the whole stack was projected onto a single image . Insulin granules were located in this image by a single particle detection algorithm ( Kiessling et al . , 2006 ) . The central pixel of fluorescence intensities of a 5-pixel by 5-pixel area around each identified center of mass were plotted as a function of time for all particles in the image series . The exact time points of the onset of fusion until the content release completed of granules docked prior to stimulation or granules that came into the TIRF field after stimulation was determined . Some granules moved after docking and before fusion . These were not included in the analysis of fast vs . slow fusion granules . 96-well glass bottom plates ( Cellvis P96-0-N ) were coated with 0 . 1% Poly-L-Lysine ( Sigma P8920 ) and washed with sterile water . GRINCH cells were detached from cell culture plates with 0 . 25% Trypsin-EDTA ( Gibco 25200–056 ) , washed with culture medium , and pelleted at 500 g for 5 min . Cells were resuspended in Fluo-4 AM solution containing buffer ( 155 mM NaCl , 4 . 5 mM KCl , 1 mM MgCl2 , 2 mM CaCl2 , 5 mM HEPES , 10 mM Glucose , pH = 7 . 4 ) , 5 uM Fluo-4 AM ( ThermoFisher F14201 ) , 0 . 02% Pluronic F-127 ( ThermoFisher P3000MP ) , and 500 μM Probenecid . Cells were plated at a density of 200 , 000 cells/well , pelleted at 500 g for 2 min , and incubated at room temp for 30 min covered from light . Fluo-4AM solution was removed and replaced with 100 μL of buffer containing 500 μM Probenecid and incubated for 5 min covered from light . Immediately before imaging each column , buffer with probenecid solution was removed and cells were washed with buffer without Probenecid . 80 μL of buffer was added and cells were placed in Flexstation three for recording . Each column of the 96-well plate was recorded separately . Calcium flux was measured on Flexstation three at excitation of 490 nm and emission of 520 nm every 2 s for 360 s at 25C and recorded on SoftMax Pro 7 . 03 software . At 31 s , 20 μL of buffer containing 4 . 5 mM , 125 mM , or 450 mM KCL was added to appropriate wells for final concentrations of 4 . 5 mM ( baseline ) , 25 mM , or 90 mM KCl respectively . At 300 s 25 μL of 1% Triton ( 0 . 2% final concentration ) was added to permeabilize cells and thus expose Fluo-4 AM to 2 mM Ca2+ containing buffer . This was used as the positive control . Data are reported as F1-F0 normalized to positive control . Secretion of C-peptide-GFP was performed as previously described ( Hussain et al . , 2018 ) . GRINCH cells were seeded at the same density and cultured in Roswell Park Memorial Institute ( RPMI ) media with 10% FBS . The cells were washed into 1 mL of basal buffer ( 145 mM NaCl , 5 . 6 mM KCl , 2 . 2 mM CaCl2 , 0 . 5 mM MgCl2 , 3 mM glucose , 15 mM HEPES , pH 7 . 4 ) and incubated at 37°C for 30 min . Cells were then stimulated by increasing the KCl concentration to 25 or 90 mM ( with a corresponding decrease in NaCl ) . The supernatant was collected and any cellular debris was pelleted . Cells were scraped , pelleted , and lysed and the supernatant was used to assay GFP fluorescence . Fluorescence at 510 nm ( emission peak of GFP ) recorded in a SpectraMax M5 plate reader ( Molecular Devices ) and was used to calculate fractional secretion ( percent of total GFP fluorescence secreted under stimulated conditions relative to the total amount of GFP in cell lysates ) . INS 832/13 cells were plated at equal density per plate and allowed to grow for 24 hr before treatments were performed . The cells were washed into 1 mL of basal buffer and incubated at 37°C for 5 min . The buffer was then collected and cells were incubated with 1 mL of 25 mM KCl or 90 mM KCl stimulation solution for 3 min 37°C . The collected stimulation media were centrifuged to pellet any residual cells . Cells were scraped , pelleted , and lysed for the protein concentration to be determined by a BCA assay ( BCA Protein Assay Kit from Thermo Fisher Scientific cat . 23227 ) . The amount of glutamate was then measured in 100 µL of the media containing the iGluSnFr . A184S fluorescence glutamate sensor variant ( Marvin et al . , 2013; Marvin et al . , 2018 ) using a Flexstation three at excitation of 490 nm and emission of 520 nm at 25°C and recorded using SoftMax Pro 7 . 03 software . The fluorescence was normalized to the amount of protein in the cell lysates to account for any difference in cell density . The relative secretion ( Figure 6A ) was determined by normalizing each sample to the amount of secretion observed for untreated cells at 90 mM KCl . INS 832/13 cells were plated at equal density per plate and allowed to grow for 24 hr before treatments were performed . The cells were washed into 1 mL of basal buffer and incubated at 37°C for 5 min . The buffer was then collected and cells were incubated with 1 mL of 25 mM KCl or 90 mM KCl stimulation solution for 3 min at 37°C . The stimulation media were centrifuged to pellet any cells . Cells were scraped , pelleted , and lysed for the protein concentration to be determined by a BCA assay ( BCA Protein Assay Kit from Thermo Fisher Scientific cat . 23227 ) . The amount of ATP was assayed using a luminescent ATP detection assay ( Abcam ) with bioluminescence being recorded on Flexstation 3 at 25°C and recorded using SoftMax Pro 7 . 03 software . The luminescence was normalized to the amount of protein in the cell lysates to account for any difference in cell density . The relative secretion ( Figure 6A ) was determined by normalizing each sample to the amount of secretion observed for untreated cells at 90 mM KCl . Cells were grown to ~70% confluency and then incubated with 50 µM Click-iT palmitic acid azide ( Molecular Probes cat . C10265 ) for 6 hr ( 10 plates of cells were used ) . Cells were then scraped into PBS and insulin granules were purified . Granules were dialyzed for 36 hr in buffer ( 120 mM potassium glutamate , 20 mM potassium acetate , and 20 mM HEPES , pH 7 . 4 ) with two buffer changes . The purified granules where solubilized using 1% Triton-X100 detergent and then the Click-iT azide group was allowed to react with an Alexa Fluor 647 Alkyne ( Thermo Fisher Scientific cat . A10278 ) for 1 hr . After reaction , the granules were split into three samples . One sample was left untreated while the other two were immunodepleted using either anti-syt7 or syt9 antibodies . The fluorescence of the supernatants was then compared to the control for loss of Alexa647 fluorescence . This experiment was repeated with three independent granule purifications . Samples of each biological replicate ( three replicates in total ) were split into three equal volumes . One sample was left untouched as a control , while the other two were immunodepleted for one or the other syt isoform . Following immunodepletion the amount of protein per sample was determined by a BCA assay ( BCA Protein Assay Kit from Thermo Fisher Scientific cat . 23227 ) , cholesterol was quantified using a colormetric assay ( Amplex Red Cholesterol Assay from Thermo Fisher Scientific , cat . A12216 ) , and sphingomyelin was quantified using an fluorometric assay ( Sphingomyelin Assay Kit from Abcam , cat . ab138877 ) . The amount of cholesterol or sphingomyelin per amount of protein in the sample was determined for control and immunodepleted samples ( Figure 4—figure supplement 1 ) . The relative amounts of cholesterol and sphingomyelin as compared to the control sample are shown ( Figure 4A ) by normalizing the immunodepleted samples to equal amounts of samples that were not immunodepleted . To purify granules labeled with EQ-sol-GFP and EQ-SM-Kate ( gifts of Christopher Burd , Yale School of Medicine ) , cells were simultaneously transfected with both constructs using an ECM 830 Electro Square Porator ( BTX , Hawthorne , NY ) . Accordingly , cells were harvested and suspended in 0 . 75 mL of a sterile cytomix electroporation buffer ( 120 mM KCl , 10 mM KH2PO4 , 0 . 15 mM CaCl2 , 2 mM EGTA , 25 mM Hepes-KOH , 5 mM MgCl2 , 2 mM ATP , and 5 mM glutathione , pH 7 . 6 ) ( van den Hoff et al . , 1992 ) . Cells were then diluted to 10 × 106 cells in 700 µL buffer with 30 µg of DNA and electroporated in a cuvette with a 4 mm gap and two 255 V 8 ms electroporation pulses were applied . Subsequently , cells were transferred to a 10-cm cell culture dish with 10 mL of normal growth medium and grown for 4 days before insulin granules were purified . Purified insulin granules were dialyzed for 36 hr ( two buffer changes ) with fusion buffer ( 120 mM potassium glutamate , 20 mM potassium acetate , and 20 mM Hepes , pH 7 . 4 ) to remove iodixanol . Dialyzed granules were divided into three samples of equal volume – one left untreated , one immunodepleted for synaptotagmin-7 , and one immunodepleted for synaptotagmin-9 . After immunodepletion the fluorescence spectra of the supernatants were recorded on a Fluorolog 3 ( model FLS-21 from Horiba ) and compared to the spectrum of the untreated sample . For fluorescence microscopy , cells were plated at 80% confluency in a 6-well plate . Then 2 . 5 µg of EQ-sol-GFP and EQ-SM-Kate plasmids were transfected with lipofectamine 3000 ( ThermoFisher Cat #L300015 ) into each well . Cells were incubated for 3 days at 37°C , plated onto cover slips , fixed in 3% formaldehyde for 10 min ( Cell Signaling Technology #12606 ) and mounted onto slides with DAPI stain ( Cell Signaling Technology #8961 ) . Cells were then imaged on a Zeiss LSM-880 confocal microscope . Image analysis for colocalization of EQ-sol-GFP and EQ-SM-Kate was quantified using the coloc2 plugin in ImageJ . Each cell region of interest was manually defined and Manders Colocalization Coefficient was generated to quantify the degree of colocalization of EQ-sol-GFP and EQ-SM-Kate puncta . M1 Red/Green defines the extent of EQ-SM-Kate colocalizing with EQ-sol-GFP . M2 Green/Red represents the extent of sol-GFP colocalizing with SM-Kate . Lipids were isolated from syt7 and syt9 granules purified from INS 832/13 cells using a modified Bligh-Dyer method . In brief , 400 μL of granules in PBS were added to 500 μL of CHCl3 ( EMD Cat . CX1050-1 ) and 1 mL methanol ( Sigma-Aldrich Cat . 34860 ) containing Splash Lipidomix Mass Spec Standard ( 1 μL , Avanti Polar Lipids , Inc , Cat . 330709 ) and centrifuged for 5 min at 787 g ( Eppendorf Centrifuge 5810 R ) . Samples were decanted and 600 μL of water ( HPLC grade , Sigma-Aldrich Cat . 270733 ) and 1 . 5 mL of CHCl3 were added to each sample . Samples were vortexed and centrifuged ( 5 min at 787 g ) . The organic layer was collected . The extraction was repeated for a total of three extractions . The combined organic layers were dried under nitrogen gas and resuspended in 400 μL of methanol . Samples were analyzed using a ThermoFisher Q Exactive mass spectrometer coupled with a Vanquish Ultra-High Performance Liquid Chromatography system . Chromatographic separation was achieved using a 10–100% gradient of Solvent A over 24 min at a constant flow rate of 400 μL/min ( Solvent A: 50% acetonitrile ( Fisher Scientific Cat . A998-4 ) in water , 10 mM ammonium formate ( Acros Organics Cat . 401152500 ) and 0 . 1% formic acid ( Fluka Cat . 56302 ) ; Solvent B: 10% acetonitrile , 88% 2-propanol ( Sigma-Aldrich Cat . 34863 ) , 2% water , 2 mM ammonium formate , and 0 . 02% formic acid ) on a ThermoFisher Acclaim 120 C18 column ( 5 μM , 120 Å , 4 . 6 × 100 mm ) . Lipid species masses were collected by ddMS2 Top5 with an inclusion list for Splash Lipidomix Mass Spec Standard lipid species . Lipid identities were assigned by LipidSearch 4 . 1 . 16 . The resulting species were normalized to d18:1-18:1 ( d9 ) sphingomyelin and protein concentration . To identify lipid species that were significantly enriched in either syt7 ( immunodepleted of syt9 ) or syt9 ( immunodepleted of syt7 ) granules , sample replicates for each lipid species were averaged and the ratio ( syt9/syt7 ) was calculated . The ratios for each lipid species were log2 transformed . Significance for each lipid species was calculated by two-tailed Student’s T-test with unequal variance . The resulting p-values were log2 transformed and -log2 significance vs . log2 ( syt9/syt7 ) , which represents fold enrichment in syt9 over syt7 granules , was plotted to generate a Volcano plot . A p-value of 0 . 05 or lower ( corresponding to -log2 > 4 . 31 ) is considered a significant change in lipid content between the two types of granules . To identify lipid classes that were enriched in either syt7 or syt9 granules , the peak areas for each class were totaled on a per sample basis . The resulting lipid class totals for each sample were normalized to the average of syt7 granules for each lipid class . The normalized data was then log2 transformed and the data represented in a bar-graph ( Figure 4G ) . Only significant changes ( p<0 . 05 ) of individual lipid species were included and compared to average class changes in Figure 4G . Monoacyl glycerides ( MG ) were excluded from this analysis as only one MG species was detected in only three of the seven syt7 and syt9 granule samples ( see Supplemental Data Tables for lipidomic data and analysis ) . Syntaxin-1a ( 1-288 full length construct ) , SNAP-25 , Munc18 , Munc13 , and complexin-1 from Rattus norvegicus were expressed in Escherichia coli strain BL21 ( DE3 ) cells under the control of the T7 promoter in the pET28a expression vector and purified as described previously ( Kreutzberger et al . , 2016; Kreutzberger et al . , 2017a; Kreutzberger et al . , 2019 ) . SNAP-25 was quadruply dodecylated through disulfide bonding of dodecyl methanethiosulfonate ( Toronto Research Company , Toronto , Ontario ) to its four native cysteines ( Kreutzberger et al . , 2016 ) . To produce the iGluSnFr . A184S fluorescence glutamate sensor , plasmid containing the GltI-cpsfGFP variant ( pRSET ) of the SF-iGluSnFR . A184S ( kindly provided by Jonathan S Marvin ) ( Marvin et al . , 2013; Marvin et al . , 2018 ) was transformed into E . coli BL21 ( DE3 ) cells ( NEB ) . Protein expression was induced in TB-medium supplemented with 0 . 3 mM IPTG and 40 µg/ml ampicillin and incubated overnight at 25°C . Cells were resuspended in ice cold extraction buffer ( 20 mM HEPES , 500 mM NaCl and 8 mM imidazole , pH 7 . 4 ) and lysed by sonication after incubation with lysozyme , DNAse , MgCl2 and EDTA-free Protease inhibitor ( Roche ) for 1 h , 4 °C . Cell debris was removed by centrifugation and the protein was subsequently purified using Ni2+-nitrilotriacetic acid beads ( Qiagen GmbH ) in 20 mM HEPES buffer containing 1 M NaCl and 150 mM imidazole , pH 7 . 4 followed by ion exchange chromatography ( HiTrap QFF , GE Healthcare ) . Planar supported bilayers with reconstituted plasma membrane SNAREs were prepared by Langmuir-Blodgett/vesicle fusion technique as described in previous studies ( Domanska et al . , 2009; Kalb et al . , 1992; Wagner and Tamm , 2001 ) . Quartz slides were cleaned by dipping in 3:1 sulfuric acid:hydrogen peroxide for 15 min using a Teflon holder . Slides were then rinsed in milli-Q water . The first leaflet of the bilayer was prepared by Langmuir-Blodgett transfer onto the quartz slide using a Nima 611 Langmuir-Blodgett trough ( Nima , Conventry , UK ) by applying the lipid mixture of 70:30:3 bPC:Chol:DPS from a chloroform solution . Solvent was allowed to evaporate for 10 min , the monolayer was compressed at a rate of 10 cm2/min to reach a surface pressure of 32 mN/m . After equilibration for 5 min , a clean quartz slide was rapidly ( 68 mm/min ) dipped into the trough and slowly ( 5 mm/min ) withdrawn , while a computer maintained a constant surface pressure and monitored the transfer of lipids with head groups down onto the hydrophilic substrate . Plasma membrane SNARE containing proteoliposomes with a lipid composition of bPC:bPE:bPS:Chol:PI:PI ( 4 , 5 ) P2 ( 25:25:15:30:4:1 ) were prepared by mixing the lipids and evaporating the organic solvents under a stream of N2 gas followed by vacuum desiccation for at least 1 hr . The dried lipid films were dissolved in 25 mM sodium cholate in a buffer ( 20 mM HEPES , 150 mM KCl , pH 7 . 4 ) followed by the addition of an appropriate volume of synatxin-1a and SNAP-25 in their respective detergents to reach a final lipid/protein ratio of 3000 for each protein . After 1 hr of equilibration at room temperature , the mixture was diluted below the critical micellar concentration by the addition of buffer to the desired final volume . The sample was then dialyzed overnight against 1 L of buffer containing Biobeads , with one buffer change after ~4 hr . To complete the formation of the SNARE containing supported bilayers , proteoliposomes were incubated with the Langmuir-Blodgett monolayer with the proteoliposome lipids forming the outer leaflet of the planar supported membrane and most SNAREs oriented with their cytoplasmic domains away from the substrate and facing the bulk aqueous region ( Domanska et al . , 2009; Kalb et al . , 1992; Kiessling et al . , 2017; Wagner and Tamm , 2001 ) . A concentration of ~77 μM total lipid in 1 . 2 mL total volume was used . Proteoliposomes were incubated for 1 hr and excess proteoliposomes were removed by perfusion with 5 mL of buffer ( 120 mM potassium glutamate , 20 mM potassium acetate 20 mM HEPES , 100 μM EDTA , pH 7 . 4 ) . Experiments examining single-granule docking and fusion events were performed on an Axiovert 35 fluorescence microscope ( Carl Zeiss ) , with a 63x water immersion objective ( Zeiss , numerical aperture , 0 . 95 ) and a prism-based TIRF illumination . The light source was an OBIS 532 LS laser or an OBIS 488 LX laser from Coherent Inc Fluorescence was observed through a 610-nm band-pass filter ( D610/60 , Chroma ) by an electron multiplying charge coupled device ( CCD ) ( DU-860E , Andor Technology ) . The electron multiplying CCD ( EMCCD ) was cooled to −70°C , and the gain was set at 200 . The prism-quartz interface was lubricated with glycerol to allow easy translocation of the sample cell on the microscope stage . The beam was totally internally reflected at an angle of 72° from the surface normal , resulting in an evanescent wave that decays exponentially with a characteristic penetration depth of ~100 nm . An elliptical area of 250 μm x 65 μm was illuminated . The laser intensity , shutter , and camera were controlled by a homemade program written in LabVIEW ( National Instruments ) . Experiments triggering secretory vesicle fusion with calcium were performed on a Zeiss AxioObserver Z1 fluorescence microscope ( Carl Zeiss ) , with objective and TIRF setups as described above . The light source was a 488 or 514 nm beamline of an argon ion laser ( Innova 90C , Coherent ) , controlled through an acousto-optic modulator ( Isomet ) , and a diode laser ( Cube 640 , Coherent ) emitting light at 640 nm . The characteristic penetration depths were between 90 and 130 nm . An OptoSplit ( Andor Technology ) was used to separate the fluorescence from the two colors . Fluorescence signals were recorded by an EMCCD ( iXon DV887ESC-BV , Andor Technology ) . The EMCCD camera was cooled to −70°C , and the electron gain was set at 200 . Planar supported bilayers containing syntaxin-1a:SNAP-25 ( bulk phase-facing leaflet lipid composition of 25:25:15:30:4:1 bPC:bPE:bPS:Chol:PI:bPIP2 ) were incubated with 0 . 5 μM Munc18 and 2 μM complexin-1 . Secretory vesicles were then injected while keeping the concentrations of Munc18 and complexin-1 constant . Secretory vesicle docking was allowed to occur for ~20 min before the chamber was placed on the TIRF microscope and the microscope was focused on the planar supported membrane . Docking was quantified by counting bound vesicles and normalizing the numbers to those obtained under a reference condition ( syx ( 183-288 ) :dSN25 ) on the same day . Fluorescence from the sample was recorded while buffer containing 100 μM calcium was injected with a soluble Alexa647 dye in the buffer to monitor the arrival of calcium at the observation site . Steady state FCS measurements were acquired on a Zeiss LSM 880 confocal microscope ( Oberkochen , Germany ) equipped with a 40x/1 . 2 M27 W Korr C-Apochromat objective . Sample temperature was monitored using a Thor Labs ( Newton , New Jersey ) TSP01 external temperature probe . Confocal volume was determined using Rhodamine B with a known diffusion coefficient of 4 . 50 × 10−5 cm2s−1 at 25 . 0°C ( Gendron et al . , 2008 ) diluted in ultra-pure water . Samples containing GFP were excited at 514 nm at an average power of 10 . 5 μW measured at the objective with an external Thor Labs PM16-120 using an argon laser and fluorescence emission was detected from 526 to 695 nm using a GaAsp PMT spectral detector . Twenty steady state FCS correlation curves were collected for 30 s intervals for each sample , that were then averaged and fit to a one-state three-dimensional diffusion model incorporating blinking , ( 1 ) Gτ=1+∑i=13fi*ni2∑i=13fi*ni21+ττd , i*1+ττd , i*ωr2ωz212*1+Tb*eττb1-Tbusing Zeiss’s Zen Black 2 . 3 FCS software extension pack where f is the fraction of molecules , n is the molecular brightness , τd is the diffusion time , ωr is the lateral focus radius , ωz is the axial focus radius , Tb is the blinking fraction and τb is the blinking relaxation . FCS data for cytokine-treated granules contained an additional fast component consistent with the presence of a triplet state ( Figure 1—figure supplement 3 ) . To fit these datasets , a linearly independent triplet state correction was added to Equation 1 to yield a one-state three-dimensional diffusion model incorporating both blinking and triplet state dynamics , ( 2 ) Gτ=1+∑i=13fi*ni2∑i=13fi*ni21+ττd , i*1+ττd , i*ωr2ωz212*1+Tb*eττb1-Tb*1+Tt*eττt1-Ttwhere Tt is the triplet state fraction and τt is the triplet state relaxation time . More complex fitting models incorporating rotational diffusion were considered and applied with no appreciable change in diffusion coefficient . Temperature was monitored throughout the experiment and correlated to each run to adjust for thermal fluctuations . The Vogel equation was used to account for changes in viscosity ( η ) as a function of temperature in Kelvins , ( 3 ) ηT=eA+BC+Twhere A , B and C are fit parameters -3 . 72 , 578 . 919 and -137 . 546 , respectively . The radius of hydration ( Rh ) was calculated using the Stokes-Einstein relation for each 20-run sample collection , ( 4 ) Rh=kT6πηTD ( T ) where the viscosity ( η ) and diffusion coefficients ( D ) are both functions of temperature in Kelvin . A minimum of three independent 20x30 second runs were performed for each sample . This triplicate was collected on minimally 3 independent days where confocal volume and temperature changes were accounted for . Final averaging of data was performed across radius of hydration Rh measurements to account for temperature and viscosity differences across samples . The reported Rh values correspond to a minimum of 20x30 second runs averaged three times for three independently run samples resulting in an average hydrodynamic radius comparison over 180 trial runs . For each sample , 3 . 5 μl of purified secretory vesicles were applied to C-flat holey carbon grids ( Electron Microscopy Sciences ) then manually blotted and plunge-frozen into liquid ethane . Images were recorded at a magnification of 25 , 000X under low electron-dose conditions using a Tecnai F20 electron microscope operating at 120kV with a 4096 × 4096 pixel CCD camera ( Gatan , Pleasanton , CA ) . Diameters of vesicles in micrographs were measured manually in Fiji ( Schindelin et al . , 2012 ) .

Diabetes is a disease that occurs when sugar levels in the blood can no longer be controlled by a hormone called insulin . People with type 1 diabetes lose the ability to produce insulin after their immune system attacks the β-cells in their pancreas that make this hormone . People with type 2 diabetes develop the disease when β-cells become exhausted from increased insulin demand and stop producing insulin . β-cells store insulin in small compartments called granules . When blood sugar levels rise , these granules fuse with the cell membrane allowing β-cells to release large quantities of insulin at once . This fusion is disrupted early in type 1 diabetes , but later in type 2: the underlying causes of these disruptions are unclear . In the laboratory , signals that trigger inflammation and molecules called fatty acids can mimic type 1 or type 2 diabetes respectively when applied to insulin-producing cells . Kreutzberger , Kiessling et al . wanted to know whether pro-inflammatory molecules and fatty acids affect insulin granules differently at the molecular level . To do this , insulin-producing cells were grown in the lab and treated with either fatty acids or pro-inflammatory molecules . The insulin granules of these cells were then isolated . Next , the composition of the granules and how they fused to lab-made membranes that mimic the cell membrane was examined . The experiments revealed that healthy β-cells have two types of granules , each with a different version of a protein called synaptotagmin . Cells treated with molecules mimicking type 1 diabetes lost granules with synaptotagmin-7 , while granules with synaptotagmin-9 were lost in cells treated with fatty acids to imitate type 2 diabetes . Each type of granule responded differently to calcium levels in the cell and secreted different molecules , indicating that each elicits a different diabetic response in the body . These findings suggest that understanding how insulin granules are formed and regulated may help find treatments for type 1 and 2 diabetes , possibly leading to therapies that reverse the loss of different types of granules . Additionally , the molecules of these granules may also be used as markers to determine the stage of diabetes . More broadly , these results show how understanding how molecule release changes with disease in different cell types may help diagnose or stage a disease .

2020

Distinct insulin granule subpopulations implicated in the secretory pathology of diabetes types 1 and 2

Members of the Tolloid family of metalloproteinases liberate BMPs from inhibitory complexes to regulate BMP gradient formation during embryonic dorsal-ventral axis patterning . Here , we determine mechanistically how Tolloid activity is regulated by its non-catalytic CUB domains in the Drosophila embryo . We show that Tolloid , via its N-terminal CUB domains , interacts with Collagen IV , which enhances Tolloid activity towards its substrate Sog , and facilitates Tsg-dependent stimulation of cleavage . In contrast , the two most C-terminal Tld CUB domains mediate Sog interaction to facilitate its processing as , based on our structural data , Tolloid curvature positions bound Sog in proximity to the protease domain . Having ascribed functions to the Tolloid non-catalytic domains , we recapitulate embryonic BMP gradient formation in their absence , by artificially tethering the Tld protease domain to Sog . Our studies highlight how the bipartite function of Tolloid CUB domains , in substrate and ECM interactions , fine-tune protease activity to a particular developmental context . Bone Morphogenetic Proteins ( BMPs ) represent a conserved family of secreted signalling proteins that regulate diverse processes during development and homeostasis ( Wu and Hill , 2009 ) . Multiple tiers of regulation within the BMP signalling pathway exist , including extracellular modulators that form inhibitory complexes with BMPs to prevent ligand interaction ( Wharton and Serpe , 2013 ) . These extracellular BMP inhibitors are crucial during dorsal-ventral axis patterning of the early embryo , where they act to redistribute broadly expressed BMP ligands to form a gradient of BMP activity . The mechanism of BMP gradient formation has been extensively studied in the early Drosophila embryo , where a heterodimer of the BMP ligands , Dpp and Scw , acts as the most potent BMP signalling species ( Shimmi et al . , 2005 ) . In dorsolateral regions , an inhibitory complex is formed , with the Dpp-Scw heterodimer bound by the extracellular antagonists Sog and Tsg ( Wharton and Serpe , 2013 ) , aided by the scaffold protein Collagen IV ( Wang et al . , 2008; Sawala et al . , 2012; Wharton and Serpe , 2013 ) . In this inhibitory complex , the Dpp-Scw heterodimer is unable to bind to receptors or Collagen IV , but is free to diffuse dorsally . The secreted Tld metalloproteinase cleaves Sog to release active BMP ligand , however in dorsolateral regions due to the high concentration of Sog , the ligand dimer is rebound . Thus it is through a cycle of complex formation , diffusion and cleavage that the ligands accumulate at the dorsal midline where , upon cleavage by Tld , Dpp-Scw is free to signal ( Wharton and Serpe , 2013 ) . A second phase of gradient formation involving intracellular positive feedback then refines this initial broad Dpp-Scw gradient into a peak of BMP receptor activation at the dorsal midline ( Wang and Ferguson , 2005; Gavin-Smyth et al . , 2013 ) . Enzymes within the Tld family have a well-characterised domain structure in which the N-terminal protease domain is followed by a series of non-catalytic CUB ( Complement-Uegf-BMP1 ) and EGF ( Epidermal Growth Factor-like ) domains ( Muir and Greenspan , 2011 ) ( Figure 1A ) . Structural analysis of mammalian Tolloids has revealed that both mammalian Tolloid ( mTld ) and Tolloid-like-1 ( Tll-1 ) form dimers in which the most C-terminal EGF and CUB domains act in a substrate exclusion mechanism to restrict enzyme activity ( Berry et al . , 2009 , 2010 ) . In contrast , the highly active , shorter BMP1 protease is monomeric ( Berry et al . , 2009 ) . Although Drosophila Tld possesses a similar domain structure to mTld , previous studies suggest that removing the three most C-terminal domains , to mimic a shorter BMP1-like form , results in a loss of activity ( Canty et al . , 2006 ) . Therefore , a substrate exclusion mechanism may not exist for Drosophila Tld . In addition , despite single amino acid mutations in four out of the five CUB domains resulting in some degree of ventralisation of the Drosophila embryo , the requirement for these domains in Drosophila Tld function is unclear ( Ferguson and Anderson , 1992; Childs and O'Connor , 1994; Finelli et al . , 1994 ) . 10 . 7554/eLife . 05508 . 003Figure 1 . Structure of Drosophila Tld . ( A ) Cartoon of Drosophila Tld domain organisation . E = EGF-like domain . SP = specific peptide . ( B ) Low resolution ab initio model generated by DAMMIN and DAMAVER from solution SAXS data shown in three orthogonal orientations . ( C ) Overlay of the ab initio model shown in ( B ) displayed as chicken-wire with a representative rigid body model generated by SASREF . ( D ) The rigid body model shown in ( C ) with Tld domains labelled . ( E ) 3D reconstruction of the Drosophila Tld monomer by single particle analysis electron microscopy shown in three orthogonal orientations . ( F ) The rigid body model was fitted into the EM volume shown in ( E ) , here displayed as chicken-wire . All scale bars = 5 nm . See also Figure 1—figure supplements 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 05508 . 00310 . 7554/eLife . 05508 . 004Figure 1—figure supplement 1 . Drosophila Tld SAXS data analysis . Drosophila Tld SAXS data showing the SAXS data curve , Kratky plot , Distance distribution function ( p ( r ) ) and Guinier plot . Maximum particle dimension is indicated by the point at which the p ( r ) function reaches zero . DOI: http://dx . doi . org/10 . 7554/eLife . 05508 . 00410 . 7554/eLife . 05508 . 005Figure 1—figure supplement 2 . Drosophila Tld TEM data . Single particle EM data ( i ) Representative class averages used in the 3D reconstruction after 10 iterations of multi-reference alignment and refinement box size shown . ( ii ) Re-projections from the Drosophila Tld 3D EM model . ( iii ) Surface views of the Tld 3D reconstruction as calculated by angular reconstitution . All box sizes = 23 × 23 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 05508 . 005 Here we combine biophysical , biochemical and genetic approaches to determine how the domain structure of Tld acts to regulate its enzymatic activity . We propose that the role of Drosophila Tld CUB domains can be segregated; the most C-terminal domains are necessary for Sog interaction whilst the N-terminal domains manage a novel interaction with Collagen IV that acts to enhance Tld activity . In addition , the Tld-Collagen IV interaction is important for the Tsg enhancement of Sog processing by Tld . We also demonstrate that the need for Tld CUB domains can be bypassed by engineering the Tld–Sog interaction in a highly efficient manner . We investigated the structure of Drosophila Tld purified from tissue culture cells using Small Angle X-ray Scattering ( SAXS ) . Drosophila Tld has a maximal particle dimension ( Dmax ) of 17 . 7 nm and a radius of gyration ( Rg ) of 4 . 6 nm ( Figure 1—figure supplement 1 ) . Ab inito modelling of the most probable shape of Drosophila Tld gives an elongated monomer with the approximate dimensions 18 × 6 . 5 × 6 . 8 nm ( Figure 1B ) . Further detail was extrapolated from the SAXS data on the location of the CUB and EGF domains of Tld by rigid body modelling ( RBM ) . The model retains similar characteristics to the SAXS bead model with the approximate dimensions of 16 . 6 × 7 × 9 . 2 nm . An overlay of the ab inito model and rigid body model ( Figure 1C ) shows that Tld displays an elongated conformation with protein curvature around the C-terminal tail ( Figure 1D ) . This brings the most C-terminal CUB domains into close proximity to the N-terminal protease domain . Single particle analysis TEM demonstrates that Drosophila Tld is a monomer , with the overall shape of the 3D reconstruction in agreement with the SAXS model ( Figure 1E , Figure 1—figure supplement 2 ) . In addition to being relatively elongated , the curvature of the EM model mimics that seen in the RBM ( Figure 1F ) . However , it is evident that the TEM model is shorter; this suggests flexibility of the most C-terminal regions that is likely to result in loss of information on this region during the averaging procedure . We investigated the role of individual CUB domains , with particular interest in the most C-terminal CUB domains , CUB4 and CUB5 , due to their position within the rigid body model structure . Each domain was independently deleted in full-length Tld and the ability of these variants , as well as the isolated metalloprotease ( MP ) domain , to cleave Sog substrate was tested ( Figure 2A , B ) . Cleavage of Sog by Tld relies on the presence of Dpp ( Marques et al . , 1997 ) ( Figure 2—figure supplement 1 ) , as such Dpp was added to all in vitro assays . Only ΔCUB2 retained the ability to cleave Sog , although the level of cleavage by ΔCUB2 is lower than full-length Tld ( Figure 2C ) . Additionally , whilst increasing levels of Tsg enhances Tld cleavage of the Sog 50 and 33 kDa fragments , especially the 33 kDa fragment as described previously ( Shimmi and O'Connor , 2003 ) , ΔCUB2 mediated cleavage is not enhanced ( Figure 2C ) . Therefore , while CUB2 is not required for Tld activity , its presence is necessary for Tsg enhancement of cleavage . 10 . 7554/eLife . 05508 . 006Figure 2 . Tld CUB4/5 domains mediate Sog binding . ( A ) Schematic of Sog showing the CR domains and the position of the Tld cleavage sites . The sizes of the C-terminal fragments liberated by cleavage , as detected by the C-terminal Myc tag , are indicated below the arrows . ( B ) Western blot ( anti-Myc ) of Sog cleavage assays carried out using normalised amounts of the Tld deletion proteins indicated , all in the presence of Dpp and absence of Tsg . C = control assay with no Tld added . MP = the isolated metalloproteinase domain . ( Ci ) A representative Western blot ( anti-Myc ) showing Sog cleavage assays carried out with normalised levels of Tld and ΔCUB2 in the presence of Dpp and either in the absence ( − ) , or with increasing amounts ( + , ++ ) of Tsg . ( Cii ) Graph shows the amount of cleavage measured as the amount of the 50 kDa and 33 kDa fragments quantified relative to cleavage calculated for wildtype Tld in the absence of Tsg , based on three different experiments including the Western blot shown in i . Error bars show SEM , n = 3 , *p < 0 . 005 , **p < 0 . 001 . ( Di ) Western blot ( anti-HA ) showing 10% input of catalytically inactive forms of Tld-HA proteins and the amount bound to Sog-Myc by immunoprecipitation . ( Dii ) Quantitation of the level of binding of each Tld protein to Sog , relative to binding of full-length Tld . Error bars are SEM , n = 4 . See also Figure 2—figure supplements 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 05508 . 00610 . 7554/eLife . 05508 . 007Figure 2—figure supplement 1 . Sog cleavage by Tld is dependent on Dpp . Western blot ( anti-Myc ) showing Sog cleavage by Tld in the presence of increasing amounts ( + , ++ , +++ ) of the Dpp ligand . No Sog cleavage by Tld is detectable in the absence of Dpp . DOI: http://dx . doi . org/10 . 7554/eLife . 05508 . 00710 . 7554/eLife . 05508 . 008Figure 2—figure supplement 2 . Tld CUB4-CUB5 domain pair is specifically required at the C-terminus . Western blot ( anti-HA ) of the indicated Tld protein levels following expression in Drosophila S2R+ cells with Sog cleavage in the presence of Dpp shown below on an anti-Myc Western blot . The 2xCUB4 and 2xCUB5 variants retain five CUB domains but have a CUB4 or CUB5 duplication at the C-terminus instead of the CUB4-5 domain pair . DOI: http://dx . doi . org/10 . 7554/eLife . 05508 . 008 In order to test whether the loss of activity of these Tld variants is due to an effect on substrate interaction , the ability of each Tld protein to bind to Sog was assessed using immunoprecipitation . To facilitate detection of the Sog–Tld interaction , for all Sog binding experiments described here we used catalytically inactive versions of Tld and the various derivatives tested , by introducing the previously described E94A mutation into the protease active site ( Garrigue-Antar et al . , 2001 ) . Analysis of Sog–Tld interactions reveals that , similar to the isolated protease domain , the CUB4 and CUB5 deletions are unable to interact with Sog , suggesting that these domains mediate Sog interaction ( Figure 2D ) . To determine whether the loss of activity for the ΔCUB4 and ΔCUB5 forms is due to a specific requirement for both CUB4 and CUB5 domains in Sog interaction rather than an issue of protein length , full-length variants of Tld were generated with CUB4 or CUB5 domain duplications at the C-terminus . However , these enzymes are also inactive , demonstrating a specific requirement for the CUB4–CUB5 domain pair at the C-terminus of Tld ( Figure 2—figure supplement 2 ) . For the other single CUB deletions , ΔCUB2 interacts with Sog to a similar extent as wildtype Tld even though its activity is reduced and , despite being inactive , the ΔCUB1 and ΔCUB3 forms can bind Sog , albeit at reduced levels ( Figure 2D ) . These data suggest that a correct protein conformation is important for Tld activity , or that CUB1 and CUB3 mediate a different protein interaction that is important for Sog binding ( see later ) . Four single amino acid mutations situated within CUB4 or CUB5 have previously been isolated from genetic screens and described in terms of the cuticle phenotype and the strength of genetic interactions with dpp mutations ( Ferguson and Anderson , 1992; Childs and O'Connor , 1994; Finelli et al . , 1994 ) ( Figure 3A ) , yet the molecular basis for the decrease in Tld activity is unclear . Therefore , based on our data that CUB4 and CUB5 mediate Sog interaction , we tested whether the molecular defect associated with these mutant Tld proteins is reduced binding to Sog . First we extended the phenotypic analysis for the tldE839K mutation , by determining the effect of this mutation on the expression of target genes responding to different thresholds of Dpp activity . Visualisation of the expression patterns of the Dpp target genes Race and u-shaped ( ush ) in tldE839K homozygous mutant embryos reveals that expression of the peak threshold gene Race is lost and expression of the lower threshold gene ush is narrower than in wild-type embryos ( Figure 3B ) , suggesting shallow gradient formation , consistent with the previous characterisation of this allele as weak . 10 . 7554/eLife . 05508 . 009Figure 3 . Altered Sog binding of Tld point mutants . ( A ) Table summarising previous experimental data regarding the CUB4 and CUB5 mutations and the severity of the resulting phenotypes ( Childs and O'Connor , 1994; Finelli et al . , 1994 ) . ( Bi ) Dorsal views of stage 6 wildtype and tldE839K homozygous embryos stained by RNA in situ hybridisation for the Dpp target genes Race and u-shaped . Phenotypes were quantified in Bii as the percentage of embryos with either wildtype or lost/narrow expression for Race and ush . ( C ) Models of Tld CUB4 and CUB5 with the single amino acid mutations highlighted in green . ( Di ) Western blot with anti-HA showing the input levels of the catalytically inactive Tld-HA proteins tested and the amounts bound to Sog-Myc in an immunoprecipitation experiment . ( Dii ) Graph shows the level of binding to Sog relative to that for full-length Tld . ( E ) Western blot ( top , anti-HA ) showing the levels of Tld proteins carrying the indicated single amino acid changes . Western blots ( anti-Myc ) show 1 hr ( Ei ) or 7 hr ( Eii ) Sog cleavage assays using the Tld proteins shown . Below the blot on the right is a longer exposure of the 50 kDa cleavage fragment . DOI: http://dx . doi . org/10 . 7554/eLife . 05508 . 009 Identification of the positions of the four mutated residues within the individual CUB domain structures reveals that all lie on the surface of the protein ( Figure 3C ) and therefore can by hypothesised to be important for substrate binding . To directly test the effect of each mutation on Sog binding and cleavage activity , we engineered these mutations into Tld-HA and expressed the mutant proteins . Each mutant shows reduced binding to Sog ( Figure 3D ) and a loss of cleavage activity when tested over a short reaction time ( Figure 3Ei ) , although upon a longer incubation time varying levels of weaker cleavage activity can be detected for all of the mutant Tld proteins tested ( Figure 3Eii ) . Together , these data provide molecular understanding of how these mutations impair Tld activity , and provide in vivo support for the importance of CUB4 and CUB5 in mediating the Tld–Sog interaction . In a pilot mass spectrometry screen for Tld binding proteins we identified Collagen IV , an extracellular matrix protein that we have previously shown to be required for Dpp gradient formation ( Wang et al . , 2008 ) . To both validate a direct interaction between Tld and Collagen IV , and determine which domains in Tld are responsible for this interaction , we tested the ability of full-length Tld and the individual CUB domain deletion variants to bind to the NC1 domain of the Viking ( Vkg ) Collagen IV protein fused to GST . The Collagen IV NC1 domain is a highly conserved globular domain that has been shown to bind Dpp and Sog previously ( Wang et al . , 2008; Sawala et al . , 2012 ) . Tld binds to Collagen IV via its N-terminal CUB domains , with deletion of CUB1 , CUB2 or CUB3 reducing binding to <25% that of full-length ( Figure 4A ) . However , removal of the Sog interacting domains , CUB4 and CUB5 , has little effect on the Tld-Collagen IV interaction . No interaction with the control GST protein was observed for any of the Tld-HA proteins ( data not shown ) . To test the effect of Collagen IV on Tld and Sog interaction , their binding was investigated in the presence of an increased amount of Collagen IV . Addition of the NC1 domains from both Drosophila Collagen IV proteins , Vkg and Dcg1 , increases the Sog–Tld interaction ( Figure 4B ) . 10 . 7554/eLife . 05508 . 010Figure 4 . Tld binds Collagen IV . ( Ai ) Western blot showing Tld-HA input proteins and the amount bound to GST-VkgNC1 in a pulldown experiment . ( Aii ) Graph shows the level of binding to Collagen IV plotted relative to binding of full-length Tld , based on data from three experiments . Error bars are SEM , n = 3 , *p < 0 . 05 . ( B ) Western blot ( anti-HA ) showing Tld input and the levels bound to Sog in the presence of Dpp , with or without ( ± ) added Collagen IV . ( Ci ) Western blot ( anti-Myc ) of Sog cleavage by full-length Tld or the ΔCUB2 mutant quantified in ( Cii ) . The x-axis labels ( – , + , ++ ) refer to the absence , or increasing amounts , of Tsg . Error bars are SEM , n = 3 . ( Di ) Western blot ( anti-Myc ) of Tld cleavage of SogCR14* , with cleavage levels quantified in ( Dii ) . Both cleavage assays were carried out over 3 hr , in the presence of Dpp , and in the absence or presence of increasing amounts of Tsg , with or without of Collagen IV , as indicated . Graphs show the fold-change in the relative levels of the 50 kDa and 33 kDa cleavage fragments released by Tld ( green ) or ΔCUB2 ( blue ) , in the presence of Collagen IV relative to its absence . See also Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05508 . 01010 . 7554/eLife . 05508 . 011Figure 4—figure supplement 1 . SogCR14* is inefficiently processed by Tld in vivo . ( i ) Confocal projections showing lateral views of stage 5 ls-sog-Myc and ls-sogCR14*-Myc transgenic embryos stained with anti-Myc ( red ) and DAPI ( blue ) . Embryos were stained in parallel and imaged with the same settings . ( ii ) Western blot ( anti-Myc ) of yw , ls-sog-Myc and ls-sogCR14*-Myc embryo extracts . All samples were analysed on the same Western blot , the cut in the blot only reflects removal of an irrelevant lane . ls-sogCR14* embryo extracts contain higher levels of unprocessed Sog at 150 kDa ( black arrowhead ) than ls-sog embryos . The non-specific band at ∼25 kDa marked by the grey arrowhead serves as a loading control . Both sog and sogCR14* transgenic lines were generated by attB/attP recombination into cytosite 86Fb on chromosome 3 to eliminate position effects . DOI: http://dx . doi . org/10 . 7554/eLife . 05508 . 011 Next , the activity of Tld was compared in the presence and absence of Collagen IV . It is likely that soluble Collagen IV , or its cleavage fragments following matrix turnover , are present in the media from transfected cells along with the test proteins . Therefore , to limit Collagen IV levels , test proteins were obtained from transfected cells that were also treated with vkg RNAi . Despite this , we expect that the ‘−Collagen IV’ condition will still contain some Collagen IV , and compare this low level to a ‘+Collagen IV’ condition in which the Vkg and Dcg1 NC1 domains were added . The addition of Collagen IV enhances Tld cleavage at both the 50 and 33 kDa sites of Sog ( Figure 4C ) , consistent with increased Sog–Tld interaction in the presence of Collagen IV ( Figure 4B ) . The addition of Tsg increases Sog cleavage , especially at the 33 kDa site in the presence of Collagen IV . However , the Collagen IV enhancement of Sog processing is reduced in the presence of high levels of Tsg ( Figure 4C ) . In contrast , little enhancement at either the 50 or 33 kDa site is observed in the presence of Collagen IV when the ΔCUB2 mutant , that is defective in Collagen IV interaction , is tested ( Figure 4C ) . As before ( Figure 2C ) , Tsg fails to enhance Sog processing by the ΔCUB2 mutant ( Figure 4C ) , suggesting that Tsg function may depend on a Tld-Collagen IV interaction . These data show that Tld binding to Collagen IV enhances the Sog-Tld interaction and Sog cleavage . To extend these findings , we investigated whether the Collagen IV-Sog interaction is also important for Sog processing by Tld . To this end , we made use of a mutant form of Sog , SogCR14* , we have described previously that harbours mutations in the CR1 and CR4 domains of Sog , resulting in very weak binding to Collagen IV ( Sawala et al . , 2012 ) . This mutant Sog is a much poorer substrate for Tld cleavage compared to wildtype Sog , and no enhancement of cleavage in the presence of added Collagen IV is observed ( Figure 4D ) , contrary to that observed with wildtype Sog ( Figure 4C ) . In addition , there is very little of the 33 kDa cleavage fragment , even when Tsg is added in increasing amounts ( Figure 4D ) , suggesting that Tsg function requires a Sog-Collagen IV interaction , in addition to a Tld-Collagen IV interaction ( Figure 4C ) . Consistent with SogCR14* being a poorer Tld substrate in vitro , analysis of embryos expressing either wildtype Sog or the SogCR14* double mutant under the control of the sog lateral stripe ( ls ) enhancer reveals that the SogCR14* mutant accumulates to a greater level than the wildtype Sog in vivo ( Figure 4—figure supplement 1 ) . Together these data show that binding of both Sog and Tld to Collagen IV is necessary for efficient Sog processing . We hypothesised that the role of CUB2 and CUB3 in Collagen IV interaction can be used to explain classical mutations situated in these domains ( Figure 5A ) . The surface position of these mutations in the CUB2 and CUB3 domains is shown in Figure 5B . The mutations were engineered into Tld-HA and we first tested for Collagen IV binding , which is reduced by at least two-fold in all CUB2/3 point mutants ( Figure 5C ) . We next tested the effect of these point mutations on Sog interaction . All Tld mutants show reduced Sog interaction , with the same severity profile as that observed for the defects in Collagen IV binding ( Figure 5Di , iii , cf with Figure 5Cii ) , consistent with the data presented in Figure 4 that Collagen IV promotes the Tld–Sog interaction . To show that the effects on Sog interaction are an indirect effect of the Tld mutant proteins being unable to bind Collagen IV , which promotes Sog–Tld binding , we uncoupled the assay from Collagen IV , again using the mutant form of Sog that is defective in Collagen IV binding ( Sawala et al . , 2012 ) . As predicted , Tld binding to this mutant form of Sog is weaker as the Collagen IV enhancement is lost ( Figure 5Dii ) . However , the four mutant Tld proteins all show equivalent levels of binding to this form of Sog , as that observed for wildtype Tld ( Figure 5Dii , iii ) . These data reveal that the CUB2/3 mutations do not affect Sog interaction per se , but rather disrupt the stimulatory effect observed through Collagen IV . In terms of cleavage activity , the four mutant Tld proteins show reduced cleavage of Sog , especially of the 33 kDa fragment ( Figure 5E ) , consistent with their reduced ability to bind Sog and the requirement for Collagen IV for Tsg enhancement of processing at the 33 kDa site . Together , these data suggest that the molecular defect underlying the phenotypes associated with the CUB2/3 point mutations is a weak ability to bind Collagen IV . 10 . 7554/eLife . 05508 . 012Figure 5 . Point mutations affect Collagen IV binding . ( A ) Overview of previous data relating to classical tld alleles carrying point mutations in the CUB2 and CUB3 domains detailing the single amino acid change , extent of ventralisation and genetic interaction with dpp ( Childs and O'Connor 1994; Finelli et al . 1994 ) . ( B ) Models of Tld CUB2 and CUB3 with the single amino acid mutations highlighted in orange . ( Ci ) Western blot showing binding of the Tld-HA proteins to GST-VkgNC1 , in the presence of Dpp , relative to the input levels . ( Cii ) Graph showing quantification of binding to Collagen IV normalised to the input levels , relative to binding of full-length Tld . Error bars are SEM , n = 3 . ( D ) Western blot with anti-HA showing input levels of the catalytically inactive Tld-HA proteins tested and the amounts bound to Sog-Myc ( Di ) and SogCR14*-Myc ( Dii ) in immunoprecipitation experiments . ( Diii ) Graph showing the level of binding to Sog ( light green ) and SogCR14* ( dark green ) relative to that for full-length Tld in each case . Error bars for Sog are SEM , n = 3 . ( E ) Western blot ( top , anti-HA ) showing Tld-HA proteins with the indicated single amino acid changes . Western blot ( lower , anti-Myc ) showing Sog cleavage assays ( 4 hr ) using the Tld enzyme indicated above each lane in the presence of Dpp . DOI: http://dx . doi . org/10 . 7554/eLife . 05508 . 012 Based on the functions we have identified for the non-catalytic domains of Tld , we hypothesised that using an artificial method to bring together Sog and the Tld MP domain would rescue protease activity , bypassing both the need for substrate interaction domains and Collagen IV enhancement . To this end leucine zipper sequences ( Z+ and Z− ) ( Luan et al . , 2006 ) were attached to the C-terminus of both the isolated Tld MP domain and full-length Sog ( Figure 6A ) . Immunoprecipitations confirmed the ability of the leucine zippers to induce enzyme-substrate interaction; SogZ+ specifically interacts with the corresponding MPZ− zipper but not the MPZ+ or unmodified MP domain controls ( Figure 6B ) . To test whether the interaction induced by the leucine zippers is sufficient to rescue the activity of the isolated protease domain , cleavage assays were carried out in cell culture . Both full-length Tld and MPZ− cleave SogZ+ to yield two cleavage fragments at 57 kDa and 40 kDa ( Figure 6C ) , equivalent to the 50 and 33 kDa fragments plus the additional zipper sequence . Using this artificial system , Sog and the Tld MP domain can interact in the absence of Dpp , although Dpp still enhances SogZ+ cleavage ( data not shown ) . 10 . 7554/eLife . 05508 . 013Figure 6 . Restoration of in vivo function to the Tld MP domain . ( A ) Cartoon of the leucine zipper strategy . ( B ) Western blot ( anti-HA ) showing the input levels of the MP-HA and MP-Zipper-HA fusion proteins tested , and binding of Sog-Zipper+-Myc following immunoprecipitation with anti-HA . MPZ+ is larger than MPZ− due to the presence of additional glycine residues preceding the zipper sequence . ( C ) Western blot ( anti-Myc ) showing cleavage assays carried out with SogZ+ transfected with the indicated Tld MP-zipper fusion proteins in the presence of Dpp . C = control assay without Tld added . ( D ) Dorsal views of stage 6 embryos , oriented with anterior to the left , stained by RNA in situ hybridisation for the Dpp target genes Race and u-shaped , and stage 9 lateral views of late Race expression . Wild-type embryos are shown for reference in ( Di ) . Embryos in ( Dii ) and ( Diii ) are also stained with a lacZ RNA probe that marks the chromosome carrying wildtype tld ( ftz-lacZ , stripes , not shown ) or chromosome lacking the zipper transgene ( hb-lacZ , anterior staining ) . All embryos shown are tld mutant , and either lack at least one zipper transgene ( Dii ) , or carry both zipper transgenes ( Diii ) . See also Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05508 . 01310 . 7554/eLife . 05508 . 014Figure 6—figure supplement 1 . SogZ + overexpression results in a thinning of Race expression . Dorsal views of stage 6 embryos stained for Race expression by RNA in situ hybridisation . Embryos that overexpress SogZ+ in a wild-type background show a thinning of Race expression in the central region of the embryo . 20% of MPZ−/SogZ+;tldB4 embryos show a similar phenotype . DOI: http://dx . doi . org/10 . 7554/eLife . 05508 . 014 As the MPZ− and SogZ+ protein system facilitated interaction and SogZ+ cleavage , we tested the ability of these proteins to rescue Dpp gradient formation in a tld null background . To this end , SogZ+ was introduced into embryos as a transgene under the control of the sog lateral stripe enhancer ( Markstein et al . , 2002; Peluso et al . , 2011 ) , whereas the MPZ− domain was expressed under tld regulatory sequences ( Kirov et al . , 1994 ) which are sufficient to rescue the null embryonic phenotype when used to drive expression of wildtype tld ( data not shown ) . In a tld null background with expression of either SogZ+ or MPZ− but not both , Race and ush expression are lost ( Figure 6Dii , cf Figure 6Di ) . In contrast , when tld null mutant embryos express both SogZ+ and MPZ− , target gene expression is rescued in 50% of cases ( Figure 6Diii ) , although ∼20% of the rescued embryos show a slight thinning of Race expression in the central region of the embryo , as observed for the addition of an extra copy of ls-sogZ+ ( Figure 6—figure supplement 1 ) . Overall , these findings indicate that this artificial tethering of the MP domain to Sog can be sufficient for BMP gradient formation in vivo . Our biophysical data provide evidence that Tld is a monomer , which efficiently cleaves Sog due to critical bipartite functions of the Tld non-catalytic domains in either substrate or ECM interaction . Thus , Tld represents the first five CUB domain family member that exists as a monomer , as Tll-1 and mTld have previously been shown to be dimers that exclude Chordin substrate resulting in low catalytic activity towards it ( Berry et al . , 2009 , 2010 ) . Thus , whereas the dimeric Tolloids require the affinity of substrate interaction to be higher than the intramolecular affinity within the dimer for activation ( Hintze et al . , 2006; Berry et al . , 2009 ) , this is not a limitation for Drosophila Tld . We show that the Tld CUB1-3 domains bind Collagen IV . Consistent with this , recently an interaction has been identified between Collagen IV and the CUB domains of PCPE-1 ( Salza et al . , 2014 ) , and Collagen IV binds the zebrafish G-protein coupled receptor GPR126 , via a region containing the CUB domain ( Paavola et al . , 2014 ) . We also show that Tsg enhancement of Sog processing by Tld is dependent on the Tld-Collagen IV interaction . Finally , we identify a strategy for artificially tethering the Tld MP domain to Sog in vivo , which can rescue BMP gradient formation . Based on our findings , we revise the model of BMP gradient formation in the Drosophila early embryo ( Holley et al . , 1996; Marques et al . , 1997; Ashe and Levine , 1999; Sawala et al . , 2012 ) , as shown in Figure 7 . In addition to Sog and Dpp-Scw , which have been previously shown to bind independently to Collagen IV ( Sawala et al . , 2012 ) , the data presented here show that Tld , via its CUB1-3 domains , also binds to Collagen IV ( Figure 7 , step 1 ) . In the model , we suggest that initially there is no interaction between Sog , Dpp-Scw or Tld bound to Collagen IV . However , in the next step , when Dpp-Scw is transferred to Sog from Collagen IV , as has been previously described ( Sawala et al . , 2012 ) , we suggest that Tld is transferred at the same time , interacting with Sog via the Tld CUB4 and CUB5 domains ( Figure 7 , step 2 ) . Tsg has previously been shown to release Dpp-Scw/Sog from Collagen IV as a Dpp-Scw/Tsg complex ( Wang et al . , 2008 ) , with Tsg recruitment or interaction potentially stabilised by the BMP binding capacity of Tsg's N-terminal domain ( Oelgeschlager et al . , 2000 ) . Therefore , based on the data obtained here , we suggest that in addition Tld is released from Collagen IV at the same time , pre-bound to the shuttling complex ( Figure 7 , step 3 ) , and is then able to fully process Sog ( step 4 ) . 10 . 7554/eLife . 05508 . 015Figure 7 . Model of Dpp gradient formation showing the Tld-ECM interaction . See text for details . DOI: http://dx . doi . org/10 . 7554/eLife . 05508 . 015 The model predicts that the enhancement of Tld cleavage of Sog that we observe in the presence of Collagen IV is due to Collagen IV acting as a scaffold to promote Tld interaction with Sog . We show reduced Collagen IV enhancement of Sog processing in vitro in the presence of high Tsg levels , relative to that detected with lower Tsg levels . As Tsg has been suggested previously to compete with Collagen IV for binding to the SogCR1 domain to release the Sog-BMP complex from Collagen IV ( Sawala et al . , 2012 ) , it is possible that there is little Sog binding to Collagen IV when Tsg is in excess , so the positive effect of Collagen IV is reduced . In the model presented , the BMP dependency for Sog cleavage by Tld arises from the requirement for Dpp-Scw to disrupt the SogCR4-Collagen IV interaction as described previously ( Sawala et al . , 2012 ) , resulting in an ‘opening’ of Sog that facilitates Tld binding . In support of this , we find that Tld is unable to bind to Sog in the absence of Dpp ( data not shown ) . In addition , our mechanism of gradient formation puts the emphasis on a balance between the distance that the Tld-BMP-Sog-Tsg complex diffuses and the rate with which Tld cleavage occurs , rather than on a limitation in terms of Tld recruitment . Although Tld levels are not limiting in the Drosophila embryo ( Eldar et al . , 2002 ) , it is possible that Tld activity is suboptimal in order to allow sufficient diffusion required for peak gradient formation . Consistent with this idea , mathematical modelling has revealed that lowering Tld processing activity favours a more defined BMP peak , due to an increase in BMP shuttling to the dorsal midline ( Peluso et al . , 2011 ) . In addition , a recent model of BMP shuttling in the sea urchin embryo finds Tld to be a dominant determinant of the system , with reduced levels of Tld relative to the fly system necessary for the best fit to the spatial extent of the BMP signalling peak in sea urchins ( van Heijster et al . , 2014 ) . In the absence of Collagen IV , or the binding of Sog or Tld to it , we observe little Tsg enhancement of Sog cleavage and liberation of the 33 kDa fragment . For example , all of the CUB2/3 point mutants are more severely compromised in their ability to cleave the 33 kDa site , compared to the 50 kDa site . Similarly , we also show that Tld cleavage of a mutant form of Sog , which is defective in Collagen IV interaction , does not release the 33 kDa fragment . As Tsg releases the BMP-Sog-Tld complex from Collagen IV in our model , we suggest that this release involves a remodelling of protein interactions that alters the interaction of Tld with Sog allowing better access to all Sog cleavage sites , but especially the 33 kDa site , which is Tsg-dependent ( Shimmi and O'Connor , 2003 ) . Moreover , we speculate that Tld can cleave the 50 kDa Sog site , that does not require Tsg , when the BMP-Sog-Tld complex is bound to Collagen IV , although the dual binding of Tld to Collagen IV and Sog holds Tld in a conformation that prevents both processing at the 33 kDa site and inactivation of Sog , prior to Tsg release . If these assumptions are correct , it follows that cleavage at the 33 kDa site within Sog is necessary for complete loss of Sog inhibition of Dpp-Scw . In support of this , the Sog C-terminal fragments generated by cleavage at the other two sites retain some BMP inhibitory activity when assayed in the Drosophila wing ( Yu et al . , 2004 ) . In addition , complete cleavage of Chordin , the mammalian Sog ortholog , is required for relief of BMP inhibition ( Troilo et al . , 2014 ) . We also note that although our zipper system can rescue BMP signalling thresholds in the embryo , cleavage at the 33 kDa equivalent cleavage fragment of the Sog-zipper fusion protein is the preferred site , consistent with cleavage at this site being key to full release of the Sog inhibitory effect . In addition , if in tsg mutants Tld is unable to completely release the BMP inhibition of Sog , this may explain the only partial rescue of tsg mutants by overexpression of Supersog ( Yu et al . , 2000 ) . Supersog consists mainly of the Sog CR1 domain and has been suggested , like Tsg , to compete with full-length Sog for binding to Collagen IV to release the shuttling complex ( Sawala et al . , 2012 ) . However , it may be that incomplete processing of Sog by Tld in the absence of Tsg does not relieve full inhibition of the BMP ligands . Such a role for Tsg would be in addition to the Sog-independent positive role with respect to BMP signalling that has also been described ( Wang and Ferguson , 2005 ) . The hypothesised roles of Drosophila Tld CUB domains in both Collagen IV and Sog interaction provide molecular insight into the phenotypes of classical tld alleles arising from point mutations in these domains . Mutations in CUB1-2 and CUB4-5 reduce Collagen IV and Sog binding , respectively , resulting in lower levels of Sog cleavage . As the shuttle-transport mechanism relies on cycles of complex diffusion , destruction and reformation , decreased Sog cleavage disrupts formation of the Dpp activity gradient resulting in ventralisation . Our in vitro data show that the CUB2 deletion behaves differently from the CUB2 point mutants in that , although the ΔCUB2 and point mutants all show reduced binding to Collagen IV , unlike the point mutants , the ΔCUB2 mutant remains able to bind Sog , although it cleaves very poorly . We suggest that the ΔCUB2 mutant is conformationally different so it can bind to Sog without prior presentation by Collagen IV , but cleavage is affected , especially of the 33 kDa fragment , and not enhanced by Tsg , which may reflect perturbation of the Tld–Sog interaction , as proposed above . The CUB2 point mutant tld alleles have been shown to be antimorphic when transheterozygous with either weak tld or dpp alleles ( Ferguson and Anderson , 1992; Finelli et al . , 1994 ) . Revertants of these antimorphic mutations have also been identified within the CUB1 , 2 and 5 domains ( Childs and O'Connor , 1994 ) , but as they reduce secretion of the mutant Tld proteins ( Lee et al . , 2009 ) , they provide only limited insight into the molecular mechanism of the antimorphic mutants . Nevertheless , based on the finding that Drosophila and vertebrate Tld CUB1/2 or CUB4/5 domain pairs can bind BMP4 and BMP7 , and inhibit BMP signalling when injected into Xenopus embryos ( with reciprocal inhibition of Tld by BMP interaction ) , it was proposed that the antimorphic effect of tld alleles may be due to an inhibition of Dpp activity ( Lee et al . , 2009 ) . However , the role of the Tld-BMP interaction in the Drosophila embryo is not fully understood and only point mutants in the CUB2 or protease domains behave in an antimorphic manner ( Ferguson and Anderson , 1992; Finelli et al . , 1994 ) . One possibility is that Tld and Dpp binding to Collagen IV largely prevents a direct Tld–Dpp interaction , but instead favours shuttling complex assembly in the presence of Sog and Tsg . In the absence of Collagen IV binding , the CUB2 point mutants could inhibit Dpp to some extent , further reducing Dpp activity that is already compromised in a background heterozygous for a weak tld or dpp allele . The differential specificity of Tolloid non-catalytic domains , and in particular ECM interaction , may be relevant to the fine tuning of Tolloid activity not only during BMP gradient formation in other organisms , but also in a wide range of developmental contexts in general . In Xenopus embryos , the Chordin/BMP ligand gradient diffuses within the Fibronectin-rich Brachet's cleft ( Plouhinec et al . , 2013 ) , and Fibronectin has also been shown to bind to Chordin and BMP1 , to enhance the processing of Chordin during DV patterning of vertebrate embryos ( Huang et al . , 2009 ) . Similarly , ONT1 , a secreted Olfactomedin-class protein , facilitates Chordin-BMP1 interaction by independently recruiting both proteins , resulting in robust BMP gradient formation during embryonic DV patterning . In addition , for procollagen III processing by BMP1 in the presence of PCPE-1 , heparin-like sulphated glycosaminoglycans may act as a scaffold for BMP1 processing of procollagen III in vivo ( Bekhouche et al . , 2010 ) . Therefore , Tld family members may generally rely on a scaffold to facilitate substrate interaction , as proposed here for Collagen IV . Using a leucine zipper based interaction system ( Luan et al . , 2006 ) we were able to show that artificially tethering the Tld MP domain to Sog is sufficient to support BMP gradient formation . It would be interesting to determine whether tethering the BMP1 protease domain to Chordin can circumvent the requirement for the BMP1 non-catalytic domains in vertebrates . Perhaps this would allow too much turnover of Chordin in vivo given that BMP1 cleavage of Chordin is independent of BMPs . In addition , BMP1 MP domain tethering would prevent interaction with secreted Frizzled-related proteins , which are recruited via the non-catalytic domains of mammalian Tolloids to mediate important further positive and negative control of activity in specific regions of vertebrate embryos ( Muraoka et al . , 2006; Kobayashi et al . , 2009; Ploper et al . , 2011; De Robertis and Colozza , 2013; Inomata et al . , 2013 ) . It is interesting that Drosophila lacks secreted Frizzled-related proteins , suggesting that ECM control of the Tld-substrate interaction is sufficient for fine control of Tld activity and the resulting Sog and BMP levels in the Drosophila embryo . This may stem from Tld processing of Sog being BMP-dependent in Drosophila , a situation that has already been shown to favour the steep BMP gradient required in this context ( Peluso et al . , 2011 ) . Given that the leucine zipper system mediates an interaction between two different proteins synthesised in distinct regions of the embryo , we suggest that it will be useful for not only determining sufficiency of other interactions , but also for determining the effect of mediating ectopic or prolonged interactions between other test proteins in vivo . The Cu-inducible SogCR14*-Myc ( Sawala et al . , 2012 ) and Sog-Myc plasmids and pGEX4T1-VkgC have been described ( Wang et al . , 2008 ) . The Cu-inducible Tsg-V5 plasmid was generated by inserting the Tsg CDS from pRmHa1-Tsg-His ( Yu et al . , 2000 ) into pMT-V5-His . Tld-HA , in pRmHA1-aTld ( Marques et al . , 1997 ) , was modified by PCR to introduce deletions and single amino acid mutations . To generate VkgC-FLAG and Dcg1C-FLAG the NC1 domain was inserted into pMTBip followed by a C-terminal FLAG tag . Leucine zipper constructs were generated by insertion of zipper sequences from pActPL-Gal4AD and pActPL-Gal4DBD ( Luan et al . , 2006 ) into Sog-Myc and MP-HA . For injection into Drosophila embryos the CDS from MPLZ- was inserted into pCasper-attB containing the tld promoter as defined previously ( Kirov et al . , 1994 ) , tld 3′UTR and 1 kb of additionally regulatory sequences . The CDS of Sog-Myc , SogCR14*-Myc or SogLZ+-Myc was inserted into pCasper-attB , together with the Sog promoter and 3′UTR taken from pCasper-ls-Sog ( Peluso et al . , 2011 ) . For large-scale protein expression Tld was catalytically inactivated by mutation E94A using Quikchange mutagenesis ( Agilent Technologies , Santa Clara , CA ) and ligated into pCEP-PU using a Not1 restriction enzyme to incorporate a 6 × His tag at the C-terminus . Tld-HA , Sog-Myc , Tsg-V5 , VkgC-FLAG and Dcg1C-FLAG were produced in Drosophila S2R+ cells using the Effectene transfection kit ( Qiagen , Venlo , Netherlands ) and 2 µg DNA per well in a six-well plate format . Protein expression was induced after 24 hr by the addition of 500 µM CuSO4 and secreted proteins harvested after 72 hr . For leucine zipper cleavage assay transfections , S2R+ cells were co-transfected with a combination of 1 . 5 µg SogLZ+-Myc , and 0 . 5 µg MPZ+/MPZ−/MP or Tld as a positive control . For RNAi-mediated knock down of Collagen IV , cells were transfected and incubated for 24 hr at 25°C . The cells were then treated for 1 hr in serum-free GIBCO media with 5 µg viking dsRNA ( generated from yw genomic DNA using MEGAscript T7 with the primers listed below ) , followed by the addition of FBS and Cu-induction 24 hr later . 5′-TAATACGACTCACTATAGGGGTCCAATAGCTCCTTGCTCG-3′ 5′-TAATACGACTCACTATAGGGATCCGATGGTAGCAAAGGTG-3′ For Tld expression used in the biophysical analysis , HEK293-EBNA cells were maintained at 37°C under 5% CO2 in DMEM/F12 10% FBS , 0 . 1 unit/ml penicillin , and 10 μg/ml streptomycin growth media . The Lipofectamine transfection reagent ( Invitrogen , Carlsbad , CA ) was used to generate HEK293-EBNA stable cell lines using pCep-PU-TldE94A . After 24 hr , selection was started with the addition of 5 μg/ml puromycin . The stable cell lines were then maintained using a lower level of selection at 1 µg/ml puromycin . Tld was harvested and purified using a HisTrap Ni-NTA column followed by size-exclusion chromatography using a Superdex200 column in the presence of 10 mM Tris HCl ( pH7 . 4 ) with 0 . 5 M NaCl , 1 mM CaCl2 and 2 M Urea . GST-VkgC and GST were expressed and purified as described ( Wang et al . , 2008 ) . SAXS data for dTld were collected on EMBL Hamburg SAXS beamline P12 at Petra III . The scattering images obtained were averaged and buffer scattering intensities subtracted using PRIMUS software . The radius of gyration and distance distribution functions were calculated using GNOM ( Svergun , 1992 ) and particle shapes were modelled using DAMMIN ( Svergun , 1999 ) . 10 models were combined to produce an average shape using DAMAVER ( Svergun , 1999 ) . Rigid-body modelling of dTld domains to the experimental scattering data was performed using SASREF ( Petoukhov and Svergun , 2005 ) . For use in rigid-body modelling Tld domains were modelled on solved crystal structures of similar domains using SWISS-MODEL ( Schwede et al . , 2003; Arnold et al . , 2006; Kiefer et al . , 2009 ) . Templates included; the BMP1 protease domain ( Mac Sweeney et al . , 2008 ) , human TSG6-CUBC and human MASP1/3 from the protein databank ( Berman et al . , 2000 ) . During the modelling procedure each domain was separated by a 4 Å spacer and numerous runs performed to elucidate the most consistent structure . Tld ( 10 µl at 34 μg/ml ) was analysed by negative stain using 2% phosphotungstic acid at pH7 . 4 . Data was collected on a FEI Tecnai Twin Transmission Electron Microscope at 120 kV . Images were recorded at 31 , 000× magnification between −0 . 5 and −1 µm defocus on a 2048 × 2048 pixel camera ( Gatan Orius SC1000 ) . The total number of particles in the dataset was 2297 . Particles were selected manually using EMAN2 ( Ludtke et al . , 1999 ) with a 96 × 96 pixel window ( 30 × 30 nm ) . Imagic5 ( van Heel et al . , 1996 ) was used to align the particles and sort them into classes by multivariate statistical analysis . The classes were then used as references to realign the dataset through various rounds of realignment . When particle classes appeared stable , euler angles were assigned to the images to generate a 3D reconstruction that was subject to further rounds of refinement . To perform cleavage assays the protein containing media from transient transfections was mixed and incubated for 1–7 hr at 25°C . Reaction volumes ranged from 30–51 μl containing: 10–30 μl Sog-Myc with 0–10 μl Tld-HA proteins and a combination of 0–10 μl Tsg-V5 , 0–10 μl VkgC-FLAG/Dcg1C-FLAG ( co-transfected ) and 1–20 nM recombinant Dpp ( rDpp ) . Recombinant Dpp was purchased from R&D systems ( Minneapolis , MN ) and reconstituted in 4 mM HCl . Total reaction volume was kept constant by the addition of mock transfected cell medium . When complete , reactions were stopped with sample buffer ( 2 . 5% glycerol , 12 . 5% SDS , 20% β-mercaptoethanol , 24 mM Tris/HCl ) and analysed by Western blot . Cleavage assays using the zipper fusion proteins were performed by co-transfecting the appropriate plasmids , as described above . Antibodies used as follows; 1:500 anti-Myc antibody ( clone 4A6 , Millipore , Darmstadt , Germany ) , 1:1000 anti-V5 tag ( monoclonal AV5-PKI , AbCam , Cambridge , UK ) , 1:1000 anti-6xHis tag ( AbCam ) , 1:1000 anti-HA ( Roche , Basel , Switzerland ) , 1:1000 anti-flagM2 ( SIGMA ) . Blots were visualised using either ECL reagent ( GE Healthcare , Buckinghamshire , UK ) or West Dura Supersignal ( Thermo Scientific , Waltham , MA ) with the ChemiDoc MP System ( BioRad , Hercules , CA ) or exposed using Biomax film ( Kodak , Rochester , NY ) . For quantification ImageJ Version 10 . 2 or ImageLab ( BioRad ) was used . Equivalent amounts of Tld-HA proteins were incubated with 300 µl Sog-Myc , or SogCR14*-Myc containing media with 30 µl 50% anti-Myc agarose matrix ( Sigma , St . Louis , MO ) prewashed in IP buffer ( 50 mM Tris pH7 . 5 , 150 mM NaCl , 0 . 02% TritonX-100 ) in the presence of 1 nM Dpp for 2 hr at 4°C . Following binding , the beads were washed multiple times in IP buffer and resuspended and heated to 95°C for 5 min in 30 µl 1XSDS sample buffer for analysis by Western blot . The same principle was followed for leucine zipper immunoprecipitations in the absence of rDpp . GST pulldown assays were carried out using 10–300 µl Tld-HA proteins ( levels normalised by Western blot ) as described ( Sawala et al . , 2012 ) . Fly stocks used were y67c23w118 , tldB4/TM3 , tld2/TM3 ( Bloomington Stock Center , Bloomington , IN ) . Embryos were collected from tld2/TM3 , and yw parents as well as crosses of MPZ−/CyO , hb-lacZ; tldB4/TM3 , ftz-lacZ to sogZ+/CyO , hb-lacZ; tldB4/TM3 , ftz-lacZ . Embryos were fixed and antibody stain or RNA in situ hybridisation with digU-labelled RNA probes was carried out using standard protocols . Width of the ush stripe was quantified using ImageJ .

The body of an animal is a highly organised structure of tissues and organs that contain cells with specialised roles . To achieve this level of organisation , it is important that the cells in the embryo know their location and receive the correct instructions on how to develop , when to divide or move . Many animals are roughly symmetrical about an imaginary line that runs from their head to their tail; a developing embryo can provide its cells with information about their position along this head-to-tail axis and the axis that runs from its front to its back . Setting up the front-to-back axis in the embryo involves a family of proteins called the bone morphogenetic proteins ( or BMPs ) . These proteins can bind to other proteins that act as signals to provide instructions to cells . However , many of the BMPs are unable to perform this job because they are trapped by inhibitory molecules that bind to them instead . Enzymes belonging to the Tolloid family can break down these inhibitors to release the BMPs . Together , the inhibitors and Tolloid enzymes create a gradient of BMP activity across the embryo . The side of the embryo with the highest levels of active BMPs sets the position of the back of the body , while the opposite side—which has the lowest levels of active BMPs—becomes the front . However , it is not clear how Tolloid is controlled to create the BMP gradient . Different parts of the Tolloid enzyme have different roles; one portion of the enzyme breaks down the inhibitory molecules , and there are also several so-called ‘non-catalytic domains’ . Winstanley et al . used a combination of approaches to study how Tolloid is controlled in fruit fly embryos . The experiments show that two non-catalytic domains at one end of Tolloid help the enzyme to bind to the inhibitory molecules . At the other end of the Tolloid enzyme , another non-catalytic domain can bind to a structural protein called Collagen IV . This enhances the ability of the enzyme to break down the inhibitory molecules and release the BMPs . These findings reveal how Tolloid's non-catalytic domains can fine-tune the activity of this enzyme to create the gradient of BMP activity that is needed to set the front-to-back direction in animal embryos . Future studies will focus on identifying other proteins that bind to the non-catalytic domains of Tolloid in order to further control its activity during development .

2015

Synthetic enzyme-substrate tethering obviates the Tolloid-ECM interaction during Drosophila BMP gradient formation

The role of pro-inflammatory macrophage activation in cardiovascular disease ( CVD ) is a complex one amenable to network approaches . While an indispensible tool for elucidating the molecular underpinnings of complex diseases including CVD , the interactome is limited in its utility as it is not specific to any cell type , experimental condition or disease state . We introduced context-specificity to the interactome by combining it with co-abundance networks derived from unbiased proteomics measurements from activated macrophage-like cells . Each macrophage phenotype contributed to certain regions of the interactome . Using a network proximity-based prioritization method on the combined network , we predicted potential regulators of macrophage activation . Prediction performance significantly increased with the addition of co-abundance edges , and the prioritized candidates captured inflammation , immunity and CVD signatures . Integrating the novel network topology with transcriptomics and proteomics revealed top candidate drivers of inflammation . In vitro loss-of-function experiments demonstrated the regulatory role of these proteins in pro-inflammatory signaling . Pro-inflammatory macrophage activation plays a prominent role in a large number of disorders including cardiovascular disease ( CVD ) ( Aikawa and Libby , 2004; Glass and Olefsky , 2012; Glass and Witztum , 2001; Gregor and Hotamisligil , 2011; Liang et al . , 2007; Randolph , 2014; Ridker and Lüscher , 2014; Tabas , 2010 ) . Established treatments for CVD such as those dependent on the cholesterol lowering effect of statins do not completely eliminate cardiovascular risk ( Aikawa and Libby , 2004; Aikawa et al . , 2001; Libby , 2005 ) , therefore alternative novel solutions are needed to tackle such residual risk by targeting pro-inflammatory activation in CVD ( Ridker et al . , 2017 ) . Characterizing the mechanisms underlying macrophage activation itself proves to be a challenging task , given the functional heterogeneity of macrophages and the complex interplay between the pro- and anti-inflammatory phenotypes ( Biswas and Mantovani , 2012; Gordon and Mantovani , 2011; Koltsova et al . , 2013; Lawrence and Natoli , 2011; Ley et al . , 2011; Moore et al . , 2013; Murray et al . , 2014; Swirski and Nahrendorf , 2013 ) . Furthermore , it is increasingly recognized that macrophage activation has many distinct types and follows a spectrum model defined by specific stimuli rather than the bipolar model of pro- and anti-inflammatory polarization that once prevailed ( Murray et al . , 2014; Nahrendorf and Swirski , 2016; Xue et al . , 2014 ) . Nevertheless , using experimental models , where cause-effect relations are well defined , within a systems-based approach might help to facilitate the discovery of specific mechanisms that can contribute to the overall balance of macrophage phenotype or new therapeutic targets . As it stands , the hunt for hitherto undiscovered mechanistic connections in macrophage activation , and therapeutic targets aimed at resultant CVD , has much to benefit from complex systems approaches emerging in medicine . Complex human diseases such as CVD are seldom the result of a perturbation of a single gene but rather arise from the concerted effects of multiple genes and their products forming complex networks of interactions in cells , collectively embodied in the ‘interactome’ concept ( Vidal et al . , 2011 ) . Network medicine has emerged as an effective quantitative framework to address the complexity of human disease ( Barabási et al . , 2011 ) . Increasing evidence suggests that disease determinants tend to be localized in the same region in the interactome and interact closely with each other , leading to their organization into ‘disease modules’ ( Barabási et al . , 2011; Goh et al . , 2007; Menche et al . , 2015 ) . The same principle is valid for functionally similar genes , which form ‘functional modules’ ( Shih and Parthasarathy , 2012 ) . Based on this understanding of biological function and disease pathogenesis , network-based methods have successfully identified candidates for therapeutic targets: Early studies that constructed the network of approved drug targets and disease genes shed light on the current trends in drug discovery ( Yildirim et al . , 2007 ) . Other network-based pharmacological studies have focused on drug target prediction and proposed drug repurposing methods ( Berger and Iyengar , 2009; Wu et al . , 2013 ) . Methods using shortest paths between drug targets ( Gottlieb et al . , 2011; Guney et al . , 2014; Lee et al . , 2012; Zhao and Li , 2012 ) and drug-disease proximity ( Guney et al . , 2016 ) in the interactome were proposed . Despite these advances , one important factor hampering the effective molecular characterization of diseases is that interactomes , which are collections of multiple types of literature-curated physical protein-protein interactions ( PPIs ) with experimental evidence from high-throughput experiments and small-scale studies , are incomplete ( Menche et al . , 2015 ) and biased toward highly studied proteins such as disease-related proteins or targets of pre-existing drugs ( Rolland et al . , 2014 ) . Most importantly , interactomes are context-independent , that is do not carry information specific to the cell type , experimental condition or pathological state but rather represent the sum of all observed interactions , which makes them inherently generic and limits their utility in diverse biological and experimental settings . Owing to its unbiased nature and high information content , high-resolution/accuracy mass spectrometry ( MS ) offers an opportunity to characterize the proteome of a specific cell state in a comprehensive way ( Mann et al . , 2013 ) . The rapid progress in proteomics technologies has instigated a growing body of works that combine large amounts of MS data with network biology methods . For instance , a recent study has used the proteomics of breast cancer cells in conjunction with literature-derived signaling networks ( Sacco et al . , 2016 ) . Protein abundance profiles from global proteomics measurements have been used to cluster co-regulated proteins ( Singh et al . , 2014 ) , as well as build ‘co-abundance’ networks to identify key driver proteins for viral replication ( McDermott et al . , 2012 ) . Thus , cell- and tissue-specific proteomic profiling could complement the incomplete and generic yet system-wide cellular picture provided by the interactome , especially in specific disease contexts . Here , we used unbiased macrophage-derived proteomics measurements to enhance the literature-curated human interactome by adding cell type- and condition-specific information to it , thereby helping address its context-independence as well as its incompleteness . While we now realize that macrophage heterogeneity is more multidimensional rather than M1/M2 dichotomy ( Murray et al . , 2014 ) , assessing molecular mechanisms still requires a model in which cause-effect relationships are well defined , as we recently demonstrated ( Iwata et al . , 2016 ) . We thus used interferon-γ ( IFNγ ) as an example of major instigators of pro-inflammatory macrophage activation . We utilized a network proximity-based prediction method to identify key drivers of macrophage activation as it pertains to CVD . Our results revealed that edges derived from macrophage-specific proteomics contributed to the less characterized parts of the interactome , reflected the respective macrophage stimulation condition in terms of pathways and biological processes , and increased the prediction performance of CVD therapeutic targets . The top-ranked candidates for regulators of macrophage activation , and hence potential CVD drug targets , also showed significant enrichment with immune system as well as cardiovascular disease related signatures . Our multi-step and multi-omics analytical pipeline resulted in the identification of Guanylate binding protein 1 ( GBP1 ) and tryptophanyl-tRNA synthetase ( WARS ) as top candidates , based on evidence from network topology , gene expression and protein expression . To validate our network-based prediction , we performed loss-of-function experiments and demonstrated that GBP1 and WARS indeed regulate the expression of the pro-inflammatory cytokine , CCL2 , and phosphorylation of STAT1 , two classical pro-inflammatory readouts . Overall , our findings suggest the utility of adding context-specific information to the generic interactome . The literature-curated human interactome , or PPI network ( see Materials and methods ) , hosts invaluable information about potential protein subnetworks related to diseases . However , it is non-specific as it describes interactions that can occur between proteins within any cell or tissue under any condition , as well as currently incomplete . We hypothesized that the introduction of phenotype-specific interactions to the PPI network would address its incompleteness and fill in the biologically less characterized portions of it . We used time-course proteomics data collected from baseline control , IFNγ stimulated/pro-inflammatory , and IL-4 stimulated anti-inflammatory/pro-resolving phenotypes of the macrophage-like cell line THP-1 , denoted by M ( - ) , M ( IFNγ ) and M ( IL-4 ) , respectively , as the phenotypic information . For each condition , we built co-abundance networks to represent the relationships between proteins that show similar abundance patterns over time by constructing the correlation matrix , setting a correlation threshold and building the network based on the edges above this threshold ( Figure 1A and Figure 1—figure supplement 1A , see Materials and methods and Supplementary file 1 for a summary of topological properties ) . While the co-abundance networks have several-fold higher edge densities than the PPI network as a whole ( Supplementary file 1 ) , their edge densities are comparable to the region of the PPI that hosts the proteins in the co-abundance networks ( 0 . 96% ) . Mirroring well-known properties of PPI networks , the co-abundance networks display broad degree distributions ( Figure 1—figure supplement 1B ) and high average clustering coefficients ( Supplementary file 1 ) . Despite these similar characteristics , in comparison with the PPI network , the co-abundance networks have higher diameters and higher average shortest path lengths , as well as higher average clustering coefficients , possibly due to similar protein expression patterns connecting local groups of proteins all at once ( Supplementary file 1 ) . Thus , co-abundance networks are more locally clustered and not as compact , compared to the PPI network , overall suggesting a complementary topology to the PPI network . We then asked what the topologies of co-abundance networks add to the PPI network . Overlaying the co-abundance networks on the PPI network , we found that each stimulation condition was denser in certain regions , and overall filled in the sparse parts of the PPI network , resulting in a combined interactome where each condition can be distinguished and their overlapping regions can be assessed ( Figure 1B ) . Thus , co-abundance networks may be instrumental in helping address the lack of biological investigation of certain parts , that is the incompleteness , of the PPI network while enhancing it with experiment- and context-specific information . To ensure that the co-abundance edges correspond to biologically meaningful interactions , we validated these edges by measuring their overlap with known physical interactions and shared functional annotations . Co-abundance edges showed significant enrichment for ( i ) shared Gene Ontology ( GO ) terms , ( ii ) the literature curated PPI network used throughout this study , and ( iii ) interactions between pairs of proteins belonging to a common protein complex , based on high-quality protein complex databases and curated co-complex association maps ( Figure 1C ) . Overlaps with large-scale binary and affinity-purification-mass spectrometry ( AP-MS ) based interactomes were not significant ( Supplementary file 2 ) , possibly due to the strict experimental criteria for these maps , while the ‘binary’ portion of the literature-curated HINT database ( Das and Yu , 2012 ) showed significant enrichment ( Figure 1C , see Materials and methods for details on the databases used ) . Thus , the co-abundance networks independently generated from MS data in an experiment-specific context still capture GO-term-based functional associations and high-confidence interactions documented in context-independent interactomes . We next inquired about how the macrophage-specific topology provided by co-abundance edges could facilitate the extraction of CVD and inflammation related drug target information from the combined network . To test this , we evaluated the effect of the addition of co-abundance edges on the shortest paths between proteins and drug targets related to the biology of the co-abundance network , and hence the combined network’s overall ‘drug target navigability . ’ We found that , for both M ( IFNγ ) and M ( IL-4 ) , the average shortest distances between the co-abundance network’s nodes and known CVD drug targets from the integrated Complex Traits Network ( iCTNet ) database ( Wang et al . , 2015 ) ( see Materials and methods , Figure 1—figure supplement 1C ) are shorter for the combined network than the PPI network only ( Figure 1D ) . Proving that this effect cannot be attributed to the mere addition of edges , the addition of co-abundance edges resulted in an average shortest distance distribution that is shifted significantly towards lower values compared to the addition of the same number of randomly chosen edges ( Figure 1D ) . Overall , these findings indicate that the co-abundance links used in conjunction with the PPI network result in a more compact network structure that makes the known drug targets more reachable by other proteins . We hypothesized that , since the macrophage-derived co-abundance network renders the PPI network more compact and navigable to drug targets , network proximity-based drug target prediction methods should fare better on this combined network than the PPI network alone , enabling us to more efficiently identify novel targets . The co-abundance networks spanned a certain portion of the entire PPI network with their edges concentrated on certain regions , while the CVD drug targets were dispersed around the combined PPI network , forming a disconnected subnetwork ( Figure 2A and Figure 2—figure supplement 1A ) , suggesting the fragmented nature of the CVD drug target information in the interactome . The proportions of CVD drug targets to non-drug targets in the PPI and co-abundance network were very similar to each other ( 0 . 0018 versus 0 . 0017 , respectively ) , suggesting that there is no preferential presence of drug targets in either network over the other one . In terms of connectivity , the degrees of drug targets were significantly higher than the degrees of other proteins in the PPI network , while there was no significant degree difference between drug targets and other proteins in the co-abundance networks ( Figure 2—figure supplement 1B ) . Furthermore , measuring the tendency of co-abundance edges to connect to hubs in the PPI network , we found that proteins connected by CoA edges ( i . e . CoA network nodes ) have significantly higher degrees in the PPI network than the other nodes in the PPI network ( Figure 2—figure supplement 1C ) . Taken together , these results point at a degree bias in the way drug targets and co-abundance edges are connected in the PPI network . To give equal opportunity to proteins not necessarily captured by proteomics measurements but contained in the PPI network , we chose to leverage the biology-specific information provided by the co-abundance network for a global prioritization of all proteins in the combined PPI network . We utilized a drug target prioritization method inspired by similar methods based on the network proximity between candidates and seeds ( Guney et al . , 2016; Krauthammer et al . , 2004 ) , which uses a metric that is a function of the average shortest distance between the candidate and the set of known CVD drug targets ( seeds ) from iCTNet ( Figure 2A , see Materials and methods ) . As a proxy of prediction performance , we measured the rate of detecting known CVD drug targets by calculating the area under the receiver operating characteristic ( ROC ) curve . Compared to the PPI network only , the area under the ROC curve ( AUROC ) was significantly increased with the addition of co-abundance links ( p=0 . 018 , p=0 . 003 and p=0 . 015 for M ( - ) , M ( IFNγ ) and M ( IL-4 ) , respectively , paired t-test following k-fold cross-validation ) ( Figure 2B , Figure 2—figure supplement 2A–C ) . The prediction performance of the combined networks also surpassed those of co-abundance networks alone ( Figure 2—figure supplement 3A ) . The distribution of known CVD drug targets ordered by prediction rank indeed showed that they were more predominantly ranked in higher quantiles in the three co-abundance enriched networks compared to the PPI network only ( Figure 2C ) . Furthermore , we found that the prediction performance can be further improved by ( a ) the systematic ‘de-noising’ of the PPI network by removing edges based on low-throughput experiments and co-complex interactions ( see Materials and methods , and Figure 2—figure supplement 3B–D ) and ( b ) modulating the ratio of weights of the PPI links and co-abundance links , specifically by giving more ‘importance’ to the co-abundance links ( see Materials and methods and Figure 2—figure supplement 4A–B ) , hence increasing the specificity of the network to the biological question . Finally , to ensure that the increase in the prediction performance is not simply the result of the bulk addition of edges , we compared the AUROCs of co-abundance enriched PPI networks to those enriched with the equivalent number of random edges . To control for the degree bias in the connectivity between co-abundance and PPI networks ( Figure 2—figure supplement 1B–C ) , we implemented degree-preserving randomization , which ensures the addition of randomly selected edges that connect proteins with similar degrees to the proteins in the co-abundance networks ( see Materials and methods ) . We found that co-abundance networks contribute to the increase in prediction performance significantly more ( empirical p-value<0 . 05 ) than the random case ( Figure 2D ) . To establish the relevance of our network-based prioritization to the therapeutic targets it is aimed at finding , we sought the enrichment of the top-ranked proteins in external datasets . As iCTNet-based CVD targets were used as seeds in the prioritization scheme , we used external datasets containing clinically approved CVD targets and chemically suitable CVD targets satisfying given small molecule activity thresholds for validation ( see Materials and methods ) . These datasets were fairly orthogonal , showing insignificant overlap ( except for one case , where the overlapping genes were removed from both datasets , see Figure 2—figure supplement 4C ) between their CVD drug target sets , thereby providing three independent sources for validation ( Figure 2—figure supplement 4C ) . We once again observed the increase in the target prediction performance with the addition of co-abundance edges ( Figure 2E , Figure 2—figure supplement 5 ) . These findings applied to all drug targets derived from the same databases as well ( Figure 2—figure supplements 6 and 7 ) , reflecting the broad impact of inflammation in human disease and suggesting the potential of macrophage-specific co-abundance edges in capturing additional inflammatory disease drug targets . We also tested how robust the prediction performance is against changes in several points in the workflow . First , to investigate whether the improvement in the prediction performance is influenced by potential shared dependencies on a third confounding factor , we implemented two correlation measures that are robust against outliers and control for confounding factors , biweight midcorrelation and partial correlation , to build the co-abundance networks ( see Materials and methods ) . The improvement in the AUROCs with the addition of co-abundance networks over PPI only was similar to the Pearson coefficient case ( Figure 2—figure supplement 8 ) . These results indicate that the outlier abundance values and the indirect effect of baseline abundances do not influence the downstream analysis substantially . Second , to explore how sensitive prediction performance is to untraversed longer-range paths between candidates and drug targets , we have considered global association measures and other types of network-based distance measures . In particular , we used dynamical prioritization methods such as random walk with restarts ( RWR ) ( Köhler et al . , 2008 ) and its degree-aware version ( DADA ) ( Erten et al . , 2011 ) , as well as other distance-based measures such as the kernel distance ( Guney et al . , 2016 ) . Relying on a random walker , RWR/DADA takes into account all possible paths , including many longer-range paths between a candidate and a target ( seed ) , whereas Kernel distance penalizes paths based on their length using an exponential function ( see Matetrials and methods ) . Both types of measures yielded similar results to average shortest path in terms of the AUROC , and the improvement with the addition of CoA edges persisted ( Figure 2—figure supplement 9 ) . To ensure that the prioritization method above reflects the macrophage biology and cardiovascular disease relation of the co-abundance networks , we measured their rate of capturing genes related to inflammation , innate immune response , and coronary artery disease signatures ( see Materials and methods ) . The performance of the co-abundance enriched PPI networks in capturing these signatures surpassed that of the PPI network alone ( Figure 2F , Figure 2—figure supplement 10 ) . The top prioritized ( empirical p-value<0 . 01 , see Materials and methods ) were significantly enriched in all three datasets for both stimulations ( Supplementary file 3 ) . The majority of the enriched pathways of the top prioritized proteins ( Figure 3—figure supplement 1—source data 1 ) were common ( Figure 3A ) . This is mainly due to the fact that the despite the different stimulation conditions , the biological processes and pathways of top-prioritized candidates are mainly dominated by the cell type , resulting in a large number of commonalities , and reflecting the predominant effect of the common macrophage biology on the resulting target prioritization . Focusing on the condition-specific pathways , we found that M ( IFNγ ) is enriched in pro-inflammatory pathways such as p38 MAPK and NF-κB activation related pathways , as well as pathways related to PI3K-Akt activation , NFAT and hypertrophy of the heart , adrenergic signaling in cardiomyocytes and interleukin signaling , whereas M ( IL-4 ) is enriched in Rho GTPase activation , SUMOylation , β-catenin , scavenger receptor and Fas ( CD95 ) signaling related pathways ( Figure 3B , see Figure 3—figure supplement 1 for full list ) . Mapping these pathways in a pathway network ( see Materials and methods ) enabled us to summarize all biological processes related to each condition whereby pathways sharing molecular elements were clustered together ( Figure 3—figure supplements 2 and 3 ) . Together , these results demonstrate that our global ranking of the entire PPI network , including proteins not in macrophage proteomics , captures the inflammatory and immune-response-related component of CVD mediated by macrophage activation . Combining proteomic and transcriptional information improves the identification of key driver molecules ( Padi and Quackenbush , 2015 ) . In a similar vein , we sought to integrate our protein abundance data with previously published gene expression data from human macrophages ( Xue et al . , 2014 ) to refine our list of prioritized proteins and focus on a much smaller subspace of highly expressed candidates for further in vitro validation . Particularly , since the network-based prioritization ranks all the proteins in the interactome indiscriminately and the prediction ROC curves show the greatest improvement at moderate ranks ( Figure 2E ) , we devised a combined filtering/ranking scheme to obtain a smaller set of final drug target candidates while maximizing the advantage offered by the performance of our prioritization method . The filtering step was used to intersect the network-based prioritization ranking with the highly expressed molecules from -omics data to detect the strongest signals that were also close to drug targets . The ranking scheme considered ( a ) the network closeness to drug targets , ( b ) the relative protein abundance difference with respect to baseline M ( - ) over all time points , and ( c ) the gene expression fold change with respect to baseline M ( - ) from human macrophage transcriptome data , and calculated a combined score based on these three criteria ( Figure 4A and Figure 4—figure supplement 1 , see Materials and methods ) . To quantify each constituent ranking’s relative contribution to the final ranking , we investigated the correlations between them . The combined ranking was positively correlated with all separate rankings . While the combined ranking was slightly more driven by the gene expression and protein abundance , the network prioritization ranking was close to them for M ( IFNγ ) and on par with them for M ( IL-4 ) ( Figure 4—figure supplement 2A–B ) . Moreover , each separate ranking was orthogonal to each other with insignificant correlations ( Figure 4—figure supplement 3A–B ) . Together , these results suggest that network-based prioritization ranking contributes in a non-trivial way to the combined ranking , and that each ranking carries its own information , contributing uniquely to the combined ranking . The final list of candidates ranked according to the combined score was significantly close in the interactome to CVD drug targets for both M ( IFNγ ) and M ( IL-4 ) in terms of the average shortest distance than what would be expected by chance ( Figure 4B , see Materials and methods ) , confirming that the filtering of candidates aligns with the original premise of network proximity to CVD drug targets . This finding was further supported in the discretized distribution of shortest path lengths ( Figure 4C , see Materials and methods ) , where shorter path lengths were significantly over-represented compared to random expectation . The top five final candidates for the M ( IFNγ ) ranking were GBP1 , WARS , TGM2 , NAMPT and STAT1 , excluding PARP14 , which was among the iCTNet drug targets . STAT1 belongs to a family of molecules known to be expressed in the heart ( Xuan et al . , 2001 ) , and to play a role in the link between coronary artery disease and inflammatory responses in vascular cells ( O'Donnell and Nabel , 2011 ) . Of these candidates , GBP1 and WARS showed the most dramatic changes with IFNγ stimulation in protein abundance and gene expression alike ( Figure 4D ) . On the other hand , the top five final candidates for M ( IL-4 ) were LIPA , TGM2 , EVL , PFKP and LRRFIP1 , which showed a broad range of expression kinetics with their highest induction at different time points , implying that the related in vitro validation should be planned accordingly ( Figure 4D ) . Finally , the top prioritized and the top combined-ranked proteins showed good agreement with the rankings found through the alternative correlation measures and prioritization methods discussed above , showing significant overlap ( Figure 2—figure supplements 8C–D , 9B–C and E–F ) . To inspect more closely the molecular paths between our top candidates and drug targets , we created subnetworks connecting the top five candidates to all CVD drug targets ( Figure 4E and F ) . STAT1 and TGM2 tended to connect to CVD drug targets through more established links in the literature-curated PPI network , whereas GBP1 , WARS and NAMPT were mostly connected to drug targets by co-abundance edges ( Figure 4E ) . Similarly for M ( IL-4 ) , we noted that TGM2 , PFKP and LRRFIP1 mostly utilize PPI network edges to link to drug targets while EVL and LIPA exploit the information provided by the co-abundance links ( Figure 4F ) . Mapping the shortest paths between the known CVD drug targets and top-ranked candidates thus presents us with an unbiased network-based means to select putative targets that preferentially leverage the novel information provided by the context-specific co-abundance links . Although a systems approach facilitates target discovery , increased or decreased genes or proteins identified by unbiased omics screening may not necessarily play causal roles . While the expression levels of certain proteins increase during the activation of human macrophages , as gauged by induction of pro-inflammatory molecules , these molecules may not contribute to this phenotypic switch . Thus , to provide mechanistic insights about these induced molecules , as well as to validate our systems approach , we performed in vitro loss-of-function experiments . Based on the shortest path mapping between top candidates and CVD targets ( Figure 4E–F ) , we chose GBP1 Guanylate Binding Protein ( 1 ) and WARS ( Tryptophanyl-tRNA Synthetase ) as candidates for novel regulators of pro-inflammatory macrophage activation and proceeded to in vitro validation experiments . Evidence had linked WARS with vascular angiogenesis and homeostasis ( Ewalt and Schimmel , 2002; Otani et al . , 2002; Wakasugi et al . , 2002 ) . Further , genome-wide linkage studies have previously implicated this molecule in myocardial infarction ( Broeckel et al . , 2002 ) , although subsequent genetic-epidemiological studies did not find significant associations between WARS and the risk of myocardial infarction ( Zee et al . , 2005 ) . A more recent study that used human primary peripheral blood mononuclear cells ( PBMCs ) demonstrated that secretion of the full-length form of WARS is induced by pro-inflammatory stimuli , including bacterial-derived lipopolysaccharides ( LPS ) , suggesting a role for WARS in the defense system against infection ( Ahn et al . , 2017 ) . Here , we investigated the potential for WARS to regulate prototypical pro-inflammatory signaling proteins and cytokines in response to IFNγ . We performed small interfering RNA ( si-RNA ) -mediated WARS loss-of-function studies on THP-1-differentiated macrophage-like cells and human primary macrophages derived from PBMC . In THP-1 cells stimulated by IFNγ , silencing of WARS enhanced the mRNA expression of the chemokine CCL2 , and the secretion of its protein ( Figure 5A ) , however no such effects were observed for the cytokine TNFα ( Figure 5A ) . The enhanced secretion of CCL2 in response to WARS silencing was confirmed in three different PBMC donors , where the increase in CCL2 secretion was significant at 12 hr of IFNγ stimulation ( Figure 5—figure supplement 1A ) . Investigating the effect of WARS on the JAK-STAT pathway , the transcriptional levels of STAT1 and JAK2 did not change with WARS silencing ( Figure 5A ) . Silencing of WARS , however , increased the phosphorylation of STAT1 at Ser701 in both THP-1-differentiated macrophage-like cells and human primary macrophages ( Figure 5B and Figure 5—figure supplement 1B ) . GBP1 is a member of the IFNγ-inducible GBP family , whose members are protective against bacterial ( Kim et al . , 2011 ) and parasite ( Selleck et al . , 2013 ) infection . Following a Western-type diet GBP3 and GBP6 expression levels increase during foam cell formation in mice , indirectly suggesting their role in the acceleration of atherosclerosis by hypercholesterolemia ( Goo et al . , 2016 ) . We performed loss-of-function studies to determine the possible mechanisms through which GBP1 influences pro-inflammatory molecules . In human PBMC-derived macrophages from four different donors stimulated with IFNγ , GBP1 silencing resulted in a significant decrease in CCL2 secretion without change to its mRNA at 24 hr of IFNγ treatment ( Figure 5C ) . The expression of TNFα and JAK2 did not change ( Figure 5C ) . We found similar results in THP-1 cells ( Figure 5—figure supplement 1C ) . To assess the effect of currently available therapeutic inhibitors targeting GBP1 , we used NSC756093 that interrupts the interaction between GBP1 and proto-oncogene serine/threonine-protein kinase ( PIM1 ) , the inhibition of which could potentially revert paclitaxel resistance in cancer cells ( Andreoli et al . , 2014 ) . Human primary macrophages were pretreated with NSCT56093 ( 10 to 100 nM ) and then stimulated with IFNγ for 24 hr . We observed a dose-dependent decrease in the secretion of both CCL2 and TNFα ( Figure 5D ) , with no change to their mRNA levels and to the mRNA levels of JAK2 and GBP1 itself ( Figure 5D ) . Silencing of GBP1 decreased the expression of JAK2 and phospho-STAT1 at 6 hr of IFNγ stimulation in THP-1 macrophage-like cells ( Figure 5—figure supplement 1D ) . With the hypothesis that molecules that arise from in silico predictions seeded with known CVD targets and found to play specific regulatory roles in pro-inflammatory macrophage activation are likely to be CVD therapeutic targets , we next sought to investigate their structural ‘druggability’ in silico . The top five candidates of pro-inflammatory macrophage activation all contained binding sites suitable for small molecules , with at least 50% tractable structures and 20% druggable structures ( see Materials and methods , Figure 5—figure supplement 2A–B ) . Among these , WARS and GBP1 were the two candidates with the highest percentage of druggable structures ( Figure 5—figure supplement 2B ) . The search for new molecular connections between inflammation and CVD is continuing at an ever-increasing pace . Network medicine approaches utilizing large-scale interactomes hold the key to the efficient identification of novel therapeutic solutions targeting pro-inflammatory macrophage activation in CVD . Currently available interactomes , however , are hindered in their usefulness since they are by design the aggregation of all possible interactions from diverse cell states , which obscures any tissue , cell type , experiment or disease-specific information . One approach to address this lack of specificity and incompleteness is to enhance the interactome with context-specific information from proteomics measurements . To capitalize on the potential for interactome networks to expedite CVD drug target discovery research , we combined our protein co-abundance networks derived from a cell-specific , macrophage activation in vitro model with the ‘all-purpose’ , generic PPI network . We interrogated the resulting combined interactome to predict and highlight new regulators of macrophage activation and potential targets of CVD as well as studied the potential the mechanisms by which they might regulate pro-inflammatory biomarkers , and experimentally validated some of these mechanisms . When proposing new sources for protein interactions , two questions must be asked: ( i ) Does the new network add novel information to the existing one ? ; and ( ii ) Is this novel information biologically relevant ? The first question is addressed by the ‘fill-in-the-blanks’ effect of the co-abundance edges , where the sparse portions of the PPI network were filled in by co-abundance networks , potentially addressing the incompleteness and investigation bias of the PPI network . Previously , co-expression networks were suggested to represent a complementary tool to the PPI network ( Vella et al . , 2017 ) , and protein abundances obtained through mass spectrometry and direct protein contacts detected by crosslinking and mass-spectrometry were found to be complementary ( Solis-Mezarino and Herzog , 2017 ) . Interestingly , each stimulation condition was distinct in regards to which part of the PPI network they filled in . We also noted that the M ( IL-4 ) co-abundance network had more overlap with M ( - ) . Addressing the second question , our assumption is that proteins that are co-regulated are more likely to be involved in similar pathways ( Kustatscher et al . , 2017 ) . Indeed , the co-abundance edges were significantly enriched in interactions based on binary interactomes , co-complex associations and shared functional annotations . While direct binary interactions were less strongly represented among co-abundance edges , the indirect co-associations evidenced by the enrichment of multiple co-complex resources suggests that the real value of co-abundance edges might lie in their use as co-complex resources and not necessarily direct physical interactions . It has been proposed that network proximity is a good proxy of therapeutic effect ( Guney et al . , 2016 ) . Indeed , we found that network distance-based measures fared better with the addition of context-specific co-abundance edges , and that co-abundance edges rendered the PPI network more navigable for the drug targets . The prediction performance of novel drug targets based on network proximity measures showed a significant improvement with the addition of co-abundance edges . The prediction performance further increased when co-abundance edges were given more prominence , which suggests that the information provided by the co-abundance network is beneficial and adds positive value to the PPI network when predicting drug targets . Remarkably , the prioritized candidates also captured disease biology , especially inflammation , innate immune response and CAD signatures . This suggests that the drug target candidates themselves may play a direct role in CVD or be involved in the inflammatory mechanisms leading to it . We note that , while co-abundance edges were built systematically , filtered with statistical rigor , and as a whole contributed to the increase in drug target prediction accuracy , they were not experimentally validated in this study and therefore are not deemed high-confidence edges individually . Aside from hallmark pro-inflammatory pathways such as MAPK and NF-κB , the top ranked candidates for M ( IFNγ ) were enriched in various pathways associated with atherosclerosis and other vascular disorders , including Wnt , FGF , adrenergic , IGF , Akt , and ras . IFNγ was suggested to have direct effects on cardiomyocytes through β-adrenergic signaling ( Levick and Goldspink , 2014 ) , PI3K-Akt pathway has been posed as a key regulator in macrophage metabolism ( Brenner et al . , 2016 ) . The enriched pathways of the top ranked candidates for M ( IL-4 ) showed a breadth of evidence for the alternatively activated , pro-resolving macrophage polarization . Among these , Rho GTPase activation has been associated with M2-like anti-inflammatory macrophage phenotype ( Aflaki et al . , 2011 ) . SUMOylation was linked to anti-inflammatory signals and suggested as a potential target pathway for the modulation of inflammation ( Leitinger and Schulman , 2013; Tugal et al . , 2013 ) . Elevated levels of Fas ( CD95 ) expression in anti-inflammatory macrophages was associated with proangiogenesis in the eye ( Zhao et al . , 2013 ) . Alternatively activated macrophages were shown to activate Wnt signaling pathway and increase β-catenin expression in epithelial cells ( Cosín-Roger et al . , 2013 ) . The expression of several scavenger receptors was increased in alternatively activated macrophages ( Canton et al . , 2013 ) . This pathway-based evidence suggests that co-abundance networks indeed influence the PPI network and the resulting drug target prioritization reflects the biology of the respective macrophage activation . The correlation between gene and protein expression may be low in multi-cellular organisms ( Kustatscher et al . , 2017; de Sousa Abreu et al . , 2009 ) , which we also observed in our datasets ( Figure 5—figure supplement 3A ) . To ameliorate this issue , we used a combined ranking of rankings to effectively normalize the ranking distributions and address the possible delay between gene expression and protein expression ( Liu et al . , 2016 ) . The resulting filtered list of candidates indeed showed the desired closeness to known CVD drug targets . When measuring the network distance between the final list of candidates and CVD targets , we used two measures that offer a complementary view of network proximity: Average shortest distance is a useful summary metric that quantifies the distance between candidates and targets while shortest distance distributions provide a more detailed view enabling us to see which number of links between candidates and targets tend to be over-represented compared to random . While minimum ( or closest ) shortest distance has been found to outperform other proximity measures ( Guney et al . , 2016 ) , we chose to use average shortest distance as the metric of choice in our drug target prioritization method since minimum shortest distances are confined to integer numbers resulting in redundancy of ranks , which convolutes the overall rankings for prioritization . Considering potential limitations of shortest path based prioritization methods such as ignoring biologically meaningful alternative longer-range paths between candidates and drug targets , we also ran our workflow on global dynamical prioritization measures , which displayed similar results , suggesting that the improvement of target prediction with context-specific information from co-abundance networks is independent of the prioritization metric used . To validate our systems approach to the discovery of new regulators of macrophage activation , we chose GBP1 and WARS from the subnetworks of top five candidates and CVD drug targets . In particular , GBP1 and WARS stood out as the candidates that were linked to the majority of drug targets by novel co-abundance links , hence potentially having many undiscovered downstream mechanisms related to CVD and inflammation . It is important to note the opposite influences of GBP1 and WARS on JAK-STAT signaling , particularly on the pro-inflammatory cytokine CCL2 and pSTAT1 . Our siRNA silencing and inhibitor treatment experiments suggest that WARS has a protective effect , suppressing these downstream pro-inflammatory markers , whereas GBP1 has an exacerbating effect on inflammation , enhancing their expression , along with another pro-inflammatory molecule , JAK2 . Overall , the downstream effect of GBP1 points at a possible feedback regulation on JAK2-STAT1-CCL2 signaling ( Figure 5—figure supplement 3B ) . Other GBP family members , GBP2 , GBP3 , GBP4 and GBP6 , were also found in our proteomics measurements , showing higher induction after 24 hr , with GBP2 displaying similar expression patterns to GBP1 , while GBP3 and GBP6 , the two family members found to be induced in macrophage foam cells ( Goo et al . , 2016 ) , have a distinct pattern characterized by a sharp drop in protein expression at 12 hr and then a sharp increase at 24 hr ( Figure 5—figure supplement 3C ) . Coupled with evidence from our network-based prioritization , this suggests that other members of the GBP family with similar expression patterns to GBP3 and GBP6 , including the top-prioritized GBP1 , might also play a role in atherosclerosis mediated by hypercholesterolemia . The present work can be improved in multiple directions . While we adopted a simple weighted integration method of the PPI network and co-abundance networks to take into account the predictive potential of each network , similar to the weighted network integration method demonstrated in the study by Valentini et al . , 2014 , many alternative filtering strategies remain . As an alternative to combining the PPI network and co-abundance networks , we explored a filtering strategy where we removed the negative correlation edges in addition to adding the positive correlation ones . However , the number of edges that were already present in the PPI was limited to a few hundreds 114 , 271 and 105 edges for M ( - ) , M ( IFNγ ) and M ( IL-4 ) , respectively ) , therefore the removal of these did not result in a substantial increase in the ROCs , with limited improvement for M ( IFNγ ) and M ( IL-4 ) in terms of AUROCs ( 86 . 77% to 86 . 78% for M ( IFNγ ) and 85 . 18% to 85 . 20% for M ( IL-4 ) . In other words , the improvement caused by the removal of a few hundred edges was mostly overshadowed by the addition of many more positive correlation edges . This suggests that the main shortcoming of the PPI is incompleteness , which is thought to be a large percentage of all possible edges , rather than noise and false positives , which make up a much smaller portion . Another constraint regarding the building of co-abundance networks is that we used only the common proteome between M ( - ) , M ( IFNγ ) and M ( IL-4 ) and applied a PSM cutoff of >10 , which reduces the size of the co-abundance networks . While this ensures that we work with the strongest signals , relaxing these criteria might result in the discovery of additional key targets . In summary , the present study demonstrates that proteomics-derived co-abundance edges introduce context specificity to the PPI network and significantly improve the prediction of drug targets related to the biology in question . Indeed , co-abundance networks derived from IFNγ- and IL-4-stimulated macrophage-like cells resulted in the network proximity-based prediction of GBP1 and WARS as potential regulators of pro-inflammatory macrophage activation . Our in vitro loss-of-function studies involving human primary macrophages verified the role of these targets in pro-inflammatory signaling as regulators of CCL2 and JAK2 as well as STAT1 phosphorylation . Overall , our workflow has general applicability and can serve as the blueprint for subsequent studies to combine proteomics data with context-independent interactomes to extract cell type- and experimental condition-specific information for the purpose of identifying targetable pathways and molecules in the context of complex pathobiologies such as CVD . To represent the current knowledge on human protein-protein interactions as a literature-based PPI network , we compiled a comprehensive list of PPIs with experimental evidence from various databases ( Menche et al . , 2015 ) including the following types of interactions: ( i ) regulatory interactions ( Matys et al . , 2006 ) , ( ii ) high-quality binary PPIs tested via high-throughput yeast two-hybrid ( Y2H ) systems , obtained from multiple publications ( Rolland et al . , 2014; Rual et al . , 2005; Stelzl et al . , 2005; Venkatesan et al . , 2009; Yu et al . , 2011 ) and public databases ( Aranda et al . , 2010; Ceol et al . , 2010 ) , ( iii ) literature-curated PPIs identified by affinity purification followed by mass spectrometry ( AP-MS ) , Y2H , low-throughput experiments , and protein three-dimensional structures ( Aranda et al . , 2010; Ceol et al . , 2010; Keshava Prasad et al . , 2009; Stark et al . , 2011; Zhang et al . , 2013 ) , ( iv ) metabolic enzyme-coupled interactions ( Lee et al . , 2008 ) , ( v ) protein complexes derived from a variety of experimental tools , from co-immunoprecipitation to co-sedimentation and ion exchange chromatography ( Ruepp et al . , 2010 ) , ( vi ) kinase-substrate interactions derived from high-throughput and literature-curated low-throughput experiments ( Hornbeck et al . , 2012 ) , and ( vii ) signaling interactions from both high-throughput and literature curation ( Vinayagam et al . , 2011 ) . After the removal of duplicated edges , the resulting PPI network contained 170 , 303 interactions between 14 , 213 proteins . For our analyses , we discarded the isolated nodes that only self-interact and used the largest connected component ( LCC ) of this network , which has 170 , 253 interactions and 14 , 115 proteins . The resulting network has an average degree <k > of 24 . 12 , an average clustering coefficient <C> of 0 . 210 , a diameter of 12 and an average shortest path length of 3 . 54 ( Supplementary file 1 ) . To address the investigation bias and eliminate noise due to indirect associations from protein complexes wherever possible within the PPI network , we systematically removed the edges from individual , low-throughput experiments from literature and co-complex interactions . The removal of low-throughput edges resulted in 13 , 604 proteins and 147 , 295 interactions and the subsequent removal of co-complex edges resulted in 13 , 568 proteins and 125 , 495 interactions . As it yields the best performance ( Figure 2—figure supplement 3 ) , this ‘de-noised’ PPI with the low-throughput and co-complex interactions removed was used throughout the manuscript . We have detailed the IFNγ and IL-4 stimulation conditions , cell culture experiments and six-plex tandem mass tagging ( TMT ) sample preparation methods previously in Iwata et al . , 2016 . The TMT peptide samples were analyzed using high-resolution and accuracy LTQ-Orbitrap Elite ( Thermo Scientific ) and subsequently annotated using the SEQUEST search algorithm via the Proteome Discoverer ( PD ) Package ( version 1 . 3 , Thermo Scientific ) ( Eng et al . , 1994 ) as described previously ( Iwata et al . , 2016 ) . Master proteins with two or more unique peptides were used for TMT reporter ratio quantification . For each peptide-spectrum match ( PSM ) , the TMT ion channel intensities were normalized to the time-zero channel . Protein abundances were then calculated by taking the median of their corresponding PSM ratios ( Dayon et al . , 2008 ) . To ensure that the subsequent co-abundance networks are built out of the proteins detected with high confidence , we further filtered the list of proteins to those with more than 10 PSMs . For each of these proteins with PSM >10 ( 2555 , 2586 and 2695 proteins for M ( - ) , M ( IFNγ ) and M ( IL-4 ) , respectively ) , we then extracted the abundance profile , which consists of six time points ( 0 , 8 , 12 , 24 , 48 and 72 hr ) . Then , for every pair of proteins , we calculated the Pearson correlation coefficient ( r ) between their abundance profiles . This resulted in a weighted , complete graph where the edge weights were given by Pearson’s r values . As a robust way of comparing these correlation values against a null model representing the expectation by chance , we created randomized datasets whereby the abundance profile vectors were shuffled for a large number of realizations , and calculated permutation-based empirical p-values . In this case , the empirical p-value ( P* ) was calculated as P* = r>/N where N = 300 is the total number of permutations performed and r> is the number of permutations where the permuted r was higher than the real r . In other words , P* is the probability of encountering a higher value of r in the permuted data than the observed r . We then adjusted the empirical p-values for multiple-testing correction using the Benjamini-Hochberg ( BH ) procedure to control for the false discovery rate ( FDR ) . This resulted in a Pearson correlation and adjusted empirical p-value pair ( r , Q* ) for all possible edges . Finally , to filter the network to preserve only the most high-confidence co-abundance edges , we set an edge weight threshold . For the selection of this threshold , we performed a sensitivity analysis where we plotted the density of the resulting co-abundance network as a function of the adjusted empirical p-value . At an FDR of 1% , we selected the highest Pearson correlation that maintained the network density , that is 0 . 90 , as the edge weight cutoff ( Figure 1—figure supplement 1A ) . As additional correlation measures , we calculated biweight midcorrelation and partial correlation using the WGCNA package ( Langfelder and Horvath , 2012 ) and ppcor package ( Kim , 2015 ) , respectively , in R . Baseline ( M ( - ) ) abundances were controlled for in the partial correlation calculation . Resulting two-sided p-values were adjusted for multiple testing using the BH procedure , and a correlation threshold of 0 . 90 ( p<0 . 01 ) was chosen for consistency . We tested the biological relevance of the co-abundance networks by quantifying their edge overlap with the external protein-protein interaction networks below: ( a ) Binary interactions: We obtained human binary interactions from HINT database ( http://hint . yulab . org/; Das and Yu , 2012 ) , which had 55 , 493 interactions ( retrieved March 2017 ) , as well as two interactomes based on Y2H assays: HI-II-14 ( Rolland et al . , 2014 ) and HI-III-16 ( Unpublished , from the Center for Cancer Systems Biology ) , which contain 11 , 603 and 48 , 229 interactions , respectively , after mapping from UniProt IDs to gene symbols . ( b ) Affinity Purification-Mass Spectrometry-based interactions: We used two recent sets of interactomes based on AP-MS measurements . BioPlex 2 . 0 ( http://bioplex . hms . harvard . edu/; Huttlin et al . , 2015 ) contained 56 , 553 interactions and ( Hein et al . , 2015 ) contained 27 , 380 interactions overall . ( c ) Co-complex interactions were obtained from HINT ( Das and Yu , 2012 ) co-complex interactome consisting of 121 , 546 interactions ( retrieved March 2017 ) and a curated co-complex association network ( Woodsmith and Stelzl , 2014 ) consisting of 74 , 131 interactions . We also derived interactions based on co-complex membership using the CORUM ( http://mips . helmholtz-muenchen . de/corum/; Ruepp et al . , 2010 ) protein complex database , which resulted in 47 , 378 interactions ( retrieved July 2017 ) . Lastly , we used hu . Map , a recent large-scale protein complex map resulting from the integration of over 9000 MS experiments , which contained 35 , 375 interactions after setting an interaction probability threshold of 0 . 265 , which is deemed high-confidence ( Drew et al . , 2017 ) . ( d ) We constructed GO term networks for the three branches of the GO , namely the Biological Process ( GO:BP ) , Molecular Function ( GO:MF ) and Cellular Component ( GO:CC ) by taking the one-mode projection of the gene-term bipartite network on the gene component , that is by connecting genes based on their shared GO terms . We excluded the inference-based evidence types IPI ( inferred from physical interaction ) and IEA ( inferred from electronic analysis ) , and ND ( not described ) . To ensure that the GO networks adhere to relatively specific GO terms , we considered terms with less than 100 associated genes . This resulted in 11 , 361 nodes and 631 , 799 edges for GO:BP , 7586 nodes and 162 , 154 edges for GO:MF , and 6044 nodes and 196 , 936 edges for GO:CC , after mapping Entrez IDs to gene symbols . The GO database was downloaded from http://www . geneontology . org/ in August 2015 . To calculate the edge overlap , we pruned each external interaction network to contain only the proteins present at the intersection of that network and the co-abundance network . We used two-sided Fisher Exact test to calculate the enrichment odds ratios , confidence intervals and p-values . We retrieved the known drug targets for cardiovascular diseases by querying iCTNet2 ( Wang et al . , 2015 ) ( http://apps . cytoscape . org/apps/ictnet2; retrieved November 2015 ) for phenotype-gene , gene-drug and disease-drug interactions . To cover a broad range of cardiovascular diseases , we selected the ‘cardiovascular system disease’ phenotype ( ID: 1287 ) and all its subcategories and included the results from both the GWAS catalog and the Online Mendelian Inheritance in Man ( OMIM ) database . For the drug associations , we queried iCTNet2 for disease-drug and gene-drug interactions from DrugBank ( Wishart et al . , 2006 ) and the Comparative Toxicogenomics Database ( CTD ) ( Mattingly et al . , 2003 ) . This resulted in a tripartite network of cardiovascular system diseases and their related genes and targets . We excluded isolated cardiovascular system diseases with no drug or gene interactions , or only one type of interaction . We then eliminated diseases and extracted the largest connected component of the remaining network to arrive at the final bipartite network of 268 drugs and 283 drug targets ( Figure 1—figure supplement 1C ) . Of the 283 drug targets 251 were mapped onto the PPI network and 52 were represented in the co-abundance networks . While the iCTNet database also has protein-protein interactions , they were not used in our study to avoid circularity . We used a drug target prioritization method based on the topological proximity of candidate genes to all seed genes in the molecular interaction network , where seed genes are defined as the known cardiovascular drug targets . In particular , the proximity score is inversely proportional to the network distance between candidate genes and seed genes , where the proximity score PS ( c ) for each candidate c is defined asPSc=∑s∈SI ( s ) dcs+1where I ( s ) is the relative weight , or importance , of seed s , dcs is the shortest network distance between candidate gene c and seed gene s and S is the set of all seed genes . The non-Euclidean network distance dcs is measured in terms of the number of links . As the drug target data lacks information about their relative importance , we weighted the seed genes equally , assigning them the relative weight I ( s ) value of 1 . 0 . Using the proximity score , we ranked all the proteins in the PPI network . As a measure of the prediction performance , that is the rate of capturing true positives , of our prioritization method , we plotted the receiver operating characteristic ( ROC ) curves and calculated the area under the ROCs ( AUROC ) . We implemented k-fold ( k = 7 was chosen for a reasonable test set size , unless otherwise noted ) cross-validation to determine the statistical significance of the difference between AUROC values , and as the datasets were partitioned into folds consistently across PPI and PPI + CoA , a paired t-test was used to compare the two cases and two-tailed p-values were reported . For the addition of random edges used as negative control in the prediction performance assessment , we adopted a degree-preserving randomization strategy . As there is typically a small number of very highly connected proteins in the PPI network , to avoid repeatedly selecting the same proteins in lieu of these highly connected nodes , all proteins were logarithmically binned according to their degree . Edges were then established between pairs of proteins uniformly randomly picked from their respective degree bin representing a pool of similar degree proteins . In addition to ROC curves , we additionally plotted precision-recall curves as they can provide additional information for imbalanced datasets where positives are rare . As additional prioritization measures , we implemented RWR and DADA as described in Erten et al . , 2011 , and Kernel distance dk ( c ) per candidate c was defined asdk ( c ) =−ln∑s∈Se− ( dcs+1 ) |S| , where dcs is the shortest network distance between candidate gene c and seed gene s and S is the set of all seed genes . In order to further optimize the prediction performance , we sought to distinguish between co-abundance and PPI links . For this , we redefined network distance dcs to accommodate the link weights such that co-abundance link weights wCoA and PPI link weights wPPI have different values , which effectively results in two different types of links . The new network distance d*cs between a candidate protein c and a seed protein s therefore becomesdcs*=∑C∈ECoAwCnC+∑P∈EPPIwPnP , where C ( P ) denotes co-abundance ( PPI ) links , ECoA ( EPPI ) is the set of all co-abundance ( PPI ) links , wC ( wP ) is the weight of co-abundance ( PPI ) links , and nC ( nP ) is the number of co-abundance ( PPI ) links in the shortest path between c and s . Since the proximity score D ( c ) is inversely proportional to the distance between candidates and seeds , the ratio between the weights of the two types of links modulates the relative importance of these link types in the prioritization scheme . Under the regime wC/wP <1 , the co-abundance network is given the advantage , influencing the enriched network topology more than PPI whereas under for wC/wP >1 , PPI has more influence than the co-abundance component . In the case of wC/wP = 1 , d*cs simplifies to dcs , corresponding to the case where co-abundance and PPI links are treated equally . With the weight ratio as our single parameter , we performed a scan where we calculated the area under the ROC curve ( AUROC ) as a function of this ratio in the range [0 , 10] , which displayed a maximum in this range ( Figure 2—figure supplement 4A ) indicating that this procedure of weighing the two types of links is amenable to optimization . We also implemented k-fold ( k = 7 was chosen for a reasonable test set size , unless otherwise noted ) cross-validation in the optimization procedure , and selected the weight ratio that yields the highest mean AUROC across all folds for each condition , to be used for subsequent analyses throughout the manuscript . These values were found to be 0 . 4 , 0 . 4 and 0 . 1 for M ( - ) , M ( IFNγ ) and M ( IL-4 ) , respectively . For the external validation of drug targets , we downloaded drug-target interactions and drug indications from the DrugCentral database ( Ursu et al . , 2017 ) ( http://drugcentral . org/; datasets were time stamped 04/25/2017 ) . After removing drugs without associated gene symbols , we obtained 2272 targets ( 1444 of which were mapped to the PPI and 324 of which were in the co-abundance networks ) for the ‘DrugCentral’ dataset representing all drug-target interactions in the DrugCentral database and 599 targets ( 229 of which were mapped to the PPI and 23 of which were in the co-abundance networks ) for the ‘T_clin’ dataset representing the targets with known mechanisms of action . To acquire the CVD related targets from DrugCentral , we mined the drug indications data for the keywords ‘arterio’ , ‘athero’ , ‘artery’ , ‘cardi’ , ‘coronary’ and ‘heart’ . This resulted in 502 CVD targets ( 330 of which were mapped to the PPI and 23 of which were in the co-abundance networks ) for DrugCentral , 115 CVD targets ( 55 of which were mapped to the PPI and 1 of which was in the co-abundance networks ) for the ‘T_clin’ dataset . In addition to the drug-target interaction and drug indication data from DrugCentral database , we downloaded detailed target information from TCRD database ( Nguyen et al . , 2017 ) ( http://juniper . health . unm . edu/tcrd/; Version 4 . 6 . 2 ) , which contains information about the top five associated diseases and targets are categorized into four categories based on their development and druggability levels . We considered the targets belonging to the ‘Clinical’ ( approved drugs ) and ‘Chemical’ ( drugs that satisfy the activity thresholds outlined in the TCRD website - http://juniper . health . unm . edu/tcrd/ ) categories . Mining the top 5 diseases of TCRD Clinical and Chemical datasets for the CVD related keywords above , we obtained 53 CVD targets ( 52 of which were mapped to the PPI and 9 of which was in the co-abundance networks ) for Clinical and 121 CVD targets ( 116 of which were mapped to the PPI and 27 of which was in the co-abundance networks ) for the Chemical category . We used three datasets to test the relevance of the prioritized proteins to inflammatory processes , innate immune response and cardiovascular disease . For inflammation , we looked for the enrichment of the top-ranked proteins in the inflammatome signature ( Wang et al . , 2012 ) , which includes common rodent inflammatory signatures from 12 expression profiling datasets corresponding to nine different tissues and 11 disease models . Overall , this dataset contained 2483 genes that comply with a consensus p-value threshold . Of these 1942 were mapped to the PPI network and 555 of which were in the co-abundance networks . The innate response genes were obtained from the InnateDB database ( http://www . innatedb . com/; Lynn et al . , 2008 ) , which is a knowledge base for the mammalian innate immune response captured through contextual manual curation . It includes over 18 , 000 interactions between over 1500 proteins ( retrieved in April 2016 ) , 872 of which were mapped to the PPI network and 233 of which were in the co-abundance networks . For cardiovascular disease signature , we used the CADGene database ( http://bioinfo . life . hust . edu . cn/CADgene/; Liu et al . , 2011 ) , which includes genes implicated by genome wide association studies ( GWAS ) as well genes obtained by manual literature curation . Overall , it included 604 genes related to coronary artery disease , 574 of which were mapped to the PPI network and 97 of which were in the co-abundance networks . Finally , to further expand our list of CAD genes , we used the ‘CAD1000G Extend’ dataset described in Zhao et al . , 2016 , containing 881 CAD genes ( 560 of which were mapped to the PPI network and 148 of which were in the co-abundance networks ) based on the GWAS catalog , additional candidate genes identified through the CARDIoGRAM-C4D study based on Metabochip data and therefore not included in the GWAS catalog , further supplemented by genes from the 1 , 000 Genomes study ( 1000 Genomes Project Consortium et al . , 2015 ) . For the pathway enrichment analysis , we used ConsensusPathDB ( Kamburov et al . , 2011 ) ( pathway data retrieved from http://consensuspathdb . org/ in February 2017 ) and considered the canonical pathways from KEGG , Biocarta , and Reactome ( Chowdhury and Sarkar , 2015 ) . This resulted in 301 pathways from KEGG comprising 7121 genes , 252 pathways from Biocarta comprising 1408 genes , and 1764 pathways from Reactome comprising 10 , 095 genes , which together defined a pathway space of 2317 pathways consisting of 11 , 447 genes . We tested the top prioritized proteins for pathway enrichment by a hypergeometric test and adjusted for multiple comparisons using the Benjamini-Hochberg method for controlling false discovery rate ( FDR ) . The cutoff to determine the top prioritized proteins was based on the permutation of p-values whereby an empirical p-value was calculated for each rank for N = 10 , 000 random realizations . Pathways with FDR adjusted p-value ( q-value ) <0 . 05 were considered significantly enriched . The pathway networks represent pathways as the nodes and the shared genes between pathways as the edges . Node size corresponds to –log ( q-value ) and edge weight ( thickness ) corresponds to the gene overlap between pairs of pathways measured by the Jaccard index J , which is defined asJ=sA∩sBsA∪sBwhere sA and sB are the sets of top prioritized proteins that belong to pathway A and pathway B , respectively . We calculated J for all enriched pathway pairs and discarded edges with J values less than 0 . 1 in the visualization for clarity . The network visualizations were made using Gephi v0 . 8 . 2 ( Bastian and Heymann , 2009 ) . To condense the list of network-prioritized candidates , which consists of all of the proteins in the interactome , further into a smaller list supported by proteomics and transcriptomics evidence , we followed a three-layer filtering procedure . The combined ranking was achieved as the result of ( i ) the rank according to the network prioritization , ( ii ) the rank according to the relative protein abundance difference with respect to baseline M ( - ) over all time points , and ( iii ) the gene expression fold change with respect to baseline M ( - ) from human macrophage transcriptome data ( Xue et al . , 2014 ) . We first determined the optimal threshold on the network-based prioritization ranking that would maximize the sum of the sensitivity ( true positive rate ) and specificity ( 1 – false positive rate ) . Based on the ROC curves obtained after optimization of weight ratios , we found these optimal threshold values to be the top 2971 candidates for M ( IFNγ ) and the top 3496 candidates for M ( IL-4 ) ( Figure 4—figure supplement 1A–B ) . Second , we filtered these top-N candidates with the top 500 candidates with the highest relative abundance in the proteomics data for each time point , with respect to the baseline M ( - ) . As the third step of the filtering procedure , we use the top 500 genes with the highest fold change for IFNγ and IL-4 stimulated macrophages from an extensive dataset of human macrophage activation transcriptomes ( Xue et al . , 2014 ) . These three steps resulted in 43 and 49 candidates for M ( IFNγ ) and M ( IL-4 ) , respectively . Finally , we re-ranked this final list of candidates using a combination score based on the network-based prioritization rank , relative abundance rank ( defined as the total difference between the relative abundance profiles between IFNγ or IL-4 stimulated and baseline macrophages ) , and expression fold change rank , such that the final combined rank is given byRcomb . = N ( Nprior . + Nabun . + Nexpr . ) /3 . We confirmed that the final combined ranking is robust with respect to the choice of top-N ranked expression and abundance ( Figure 4—figure supplement 1C ) . The final combined ranking was used in candidate selection for in vitro silencing experiments . The network closeness of the candidate proteins filtered with the combined ranking to CVD drug targets was measured in terms of the average shortest distance . The average shortest distance D to CVD drug targets was measured by calculating the average shortest distance between each candidate protein c and all drug targets t and then averaging over all candidate proteins c such thatdc=1Nt∑t∈TdctandD=1Nc∑c∈Cdc , where dct is the shortest distance between c and t and C and T are the sets of proteins in the target candidates and CVD drug targets , respectively . To compare this average shortest distance value to what would be expected by chance , the average shortest distance of the same number of randomly selected proteins to CVD drug targets was calculated for N = 1000 realizations . To control for degree bias , the random protein selection was done in a degree-preserving manner where all proteins were binned according to their degree and random proteins were selected uniformly at random from their corresponding degree bin . Finally , z-scores and empirical p-values were calculated byz= D-DrσDrandpemp . =P ( Dr<D ) , respectively , where Dr is the average shortest distance of a randomized instance , Dr is the mean of the average shortest distance of all randomized instances , and σDr is their standard deviation . Network measures including shortest distances and centralities were calculated using the NetworkX package ( Hagberg et al . , 2008 ) v1 . 9 in Python v2 . 7 . 10 . To assess the potential of candidates of interest to be drug targets , we used the DrugEBIlity database ( https://www . ebi . ac . uk/chembl/drugebility/ ) ( version 3 . 0 ) , which predicts the structural druggability of a molecule by how suitable its binding sites are for small molecules under the Lipinski's Rule of 5 requiring at most 10 hydrogen bond acceptors , at most five hydrogen bond donors , and a molecular weight 500 Da or less . ‘Tractability’ is a more relaxed criterion compared to druggability , requiring at most 15 hydrogen bond acceptors , at most eight hydrogen bond donors , and a molecular weight between 200 Da and 800 Da . ‘Ensemble druggability’ is the strictest criterion where the average of druggability score is calculated under different machine learning models . Silencer Select validated siRNA for human WARS was purchased from Thermo Fisher Scientific ( Catalog# 439085 ) . SMARTpool ON-TARGET plus Human GBP1 siRNA oligos were from GE Healthcare Dharmacon ( L-005153 ) . Si-RNA transfection on THP-1 cells or PBMC-derived macrophages was performed by using Magnetofection SilenceMag ( OZBIOSCIENCES , San Diego ) at final concentration of 50 nM . 48 hr after transfection , macrophages were stimulated with human 10 ng/ml IFNγ ( R and D systems ) for 6–24 hr before further experiments . Total RNA were extracted using an Illustra RNAspin Mini kit ( GE Healthcare , Piscataway , NJ ) and cDNAs were synthesized using a high capacity cDNA reverse transcription kit ( Applied Biosystems , Carlsbad , CA ) . Real-time PCR was performed using Taqman probes for WARS , GBP-1 , CCL2 , TNFα , JAK2 , STAT1 , and GAPDH on a 7900HT fast real-time PCR system ( Applied Biosystems ) . Relative expression of each gene was normalized by GAPDH . CCL2 and TNFα proteins in culture medium from macrophages were detected by ELISA kit purchased from R and D systems ( Minneapolis , MN ) . Macrophages whole cell lysate were prepared using RIPA buffer containing protease inhibitor ( Roche ) . Total protein was separated by 4–20% Mini-PROTEAN TGX Precast Gel and transferred using the iBlot Western blotting system ( Life Technologies ) . Primary antibodies against human GBP-1 ( Abcam , Catalog# ab131255 ) , WARS ( Thermo Fisher Scientific , Catalog# PA5-29102 ) , STAT1 ( Cell signaling , Catalog# 9172 ) , phosphorylated STAT1 at Y701 ( Cell signaling , Catalog #9167 , ) JAK2 ( Cell signaling , Catalog#3230 ) , and β-actin ( Novus ) were used . Protein expression was detected using Pierce ECL Western Blotting substrate reagent ( Thermo Scientific ) and ImageQuant LAS 4000 ( GE Healthcare ) .

When human cells or tissues are injured , the body triggers a response known as inflammation to repair the damage and protect itself from further harm . However , if the same issue keeps recurring , the tissues become inflamed for longer periods of time , which may ultimately lead to health problems . This is what could be happening in cardiovascular diseases , where long-term inflammation could damage the heart and blood vessels . Many different proteins interact with each other to control inflammation; gaining an insight into the nature of these interactions could help to pinpoint the role of each molecular actor . Researchers have used a combination of unbiased , large-scale experimental and computational approaches to develop the interactome , a map of the known interactions between all proteins in humans . However , interactions between proteins can change between cell types , or during disease . Here , Halu et al . aimed to refine the human interactome and identify new proteins involved in inflammation , especially in the context of cardiovascular disease . Cells called macrophages produce signals that trigger inflammation whey they detect damage in other cells or tissues . The experiments used a technique called proteomics to measure the amounts of all the proteins in human macrophages . Combining these data with the human interactome made it possible to predict new links between proteins known to have a role in inflammation and other proteins in the interactome . Further analysis using other sets of data from macrophages helped identify two new candidate proteins – GBP1 and WARS – that may promote inflammation . Halu et al . then used a genetic approach to deactivate the genes and decrease the levels of these two proteins in macrophages , which caused the signals that encourage inflammation to drop . These findings suggest that GBP1 and WARS regulate the activity of macrophages to promote inflammation . The two proteins could therefore be used as drug targets to treat cardiovascular diseases and other disorders linked to inflammation , but further studies will be needed to precisely dissect how GBP1 and WARS work in humans .

2018

Context-enriched interactome powered by proteomics helps the identification of novel regulators of macrophage activation

Natural killer ( NK ) cells belong to the innate immune system; they can control virus infections and developing tumors by cytotoxicity and producing inflammatory cytokines . Most studies of mouse NK cells , however , have focused on conventional NK ( cNK ) cells in the spleen . Recently , we described two populations of liver NK cells , tissue-resident NK ( trNK ) cells and those resembling splenic cNK cells . However , their lineage relationship was unclear; trNK cells could be developing cNK cells , related to thymic NK cells , or a lineage distinct from both cNK and thymic NK cells . Herein we used detailed transcriptomic , flow cytometric , and functional analysis and transcription factor-deficient mice to determine that liver trNK cells form a distinct lineage from cNK and thymic NK cells . Taken together with analysis of trNK cells in other tissues , there are at least four distinct lineages of NK cells: cNK , thymic , liver ( and skin ) trNK , and uterine trNK cells . Immune cells migrate throughout the body where they search tissues for pathological events induced by invading pathogens or emerging tumors , for example . Upon encounter with these events , circulating immune cells stop and respond in collaboration with yet other immune cells , often in organized lymphoid tissues . The subsequent orchestrated host immune response controls the pathological process by directing relevant immune cells to the damaged tissue . In contrast to circulating immune cells are tissue-resident immune cells which already reside in selected organs where they appear to be poised to deliver immune responses . However , how tissue-resident immune cells contribute to host responses is less well understood and will be aided by identifying factors that distinguish circulating and tissue-resident immune cells . Natural killer ( NK ) cells are components of the innate immune system ( Yokoyama , 2013 ) . Initially described on the basis of their inherent capacity to kill tumor cells without prior sensitization , NK cells are now known to participate in a wide variety of immune responses , such as early control of viral infections . In addition , they can respond to pro-inflammatory cytokines by producing yet other inflammatory cytokines , such as interferon-γ ( IFNγ ) , their signature cytokine that can influence adaptive immune cells . NK cells require IL-15 and its cognate receptor , IL-15R , for development ( Cao et al . , 1995; DiSanto et al . , 1995; Suzuki et al . , 1997; Lodolce et al . , 1998; Kennedy et al . , 2000 ) . In knockout mice lacking IL-15 or any chain of the trimeric IL-15R ( α , β , γ ) , splenic NK cells are absent . The development of splenic NK cells is thought to occur largely if not exclusively in the bone marrow ( BM ) where cells committed to the NK cell lineage undergo a series of putative developmental stages , characterized by acquisition and loss of various surface markers , including cytokine receptors , NK cell receptors , and integrins ( Kim et al . , 2002; Yokoyama et al . , 2004; Di Santo , 2006 ) . Out in the periphery , mature splenic NK cells can be further distinguished by differential expression of CD11b and CD27 ( Kim et al . , 2002; Hayakawa and Smyth , 2006 ) . Thus , conventional splenic NK cells display developmental markers associated with maturation . NK cells require certain transcription factors for development ( Hesslein and Lanier , 2011 ) , in particular NFIL3 ( E4BP4 ) , described as the NK cell-specification factor ( Di Santo , 2009 ) . Mice lacking NFIL3 have essentially no splenic NK cells though other organs were not thoroughly examined ( Gascoyne et al . , 2009; Kamizono et al . , 2009; Kashiwada et al . , 2010 ) . The related t-box transcription factors , Tbet ( Tbx21 ) and eomesodermin ( Eomes ) , play more complex roles in NK cell development ( Townsend et al . , 2004; Gordon et al . , 2012 ) . In the absence of Tbet , splenic NK cells display an immature phenotype , and a subpopulation of NK cells in the liver is absent , consistent with overlapping and cooperative roles of Tbet and Eomes in NK cell development . Alternatively , Tbet may direct the development of a separate lineage of NK cells . Current data cannot definitively distinguish between these possibilities for the role of Tbet and Eomes in NK cell development . Classically studied in the mouse spleen and expressing the NK1 . 1 antigen on CD3ε-negative cells , NK cells are also present in solid organs , such as the thymus , uterus , and liver ( Yokoyama , 2013 ) . Like the conventional NK ( cNK ) cells in the spleen , thymic NK cells are cytotoxic and require IL-15 but they differ from cNK cells by their characteristic expression of CD127 ( IL-7 receptor α ) and requirement for a thymus where they can arise from early thymocyte precursors ( Vosshenrich et al . , 2006; Ribeiro et al . , 2010; Vargas et al . , 2011 ) . Furthermore , thymic NK cells uniquely require the transcription factor GATA-3 for development ( Vosshenrich et al . , 2006 ) . NK cells are normally present in the non-pregnant uterus ( Parr et al . , 1991; Yadi et al . , 2008; Mallidi et al . , 2009 ) but have been mostly studied after they expand at the site of embryo implantation during pregnancy ( Moffett and Loke , 2006; Hatta et al . , 2012 ) . Like cNK cells , uterine NK ( uNK ) cells require IL-15 for development ( Ashkar et al . , 2003 ) . In addition , they are cytotoxic as they express perforin and granzymes , and they produce IFNγ ( Parr et al . , 1990; Ashkar et al . , 2000 ) . Interestingly , however , uNK cells appear relatively normal in Tbet-deficient mice ( Tayade et al . , 2005 ) and recent studies suggest that a subset of uNK cells can be distinguished from cNK cells ( Yadi et al . , 2008 ) . Thus , NK cell subsets can be identified in different tissues that appear to be distinguishable from cNK cells . In the liver , we recently showed that there are two populations of NK cells , distinguished by mutually exclusive expression of CD49a and DX5 ( Peng et al . , 2013 ) . Phenotypically , CD49a−DX5+ are very similar to splenic cNK cells whereas CD49a+ DX5– are unlike splenic cNK cells . In parabiotic mice , the host liver contains CD49a+ DX5− NK cells of host origin and circulating CD49a−DX5+ NK cells derived from both host and the other parabiont , indicating that the CD49a+ DX5− cells are tissue-resident NK ( trNK ) cells whereas the CD49a−DX5+ cells are cNK cells . The trNK cells appear similar to immature cNK cells because they express similar markers ( NK1 . 1 , NKp46 ) but low levels of CD11b , are DX5− , and display high levels of TNF-related apoptosis-inducing ligand ( TRAIL ) ( Kim et al . , 2002; Di Santo , 2006; Peng et al . , 2013 ) . However , the liver trNK cells appear to be distinct from immature BM cNK cells because they are cytotoxic and express CD49a . Nonetheless , the liver trNK cells could represent an intermediate cell stage in the development of mature splenic cNK cells from NK cell precursors . Consistent with this possible developmental intermediate stage , there is a complete absence of TRAIL+ NK cells in the livers of Tbet-deficient mice ( Gordon et al . , 2012 ) . Importantly , this analysis was done before the CD49a+ DX5− trNK cells in the liver were described . Therefore , while the subset of TRAIL+ liver NK cells lacking in Tbet-deficient mice could represent a failure of maturation of cNK cells , an alternative interpretation of these results is that there may be two different NK cell lineages , one of which is completely Tbet-dependent . The current data therefore suggest that there are three possible origins of the liver trNK cells . ( 1 ) They could be related to thymic NK cells because GATA-3-deficient NK cells are defective in homing to the liver ( Samson et al . , 2003 ) . ( 2 ) They could represent an intermediate stage in the development of NK cell precursors into mature splenic NK cells . ( 3 ) They could represent an alternative NK cell lineage , distinct from both cNK and thymic NK cells . If these relationships can be resolved , it will then be important to determine how liver trNK cells are related to other trNK cells , such as uNK cells , and to the expanding list of innate lymphoid cells ( ILCs ) ( Spits and Cupedo , 2012; Spits et al . , 2013 ) . Herein we examined the liver trNK cells in detail with respect to thymic NK , splenic cNK , and trNK cells in other organs . Based on phenotypic characteristics and transcription factor requirements , our studies support the presence of at least four populations of NK cells: cNK ( spleen and circulating ) , thymic , liver ( and skin ) trNK cells , and uNK cells . Furthermore , these studies indicate that trNK cells can be distinguished from each other and from ILCs . Thymic NK cells require GATA-3 ( Vosshenrich et al . , 2006 ) , and GATA-3-deficient NK cells have a defect in liver migration and IFNγ production ( Samson et al . , 2003 ) , raising the possibility that the liver trNK cells may be related to thymic CD127+ NK cells . However , as compared to thymic NK cells , liver trNK ( CD49a+ DX5− ) cells did not clearly express the prototypic thymic NK cell marker , CD127 ( Figure 1A ) . To more definitively determine the relationship between these NK cell subsets , we examined athymic nude mice ( Foxn1−/− ) which lack thymic NK cells but possess liver trNK cells at a somewhat higher percentage when compared to wildtype ( WT ) controls ( Figure 1B , C ) . Also , liver trNK cell numbers are preserved or higher in nude mice ( Figure 1D ) . In addition , we examined mice with hematopoietic cell deficiency of GATA-3 due to expression of Cre under control of Vav which is expressed in all hematopoietic cells . The liver trNK cells from Gata3fl/fl-Vav-Cre mice were present ( Figure 1E ) whereas CD127+ thymic NK cells were absent in these mice ( Figure 1F ) , as previously shown ( Vosshenrich et al . , 2006 ) . Taken together , these data indicate that the trNK cells from the liver develop independent of the thymus or GATA-3 , and can be clearly distinguished from thymic NK cells . 10 . 7554/eLife . 01659 . 003Figure 1 . CD49a+ DX5− trNK cells in the liver are distinct from the CD127+ thymic NK cells . ( A ) CD127 is poorly expressed on liver trNK cells . Thymi and livers were isolated from WT C57BL/6NCr mice , stained , and flow cytometry was performed . The histogram displays the expression level of CD127 on thymic NK cells and CD49a+DX5− and CD49a−DX5+ NK cells in the liver . Gated on live CD3−CD19−NK1 . 1+ cells . ( B–D ) Liver trNK cells are present in nude mice . Spleens and livers were isolated from WT C57BL/6NCr and Foxn1−/− mice , stained , and flow cytometry performed . The dot plots ( B ) were gated on live CD3−CD19−NK1 . 1+ cells and display the percentage expressing CD49a and DX5 in each mouse strain in the liver and spleen , as indicated . Stacked bar graphs represent the percentage ( C ) and total number ( D ) of CD3−CD19−NK1 . 1+ cells that express CD49a and DX5 in the liver and spleen of the WT and Foxn1−/− mice . Experiments were performed three independent times . ( E ) Liver trNK cells are present in GATA-3 conditional-deficient mice . Spleens and livers were isolated from WT and Gata3fl/fl-Vav-Cre mice , stained , and flow cytometry was performed . The dot plots were gated on live CD3−CD19−NK1 . 1+ cells and display the percentage expressing CD49a and DX5 in each mouse strain in the liver ( top panels ) and the spleen ( bottom panels ) . Dot plots represent one of two independent experiments performed . ( F ) Thymic NK cells are absent in Gata3fl/fl-Vav-Cre mice . Cells from the thymi of WT and Gata3fl/fl-Vav-Cre mice were isolated , stained , and flow cytometry performed . The histogram displays the expression level of CD127 on thymic NK cells gated on live CD3−CD19−NK1 . 1+ cells . Histogram represents one of two independent experiments performed . DOI: http://dx . doi . org/10 . 7554/eLife . 01659 . 003 To further define the molecular similarities and differences between trNK cells in the liver , and circulating liver and splenic cNK cells , we performed RNA deep sequencing ( RNA-seq ) of sorted NK cells from Rag1−/− mice . Basic hierarchical clustering of the three populations revealed that liver trNK cells are distinct from the liver and splenic cNK cells , which in turn , are more closely related to each other than to liver trNK cells ( Figure 2A ) . ( A higher resolution figure displaying genes most differentially expressed on liver trNK cells is shown in Figure 2—figure supplement 1 . ) Liver trNK cells could be dissimilar from liver and splenic cNK cells perhaps due to their ‘immature’ phenotype , that is , CD27+CD11blow ( Figure 2B ) and TRAIL+ ( Figure 2C ) . To further explore this possibility , we obtained transcriptome RNA-seq profiles from ‘immature’ bone marrow ( BM ) CD49a+DX5− and CD49a−DX5+ NK cell samples and extracted CD49a+DX5− and CD49a−DX5+-specific signatures . When we compared the levels of genes specific to immature CD49a+DX5− NK cell in the BM ( Supplementary file 1 ) to levels of these genes in the other cell subsets , we found that the CD49a+DX5– cells in the liver did not display this cell-type specific pattern ( Figure 2D ) . Since there was some variability in the relative expression levels of individual genes , further examination was performed by gene set enrichment analysis ( GSEA ) in which the genes differentially expressed in the DX5− BM subset was compared to the other NK cell subsets and plotted ( Figure 2—figure supplement 2 ) . Similar analysis was done for genes differentially expressed in the DX5+ BM subset . These relative comparisons to the DX5− subsets showed that the CD49a−DX5+ subsets in BM , liver and spleen were similar to each other ( p=0 . 01 ) . However , the CD49a+DX5− BM NK cells and liver CD49a+DX5− trNK cells were dissimilar ( p=0 . 5 ) ( Figure 2—figure supplement 2 ) . Taken together with our previous analysis with a much more limited panel of cell surface markers ( Peng et al . , 2013 ) , these transcriptome data strongly suggest that the trNK cells in the liver are not a population of immature NK cells that are transitioning to mature cNK cells rather they represent a separate lineage of NK cells and indicate the CD49a−DX5+ subsets in BM , liver , and spleen likely represent the same recirculating population of cNK cells . 10 . 7554/eLife . 01659 . 004Figure 2 . Distinct lineages of NK cells suggested by RNA-seq analysis . ( A ) Heat map showing cluster analysis of the entire gene set between the liver trNK cells and the cNK cells from the liver and spleen . We obtained expression profiles from sorted CD49a+DX5− NK cells in the liver and CD49a−DX5+ cNK cells in the liver and spleen using the same small input RNA-seq approach . ( For a heat map of a smaller number of genes highly expressed in liver trNK cells , see Figure 2—figure supplement 1 . ) ( B ) The trNK cells in liver display an ‘immature’ phenotype by flow cytometry . Cells from the liver and spleen were isolated , stained , and flow cytometry was performed . Dot plots were gated on live CD3−CD19−NK1 . 1+ cells and percentages in each dot plot represent the percentages of the subpopulations , CD49a+DX5− cells in the liver and CD49a−DX5+ in the liver and spleen , that express CD11b and CD27 . Dot plot profiles are representative of at least three experiments . ( C ) Liver trNK CD49a+DX5- cells express TRAIL . Spleens and livers were isolated from WT C57BL/6NCr mice , stained , and flow cytometry was performed . The histogram was gated on live CD3−CD19−NK1 . 1+ cells and displays the expression level of TRAIL on CD49a+DX5− liver trNK cells and CD49a−DX5+ cNK cells in the liver and spleen . ( D ) Expression of genes specific to DX5– population of BM cells in DX5− liver NK cells and DX5+ liver and spleen cNK cells shows non-specific pattern . Shown are ∼200 genes most highly expressed in DX5− BM NK cells as compared to DX5+ BM NK cells ( genes shown in Supplementary file 1 ) . Approximately half of the DX5– BM NK cell-specific genes are upregulated in the liver CD49a+DX5– NK cells and the other half is upregulated in the CD49a−DX5+ BM NK cells . Gene set enrichment analysis also shows non-significant relationship between BM DX5− cells and CD49a+ NK cells ( Figure 2—figure supplement 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01659 . 00410 . 7554/eLife . 01659 . 005Figure 2—figure supplement 1 . Heat map showing the top differentially expressed genes between the liver trNK cells and the cNK cells from the liver and spleen . DOI: http://dx . doi . org/10 . 7554/eLife . 01659 . 00510 . 7554/eLife . 01659 . 006Figure 2—figure supplement 2 . Gene set enrichment analysis ( GSEA ) indicates relationships between BM NK cell and cNK cells in liver and spleen . GSEA comparing transcriptional profiles of liver trNK cells and cNK cells from the liver and spleen against signature genes for DX5+ or DX5- BM NK cells ( black ladder ) . The heat map is ordered by degree of differential expression between these two subsets . Enrichment curves computed by GSEA are shown in green . Genes expressed in cNK cells from the liver and spleen show statistically significant similarities with genes expressed in the DX5+ BM NK cells ( p = 0 . 01 ) , while DX5− BM NK cells do not show statistically significant similarities with the liver trNK cells ( p = 0 . 5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01659 . 006 To further analyze the similarities and differences between the NK cells of interest with greater precision , we examined their detailed phenotypes by flow cytometry . First , we examined the expressed repertoire of Ly49 receptors which are stably expressed on peripheral NK cells . The liver trNK cells express Ly49A , C/I , I , F , and G2 at much lower frequencies than cNK cells from both the liver and spleen ( Figure 3A , B ) . Indeed , some receptors , such as the Ly49D and Ly49H activation receptors , are not expressed on trNK cells . However the frequencies of liver trNK cells expressing Ly49E ( Ly49EF+ but Ly49F− ) and NKG2A are much higher than liver and splenic cNK cells . In addition , the liver trNK cells appear larger and more granular by scatter parameters when compared to the liver and splenic cNK cells ( Figure 3C ) . Since these parameters are associated with activated cNK cells , we examined expression of other activation markers . In naïve mice , the liver trNK cells express higher levels of CD69 , CD44 , CD160 and do not express CD62L as compared to the cNK cells in the liver and spleen , consistent with a more activated state . 10 . 7554/eLife . 01659 . 007Figure 3 . Liver trNK cells are phenotypically distinct from cNK cells . ( A and B ) Differential expression of NK cell receptors on liver trNK and splenic and liver cNK cells . Spleens and livers were isolated from WT C57BL/6NCr mice , stained , and flow cytometry performed . The histograms were gated on live CD3−CD19−NK1 . 1+ cells and display the expression level of NK receptors on CD49a+DX5- liver trNK cells and CD49a−DX5+ cNK cells in the liver and spleen ( A ) . A summary bar graph of the percentage of CD49a+DX5− liver trNK cells and CD49a−DX5+ cNK cells in the liver and spleen that express the various NK receptors ( B ) . ( C ) The liver trNK cells display an activated phenotype . Spleens and livers were isolated from WT C57BL/6NCr mice , stained , and flow cytometry performed . The histograms were gated on live CD3−CD19−NK1 . 1+ cells and display the expression level of activation markers that are expressed on CD49a+DX5− liver trNK cells and CD49a−DX5+ cNK cells in the liver and spleen . DOI: http://dx . doi . org/10 . 7554/eLife . 01659 . 007 To assess the cytokine profile of the liver trNK cells , we stimulated total liver lymphocytes with PMA and ionomycin and measured cytokine production ( Figure 4A–C ) . The liver trNK cells specifically made easily detectable tumor necrosis factor-α ( TNFα ) and granulocyte macrophage colony-stimulating factor ( GM-CSF ) while liver and splenic cNK cells produced these cytokines at much lower levels , if at all ( Figure 4A ) . By contrast , the liver trNK cells made similar amounts of interferon-γ ( IFNγ ) when compared to liver and splenic cNK cells . Interestingly when we analyzed IFNγ and TNFα simultaneously , we found a significant proportion of the liver trNK cells were IFNγ+TNFα+ double producers ( Figure 4B , C ) , a population essentially absent among the cNK cells from the liver and spleen . Finally , we examined the responses of liver trNK cells to YAC-1 targets ( Figure 4D ) . Although both trNK and cNK cells showed degranulation , as evidenced by CD107a expression , the degranulating liver trNK cells also produced TNFα that was rarely seen among the responding cNK cells . Thus , liver trNK cells display phenotypic differences from liver and spleen cNK cells whereas both cNK cell populations are similar , confirming and extending the transcriptome analysis . 10 . 7554/eLife . 01659 . 008Figure 4 . CD49a + DX5- trNK cells of the liver have a unique cytokine profile . ( A ) The liver trNK cells differentially express cytokines when activated . Spleens and livers were isolated from WT C57BL/6NCr mice stimulated with PMA/ionomycin for 4 hr , stained , and IFNγ , TNFα , and GM-CSF cytokine production was analyzed by flow cytometry . The cells were gated on live CD3−CD19−NK1 . 1+ and the graphs represent the percentage of indicated cytokine produced by CD49a+DX5− liver trNK cells and CD49a−DX5+ cNK cells in the liver and spleen 4 hr post stimulation . ( B and C ) The liver trNK cells produce both IFNγ and TNFα when stimulated . Shown are dot plots of cells prepared and stimulated as in A . Each dot plot ( B ) was gated on live CD3−CD19−NK1 . 1+ cells and further gated on CD49a+DX5- liver trNK cells and CD49a−DX5+ cNK cells in the liver and spleen . Cells were co-stained for IFNγ and TNFα and the percentage produced of each cytokine is presented in each quadrant . Bar graphs ( C ) indicate the percentage of IFNγ and TNFα-double producers shown in B . ( D ) The liver trNK cells degranulate and produce TNFα upon stimulation with YAC-1 targets . Liver lymphocytes were co-cultured at a 1:1 ratio with YAC-1 target cells for 6 hr . The cells were stained for the indicated markers and flow cytometry performed . The dot plots were gated on live CD3−CD19−NK1 . 1+ CD49a+DX5− liver trNK cells or CD3−CD19−NK1 . 1+ CD49a−DX5+ cNK cells in the liver for CD107a degranulation and TNFα production . DOI: http://dx . doi . org/10 . 7554/eLife . 01659 . 008 We analyzed liver trNK cells in IL-15Rα-deficient mice which have a deficiency in splenic cNK cells ( Lodolce et al . , 1998 ) . The development of all NK1 . 1+ CD3− cells was negatively impacted by the IL-15Rα deficiency , resulting in an absence of liver trNK cells as well as liver and splenic cNK cells ( Figure 5A ) . Thus , despite their differences , the requirement for IL-15Rα suggests that the trNK cells and cNK cells are more related to each other than the non-NK cell members of the ILC family that do not require IL-15 ( Spits and Cupedo , 2012; Spits et al . , 2013 ) . 10 . 7554/eLife . 01659 . 009Figure 5 . Liver trNK cells have different transcription factor requirements than cNK cells . ( A ) All liver NK cells require IL-15Rα . Spleens and livers were isolated from WT mice and Il15ra−/− mice , stained , and flow cytometry performed . The bar graph displays the percentage of CD3−CD19−NK1 . 1+ cells in the liver and spleen of each strain of mice . ( B ) Eomes transcripts are expressed at lower levels in liver trNK cells . Spleens and livers were isolated from Rag1−/− mice and NK1 . 1+ cells sorted for CD49a+DX5− liver trNK and liver and spleen CD49a−DX5+ cNK cells . RNA-seq was performed on the sorted populations and the expression levels of indicated transcription factors plotted . RPKM = reads per kilobase per million mapped reads . Normalization of read counts by length of transcripts allowed comparison of expression levels of different genes . One of two independent experiments . ( C ) Eomes protein expression is decreased in liver trNK cells . Spleens and livers were isolated from Rag1−/− mice , stained , and flow cytometry performed for indicated transcription factors . The histograms were gated on live CD3−CD19−NK1 . 1+ cells and display the expression level of transcription factors expressed in CD49a+DX5− liver trNK cells and CD49a−DX5+ cNK cells in the liver and spleen . Histogram plots are representative of three independent experiments . ( D and E ) The liver trNK cells are present in NFIL3-deficient mice but absent in Tbx21 ( Tbet ) -deficient mice . Spleens and livers were isolated from WT , Nfil3−/− , and Tbx21−/− mice , stained , and flow cytometry performed . Representative dot plots ( D ) were gated on live CD3−CD19−NK1 . 1+ and display the expression level of CD49a and DX5 in the liver ( top panels ) and the spleen ( bottom panels ) . Bar graphs ( E ) display the percentages ( left column ) and total NK cell number ( right column ) of CD3−CD19−NK1 . 1+ cells that express CD49a and DX5 in the liver and spleen of WT , Nfil3−/− , and Tbx21−/− mice . ( F ) The Nfil3−/− liver trNK cells display an activated phenotype , like liver trNK cells in WT mice . Spleens and livers were isolated from WT C57BL/6NCr and Nfil3−/− mice , stained , and flow cytometry was performed . The histograms were gated on live CD3−CD19−NK1 . 1+ cells and display the expression level of activation markers expressed on CD49a+DX5− liver trNK cells and CD49a−DX5+ cNK cells in the liver and spleen in WT compared to the trNK cells from the Nfil3−/− mice . ( G ) Nfil3−/− liver trNK CD49a+DX5− cells do not express Eomesodermin . Livers were isolated from WT C57BL/6NCr and Nfil3−/− mice , stained , and flow cytometry was performed . The histogram was gated on live CD3−CD19−NK1 . 1+ cells and displays the expression level of Eomes on WT and Nfil3−/− CD49a+DX5− liver trNK cells and CD49a−DX5+ cNK cells in the liver . ( H ) The Nfil3−/− trNK cells in liver display an ‘immature’ phenotype by flow cytometry , similar to trNK cells in WT mice . Cells from the liver were isolated , stained , and flow cytometry was performed . Dot plots were gated on live CD3−CD19−NK1 . 1+ cells and numbers in each dot plot represent the percentages of the subpopulations , that is , liver CD49a+DX5- cells and CD49a−DX5+ that express CD11b and CD27 . Dot plot profiles are representative of two experiments . ( I ) Stimulated Nfil3−/− liver trNK cells produce cytokines similar to WT liver trNK cells . Livers were isolated from WT C57BL/6NCr and Nfil3−/− mice stimulated with PMA/ionomycin for 4 hr , and cells were co-stained for IFNγ and TNFα and analyzed by flow cytometry . The graphs were gated on live CD3−CD19−NK1 . 1+ cells and represent the percentage of cytokine+ ( or cytokine− ) cells among the CD49a+DX5− liver trNK cells from WT and Nfil3−/− mice , as indicated . Dot plot profiles are representative of two experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 01659 . 009 While our RNA-seq and phenotypic analysis strongly suggest that the liver trNK cells represent a different NK cell lineage from cNK cells , an alternative hypothesis is that the liver trNK cells could be immature NK cells that ultimately become cNK cells , as suggested by their ‘immature’ phenotype ( Figure 2B ) and previous publications ( Townsend et al . , 2004; Gordon et al . , 2012; Peng et al . , 2013 ) . To directly test these hypotheses and gain insight into the developmental relationship of liver trNK cells to cNK cells , we analyzed the mRNA levels of several transcription factors involved in cNK cell development . Particularly informative was decreased transcript levels of eomesodermin ( Eomes ) in the liver trNK cells as compared to liver and splenic cNK cells ( Figure 5B ) . The differences in Eomes transcript levels were verified by flow cytometry which showed lower levels of Eomes protein in liver trNK as compared to cNK cells ( Figure 5C ) , consistent with previous reports ( Gordon et al . , 2012 ) . By contrast , there was no relative difference between liver trNK cells and liver and splenic cNK cells in Tbx21 ( Tbet ) expression at either the transcript or protein level ( Figure 5B , C ) . Transcription factor staining was performed as previously described for Tbx21 ( Sojka and Fowell , 2011 ) . Thus , liver trNK cells express different levels of Eomes compared to spleen cNK cells , suggesting different requirements for the related Tbox transcription factor , Tbet , which plays an overlapping role with Eomes in cNK cell development ( Gordon et al . , 2012 ) . To directly test specific transcription factor requirements , and to gain additional insight into the developmental relationship between the liver trNK cells vs liver and spleen cNK cells , we analyzed these cells in mice lacking NFIL3 and Tbet ( Tbx21 ) . Surprisingly , in NFIL3-deficient mice , liver trNK cells were present in total cell number ( Figure 5D , E ) . Conversely , as previously reported , NFIL3-deficient mice had no cNK cells in the spleen ( Gascoyne et al . , 2009; Kamizono et al . , 2009; Kashiwada et al . , 2010 ) . As well , there were no cNK cells in the liver ( Figure 5D , E ) , consistent with their relationship to splenic cNK cells and raising the percentage of liver trNK cells . By contrast , liver trNK cells were absent in Tbet-deficient mice in both percentage and total cell number , when compared to WT controls and in contradistinction to both cNK cell populations ( Figure 5D , E ) . The requirement for Tbet is consistent with an intrinsic requirement for Tbet in development of TRAIL+ liver NK cells ( Gordon et al . , 2012 ) . Thus , these data demonstrate a unique developmental pathway of liver trNK cells that does not require NFIL3 but depends on Tbx21 , indicating that the liver trNK cells are not precursors of cNK cells . To assess the phenotype and function of the trNK cells in the liver of the NFIL3-deficient mice we assessed the cells at steady state and upon activation by flow cytometry . We found that much like the liver trNK cells from a WT mouse , the trNK cells from Nfil3−/− mice also express markers correlated with activation , that is , higher FSC and SSC , higher levels of CD69 and CD44 and lower levels of CD62L ( Figure 5F ) and do not express Eomes ( Figure 5G ) . They have an ‘immature’ phenotype similar to the WT trNK cells in that they express CD27 and generally lack CD11b expression on most cells ( Figure 5H ) . Upon stimulation , the Nfil3−/− trNK cells in the liver produce both IFNγ+ and TNFα+ simultaneously in a manner comparable to WT trNK cells ( Figure 5I ) . Thus , NFIL3 is not required for normal trNK cells in the liver . Taken together , these data indicate that liver trNK cells require different transcription factors than liver and splenic cNK cells , providing strong evidence that the liver trNK cells are a distinct lineage from the liver and splenic cNK cells , whereas both cNK cell populations are likely to be developmentally identical . Inasmuch as NK cells have been described in other solid organs , we next determined if CD49a could also mark trNK cells in other tissues . In addition to the liver , we found large percentages of CD49a+DX5− NK cells in the uterus and skin but not in other organs examined ( Figure 6A , B ) , suggesting that CD49a may be a marker of trNK cells in these organs . The CD49a+ NK cells in the uterus and skin also generally lacked expression of DX5 ( Figure 6B ) and displayed constitutive expression of CD69 , much like the liver trNK cells ( Figure 6C ) . To determine if the CD49a+DX5– cells were resident to the uterus and skin , we studied parabiotic mice as we had done previously to help establish the tissue residency of CD49a+ DX5− NK cells in the liver ( Peng et al . , 2013 ) . At 2 weeks post surgery , we assessed both the trNK and cNK cells in the uterus and skin in each parabiont by flow cytometry by gating on the appropriate CD45 . 1 or CD45 . 2 congenic marker in order to differentiate host-derived cells from circulating cells from the other parabiont . In both the uterus and skin , host CD49a+ DX5− NK cells were primarily found in the indicated host tissue ( Figure 6D , F ) , similar to the CD49a+ DX5− trNK cells in the liver ( Peng et al . , 2013 ) . By contrast , CD45 allotype-disparate mice reached nearly complete chimerism in the spleen ( Figure 6E ) ( Wright et al . , 2001; Peng et al . , 2013 ) . Moreover , the CD49a−DX5+ NK cells of both allotypes were found in both parabionts , indicating that they were circulating NK cells , akin to the CD49a−DX5+ cNK cells in the liver ( Figure 6D , F ) . Thus , the uterus and skin contain both trNK ( CD49a+ DX5− ) and circulating cNK ( CD49a−DX5+ ) cNK cells , with the trNK cells appearing to dominate the uterine NK cell population . 10 . 7554/eLife . 01659 . 010Figure 6 . Tissue-resident NK cells in other organs . ( A ) CD49a+DX5− NK cells are present in liver , skin , and uterus . Various organs were isolated from WT C57BL/6NCr mice , stained , and flow cytometry performed . The stacked bar graph represents the percentage of live CD3−CD19−NK1 . 1+ cells that either express CD49a or DX5 in the indicated organs . ( B ) CD49a expression on skin and uterine trNK cells . The dot plot displays cells that were isolated from the uterus and skin of WT C57BL/6NCr mice . The dot plots were gated on live CD3−CD19−NK1 . 1+ cells and the percentage of cells expressing CD49a and DX5 are shown . ( C ) CD49a+DX5– NK cells in liver , skin , and uterus express higher levels of CD69 . Spleen , liver , uterus , and skin were isolated from WT C57BL/6NCr mice , stained , and flow cytometry performed . The histograms were gated on live CD3−CD19−NK1 . 1+ cells and display the expression level of CD69 on CD49a+DX5− trNK cells and CD49a−DX5+ cNK cells in the spleen , and indicated tissues . ( D and F ) CD49a+DX5– NK cells in uterus and skin are tissue-resident as revealed by parabiotic mice . The uterus ( left two panels ) and skin ( right two panels ) were isolated from day 14 parabiosed mice . ( E ) Chimerism in the spleen . The spleen was analyzed for the degree of chimerism by analyzing the percentage of gated live CD45 . 1 and CD45 . 2 cells in each parabiont on day 14-post parabiosis surgery . These data correspond to data shown in Figure 6D , F . ( F ) Each parabiont was analyzed for its host and migratory cells using the congenic markers CD45 . 1 and CD45 . 2 . The cells were further gated on live CD3−CD19−NK1 . 1+ cells and the percentages of cells expressing CD49a and DX5 are shown in the representative dot plots ( D ) . Stacked bar graphs ( F ) show the percentages of cells expressing CD49a and DX5 , the gated populations in ( D ) . The experiment was performed two independent times with four parabiosed animals in each experiment ( total of 8 pairs ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01659 . 010 Although uterine NK cells have not been studied previously in the context of trNK cells , they are known to be IL-15-dependent ( Ashkar et al . , 2003 ) and thus IL-15Rα-dependent . Since little is known about NK cells in the skin , we analyzed them further . The skin trNK cells do not express Eomes and are absent in IL-15Rα-deficient mice ( Figure 7A , B ) . Furthermore the trNK cells in the uterus and skin produce more IL-2 than cNK cells in the liver and spleen ( Figure 7C ) . These features are similar to liver trNK cells . 10 . 7554/eLife . 01659 . 011Figure 7 . trNK cells in the liver resemble trNK cells in the skin and not the uterus , and all develop independent of a thymus . ( A ) Eomes is poorly expressed in skin trNK cells . Skin was isolated from WT C57BL/6NCr mice , stained , and flow cytometry performed . The histogram was gated on live CD3−CD19−NK1 . 1+ cells and displays the expression level of Eomes on CD49a+DX5− skin trNK cells and CD49a−DX5+ cNK cells in the skin and spleen . ( B ) Skin trNK cells are absent in IL-15Rα-deficient mice . Skin was isolated from WT mice and Il15ra−/− mice , stained , and flow cytometry performed . The bar graph displays the number of events captured in the CD3−CD19−NK1 . 1+ gate . ( C ) CD49a+DX5− trNK cells of the liver and skin produce IL-2 . Spleen , liver , uterus and skin were isolated from WT C57BL/6NCr mice stimulated with PMA/ionomycin for 4 hr , stained , and IL-2 production was analyzed by intracellular staining and flow cytometry . To obtain the percentage of IL-2+ cells in each population , the graphs were derived from gated live CD3−CD19−NK1 . 1+ cells and represent the percentage of IL2+ cells among the CD49a−DX5+ cNK cells in the liver and spleen and the CD49a+DX5− liver , uterus and skin trNK cells . ( D ) Tissue-resident CD49a+DX5− cells in the liver , uterus and skin do not express CD127 . Thymus , liver , uterus and skin were isolated from WT C57BL/6NCr mice , stained , and flow cytometry was performed . The histograms are from gated live CD3−CD19−NK1 . 1+ cells and display the expression levels of CD127 on CD49a+DX5− liver , uterus and skin trNK cells and CD49a−DX5+ cNK cells in the liver , uterus and skin and NK1 . 1+ thymic NK cells . ( E ) Tissue-resident CD49a+DX5− cells in the liver , uterus and skin develop independent of a thymus . The uterus and skin were isolated from WT and Foxn1−/− mice , stained , and flow cytometry was performed . Bar graphs display the total number of CD3−CD19−NK1 . 1+ cells that express CD49a and DX5 in the uterus and skin of WT and Foxn1−/− mice . Experiments were performed three independent times . DOI: http://dx . doi . org/10 . 7554/eLife . 01659 . 011 We evaluated whether the trNK cells in the uterus and skin are related to the CD127+ thymic NK by two independent ways . First , we found no detectable CD127 expression on uterine and skin trNK cells when compared to the CD127 expression on the thymic NK cells ( Figure 7D ) . Second , we found uterine and skin trNK cells are present in Foxn1−/− ( athymic ) mice ( Figure 7E ) , indicating that they develop independent of a thymus , unlike thymic NK cells . To determine if the trNK cells in the liver , uterus , and skin are related lineages , we examined the NFIL3- and Tbet-deficient mice . Strikingly , trNK cells were present in the liver , uterus and skin of NFIL3-deficient mice ( Figure 8A , B ) , indicating that these trNK cells represent a distinct lineage from cNK cells . Moreover , the circulating CD49a−DX5+ NK cells were absent in these tissues , indicating that these CD49a−DX5+ NK cells are related to the cNK cells in the liver and spleen . In Tbet-deficient mice , the trNK cells were absent in skin ( Figure 8A , B ) akin to liver trNK cells ( Figure 5D , E ) , implying a developmental link between the trNK cells in the liver and skin . Interestingly , however , the uterine trNK cells were still present in Tbet-deficient mice ( Figure 8A , B ) , strongly suggesting that uterine trNK cells represent yet another NK cell lineage . 10 . 7554/eLife . 01659 . 012Figure 8 . Liver , uterus , and skin trNK cells have different requirements for NFIL3 and Tbet ( Tbx21 ) . ( A ) The liver , uterus and skin were isolated from WT , Nfil3−/− , and Tbx21−/− mice , stained , and flow cytometry performed . Representative dot plots were gated on live CD3−CD19−NK1 . 1+ cells and display the expression level of CD49a and DX5 in the liver ( top panels ) and the uterus ( middle panels ) and skin ( bottom panels ) . Percentages indicate the gated populations . Bar graphs ( B ) display the total number of CD3−CD19−NK1 . 1+ cells that express CD49a and DX5 in the liver , uterus and skin of WT , Nfil3−/− , and Tbx21−/− mice . Experiments were performed five independent times . DOI: http://dx . doi . org/10 . 7554/eLife . 01659 . 012 Many immune cell lineages migrate throughout the body via the circulatory system . However , emerging data indicate that a number of different immune cell types appear to be tissue-resident and rarely recirculate . Here we demonstrate that NK cells are also comprised of circulating and several tissue-resident cell types in the liver , skin and uterus . Our comprehensive transcriptome and FACS analyses on liver trNK cells and liver and splenic cNK cells suggested that they may be distinct lineages of NK cells . Indeed , circulating liver and splenic cNK cells were absent in NFIL3-deficient mice , further demonstrating their lineage relationship to each other . On the other hand , NFIL3-deficient mice still possessed trNK cells in the liver , skin , and uterus . Moreover , Tbet-deficient mice lacked trNK cells in the liver and skin but uterine trNK cells and cNK cells in the liver and spleen were largely intact . Finally , thymic NK cells represent a distinct NK cell lineage because they are absent in nude and GATA-3-deficient mice which generally possess cNK and trNK cells . Taken together , these data suggest that there are at least four lineages of NK cells: cNK cells circulating in spleen , blood and other organs , thymic , and two trNK cell lineages—liver ( and skin ) , and uterine . The trNK cells can be distinguished from cNK cells in several different ways . Most importantly , trNK cells do not require the putative NK cell specification factor , NFIL3 ( Di Santo , 2009; Gascoyne et al . , 2009; Kamizono et al . , 2009; Kashiwada et al . , 2010 ) , indicating that they belong to a different cell lineage . In the liver , skin , and uterus , they differentially express CD49a , though it is possible that trNK cells in other organs preferentially express another marker . Concomitantly , they lack DX5 expression which is generally expressed as a late maturation marker on most cNK cells ( Kim et al . , 2002; Peng et al . , 2013 ) . Interestingly , trNK cells from unmanipulated mice display markers associated with cNK cell activation , such as CD69 , in contrast to cNK cells which do not express CD69 until activated ( Karlhofer and Yokoyama , 1991; Wang et al . , 2000 ) . The trNK cells also express a different repertoire of Ly49 receptors , suggesting that they may be tolerant to self by using mechanisms that do not strictly employ MHC class I-dependent licensing by Ly49 receptors , as has been shown for cNK cells in the spleen ( Kim et al . , 2005; Elliott and Yokoyama , 2011 ) . The trNK cells efficiently produce other cytokines , particularly TNFα and GM-CSF which may contribute to inflammatory conditions in a manner distinct from cNK cells which predominantly produce IFNγ . TNFα production occurs following stimulation by target cells and also occurs in trNK cells producing IFNγ . Finally , trNK cells utilize a separate set of transcription factors which may endow trNK cells with other unique functions , in addition to lineage commitment . As such , due to their tissue localization and these distinctive features , trNK cells most likely play important roles in innate defense in ways that should be distinguishable from cNK cells . Future investigations to re-examine the contribution of NK cells in different organs with a focus on trNK cells should reveal these roles . Among the various NK cell lineages , the liver and skin trNK cells appear to be highly related . They are phenotypically similar and both are absent in Tbet-deficient mice . Although it remains to be formally shown that a common precursor can give rise to liver and skin trNK cells but not other NK cell lineages , these data indicate that they are more related to each other than to other NK cell lineages , including cNK cells . uNK cells appear distinct from the liver and skin trNK cells because they are Tbet-independent , even though all are NFIL3-independent . Perhaps this is due to concomitant expression of Eomes at high levels in uNK cells ( Tayade et al . , 2005 ) . Alternatively , uterine trNK cells may depend on other transcription factors unique for this NK cell lineage . The liver , skin , and uterine trNK cell lineages can be distinguished from thymic NK cells in several ways . Both trNK ( as shown here ) and cNK cells do not express high levels of CD127 that is characteristic of thymic NK cells ( Vosshenrich et al . , 2006; Yadi et al . , 2008 ) . Moreover , thymic NK cells are absent in nude mice and in mice lacking GATA-3 which possess cNK cells , as recapitulated here . By contrast , we showed liver trNK cells are present in nude and GATA-3 deficient mice , suggesting that skin trNK cells will show these features , although the skin and uterine trNK cells need to be further studied in these mice . The trNK cells in the uterus in virgin mice as studied here will also need to be studied with respect to other subsets of uNK cells which have been described previously , including endometrial vs decidual NK cells , and NK1 . 1– DX5– uterine NK cells ( Yadi et al . , 2008; Mallidi et al . , 2009 ) . Inasmuch as liver , skin , and uterine trNK cells can be distinguished from cNK cells , it is important to note that they do share some important similarities , in addition to surface expression of NK1 . 1 and NKp46 but not CD3ε . As shown here and previously published , cNK , uterine and liver trNK cells can respond to YAC-1 targets by degranulation and killing ( Kiessling et al . , 1975; Linnemeyer and Pollack , 1991; Peng et al . , 2013 ) , though this needs to be further detailed for the skin trNK cells described here . Moreover , cNK cells require IL-15 or IL-15Rα ( Lodolce et al . , 1998; Kennedy et al . , 2000 ) and here we showed that IL-15Rα-deficient mice have no cNK or trNK cells , such that no NK1 . 1+ CD3− cells were identifiable in spleen , liver , and skin of the knockout mice . Similarly , IL-15-deficient mice lack NK1 . 1+ CD3− cells in the uterus ( Ashkar et al . , 2003; Barber and Pollard , 2003 ) , indicating that trNK cells in the uterus as well as liver and skin are IL-15Rα-dependent . While more detailed analysis of the progenitors for NK cells will be required to establish the precise lineage relationship of trNK and cNK cells to each other , current data , particularly cytotoxic potential and IL-15-dependence , support their intimate relationship , despite differences in transcription factor requirements . Recent studies indicate that cNK cells are related to innate lymphoid cells ( ILCs ) ( Spits and Cupedo , 2012; Spits et al . , 2013 ) . While this area of research is fluid with potentially more cells to be identified and additional characteristics to be discovered , it now appears that ILCs can be separated into three groups , based on shared characteristics ( Spits et al . , 2013 ) . cNK cells are now classified as belonging to the group 1 ILCs , due to their shared production of IFNγ . cNK cells can be distinguished from other group 1 ILCs by their cytotoxic capacities , dependence on IL-15 , and general lack of IL7Rα expression , features shared between cNK and liver , skin , and uterine trNK cells . Recently , a new IFNγ-producing ILC1 cell was identified which demonstrated dependence on NFIL3 , Tbx21 , and IL-15 but not IL-15Rα ( Fuchs et al . , 2013 ) , unlike the trNK cells described here . Finally , while the relationship of thymic NK cells expressing CD127 to ILCs which also express CD127 may need further refinement , current knowledge suggest that the liver ( skin ) and uterine trNK cells are more related to cNK cells than ILCs . Future studies , however , may continue to blur the lines between ILCs and NK cells . Already , it has been noted that markers previously thought to be exclusively expressed on NK cells , such as NKp46 , are expressed on ILCs that now can be distinguished from classical NK cells ( Walzer et al . , 2007; Cella et al . , 2009; Spits et al . , 2013 ) . A further confounding issue is the apparent plasticity of ILCs with potential interconversion between ILC groups ( Cella et al . , 2010; Vonarbourg et al . , 2010 ) , indicating the increasing complexities of definitively identifying ILC types and potentially distinguishing ILCs from other immune cells . Thus , it is possible that trNK cells may be more closely related to certain ILCs than currently appreciated . Tissue-resident NK cells are more closely related to ILCs in that both the trNK cells and the ILCs have extended their place of residency outside the traditional secondary lymphoid organs . Originally most ILCs were considered in the context of gut or lymphoid tissues ( Spits and Cupedo , 2012 ) but recently they have been identified in other nonlymphoid tissues such as the skin and the uterus . ILC2s in the skin have been reported to regulate cutaneous inflammation but they are unaffected by IL15-deficiency ( Roediger et al . , 2013 ) , unlike trNK cells as shown here . In the uterus , immature NK cells have been described as well as a phenotypically related but distinct population of cells that express the transcription factor RORC and produce IL-22 , suggesting that they may be ILC3s rather than NK cells ( Crellin et al . , 2010; Male et al . , 2010 ) . Additional studies will be needed to more finely localize the anatomic location of trNK cells in these tissues and the transcription factors required for their development and maintenance , work that will be aided by development of new tools to distinguish trNK cells from cNK cells and ILCs . The trNK cells should also be considered in the context of other non-lymphoid tissue-resident immune cells , including NKT cells , tissue-resident memory T cells , γδ T cells , B1 B cells , and tissue macrophages , among others . Of these , NKT cells are of special interest because they share several characteristics with trNK cells in the liver . NKT cells were first identified because they express markers associated with NK cells , such as NK1 . 1 , but are not cNK cells because they express rearranged TCR genes ( Bendelac et al . , 2007 ) . While they are now known to recognize glycolipids presented by CD1 molecules , a large number of NKT cells also reside in the liver , though they can be found in other tissues . In the liver , NKT cells crawl within the sinusoidal space by patrolling sinusoidal endothelial cells ( Geissmann et al . , 2005 ) . This is reminiscent of early electron microscopy studies of the rat liver which identified ‘pit cells’ , now recognized as NK cells ( Wisse et al . , 1976; Bouwens et al . , 1987 ) . Pit cells are found in the sinusoidal space usually adjacent to sinusoidal endothelial cells . Although liver trNK cells have not been formally examined vis-à-vis pit cells , it seems likely that they are related , if not equivalent . The trNK cell populations also are potentially related to tissue-resident memory T cells which had been previously activated by antigen-stimulation and differentiated into effector cells that then reside in non-lymphoid tissues ( Sheridan and Lefrancois , 2011; Gebhardt et al . , 2013 ) . Our studies indicate that trNK cells appear to have an activated phenotype with expression of CD69 and other activation markers as well as other phenotypic changes . However , it is not yet clear if trNK cells differentiate from a circulating precursor that then differentiates into a cell taking up tissue residency , akin to tissue-resident memory T cells ( Masopust et al . , 2010 ) . Our studies do not rule out this possibility although the transcription factor requirements make it unlikely that a putative circulating precursor is an NK cell that had previously differentiated into a cNK cell . An alternative possible origin of trNK cells is a progenitor that seeds the peripheral organs during embryonic or fetal life . In particular , due to early hematopoiesis occurring in the fetal liver , liver trNK cell precursors may seed the liver during embryonic life . In this regard , it is interesting to note that Ly49E was first noted to be almost exclusively expressed on fetal NK cells ( Toomey et al . , 1998 ) . However , recent findings ( Filtjens et al . , 2013 ) and those reported here indicate that liver trNK cells in adult mice also selectively express Ly49E , raising the possibility that Ly49E+ NK cells in the fetus reflect trNK cells already present in the embryonic liver , at a time when cNK cells are poorly developed ( Bukowski et al . , 1985; Dorfman and Raulet , 1998 ) . As such , since the liver is the major site for fetal hematopoiesis , Ly49E+ NK cells may not reflect fetal NK cells per se but instead the possibility that precursor of trNK cells seeds the liver during early life . Consistent with this possibility , we recently showed that precursors of liver trNK cells are present in the adult liver ( Peng et al . , 2013 ) . Moreover , adult BM hematopoietic stem cells do not fully reconstitute the liver trNK cell population in irradiated mice . Thus , precursors of trNK cells , like γδT cells and certain tissue macrophages ( Havran and Allison , 1990; Schulz et al . , 2012 ) , may seed tissues early in life , a topic that will require further evaluation . In conclusion , we provide molecular evidence that trNK cells are distinct from circulating cNK cells in their cytokine profiles , expression of NK cell receptors and most importantly their transcription factor requirements . Their tissue residency feature suggests that they have tissue-specific homeostatic functions . Our studies should prompt a re-examination of the myriad roles of NK cells in many immune responses which could be due to trNK cells , rather than cNK cells . Finally , our studies may also be applicable to emerging data on other hematopoietic cells that circulate throughout the body in that there may be related but distinct cell types that are tissue-resident . All mice were housed in a pathogen-free facility and procedures performed in accordance with the animal protocol approved by the Washington University School of Medicine ( WUSM ) Animal Studies Committee or the NIAID Animal Care and Use Committee . WT C57BL/6NCr and B6-LY5 . 2/Cr mice were purchased from the National Cancer Institute ( Frederick , MD ) . Foxn1−/− ( nude ) , Tbx21−/− ( Tbet-deficient ) , Il15ra−/− and Rag1−/− were purchased from Jackson Laboratory ( Bar Harbor , ME ) . The Il15ra−/− were backcrossed to C57BL/6 background by speed congenics ( Wakeland et al . , 1997 ) . Nfil3−/− and Gata3fl/fl mice were previously described ( Zhu et al . , 2004; Kashiwada et al . , 2010 ) . Gata3fl/fl mice were bred onto C57BL/6 background for at least nine generations before they were bred with Vav-Cre transgenic mice ( JAX line 8610 ) on a C57BL/6 background to generate Gata3fl/fl-Vav-Cre line . Single cell suspensions were made from spleen , bone marrow , peripheral and mesenteric lymph nodes , spinal cord , omentum , peritoneal cavity , pancreas , lung , liver , skin , and uterus for flow cytometry analysis . Briefly , all organs isolated were mechanically dissociated and put through a 100 µm or 70 µm cell strainer . The liver single cell suspension was resuspended in 40% Percoll and centrifuged for 20 min at 2000 rpm at room temperature . Bone marrow was isolated from femurs . Spinal cord , omentum , pancreas , lung , uterus and skin were all incubated in 167 µg/ml liberase TL ( Roche Applied Science , Indianapolis , IN ) and 100 µg/ml of DNase ( Roche Applied Science ) for 1 hr at 37°C in shaking incubator . Skin lymphocytes were isolated from ears; each ear was split into dorsal and ventral sides prior to liberase treatment . Single cell suspensions of spleen and liver were cultured in 10% heat inactivated fetal calf serum and RPMI 1640 and stimulated with phorbol 12-myristate 13-acetate ( PMA , 200 ng/ml ) and Ionomycin ( 5 µg/ml ) for 4 hr in the presence of Brefeldin A ( BD Pharmingen , San Jose , CA ) to analyze cytokine production . For the degranulation assay , a suspension of 105 liver cells was plated with YAC-1 target cells at an effector:target ( E:T ) ratio of 1:1 in 96-well V-bottom plates . Anti-CD107a antibody , Brefeldin A , and monensin ( eBioscience , San Diego , CA ) were added to each well before incubation . Plates were incubated for 6 hr at 37°C , after which surface staining followed by intracellular staining for TNFα was performed for analysis by flow cytometry . All cells were stained with Fixable Viability Dye , prior to blocking Fc receptors with 2 . 4G2 , per manufacturer’s instructions ( eBioscience ) . The cells were stained with cell surface antibodies . To measure cytokine production the cells were fixed and permeabilized using the BD cytofix/cytoperm reagents and the forkhead box P3 ( Foxp3 ) fix/perm reagents were used to assay transcription factors ( eBioscience ) . Events were acquired on the Canto ( BD Biosciences ) using the FACSDiva software ( BD Biosciences ) and analyzed with FlowJo software ( Treestar ) . The following antibodies were purchased from eBioscience: anti-CD3 ( clone 145-2C11 ) , anti-CD19 ( eBio1D3 ) , anti-NK1 . 1 ( PK136 ) , anti-CD49b ( DX5 ) , anti-CD27 ( LG . 7F9 ) , anti-CD11b ( M1/70 ) , anti-CD69 ( H1 . 2F3 ) , anti-CD62L ( MEL-14 ) , anti-CD44 ( IM7 ) , anti-CD160 ( eBioCNX46-3 ) , anti-IFNγ ( XMG1 . 2 ) , anti-TNFα ( MP6-XT22 ) , anti-GM-CSF ( MP1-22E9 ) , anti-CD107a ( eBio1D4B ) anti-Eomesodermin ( Dan11mag ) , anti-CD45 . 1 ( A20 ) , anti-CD45 . 2 ( 104 ) , anti-TRAIL ( N2B2 ) , anti-CD127 ( A7R34 ) anti-Ly49A ( A1 ) , anti-Ly49 E/F ( CM4 ) , anti-Ly49D ( eBio4E5 ) , anti-Ly49G2 ( eBio4D11 ) , anti-Ly49H ( 3D10 ) , anti-Ly49I ( YLI-90 ) , anti-NKG2A ( 16a11 ) , and anti-NKG2D ( CX5 ) . Antibodies purchased from BD were anti-CD49a ( Ha31/8 ) , anti-Ly49F ( HBF-719 ) and anti-Ly49C/I ( 5E6 ) . Anti-Tbet ( 4B10 ) was purchased from Biolegend ( San Diego , CA ) and used as previously described ( Sojka and Fowell , 2011 ) . Spleen , liver and bone marrow cells were isolated from Rag1−/− mice and sorted for CD3−CD19−NK1 . 1+CD49a+DX5− or CD3−CD19−NK1 . 1+CD49a−DX5+ which were lysed . mRNA was extracted from cell lysates using oligo-dT beads ( Invitrogen , Grand Island , NY ) . For cDNA synthesis , we used custom oligo-dT primer with a barcode and adapter-linker sequence ( CCTACACGACGCTCTTCCGATCT—XXXXXXXX-T15 ) . After first strand synthesis samples were pooled together based on Actb qPCR values and RNA-DNA hybrids were degraded using consecutive acid-alkali treatment . Then , second sequencing linker ( AGATCGGAAGAGCACACGTCTG ) was ligated using T4 ligase ( New England Biolabs , Ipswich , MA ) and after SPRI clean-up , mixture was PCR enriched 14 cycles and SPRI purified to yield final strand specific 3′end RNA-seq libraries . Data were sequenced on HiSeq 2500 instrument ( Illumina , San Diego , CA ) using 50 bp × 25 bp pair-end sequencing . Second mate was used for sample demultiplexing , at which point individual single-end fastqs were aligned to mm9 genome using TopHat with following options -G mm9 . mrna . 10 . 31 . gtf–prefilter-multihits–segment-length 20 –max-multihits 15 . Gene expression was obtained using ESAT software tool ( http://garberlab . umassmed . edu/software/esat/ ) focused on analysis of 3′end targeted RNA-Seq . The following parameters were used: task ‘score3P’ , normalizedOutput , window 1000 , maxExtension 3000 , maxIntoGene 2000 , stranded , collapseIsoforms . Parabiosis surgery was performed as previously described ( Peng et al . , 2013 ) . Briefly , matching longitudinal skin incisions were made on the flanks of C57BL/6NCr ( Ly5 . 2 ) and B6-LY5 . 1/Cr female mice . Their elbows and knees were then joined with dissolvable sutures and the incisions were closed with wound clips . Postoperative care included administration of buprenex compound for pain management , 5% dextrose and 0 . 9% sodium chloride . Nutritional gel packs were provided in each cage and antibiotics ( Sulfatrim ) in the drinking water for the duration of the experiment .

Our immune system has white blood cells that migrate throughout the body in search of invading microbes or diseased and damaged cells . When these events are encountered , the white blood cells move into the affected tissue and launch an immune response to eliminate the threat . Natural killer cells are white blood cells that kill cells that are infected with viruses or are cancerous . Most of what is known about conventional natural killer cells is derived from studying the spleen , which filters the blood and contains many immune cells . Natural killer cells also circulate around the body or are found within other tissues , and it was thought that both types of cells were either the same , or that one type could develop into the other . However , the thymus—an organ that is another source of white blood cells—contains a sub-population of natural killer cells that are distinct from the conventional splenic natural killer cells . Furthermore , recent work revealed the existence of two types of natural killer cells within the liver: some of these cells were similar to the conventional splenic natural killer cells that circulate throughout the body , while others appeared to be ‘tissue-resident’ natural killer cells that were poised to deliver an immune response . Now Sojka et al . show that the tissue-resident natural killer cells found in the liver are a distinct lineage of cells . These cells mature independently from the conventional natural killer cells found in the spleen , and the natural killer cells found in the thymus . Moreover , the skin contains tissue-resident natural killer cells similar to those in the liver; whilst natural killer cells that had previously been discovered in the uterus were shown to contain a fourth distinct tissue-resident lineage . The work of Sojka et al . will encourage a full re-evaluation of the roles played by natural killer cells to determine which populations of these cells are responsible for implementing immune responses . Furthermore , a more thorough understanding of how tissue-resident natural killer cells function to eliminate diseased or damaged cells , such as cancerous cells , could also contribute to future efforts to develop new anti-cancer treatments .

2014

Tissue-resident natural killer (NK) cells are cell lineages distinct from thymic and conventional splenic NK cells

Autocatalytic activation of epidermal growth factor receptor ( EGFR ) coupled to dephosphorylating activity of protein tyrosine phosphatases ( PTPs ) ensures robust yet diverse responses to extracellular stimuli . The inevitable tradeoff of this plasticity is spontaneous receptor activation and spurious signaling . We show that a ligand-mediated switch in EGFR trafficking enables suppression of spontaneous activation while maintaining EGFR’s capacity to transduce extracellular signals . Autocatalytic phosphorylation of tyrosine 845 on unliganded EGFR monomers is suppressed by vesicular recycling through perinuclear areas with high PTP1B activity . Ligand-binding results in phosphorylation of the c-Cbl docking tyrosine and ubiquitination of the receptor . This secondary signal relies on EGF-induced EGFR self-association and switches suppressive recycling to directional trafficking . The re-routing regulates EGFR signaling response by the transit-time to late endosomes where it is switched-off by high PTP1B activity . This ubiquitin-mediated switch in EGFR trafficking is a uniquely suited solution to suppress spontaneous activation while maintaining responsiveness to EGF . Signaling by the epidermal growth factor receptor ( EGFR ) converts diverse external stimuli into specific cellular responses . EGFR signaling is implicated in embryonic development , tissue homeostasis and wound healing ( Yu et al . , 2010; Sibilia et al . , 2007 ) , while EGFR overexpression and hyper-activation through genetic alterations have been linked to malignant transformation ( Rowinsky , 2004 ) . Mutation-induced loss of autoinhibitory interactions or enhanced receptor expression levels , frequently elevate the basal phosphorylation and activation status of EGFR ( Arteaga and Engelman , 2014 ) . The autoinhibitory interactions include the tethered conformation of the extracellular domain ( Ferguson et al . , 2003 ) as well as receptor-membrane interactions and the local intrinsic disorder of the αC helix in the N-lobe of the kinase domain ( Arkhipov et al . , 2013; Endres et al . , 2013; Shan et al . , 2012 ) . This creates an energy barrier for EGFR self-association that is surpassed upon ligand stimulation . Ligand binding leads to receptor dimerization ( Yarden and Schlessinger , 1987; Cochet et al . , 1988 ) and the formation of an asymmetric dimer of the intracellular kinase domains ( Zhang et al . , 2006 ) . This triggers phosphorylation in trans of regulatory and signaling tyrosine residues in the intracellular part of the receptor , and a subsequent recruitment of adaptor proteins that contain Src homology 2 domains ( SH2 ) or phosphotyrosine-binding domains ( PTB ) such as c-Cbl ( Y1045 ) or Grb2 ( Y1068 and Y1086 ) ( Ushiro and Cohen , 1980; Moran et al . , 1990; Levkowitz et al . , 1998; Waterman et al . , 2002; Lemmon and Schlessinger , 2010 ) . Despite these EGFR structure intrinsic safeguards , the receptor can still attain an active conformation in the absence of ligand due to thermal fluctuations ( Lemmon and Schlessinger , 2010 ) , necessitating only low protein tyrosine phosphatase ( PTP ) activity to suppress phosphorylation due to this ‘leaky’ kinase activity . However , phosphorylation of the conserved regulatory tyrosine Y845 in the activation loop of the EGFR kinase domain leads to an acceleration of its phosphorylation , potentiating EGFR kinase activity in an autocatalytic fashion ( Shan et al . , 2012 ) . Such an autocatalytic activation system that is coupled to PTP activity , by for example a double negative feedback , offers robustness against biological noise and conveys external stimuli into threshold-activated responses ( Grecco et al . , 2011 ) . Autocatalysis can lead to amplified self-activation of the receptor in the absence of a cognate ligand ( Verveer , 2000; Endres et al . , 2013 ) , requiring high PTP activity at the plasma membrane ( PM ) to suppress . Such PTPs that act on EGFR with high catalytic efficiency ( ~2 orders of magnitude higher than EGFR ) are PTP1B and TCPTP ( Zhang et al . , 1993; Romsicki et al . , 2003; Fan et al . , 2004 ) . These PTPs are , however , segregated from the PM by association with the cytoplasmic membrane leaflet of the endoplasmic reticulum ( ER ) , and therefore mostly dephosphorylate endocytosed ligand-bound EGFR . After ligand binding , endocytosed receptor-ligand complexes contained in clathrin-coated vesicles ( CCVs ) enter early endosomes ( EEs ) by fusion ( Vieira et al . , 1996; Bucci et al . , 1992; Goh and Sorkin , 2013 ) , further maturing in the perinuclear area to late endosomes ( LEs ) and eventually fusing to lysosomes where receptors are degraded ( Rink et al . , 2005; Ceresa , 2006; Vanlandingham and Ceresa , 2009; Levkowitz et al . , 1999 ) . Although EGFR vesicular trafficking was extensively studied after ligand stimulation , little is known about the role of vesicular trafficking in suppressing spontaneous EGFR activation as well as regulating its signaling response . To assess how vesicular membrane dynamics modulates spontaneous and ligand-induced phosphorylation of EGFR , we studied three phosphorylation sites on EGFR with distinct functionality: 1 ) Y845—a regulatory autocatalytic tyrosine whose phosphorylation increases EGFR activity ( Shan et al . , 2012 ) , 2 ) Y1045—a site that upon phosphorylation affects vesicular trafficking of EGFR by binding the E3 ligase c-Cbl that ubiquitinates the receptor ( Levkowitz et al . , 1998 ) , and 3 ) Y1068—a site that upon phosphorylation binds the adapter Grb2 via its SH2 domain to propagate signals in the cell ( Okutani et al . , 1994 ) . We show that spontaneously and ligand-induced EGFR activation gives rise to distinct molecular states that are recognized and processed differently by the endocytic machinery . While unliganded monomeric receptors continuously recycle to the PM to suppress autocatalytic activation , ligand-bound dimeric receptors are ubiquitinated by the E3-ligase c-Cbl that commits them to unidirectional vesicular trafficking toward lysosomes . This route through perinuclear endosomes enables their efficient dephosphorylation by high local PTP activity to produce a finite signaling response to growth factors . We demonstrate by a compartmental model that ligand-responsive EGFR signaling can only occur in conjunction with suppression of spontaneous autocatalytic EGFR activation if a ligand-induced switch in EGFR trafficking changes its cyclic interaction with spatially partitioned PTPs to a sustained one . To investigate how EGFR auto-phosphorylation depends on its cell surface density , we quantified the relative phosphorylation ( pY/EGFR ) of three tyrosine residues with distinct regulatory functionality of autocatalysis , signaling , and trafficking in single COS-7 cells as a function of EGFR-mCitrine expression level . The variance in ectopic expression of EGFR-mCitrine was thereby exploited to sample a broad range of receptor expression levels . The EGFR-mCitrine expression level in single cells was determined relative to endogenous EGFR by an independent immunofluorescence experiment where the level of endogenously expressed EGFR was quantified from the abscissa-intercept of a linear fit to an EGFR-mCitrine intensity versus anti-EGFR antibody intensity plot ( Figure 1—figure supplement 1A , B ) . This analysis showed that most COS-7 cells expressed EGFR-mCitrine at similar level as endogenous EGFR , whereas the expression varied by a factor 6 ( from ~0 . 5–3 ) ( Figure 1A , X-axis ) . The phosphorylation level of the autocatalytic Y845 site as well as the Cbl ( Y1045 ) and Grb2 ( Y1068 ) docking sites was determined for single cells as well as for cell populations by quantifying immunofluorescence staining ( Figure 1A , B , Figure 1—figure supplement 1C ) and Western blots ( Figure 1C–E , Figure 1—figure supplement 2 , 3 ) , respectively . To determine the levels of spontaneous EGFR phosphorylation by Western blot analysis , its expression was controlled by the amount of transfected EGFR-mCitrine encoding cDNA and quantified by the EGFR-mCitrine band intensity relative to the band of the maximally applied cDNA amount ( 3 μg ) ( Figure 1C–E , Figure 1—figure supplement 2 , 3 ) . Single-cell immunofluorescence analyses showed that a wide range of auto-phosphorylation levels occur for similar EGFR expression in individual cells ( Figure 1A ) . Despite this cell-to-cell variance in response , clear trends in the average phosphorylation of the individual sites as function of EGFR-mCitrine expression were observed . The average amount of spontaneously phosphorylated Y845 on EGFR increased ~ sixfold with the density of EGFR in cells ( Figure 1A , left panel ) . Upon EGF-stimulation , the relative phosphorylation of Y845 increased , exhibiting a clear dependency on the EGFR expression levels as well ( elevated by ~ threefold for a ~ sixfold increase in EGFR ) , corroborating the autocatalytic function of Y845 ( Shan et al . , 2012 ) . The c-Cbl docking site Y1045 exhibited a very different response profile: spontaneous phosphorylation was weakly increased with EGFR expression ( ~ twofold increase ) , whereas EGF stimulation clearly resulted in phosphorylation levels that were independent of EGFR expression ( Figure 1A , middle panel ) . On the other hand , the average amount of spontaneously phosphorylated Y1068 strongly increased ( by ~ eightfold ) with the EGFR density in cells ( Figure 1A , right panel ) . However , in contrast to Y845 , EGF stimulation increased its average phosphorylation independent of EGFR expression ( Endres et al . , 2013 ) . 10 . 7554/eLife . 12223 . 003Figure 1 . Spontaneous tyrosine phosphorylation in response to EGFR expression . ( A ) Scatter plots show the relative phosphorylation ( mean fluorescence intensity of pYN-antibody/EGFR-mCitrine ) of three tyrosine residues ( N= 845 , 1045 , or 1068 ) versus EGFR-mCitrine expression in multiples of mean endogenous EGFR levels ( see Figure 1—figure supplement 1A , B ) . Points ( black: pre-; red: 2-min post-stimulation with 100 ng/ml EGF ) represent single COS-7 cells ( see Figure 1—figure supplement 1C ) and thick lines indicate mean values of binned data . ( B , C ) Spontaneous phosphorylation of Y845 , Y1045 and Y1068 . ( B ) Immunofluorescence data showing mean fluorescence intensity of pYN-antibody over EGFR-mCitrine ( pre ) normalized to mean fluorescence intensity of pYN-antibody over EGFR-mCitrine , 2-min post-stimulation with 100 ng/ml EGF ( 2’ EGF ) . YN ( number of cells pre , 2’ EGF ) : Y845 ( 119 , 108 ) ; Y1045 ( 145 , 109 ) ; Y1068 ( 127 , 87 ) . ( C ) Western blot analysis of COS-7 lysates probed for anti-GFP antibody and either anti-pY845 ( n=5 blots ) , anti-pY1045 ( n=3 ) or anti-pY1068 ( n=3 ) show pYN-antibody band over anti-GFP band ( pre ) normalized to the same fraction 5-min post-stimulation with 100 ng/ml EGF ( 5’ EGF ) . Data correspond to the highest levels of expressed EGFR-mCitrine ( 3 µg cDNA , Figure 1—figure supplement 2 ) . ( D , E ) Autonomous phosphorylation of Y845 , Y1045 , and Y1068 upon pervanadate ( PV ) treatment . Western blot analysis of COS-7 lysates probed for anti-pY845 , anti-pY1045 , anti-pY1068 and anti-GFP antibody ( all n=3 blots ) . ( D ) pYN-antibody band over anti-GFP band 5 min post-addition of PV ( 0 . 33 mM ) normalized to relative phosphorylation 5-min post-stimulation with 100 ng/ml EGF ( Figure 1—figure supplement 3 ) . ( E ) pYN-antibody band over anti-GFP band 5-min post-addition of PV ( 0 . 33 mM ) normalized to relative phosphorylation before stimulation ( pre ) . ( F ) Dependence of EGFR phosphorylation on its expression level . Western blot analysis of COS-7 lysates transfected with increasing amounts ( 0 . 5 , 1 . 5 , and 3 µg cDNA ) of EGFR-mCitrine . Blots were probed for anti-pY845 ( n=5 blots ) , anti-pY1045 ( n=3 ) , anti-pY1068 ( n=3 ) , and anti-GFP for EGFR-mCitrine ( Figure 1—figure supplement 2 ) . The ordinate displays the fraction of EGFR-band over tubulin band in each lane relative to 3 µg cDNA and bar diagram shows relative EGFR phosphorylation of the three tyrosine residues ( see 'Materials and methods' ) . ( G , H ) Dependence of Akt and Erk activation on EGFR expression level . Western blots of COS-7 lysates transfected with increasing amounts ( 0 . 5 , 1 . 5 , and 3 µg cDNA ) of EGFR-mCitrine and probed for phosphorylated Ser473 on Akt , total Akt , phosphorylated Thr202/Tyr204 on Erk1/2 , and total Erk1/2 levels . Data represents 'ratio of fractions' of either Akt phosphorylation ( E ) or Erk phosphorylation as a function of EGFR-mCitrine expression level as described in ( D ) ( n=3 blots , Figure 1—figure supplement 4 ) . All error bars correspond to standard error of the mean . EGFR , epidermal growth factor receptor . DOI: http://dx . doi . org/10 . 7554/eLife . 12223 . 00310 . 7554/eLife . 12223 . 004Figure 1—figure supplement 1 . Relating ectopic EGFR-mCitrine expression with endogenous EGFR expression levels and dependency of autonomous EGFR activation on EGFR expression levels . ( A ) Left panel: Exemplary COS-7 cell expressing EGFR-mCitrine ( fluorescence intensity displayed in upper image ) that was immunostained against EGFR ( antibody fluorescence displayed in middle image ) . The background-corrected ratio of intensities ( lower image ) clearly shows areas of mismatch , where antibody staining failed to reach all available EGFR-mCitrine ( red-yellow-white , ratio > 2 ) . We therefore quantified the average anti-EGFR fluorescence intensity in the image masked with the thresholded ratio-image ( 0 . 15 < ratio mCitrine / anti-EGFR < 1 . 5 ) . This average anti-EGFR fluorescence intensity is plotted against the average whole-cell EGFR-mCitrine intensity ( right panel , upper graph ) for transfection with 0 . 5 µg ( red ) and 3 µg ( green ) cDNA . To be able to relate mCitrine fluorescence intensity with endogenous EGFR expression levels , this data was fitted with a linear model ( straight lines with dashed lines for 95% confidence intervals ) , yielding a slope of m=1 . 15 ± 0 . 13 and an abscissa intercept of b=190 ± 25 ( all data ) . This intercept corresponds to the anti-EGFR fluorescence without ectopic EGFR-mCitrine expression and m/b is the factor that translates mCitrine intensity to n-fold ectopic expression over endogenous levels . This quantification is almost irrespective of the amount of cDNA used for EGFR-mCitrine transfection . Compared to transfection with 0 . 5 µg cDNA ( red ) , the frequency of a higher average EGFR-mCitrine intensity occurring ( right panel , lower graphs ) clearly increases when 3 µg ( green ) of cDNA were used , however , not by a factor of 6 that could be expected . For one , a larger amount of cDNA not only increases the amount of cDNA per cell , but also the number of transfected cells ( transfection efficiency ) . Secondly , selection of cells under the microscope suffers from a bias toward transfected cells . However , we used all cells in a field of view to also incorporate weakly expressing cells in our results . ( B ) For single-cell immunofluorescence experiments ( Figure 1A ) that were stained with antibodies against EGFR phosphorylation , we matched the cumulative EGFR-mCitrine fluorescence intensity histograms ( red curves in upper three plots , for the phosphotyrosine-specific antibody indicated ) by normalizing with a scalar to that of anti-EGFR immunostaining ( green curve , same as green curve in A ) . Thereby , the median values of the mCitrine intensity distributions ( blue vertical lines ) coincide ( lowest plot ) and the n-fold ectopic EGFR expression over mean endogenous level can be calculated for each series of experiments . ( C ) Expression level dependence of EGFR-mCitrine phosphorylation at Y845 ( left panel ) , Y1045 ( middle panel ) , and Y1068 ( right panel ) . Immunostaining with specific pYN-antibodies ( second column ) in COS-7 cells ectopically expressing EGFR-mCitrine ( first column ) and corresponding green/magenta overlay ( third column ) before or after 2-min stimulation with 100 ng/ml EGF ( Figure 1A ) . All scale bars: 10 μm . EGFR , epidermal growth factor receptor . DOI: http://dx . doi . org/10 . 7554/eLife . 12223 . 00410 . 7554/eLife . 12223 . 005Figure 1—figure supplement 2 . Dependency of autonomous EGFR activation on EGFR expression levels . ( A-–C ) COS–7 cells were transfected with increasing amounts of EGFR-mCitrine cDNA ( 0 . 5 , 1 . 5 , and 3 µg ) . Lysates were collected pre or 5-min post-stimulation with 100 ng/ml EGF and used in Western blot analysis and phosphorylation of Y845 ( A ) , Y1045 ( B ) , and Y1068 ( C ) was determined with specific pYN-antibodies . EGFR-mCitrine was detected with an anti-GFP antibody . Anti-tubulin was used as a loading control . ( D ) Scatter plots shows pYN-antibody band over anti-GFP band ( pre ) normalized to the same fraction 5-min post-stimulation with 100 ng/ml EGF for individual bands versus the relative EGFR expression level normalized to the tubulin band relative to 3 µg cDNA ( see 'Materials and methods' ) . EGFR , epidermal growth factor receptor . DOI: http://dx . doi . org/10 . 7554/eLife . 12223 . 00510 . 7554/eLife . 12223 . 006Figure 1—figure supplement 3 . EGFR phosphorylation induced by PV-mediated PTP inhibition and EGF stimulation . ( A- ( –C ) COS–7 cells expressing EGFR-mCitrine were stimulated with EGF ( 100 ng/ml ) or treated with PV ( 0 . 33 mM ) for 5 min . The collected lysates were used in Western blot analysis , and the phosphorylation of Y845 ( A ) , Y1045 ( B ) , and Y1068 ( C ) was determined by using specific pYN-antibodies . EGFR-mCitrine was detected with an anti-GFP antibody . Anti-tubulin was used as a loading control . EGFR , epidermal growth factor receptor; PV , pervanadate; PTP , protein tyrosine phosphatase . DOI: http://dx . doi . org/10 . 7554/eLife . 12223 . 00610 . 7554/eLife . 12223 . 007Figure 1—figure supplement 4 . Dependency of downstream EGFR signaling on EGFR expression levels . COS–7 cells were transfected with increasing amounts of EGFR-mCitrine cDNA ( 0 . 5 , 1 . 5 , and 3 µg ) . Lysates were collected pre or 5-min post-stimulation with 100 ng/ml EGF and used in Western blot analysis . Blots were probed with phospho-specific antibodies for Akt and Erk and total Akt and total Erk antibodies . Anti-tubulin was used as a loading control . EGFR , epidermal growth factor receptor . DOI: http://dx . doi . org/10 . 7554/eLife . 12223 . 007 To compare the level of spontaneous phosphorylation for the three sites , the fraction of auto-phosphorylated tyrosine relative to EGFR expression was normalized to the corresponding fraction after EGF stimulation ( the maximally attainable phosphorylation at a given EGFR expression level ) . This 'ratio of fractions' ( pre/EGF ) gives a comparative measure of the auto-phosphorylation level of tyrosine residues as determined by immunofluorescence and Western blots . This analysis revealed that the extent of phosphorylation differed for the individual tyrosine residues , with a high correspondence between Western blot and immunofluorescence analysis ( Figure 1B , C ) . In both experiments , the c-Cbl docking Y1045 site was significantly less phosphorylated as compared to the autocatalytic ( Y845 ) and the signaling site ( Y1068 ) . In order to disentangle the contributions of EGFR kinase from PTP activity in generating this spontaneous phosphorylation profile , phosphorylation of EGFR was induced by pervanadate ( PV ) -mediated inhibition of PTPs ( Huyer et al . , 1997; Haj , 2002 ) . In this way , the phosphorylation profile reflects the catalytic efficiency of EGFR kinase for the three tyrosine residues . The autocatalytic Y845 site was approximately 2 . 5 times more phosphorylated at 5’ PV stimulation as compared to 5’ EGF stimulation ( Figure 1D ) . This rapid phosphorylation is consistent with a self-amplifying , autocatalytic phosphorylation . Both the Grb2 ( Y1068 ) and c-Cbl ( Y1045 ) docking sites were phosphorylated to a ~2 . 5 times lower level than Y845 ( Figure 1D ) . The ratio of PV-induced over the spontaneous phosphorylation profiles ( Figure 1C ) allowed an estimation of the relative contribution of PTP activity in suppressing phosphorylation of the three tyrosines ( Figure 1E ) . This analysis shows that the autocatalytic Y845 site requires higher PTP activity to be maintained in check as compared to the other two tyrosines . Moreover , the similar PTP activity that acts on Y1045 and Y1068 shows that the lower phosphorylation of Y1045 as compared to Y1068 is mostly due to differences in the catalytic efficiency of the EGFR kinase for these sites . All three tyrosines exhibited a switch-like spontaneous phosphorylation response as function of EGFR expression ( Figure 1F , Figure 1—figure supplement 2–4 ) , indicating that the autocatalytic activation of EGFR is suppressed by PTP activity only up to a threshold of EGFR kinase activity . The phosphorylation of Y1045 was the least responsive to EGFR expression , which is consistent with a low catalytic efficiency of EGFR kinase for this site in the absence of ligand . Spontaneously phosphorylated EGFR largely remained at the PM ( Figure 1—figure supplement 1C ) , which could result from the lack of c-Cbl binding to this inefficiently phosphorylated site . This steady state distribution is distinct from the EGF-induced internalized EGFR , which manifests in down-stream signaling where an increase in spontaneously phosphorylated EGFR clearly activates Akt but not Erk ( Figure 1G , H , Figure 1—figure supplement 4 ) . These results show that spontaneously and ligand-activated receptors behave as distinct signaling entities . We investigated whether the distinct phosphorylation pattern and signaling behavior of spontaneously activated EGFR arises from a different self-association state as compared to ligand-bound receptors . To this end , fluorescence anisotropy microscopy was used to detect self-association of EGFR by homo-FRET ( Squire et al . , 2004; Varma and Mayor , 1998 ) . For this purpose , an EGFR construct where mCitrine is inserted via a linker between amino acids Q958 and G959 ( EGFR-QG-mCitrine ) was generated that displayed a similar phosphorylation response to EGF as C-terminally tagged EGFR ( EGFR-mCitrine ) in COS-7 cells ( Figure 2—figure supplement 1A ) . EGFR-QG-mCitrine exhibited homo-FRET upon dimerization as apparent from the EGF-induced decrease in anisotropy for all EGFR expression levels ( Figure 2A ) . In order to assess EGFR phosphorylation in the same experiment , mCherry tagged PTB ( PTB-mCherry ) ( Batzer et al . , 1995; Offterdinger et al . , 2004 ) that targets the phosphorylated signaling tyrosines 1148/1086 was co-expressed and its co-localization with EGFR-QG-mCitrine was determined at the cell periphery . At high EGFR-QG-mCitrine expression levels , co-localization with PTB-mCherry could not be enhanced by EGF stimulation , demonstrating the auto-phosphorylation of EGFR in the absence of ligand ( Figure 2B ) . This spontaneously phosphorylated receptor remained mostly monomeric as shown by its high anisotropy ( Figure 2A ) , even after PV-treatment that fully phosphorylates the receptor ( Figure 2C , 1D , Figure 2—figure supplement 1B ) . However , spontaneously activated EGFR at high expression levels still dimerizes upon ligand binding as apparent from the drop in anisotropy upon addition of EGF . These data are therefore consistent with mostly monomeric , spontaneously activated EGFR that is distinct from ligand-activated , self-associated EGFR . 10 . 7554/eLife . 12223 . 008Figure 2 . Autonomously and ligand-activated EGFR are different molecular states . ( A ) EGFR-QG-mCitrine anisotropy in COS-7 cells . Representative fluorescence images of EGFR-QG-mCitrine ( first column ) , PTB-mCherry ( second column ) , and anisotropy of EGFR-QG-mCitrine ( third column ) upon EGF stimulation for the indicated time . Scale bar: 10 μm . Graph shows the anisotropy of EGFR-QG-mCitrine versus its binned mean fluorescence intensity ( F of EGFR ) per pixel ( black: pre- , red: 5-min post-stimulation with 100 ng/ml EGF ) . ( B ) Corresponding phosphorylation of EGFR-QG-mCitrine . Graph shows the recruitment ( R , see 'Materials and methods' ) of PTB to EGFR versus the binned mean fluorescence intensity ( F of EGFR ) of EGFR-QG-mCitrine per pixel . ( C ) Anisotropy of spontaneously activated EGFR-QG-mCitrine upon PV-mediated PTP inhibition . Graph shows the anisotropy of EGFR-QG-mCitrine versus its binned mean fluorescence intensity ( F of EGFR ) per pixel ( black: pre- , red: 5-min post-treatment with PV 0 . 33 mM; see Figure 2—figure supplement 1B ) . All error bars correspond to ( standard error of the mean . EGFR , epidermal growth factor; PTB , phosphotyrosine-binding domain; PV , pervanadate . DOI: http://dx . doi . org/10 . 7554/eLife . 12223 . 00810 . 7554/eLife . 12223 . 009Figure 2—figure supplement 1 . EGF-induced EGFR-QG-mCitrine phosphorylation and anisotropy upon PTP inhibition by PV . ( A ) COS–7 cells expressing EGFR-mCitrine or EGFR-QG-mCitrine were stimulated with EGF ( 100 ng/ml ) . The collected lysates were used in western blot analysis and probed with the generic phospho tyrosine antibody PY72 and anti-GFP ( EGFR-mCitrine ) . Anti-tubulin was used as a loading control . The bar diagram shows the relative phosphorylation of EGFR-mCitrine and EGFR-QG-mCitrine for the indicated time points ( mean ± SEM , n=3 ) . ( B ) EGFR-QG-mCitrine anisotropy upon PV treatment . Representative fluorescence images of COS–7 cells co-expressing EGFR-QG-mCitrine and PTB-mCherry and the corresponding anisotropy images of EGFR-QG-mCitrine . Upper row: Pre-stimulation , lower row: 5-min post-stimulation with 0 . 33 mM PV . Scale bar: 10 μm . EGFR , epidermal growth factor receptor; PTB , phosphotyrosine-binding domain PV , pervanadate; SEM , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 12223 . 009 To assess whether EGFR is maintained at the PM by vesicular recycling in COS-7 cells , we investigated if receptors partition in the Rab11-positive pericentriolar recycling endosome ( RE ) ( Ullrich , 1996 ) . Immunofluorescence staining showed co-localization of endogenous EGFR with Rab11 in unstimulated cells , which was also observed upon ectopic expression of EGFR-mCitrine and BFP-Rab11a ( Figure 3A ) . However , ectopic expression of BFP-Rab11a enhanced the biogenesis of the RE ( Ullrich , 1996 ) , which shifted the distribution of EGFR to this compartment . Blocking protein synthesis with cycloheximide did not affect the apparent co-localization of Rab11 with EGFR showing that it did not originate from newly synthesized EGFR transiting through the Golgi via the secretory pathway ( Figure 3 ) . 10 . 7554/eLife . 12223 . 010Figure 3 . EGFR continuously recycles through the RE . ( A ) Co-localization of EGFR and Rab11 . Left panel , first row: Immunostaining of endogenous EGFR ( left ) , Rab11 ( middle ) and corresponding green/magenta overlay ( right ) in fixed COS-7 cells . Left panel , second row: Fluorescence images of EGFR-mCitrine ( left ) , BFP-Rab11a ( middle ) , and corresponding green/magenta overlay ( right ) in living COS-7 cells . Right panel: Quantification of co-localization between EGFR and Rab11 in A ( n=15 cells ) , B ( n=12 ) by Pearson’s correlation coefficient . ( B ) Fluorescence redistribution after photoactivation of EGFR-paGFP on the recycling endosome . First row: EGFR-paGFP fluorescence at the indicated time in seconds ( photoactivation area: black rectangle in the pre-activation image ) , second row: corresponding EGFR-mCherry fluorescence and image in third row: BFP-Rab11a fluorescence . Right graphs: Loss of EGFR-paGFP fluorescence at the RE normalized to EGFR-mCherry fluorescence ( left ) and corresponding gain of EGFR-paGFP fluorescence at the PM ( right ) , normalized to EGFR-mCherry fluorescence . Average trace ( black ± SEM . ; n=6 cells ) was fitted to an exponential function ( red ) to retrieve time constants ( τ ) of EGFR fluorescence intensity lossat the RE and its gain at the PM . ( C ) Fluorescence redistribution of TAMRA-labeled-EGFR in HEK293 cells . Representative fluorescence images of EGFR- ( 128TCOK ) -GFP ( first row ) , TAMRA-labeled EGFR- ( 128TCOK ) -GFP ( second row ) , and BFP-Rab11a ( third row , left panel ) or LAMP1-BFP ( third row , right panel ) for the indicated time ( min ) in the respective columns ( see 'Materials and methods' ) . ( D ) Co-localization of ligand-activated EGFR and Rab5 in COS-7 cells . Representative fluorescence images of EGFR-mCitrine ( first column ) , Rab5-mCherry ( second column ) , and green/magenta overlay of selected ROIs ( third column ) , pre- and 15-min post- stimulation with EGF ( 100 ng/ml ) . ( E ) Co-localization of spontaneously activated EGFR and Rab5 in COS-7 cells . Representative fluorescence images of EGFR-mCitrine ( first column ) , Rab5-mCherry ( second column ) , and green/magenta overlay of selected ROIs ( third column ) , pre- and 15-min post- treatment with 0 . 33 mM PV . ( F ) Quantification of co-localization between EGFR-mCitrine and Rab5-mCherry with Pearson’s correlation coefficient upon EGF stimulation ( 100 ng/ml , n=5 cells ) or PV treatment ( 0 . 33 mM , n=7 cells ) for the indicated time points . All scale bars: 10 μm . EGFR , epidermal growth factor receptor; PM , plasma membrane; PV , pervanadate; RE , recycling endosome; SEM , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 12223 . 01010 . 7554/eLife . 12223 . 011Figure 3—figure supplement 1 . Vesicular trafficking of autonomously and ligand-activated EGFR . ( A ) Co-localization of EGFR and Rab11 upon inhibition of protein synthesis . Representative fluorescence images of three individual cells co-expressing of EGFR-mCitrine ( first row ) and BFP-Rab11a ( second row ) , and corresponding green/magenta overlay ( third row ) after treatment with cyclohexamide ( 10 μg/ml ) for app . 20 hr . ( B ) Fluorescence recovery after photobleaching EGFR-mCitrine in COS–7 cells co-expressing BFP-Rab11a . Representative fluorescence images of EGFR-mCitrine and BFP-Rab11a for the indicated time points ( left panel ) . EGFR-mCitrine was photobleached in the indicated region at the RE ( red square ) . The normalized mean fluorescence intensity of the bleached area is plotted as a function of time ( red line: average curve , grey lines: individual recovery curves , n=16 cells , right panel ) . ( C ) EGFR trafficking from the PM to the RE . Fluorescence images of EGFR-TCOK ( 128 ) -GFP , TAMRA-EGFR , and BFP-Rab11a with white lines for the profile analysis . Line plots show the normalized fluorescence intensity of BFP-Rab11a ( blue ) , TAMRA-EGFR ( red ) , and EGFR-GFP for the upper ( middle panel ) and lower cell ( right panel ) . The highest intensity of each channel was normalized to 1 . ( D ) Fluorescence redistribution of TAMRA-labeled EGFR upon EGF stimulation . Representative fluorescence images of HEK293 cells ectopically expressing EGFR- ( 128TCOK ) -GFP and BFP-Rab11a ( left panel ) or LAMP1-BFP ( right panel ) for the indicated time points . EGFR- ( 128TCOK ) -GFP was selectively labeled on the plasma membrane using TAMRA and simultaneously stimulated with 100 ng/ml EG ( F ) . All scale bars: 10 μm . EGFR , epidermal growth factor receptor; PM , plasma membrane; RE , recycling endosome . DOI: http://dx . doi . org/10 . 7554/eLife . 12223 . 01110 . 7554/eLife . 12223 . 012Figure 3—figure supplement 2 . Vesicular trafficking of autonomously-activated EGFR . ( A ) Immunostaining of endogenous EGFR ( upper row ) , Rab5 ( middle row ) and green/magenta overlay of selected ROIs ( lower row ) in fixed COS–7 cells after stimulation with EGF ( 100 ng/ml ) or treatment with PV ( 0 . 33 mM ) for the indicated time . ( B ) Fluorescence redistribution after photoactivation of EGFR-paGFP . Representative fluorescence images of COS-7 cells co-expressing EGFR-paGFP and Rab5-mCherry for the indicated time points . EGFR-paGFP was photoactivated at the basal membrane in TIRF microscopy and its redistribution and co-localization with Rab5-mCherry was followed over time in TIRF and widefield microscopy . All scale bars: 10 μm . EGFR , epidermal growth factor receptor; PV , pervanadate . DOI: http://dx . doi . org/10 . 7554/eLife . 12223 . 012 Fluorescence loss after photoactivation ( FLAP ) and fluorescence recovery after photobleaching ( FRAP ) in the perinuclear area were performed to assess the extent of vesicular recycling of unliganded EGFR . Due to the shift in EGFR steady state distribution by ectopic Rab11a-BFP expression , most of the photo-activated or –bleached EGFR resided on the RE . The fluorescence loss after photoactivation of EGFR fused to photoactivatable GFP ( EGFR-paGFP ) ( Patterson , 2002 ) on the RE ( τ= 273 ± 15 s ) , and the concomitant gain at the PM ( τ= 167 ± 33 s ) shows that EGFR is recycled from the RE to the PM ( Figure 3B ) . The difference in τ stems from the difference in surface area of the RE in relation to the PM , which determines the rate of RE-dissociation and PM-association . The residence time of EGFR at the RE was estimated to be ~7 . 6min ( koff = 2 . 22 10–3 s-−1 ) based on fitting the FLAP data to a two-compartment model ( see 'Materials and methods' ) . From the fluorescence recovery after photobleaching EGFR-mCitrine on the RE ( Figure 3—figure supplement 1B , τ= 540 ± 47 s ) , it was apparent that EGFR also enters the RE . The FLAP and FRAP data thus show that there is constitutive vesicular trafficking of EGFR to and from the RE . To ascertain that EGFR-mCitrine on the RE originates from the PM , a pulse/chase experiment was performed by conjugating tetrazine-TAMRA in a fast bioorthogonal chemical reaction to EGFR-GFP that contained a trans-cyclo-octene derivative of lysine ( TCOK ) ( Lang et al . , 2012 ) . This unnatural amino acid was site-specifically and co-translationally incorporated by genetic code expansion in HEK293 cells into the extracellular part of EGFR at position 128 ( Lang et al . , 2012 ) . TAMRA fluorescence of EGFR- ( 128TCOK ) -GFP that was pulse labeled at the PM ( Figure 3C , t=0 min ) appeared 20 min later in Rab11a positive perinuclear endosomes , but not in LAMP1-BFP marked lysosomes ( Figure 3C , Figure 3—figure supplement 1C ) . EGF stimulation , however , redirected a fraction of TAMRA-labeled EGFR- ( 128TCOK ) -GFP to lysosomes ( Figure 3—figure supplement 1D ) . This indicates that vesicular recycling between the PM and the RE maintains a steady state with the majority of EGFR at the PM , whereas EGF stimulation redirects self-associated EGFR towards lysosomes thereby depleting it from the PM . We then investigated where this EGF-induced switch in vesicular trafficking occurs . Co-localization of EGFR-mCitrine with Rab5-mCherry in quiescent cells showed that unliganded EGFR is localized in EEs ( Figure 3D , E , 0 min ) . Upon stimulation with EGF , the co-localized Rab5-mCherry and EGFR-mCitrine signals increased , confirming that liganded receptors transit via EEs to LE/Lysosomes and further increase the size and number of Rab5-mCherry-positive EEs ( Barbieri , 2000 ) ( Figure 3D , F ) . However , 15’ PV-mediated PTP inhibition only marginally affected the detectable amount and size of Rab5-mCherry positive EEs with phosphorylated EGFR ( Figure 3E , F , Figure 3—figure supplement 2A ) . Photoactivation of EGFR-paGFP at the basal PM by TIRF microscopy , and the concurrent appearance of its fluorescence in Rab5-mCherry fluorescent vesicles confirmed that PM-residing EGFR recycles by transit through EEs ( Figure 3—figure supplement 2B ) . These experiments show that recycling of monomeric EGFR via the EE compartment maintains its high steady-state distribution at the PM irrespective of phosphorylation state , whereas the different vesicular trafficking fates of unliganded versus liganded EGFR are decided at the level of the EE . We next investigated the coupling between EGFR auto-phosphorylation and vesicular recycling by decreasing/increasing the biogenesis of the pericentriolar RE . Decreasing RE biogenesis by knockdown of Rab11a with two independent siRNAs increased the phosphorylation of the autocatalytic Y845 ~ threefold and to a lesser extent the Y1045 ( ~1 . 5 fold ) , with no effect on EGFR-expression level ( Figure 4—figure supplement 1B ) . However , no change in autophosphorylation for Y1068 was observed ( Figure 4A , Figure 4—figure supplement 1A–E ) . Increasing RE biogenesis by overexpression of BFP-Rab11a gave the complementary result: phosphorylation of Y845 and Y1045 was lowered , whereas no change in phosphorylation was observed for Y1068 ( Figure 4B , Figure 4—figure supplement 2A ) . The difference in Y845 phosphorylation between the PM and RE indicated that spontaneously activated EGFR is dephosphorylated on the RE ( Figure 4C ) . These experiments are consistent with suppression of spontaneous EGFR phosphorylation at the perinuclear RE during recycling of the receptor . 10 . 7554/eLife . 12223 . 013Figure 4 . Suppression of spontaneous autocatalytic EGFR activation by vesicular recycling . ( A ) Effect of Rab11a knockdown on EGFR-mCitrine phosphorylation in COS-7 cells . The bar diagram shows the relative phosphorylation ( pY/EGFR ) of EGFR-mCitrine on Y845 , Y1045 , and Y1068 upon siRNA-mediated Rab11a knockdown normalized to pY/EGFR for cells transfected with non-targeting siRNA . For non-targeting siRNA: Y845 ( n=69 cells ) , Y1045 ( n=82 ) , Y1068 ( n=75 ) . For Rab11a siRNA: Y845 ( n=75 cells ) , Y1045 ( n=74 ) , Y1068 ( n=83 ) ( see Figure 4—figure supplement 1A , B ) . ( B ) Effect of BFP-Rab11a expression on EGFR-mCitrine phosphorylation in COS-7 cells . The bar diagram shows the relative phosphorylation ( pY/EGFR ) of Y845 , Y1045 and Y1068 on EGFR-mCitrine upon ectopic expression of BFP-Rab11a normalized to pY/EGFR in the presence of empty pcDNA 3 . 1 . For BFP-Rab11a ectopic expression: Y845 ( n=52 cells ) , Y1045 ( n=34 ) , Y1068 ( n=38 ) and for pcDNA 3 . 1: Y845 ( n=125 cells ) , Y1045 ( n=110 ) , Y1068 ( n=138 ) ( see Figure 4—figure supplement 2A ) . ( C ) Spatial distribution of spontaneously phosphorylated Y845 in COS-7 cells . Left panel: Representative fluorescence images of EGFR-mCitrine ( top left ) , immunostaining of pY845 ( top right ) , BFP-Rab11a ( bottom left ) , and a ratio image of pY845/EGFR-mCitrine ( bottom right ) . Right panel: Graph shows the difference in Y845 auto-phosphorylation between the PM and RE ( Δ pY845 over EGFR , see materials and methods ) as a function of overall EGFR phosphorylation level in individual cells ( pY845/EGFR ) . ( D ) Effect of ectopically expressed BFP-PTP1B on EGFR-mCitrine phosphorylation in COS-7 cells . The bar diagram shows the relative phosphorylation ( pY/EGFR ) of EGFR-mCitrine on Y845 , Y1045 , and Y1068 upon ectopic expression of BFP-PTP1B normalized to pY/EGFR upon transfection with empty pcDNA 3 . 1 . For BFP-PTP1B ectopic expression: Y845 ( n=69 cells ) , Y1045 ( n=59 ) , Y1068 ( n=61 ) and for pcDNA 3 . 1: Y845 ( n=54 cells ) , Y1045 ( n=57 ) , Y1068 ( n=51 ) ( see Figure 4—figure supplement 2B ) . ( E ) Spatial distribution of the interacting fraction ( α , third column ) of EGFR-mCitrine ( first column ) with mCherry-PTP1B D/A ( second column ) as detected by FLIM-FRET , with ( upper row ) or without ( lower row ) ectopic expression of BFP-Rab11a ( 4th column ) . Graph shows average α in regions of high EGFR-mCitrine intensity as a function of mean fluorescence ( F of EGFR ) with ( blue , n=28 cells ) or without ectopic expression of BFP-Rab11a ( black , n=20 ) . Lower row: percentage of EGFR/PTP1B D/A interactions in the vicinity of the RE was retrieved from the overlap ( III ) between areas with high α values ( I ) and areas with high intensity of BFP-Rab11a fluorescence ( II ) ( see 'Materials and methods' and Figure 4—figure supplement 3 ) All scale bars: 10 μm . All error bars correspond to standard error of the mean . EGFR , epidermal growth factor receptor . DOI: http://dx . doi . org/10 . 7554/eLife . 12223 . 01310 . 7554/eLife . 12223 . 014Figure 4—figure supplement 1 . Suppression of autonomous EGFR activation by recycling through perinuclear membranes . ( A ) Effect of Rab11a knockdown on EGFR phosphorylation at Y845 , Y1045 , and Y1068 . COS-7 cells expressing EGFR-mCitrine were immunostained with specific pYN-antibodies to detect phosphorylation of Y845 , Y1045 , and Y1068 upon transfection with non-targeting siRNA ( Ctrl , upper columns ) or upon siRNA-mediated Rab11a knockdown ( Rab11a , lower columns ) . ( B ) Left: Western blot of COS-7 lysates 72 hr post-transfection with non-targeting siRNA ( Ctrl , first column ) or Rab11a siRNA ( siRab11a , second column ) . Blots were probed for Rab11a and GAPD ( H ) Middle: Bar diagram represents the ratio of Rab11a to GAPDH ( n=5; mean ± SEM ) . Right: EGFR-mCitrine expression upon Rab11a knockdown . Western blot of COS-7 cells expressing EGFR-mCitrine after transfection with non-targeting siRNA ( Ctrl , first column ) or Rab11a siRNA ( siRab11a , second column ) . Blot was probed for Rab11a and GAPDH and an anti-GFP antibody to detect EGFR-mCitrine . ( C ) Effect of Rab11a/b knockdown on EGFR phosphorylation at Y845 , Y1045 and Y1068 . COS-7 cells expressing EGFR-mCitrine were immunostained with specific pYN-antibodies to detect phosphorylation of Y845 , Y1045 , and Y1068 upon transfection with non-targeting siRNA ( Ctrl , upper columns ) or upon siRNA-mediated Rab11a/b knockdown ( Rab11a/b , lower columns ) . ( D ) Western blot of COS-7 lysates 72 hr post-transfection with non-targeting siRNA ( Ctrl , first column ) or Rab11a siRNA ( siRab11a , second column ) or Rab11a/b siRNA ( siRab11a/b , third column ) . Blots were probed for , Rab11a/b and GAPDH . ( E ) Effect of Rab11a/b knockdown on EGFR-mCitrine phosphorylation . The bar diagram shows the relative phosphorylation pY/EGFR of EGFR-mCitrine on Y845 , Y1045 , and Y1068 upon transfection with non-targeting siRNA or upon siRNA-mediated Rab11a/b knockdown . For non-targeting siRNA:Y845 ( n=95 cells ) , Y1045 ( n=97 ) Y1068 ( n= 84 ) . For Rab11a/b siRNA: Y845 ( n=147 ) , Y1045 ( n=96 ) , Y1068 ( n=88 ) . All scale bars: 10 μm . All error bars correspond to standard error of the mean . EGFR , epidermal growth factor receptor . DOI: http://dx . doi . org/10 . 7554/eLife . 12223 . 01410 . 7554/eLife . 12223 . 015Figure 4—figure supplement 2 . Suppression of autonomous EGFR activation by recycling through perinuclear membranes . ( A ) Effect of ectopic BFP-Rab11a expression on EGFR phosphorylation at Y845 , Y1045 , and Y1068 . COS-7 cells expressing EGFR-mCitrine were immunostained with specific pYN-antibodies to detect phosphorylation of Y845 , Y1045 , and Y1068 in the presence of empty pcDNA 3 . 1 or BFP-Rab11a . ( B ) Effect of ectopic BFP-PTP1B expression on EGFR phosphorylation at Y845 , Y1045 and Y1068 . COS-7 cells expressing EGFR-mCitrine were immunostained with specific pYN-antibodies to detect phosphorylation of Y845 , Y1045 , and Y1068 in the presence of empty pcDNA 3 . 1 ( upper columns ) or presence of BFP-PTP1B ( lower columns ) . All scale bars: 10 μm . EGFR , epidermal growth factor receptor . DOI: http://dx . doi . org/10 . 7554/eLife . 12223 . 01510 . 7554/eLife . 12223 . 016Figure 4—figure supplement 3 . Interaction of EGFR and PTP1B D181A in perinuclear areas . Representative fluorescence images of living COS–7 cells co-expressing EGFR-mCitrine , PTP1B D181A-mCherry , BFP-Rab11a and α map showing the fraction of interaction between EGFR-mCitrine and PTP1B D181A-mCherry . Scale bars: 10 μm . EGFR , epidermal growth factor receptor . DOI: http://dx . doi . org/10 . 7554/eLife . 12223 . 016 The ER-associated , catalytically efficient PTP1B dephosphorylates internalized EGF-EGFR complexes mostly on perinuclear membranes ( Haj , 2002; Yudushkin et al . , 2007 ) . The lowered phosphorylation of the three tyrosine sites upon ectopic expression of BFP-PTP1B showed that spontaneously phosphorylated EGFR is also dephosphorylated by PTP1B ( Figure 4D , Figure 4—figure supplement 2B ) . To determine where this dephosphorylation occurs , we examined the interaction between EGFR-mCitrine and an mCherry-tagged substrate-trapping mutant of PTP1B ( mCherry-PTP1B-D181A ) by FLIM-FRET ( Haj , 2002; Sabet et al . , 2015 ) . The spatially resolved fraction ( α ) of EGFR-mCitrine interacting with mCherry-PTP1B-D181A was computed using global analysis of FLIM data ( Verveer , 2000; Grecco et al . , 2009 ) . An interacting fraction of EGFR-mCitrine and mCherry-PTP1B-D181A was only observed in the perinuclear region ( Figure 4E ) . The mean spatial coincidence between this EGFR-mCitrine and mCherry-PTP1B-D181A interaction and BFP-Rab11a ( 63 ± 29% ; n= 20 cells; see 'Materials and methods' ) is consistent with EGFR being dephosphorylated by PTP1B on perinuclear membranes in the vicinity of the RE . BFP-Rab11a expression led to an elevated interacting fraction ( α ) at the RE due to the enhanced partitioning of EGFR in this compartment ( Figure 4E , Figure 4—figure supplement 3 ) . This shows that the steady state distribution of EGFR as determined by the anterograde and retrograde rates of recycling dictates the effectiveness of spontaneous phosphorylation suppression . Constitutive recycling of EGFR suppresses autonomous activation , but maintains a steady state distribution of EGFR at the PM irrespective of its activation state . In contrast , EGF stimulation leads to concomitant receptor depletion from the PM and its vesicular trafficking towards perinuclear endosomes ( Figure 5A ) . At early times after stimulation , a transient translocation of EGFR-mCitrine to the PM and a parallel loss at the RE can be observed ( Figure 5B–D ) . This transient shift in the steady-state distribution of EGFR likely occurs because recycling from the RE to the PM continues after EGF stimulation , whereas endocytosis of ligand-bound receptors at the PM is delayed . 10 . 7554/eLife . 12223 . 017Figure 5 . Ubiquitin-mediated switch in ligand-activated EGFR trafficking . ( A ) Time-lapse of EGFR and Rab11a co-localization after EGF stimulation of COS-7 cells with Alexa647-labeled EGF ( 5 ng/ml ) . Representative fluorescence images of EGFR-mCitrine ( first row ) , c-Cbl-mCherry ( second row ) , and BFP-Rab11a merge with Alexa647-labeled EGF ( third row ) at the indicated time in minutes . ( B ) Change in spatial distribution of EGFR-mCitrine upon EGF stimulation calculated as the difference between each time point and the one acquired prior to it , indicates areas with increased ( cyan ) or decreased ( magenta ) fluorescence . ( C ) Plot shows the normalized average fraction ± SEM of EGFR-mCitrine at the PM ( PM-EGFR over total EGFR ) in the presence ( red , n=10 cells ) or absence ( black , n=6 cells ) of c-Cbl-mCherry over time ( see 'Materials and methods' ) . ( D ) Fraction of EGFR-mCitrine fluorescence at the RE with ( n=10 cells ) and without ( n=6 cells ) ectopic expression of c-Cbl-mCherry upon EGF stimulation for the indicated time . ( E ) Differential ubiquitination of ligand- and autonomously-activated EGFR . COS–7 lysates immunoprecipitated with anti-EGFR ( left panel ) or blotted for total proteins ( middle panel ) . IP was probed with anti-HA ( HA-ubiquitin ) and anti-GFP ( EGFR-mCitrine ) and total lysates were probed with anti-GFP ( EGFR-mCitrine ) , anti-Cbl ( c-Cbl-mCherry ) , anti-tubulin ( Tubulin ) , and anti-pY1045 . Right panel: Quantification shows relative EGFR ubiquitination ( Ub over EGFR ) in the upper graph and degradation ( EGFR over Tubulin ) in the lower graph upon EGF stimulation ( 100 ng/ml ) or PV treatment ( 0 . 33 mM ) for the indicated time in minutes ( see Figure 5—figure supplement 1A , B ) . ( F ) Fraction of EGFR-mCitrine fluorescence at the RE upon stimulation with 100 ng/ml EGF ( n=12 cells ) or 0 . 33 mM PV ( n=14 cells ) for the indicated time ( see Figure 5—figure supplement 2 ) . ( G ) Time-lapse of EGFR-Y1045F-mCitrine and BFP-Rab11a co-localization after EGF stimulation of COS-7 cells with Alexa647-labeled EGF ( 5 ng/ml ) . Representative fluorescence images of EGFR-Y1045F-mCitrine ( first row ) , c-Cbl-mCherry ( second row ) , and BFP-Rab11a merge with Alexa647-labeled EGF ( third row ) at the indicated time in minutes . ( H ) Fraction of EGFR-Y1045F-mCitrine ( n=9 cells ) or EGFR-Y1045/1068/1086F-mCitrine ( n=9 cells ) fluorescence at the RE upon EGF stimulation for the indicated time ( min ) in the presence of ectopically expressed c-Cbl-mCherry ( see Figure 5—figure supplement 3A ) . ( I ) Sustained phosphorylation of EGFR-Y1045F-mCitrine and EGFR-Y1045/1068/1086F-mCitrine as compared to EGFR-mCitrine upon EGF stimulation . COS-7 lysates were probed with generic phosphotyrosine ( PY72 ) and anti-GFP ( EGFR-mCitrine ) and bar diagram shows relative EGFR phosphorylation ( PY72/EGFR ) upon EGF stimulation ( 100 ng/ml ) for the indicated time ( see Figure 5—figure supplement 3B ) . All blots are n=3 ( mean ± SEM ) . All scale bars: 10 μm . EGFR , epidermal growth factor receptor; SEM , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 12223 . 01710 . 7554/eLife . 12223 . 018Figure 5—figure supplement 1 . EGF-induced vesicular trafficking of wild-type EGFR-mCitrine and differential ubiquitination of autonomously and ligand-activated EGFR . ( A ) Ubiquitination of EGFR . Western blot of COS–7 total cell lysates ( total ) ( WB ) and immunoprecipitated EGFR ( IP:EGFR ) ( middle panel ) showing time course of ubiquitination and phosphorylation of Y845 after incubation with EGF ( 100 ng/ml ) or PV ( 0 . 33 mM ) . HA-ubiquitin and HA-c-Cbl-mCherry were co-expressed . Total lysates were probed with anti-GFP ( EGFR-mCitrine ) , anti-Cbl ( c-Cbl-mCherry ) , anti-tubulin ( Tubulin ) , and anti-phospho-tyrosine 845 ( pY845 ) . IP was probed with anti-HA ( HA-ubiquitin ) and anti-GFP ( EGFR-mCitrine ) . ( B ) Quantification shows the relative phosphorylation ( mean ± SEM ) on Y845 ( pY845/EGFR ) upon EGF stimulation and PV treatment for the indicated time points . EGFR , epidermal growth factor receptor; PV , pervanadate; SEM , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 12223 . 01810 . 7554/eLife . 12223 . 019Figure 5—figure supplement 2 . Vesicular trafficking of autonomously versus ligand-activated EGFR to the RE . Representative fluorescence images of COS–7 cells co-expressing EGFR-mCitrine ( upper rows ) , BFP-Rab11a ( middle rows ) and c-Cbl-mCherry ( lower rows ) upon 0 . 33 mM PV treatment ( left panel ) or 100 ng/ml EGF stimulation ( right panel ) for the indicated time in minutes . Scale bars: 10 μm . EGFR , epidermal growth factor receptor; PV , pervanadate; RE , recycling endosome . DOI: http://dx . doi . org/10 . 7554/eLife . 12223 . 01910 . 7554/eLife . 12223 . 020Figure 5—figure supplement 3 . EGF-induced vesicular trafficking of EGFR-Y1045/Y1068F/Y1086F-mCitrine and c-Cbl-mediated degradation profiles of EGFR upon EGF stimulation . ( A ) Representative fluorescence images of COS–7 cells co-expressing the ubiquitination-impaired mutant EGFR-Y1045/1068/1086F-mCitrine , BFP-Rab11a , and c-Cbl-mCherry upon stimulation with EGF-Alexa647 ( 5 ng/ml ) . Merged images of EGF-Alexa647 ( green ) and BFP-Rab11a ( red ) are shown . Scale bar: 10 μm . ( B ) Lysates of COS–7 cells co-expressing EGFR-mCitrine or EGFR-Y1045F-mCitrine or EGFR-Y1045/1068/1086F-mCitrine and c-Cbl-mCherry and stimulated with EGF ( 100 ng/ml ) for the indicated time points were used in western blot analysis . Blots were probed with the generic phospho tyrosine antibody PY72 , anti-GFP ( EGFR-mCitrine ) and anti-c-Cbl . Anti-tubulin was used as a loading control . ( C ) Ratiometric quantification of total EGFR-mCitrine/tubulin for EGFR wild type and the ubiquitination-impaired EGFR mutants for the indicated time points ( mean ± SEM , n=3 blots ) . EGFR , epidermal growth factor receptor; SEM , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 12223 . 02010 . 7554/eLife . 12223 . 021Figure 5—figure supplement 4 . Interaction of EGFR and PTP1B D181A in perinuclear areas upon EGF stimulation . Representative fluorescence images of living COS–7 cells co-expressing EGFR-mCitrine , PTP1B D181A-mCherry , BFP-Rab11a and α map showing the fraction of interaction between EGFR-mCitrine and PTP1B D181A-mCherry upon EGF stimulation ( 100 ng/ml ) for the indicated time ( min ) . Scale bar: 10 μm . EGFR , epidermal growth factor receptor . DOI: http://dx . doi . org/10 . 7554/eLife . 12223 . 021 We investigated the cause of this ligand-induced switch in receptor trafficking and how this affects its phosphorylation response . EGF stimulation induces EGFR ubiquitination by the E3-ligase c-Cbl ( Levkowitz et al . , 1998; Waterman et al . , 2002 ) and we therefore asked whether c-Cbl-mediated ubiquitination is the key signal to differentiate trafficking of autonomously from ligand-activated EGFR . Ectopically expressed c-Cbl-mCherry was rapidly recruited to the EGF/EGFR complexes at the PM and subsequently observed on internalized EGFR-positive endosomes ( Figure 5A ) . This ectopic expression strongly enhanced the rate of internalization after EGF stimulus ( Figure 5C ) , which was concurrent with a diminished fraction of recycling EGF-EGFR ( Figure 5D ) . EGF-stimulation led to ubiquitination and degradation of liganded receptors ( Figure 5E , Figure 5—figure supplement 1A ) , whereas PV-mediated spontaneously activated EGFR exhibited much lower ubiquitination , was not degraded and did not alter its cyclic trafficking ( Figure 5E , F ) . This shows that c-Cbl-mediated ubiquitination of EGFR causes a switch from recycling to unidirectional receptor trafficking toward the perinuclear cytoplasm . An EGFR mutant that is impaired in direct c-Cbl binding ( EGFR-Y1045F-mCitrine ) ( Grøvdal et al . , 2004 ) further corroborated this conclusion by continuing to recycle via the RE after ligand stimulation , thereby preventing its degradation ( Figure 5G , H , Figure 5—figure supplement 3A–C ) . This constitutive recycling also resulted in an equally sustained phosphorylation in response to EGF , as compared to the adaptive phosphorylation response of wild type EGFR ( Figure 5I , Figure 5—figure supplement 3B ) . The same stabilizing features were additionally observed for an EGFR mutant ( EGFR-Y1045F/Y1068F/Y1086F-mCitrine ) that is impaired in both direct and indirect c-Cbl binding via Grb2 ( Figure 5H , I , Figure 5—figure supplement 3A–C ) . This shows that the EGF-mediated ubiquitin-switch in vesicular traffic that imposes a finite EGFR signal duration is mediated by c-Cbl binding to phosphorylated Y1045 . However , PV-mediated spontaneous EGFR activation caused a similar or higher phosphorylation of Y1045 without c-Cbl recruitment ( Figure 5—figure supplement 2 ) , leading to inefficient receptor ubiquitination ( Figure 5E , Figure 5—figure supplement 1B ) . This indicates that phosphorylation of Y1045 alone is not sufficient for EGFR ubiquitination by c-Cbl and that EGF-induced receptor association is an essential requirement . EGF-induced ubiquitination thus is a secondary signal that redirects EGFR toward perinuclear areas with high PTP1B activity to inactivate signaling of the liganded receptor ( Figure 5—figure supplement 4 ) and further commit it for lysosomal degradation to avoid reactivation at the PM . In order to better understand why vesicular EGFR trafficking switches from recycling to unidirectional trafficking upon EGF stimulation , we incorporated our experimental findings and previous knowledge into a minimal model in which the effect of spatial partitioning of phosphates and EGFR trafficking on its activity could be investigated . The set of reactions in this model ( Figure 6A ) encompassed the following precepts:10 . 7554/eLife . 12223 . 022Figure 6 . Compartmental model for EGFR spontaneous autocatalytic activation supression . ( A ) The set of conversion reactions based on which ordinary differential equations were numerically integrated in Mathematica ( see 'Materials and methods' ) . EGFR fluctuates between an inactive and an active conformation – denoted by star ( 1 ) . The active conformation phosphorylates EGFR with the rate constant kautonom that encompasses the fraction of active conformation ( 2 ) . Here , only phosphorylation of an autocatalytic tyrosine is considered . Phosphorylated EGFR is dephosphorylated ( kPTP; 3 ) and can autocatalytically phosphorylate EGFR ( kautocat; 4 ) similar to ( 2 ) . All species of EGFR exchange between two compartments ( from PM to internal – kendocyt – and back – krecyc; 5 ) . kPTP may vary between both compartments . Ligand-binding locks EGFR in the active conformation , irrespective of its phosphorylation state ( kkinase over kPTP ) . ( B-–D ) Time traces of EGFR phosphorylation and bar diagrams of the corresponding phospho-EGFR steady states at the PM for different parameters ( green – phospho-EGFR >80% , red– phospho-EGFR ~50% , black – phospho-EGFR <10% ) : ( B ) without autocatalytic feedback or recycling and increasing values of kPTP; ( C ) with autocatalysis , but no recycling and increasing values of kPTP; and ( D ) autocatalysis , recycling and kPTP=15 at the PM versus kPTP=200 for the internal EGFR fraction with increasing steady state partitioning of EGFR towards the internal fraction . Dashed line in ( C ) is in the absence of phosphatase activity ( phospho-EGFR rapidly approaches 100% ) . ( E-–G ) Steady state phosphorylation of EGFR at the PM for the situations displayed in ( B–D ) upon varying the total amount of EGFR per cell ( expression level ) . Colors ( green , red , black ) correspond to the parameters in ( B–D ) . ( H ) 2D plots of EGFR phosphorylation ( color code displayed on top ) analogous to the last line plot in ( E ) for different expression levels of EGFR ( x-axis ) versus increasing phosphatase activity acting on the internal EGFR fraction ( y-axis ) . Each image corresponds to a different steady state partitioning of EGFR towards the PM ( 10:1 , 5:1 , 3:1 , 2:1 , 1:1 ) ; lines in each image mark 50% EGFR phosphorylation ( red color ) for the corresponding steady state EGFR partitioning . ( I ) Simulation of EGF-induced activation of EGFR . In the absence of recycling ( upper time traces ) , zeroth order phosphorylation rate of liganded EGFR locked in the active conformation results in a stable phosphorylation of all EGFR for low PTP activity ( red curve ) or an extremely rapid dephosphorylation for high PTP activity ( black curve ) . Recycling in the presence of partitioned PTP activity ( as in ( D ) , kPTP=15 at the PM and kPTP=200 on endomembranes ) results in 100% phosphorylation at the PM and low phosphorylation of EGFR on endomembranes . The total phosphorylation of EGFR in the cell ( abscissa ) is then determined by steady state EGFR partitioning ( ~50% for equipartitioning , red curve ) . Only degradation of the internal fraction with ongoing endocytosis allows finite signaling whose duration is determined by the kinetics of endocytosis ( black curve ) . ( J ) Schematic of switch in EGFR vesicular traffic . Vesicular recycling ( thick blue arrows ) of EGFR monomers from the PM through the pericentriolar RE within areas of high PTP activity and back continuously suppresses autonomous receptor phosphorylation . Horizontal black arrows: Chemical conversions , vertical red arrows: Causalities ( left ) . Following EGF-binding to the receptor , fully active and phosphorylated dimers share the same entry into early endosomes as the monomer . However , ligand-induced ubiquitination of EGFR clusters mediates a switch in the endocytic trafficking from continuous recycling to unidirectional trafficking . By depleting the pool of monomeric EGFR at the RE , this results in a transient translocation of the receptors to the PM reinforcing activation , and a subsequent depletion from the PM by delayed endocytosis . During endosomal transit from the PM to LEs , receptors signal in the cytoplasm until their dephosphorylation in perinuclear areas with high PTP activity ( right ) and subsequent degradation in lysosomes . DOI: http://dx . doi . org/10 . 7554/eLife . 12223 . 022 In a system where autonomous phosphorylation of EGFR results purely from leaky kinase activity due to conformational fluctuations , low phosphatase activity suffices to prevent all EGFR from eventually being phosphorylated . We normalized the phosphatase rate constant to a unit 1 as that phosphatase activity , which allows an autonomous phosphorylation of 50% EGFR in steady state . This phosphatase activity balances the 1% of all EGFR spontaneously attaining an active conformation . Under these conditions , a kPTP = 15 units suppresses phosphorylation to a level below 10% ( Figure 6B ) . Introducing autocatalysis results in a massive acceleration of phosphorylation . If phosphatase activity was absent in this system , all EGFR would be phosphorylated 250 times more rapidly than without the autocatalytic feedback ( compare Figure 6C , dashed curve with Figure 6B , green curves ) . In combination with the low phosphatase activity of 15 that maintains autonomous phosphorylation below 10% ( black curves in Figure 6B ) , autocatalysis causes a highly phosphorylated fraction . In this case , a phosphatase activity of 200 is required to suppress autocatalytic phosphorylation below 10% ( Figure 6C ) . However , this level of phosphatase activity rapidly suppresses phosphorylation of ligand-bound EGF-EGFR ( Figure 6I , upper graph ) . We then posed the question , whether steady state EGFR distribution caused by recycling and a partitioning of phosphatase activity between PM and RE can counteract autocatalytic phosphorylation while maintaining responsiveness of EGFR to EGF . We therefore started from a condition where EGFR at the PM is subjected to low phosphatase activity of ( kptp = 15 ) that is sufficient to prevent autonomous phosphorylation without suppressing phosphorylation of liganded EGFR , nor autocatalytic phosphorylation ( compare Figure 6B , C ) . Additionally , a PTP activity that is sufficient to counter autocatalytic phosphorylation ( kptp = 200 ) was assigned to intracellular membranes . Upon exchange of EGFR between both compartments by recycling , the relative amount of EGFR at the PM determines its phosphorylation level ( Figure 6D ) . EGFR phosphorylation is high at a 10:1 steady state distribution , while equipartitioning of EGFR leads to a phosphorylation well below 10% . Vesicular recycling thus acts as an additional , effective phosphatase activity that can suppress autocatalytic phosphorylation . Why is the system using recycling in order to generate an effective phosphatase activity at the PM ? Since autocatalytic activity is a non-linear process that is strongly dependent on the density of EGFR at the PM , we also compared the phosphorylation of EGFR as a function of EGFR expression level in the three cases described above: no autocatalysis , no recycling , recycling , and partitioning of phosphatase activity ( Figure 6E–G ) . Without autocatalytic phosphorylation , the second-order rate of autonomous phosphorylation causes a square-root dependence of phosphorylation on EGFR expression . For the moderate phosphatase activity of 15 that suppresses autonomous phosphorylation , this dependence is stretched out and it appears almost linear ( black curve in Figure 6E ) . However , this amount of phosphatase activity combined with autocatalytic phosphorylation gives rise to a low threshold ( < 0 . 1 ) in EGFR expression , beyond which phosphorylation increases sharply as a function of EGFR levels . This threshold increases with phosphatase activity ( expression level of 2 for a phosphatase activity of 200 , compare Figure 6F ) . In the case of partitioned phosphatase activity , the dependence of EGFR phosphorylation on expression is similar to the scenario without recycling ( Figure 6G ) . This shows that the system of partitioning phosphatase activity and EGFR trafficking poses a valid alternative to a purely PM-based solution of keeping autocatalytic phosphorylation in check . The essential difference is that a partitioned phosphatase activity achieves this while maintaining a low phosphatase activity at the PM , where ligand-mediated activation of EGFR occurs . So far , we exemplary considered a partitioning of phosphatase activities of 15:200 between PM and endomembranes . We find that the autocatalytic threshold ( black and white lines in Figure 6H ) depends on this phosphatase partitioning , as well as on the EGFR distribution . Even if there is 10 times more EGFR on the PM , strongly asymmetric phosphatase partitioning hardly shifts the threshold . However , for more equal distributions of EGFR ( >2:1 ) , a concurrent asymmetric phosphatase partitioning provides a large shift in the threshold . This corresponds to a safeguard against autocatalytic phosphorylation even at high EGFR expression . A low phosphatase activity at the PM suffices in combination with the suppressive recycling to maintain the level of EGFR phosphorylation below the autocatalytic threshold . For liganded EGFR , vesicular recycling leads to efficient dephosphorylation on perinuclear membranes with their high local phosphatase activity . However , the stabilization of the active EGFR conformation by the ligand causes its kinase activity to surpass suppression by low local phosphatase activity once liganded EGFR recycles to the PM ( Figure 6I , lower graph , red curve ) . This is in stark contrast to unliganded dephosphorylated EGFR returning from the perinuclear area that requires slow autonomous phosphorylation as the trigger to return to the active state . The slow rate of recycling thus proves ineffective to maintain ligand-activated EGFR in a dephosphorylated state , while it can suppress autonomous phosphorylation ( compare Figure 6D with Figure 6I , lower graph ) . The system is therefore left with one way to shut down signaling after ligand stimulation: reducing the amount of EGF-EGFR complexes in the whole cell by lysosomal degradation to reset the kinase-phosphatase balance to a level below the autocatalytic threshold . The scheme in Figure 6J summarizes this switch from continuous suppressive EGFR recycling to directional trafficking leading to finite signaling Ligand-induced EGFR signaling is involved in numerous cellular responses in health and disease , with a documented link between aberrant ligand-independent EGFR activation through overexpression or mutations and tumorigenesis ( Arteaga and Engelman , 2014 ) . It remains unclear how spontaneous auto-activation of EGFR is suppressed , while still maintaining responsiveness to physiological stimuli . Here , we show that a switch from cyclic to unidirectional vesicular traffic of EGFR enables its interaction with spatially organized PTPs to suppress spontaneous auto-activation while allowing for ligand-induced signal propagation . Otherwise , suppression of spontaneous autocatalytic activation at the PM would be incompatible with ligand-induced signaling of EGFR . Partitioning high PTP activity away from the PM allows these opposing requirements , because of two modes of spontaneous EGFR phosphorylation: 1 ) slow autonomous phosphorylation originating from a low fraction of EGFR that is in the active conformation due to thermal fluctuations; and 2 ) rapid autocatalytic phosphorylation due to phosphorylation of the autocatalytic site Y845 . Low PTP activity at the PM is sufficient to suppress autonomous EGFR phosphorylation and can be overcome by ligand-activated EGFR . In contrast , the extremely fast kinetics of autocatalytic phosphorylation can only be suppressed by PTPs with high catalytic efficiency . This would hinder EGF-induced phosphorylation of EGFR at the PM and thereby prevent its signaling . This is the reason why shutdown of EGFR signaling occurs after its internalization by vesicular trafficking to the perinuclear area , where high PTP activity dephosphorylates the liganded receptor without interfering with EGF-responsiveness at the PM . This feature of the perinuclear area is also capitalized by monomeric , spontaneously phosphorylated EGFR through continuous vesicular trafficking into the same spatial location . The two different 'species' of phosphorylated EGFR – —spontaneously and ligand-activated – —are distinguished by their ubiquitination that depends on phosphorylation of tyrosine 1045 and self-association of EGFR . EGF-EGFR complexes travel in a unidirectional way to this perinuclear area due to their ubiquitination and will not leave it again , whereas spontaneously phosphorylated EGFR will recycle to the PM after dephosphorylation at the RE . Slow spontaneous autonomous receptor phosphorylation will then start again at the PM . However , before the threshold is reached , where rapid autocatalysis would accelerate phosphorylation of Y845 , trafficking to the RE restarts the suppressive cycle . The primary function of EGFR recycling is therefore to suppress spontaneous phosphorylation of Y845 that leads to autocatalysis . This notion is consistent with our finding that – —out of the three investigated tyrosines – —phosphorylation of the autocatalytic Y845 site was the most efficiently suppressed by vesicular recycling of EGFR ( Figure 4A ) . This has implications for spontaneously versus EGF-activated EGFR- signaling . Since auto-activated EGFR is maintained at the PM , it can activate Akt via PIP3 production by PI3K at the PM ( Cantley , 2002 ) , resulting in survival signaling . In contrast , we observed unresponsiveness of Erk to spontaneous auto-activation of EGFR . Possible reasons for this could be that negative feedback in the Erk circuitry ( Fritsche-Guenther et al . , 2011 ) results in an adaptive response to sustained EGFR activity at the PM or that efficient activation of Erk only occurs upon endocytosis of the receptor . In general , our data show that interference with vesicular trafficking by knock down of Rab proteins can have substantial pleiotropic effects on growth factor receptor signaling activity . We have shown here that the ER-associated PTP1B efficiently dephosphorylates the autocatalytic Y845 . The highest intrinsic activity of PTP1B has been demonstrated to be on perinuclear membranes ( Yudushkin et al . , 2007 ) , whereas only a small fraction of its activity can reach the PM at points of cell–cell contacts ( Haj , 2002 ) . In addition to PTP1B , TCPTP is also spatially segregated from PM-localized EGFR by its association with the ER and has been shown to dephosphorylate Y845 ( Mattila et al . , 2005 ) . Thus , the localization of these ER-anchored PTPs with a high catalytic efficiency toward phosphorylated Y845 in the perinuclear area is compatible with the presented model of maintaining unliganded EGFR phosphorylation at the PM below the autocatalytic threshold . In this way , even a marginal PTP activity at the PM suffices to silence the signaling tyrosines . In this regard , several RPTPs ( i . e . RPTPJ , G , A ) that dephosphorylate EGFR at the PM ( Rotin et al . , 1992; Tarcic et al . , 2009; Leung Cheung et al . , 2015 ) exhibit a high preference for the signaling tyrosines Y1068 and Y1086 ( Leung Cheung et al . , 2015; Tarcic et al . , 2009 ) . In addition , the catalytic efficiency of RPTPA for phosphorylated peptide substrates is three orders of magnitude lower than that of PTP1B or TC-PTP ( Selner et al . , 2014 ) . This agrees with our model where low PTP activity at the PM can only suppress slow autonomous phosphorylation in case that high perinuclear PTP activity suppresses autocatalytic activation . This is substantiated by the fact that inhibiting internalization of EGFR results in its prolonged activity at the PM ( Offterdinger and Bastiaens , 2008 ) , whereas inhibition of EGFR kinase activity with an ATP-binding site competitor results in rapid dephosphorylation of signaling tyrosines ( Offterdinger et al . , 2004 ) . In the latter case , there is no autocatalytic maintenance of the kinase activity and the PM-resident PTP activity suffices to dephosphorylate EGFR on signaling tyrosines . We argue that the unidirectional vesicular trafficking of liganded receptors to the perinuclear area determines the duration of EGFR signaling in response to EGF . During vesicular transit of EGF-EGFR complexes to perinuclear areas with the highest PTP1B activity ( Yudushkin et al . , 2007 ) , growth factor signals are propagated in the cytoplasm ( Lai , 1989; Di Guglielmo et al . , 1994; Wouters and Bastiaens , 1999; Sorkina et al . , 1999 ) , indicating a link between vesicular trafficking and duration of EGFR signaling response . Such a response typically occurs on the tens-of-minutes time-scale , corresponding to characteristic vesicular trafficking rates . This time-scale of EGFR response cannot be achieved by cytosolic , regulated PTPs ( i . e . SHP1/2 ) that couple to EGFR phosphorylation in a negative feedback manner ( Grecco et al . , 2011 ) . The recruitment of these diffusible PTPs to the receptor on the time scale of seconds would dephosphorylate EGFR too rapidly to signal on a time scale of minutes . By the same argument , suppression of spontaneous autocatalytic activation at the PM by regulated PTPs is incompatible with prolonged EGFR signaling . Both , EGF-EGFR complexes as well as recycling receptors transit via Rab5 positive EEs , into the LE and RE , respectively ( Figure 3D , E ) . This suggests that sorting of monomeric EGFR to the RE and self-associated , ubiquitinated EGFR to the LE happens at the early endosomal compartment . Why is it necessary to re-route ligand-activated receptors by endosomal sorting away from the unliganded EGFR that keeps recycling ? The answer to this question lies in the stability of the ligand-receptor complexes . Liganded EGFR maintains an active conformation even after dephosphorylation in the perinuclear area , and therefore will be rapidly re-phosphorylated in the environment of low PTP activity at the PM . This would sustain signaling as it has been observed for the c-Cbl-binding-impaired mutants , which continued recycling upon EGF stimulation ( Figure 4B–D ) . Importantly , this reactivation does not occur for spontaneously activated EGFR because dephosphorylation of Y845 inactivates the conformation that results in rapid autocatalytic phosphorylation . The protective recycling can be exploited by oncogenic versions of EGFR ( Chung et al . , 2009 ) that show impaired c-Cbl association ( Shtiegman et al . , 2007 ) , or c-Cbl mutations that abolish interaction with EGFR , or within cells with altered c-Cbl activity , all leading to insufficient EGFR ubiquitination , subsequent recycling and sustained downstream signaling ( Thien and Langdon , 1997; Peschard and Park , 2003; Ravid et al . , 2004 ) . Recycling also occurs at low doses of EGF stimulation ( Sigismund et al . , 2005 ) , where some receptors might autocatalytically activate but fail to be ubiquitinated due to lack of stable EGF-induced self-association . However , as shown in this study , c-Cbl is a limiting factor in rerouting EGF-EGFR complexes to lysosomes under saturating doses of EGF . Therefore , the level of c-Cbl expression in different cells might dictate the fraction of EGF-EGFR complexes that keep recycling . Even though re-routing of liganded EGFR occurs on EEs , EGF stimulation leads to an apparent transient translocation of EGFR to the PM , followed by its depletion ( Figure 5A–C ) . This shows that EGF-induced receptor clustering or ubiquitination results in a delayed internalization , where continued trafficking from the RE results in this perceived net translocation . The delayed internalization might point at a different internalization and trafficking route to EEs of monomeric receptors as compared to ubiquitinated EGF-EGFR complexes ( Sigismund et al . , 2013 ) . One of the decisive factors for this differential sorting of spontaneously and ligand-activated receptors is phosphorylation of the c-Cbl binding site Y1045 . We could show that - —out of the three investigated tyrosine residues - —it was the least efficiently phosphorylated by spontaneous EGFR kinase activity ( Figure 1B–D ) . This is in accordance with data showing that Y1045 is inefficiently auto-phosphorylated under oxidative stress , in contrast to Y845 ( Ravid et al . , 2002 ) . The fact that monomeric EGFR was inefficiently ubiquitinated upon PTP inhibition ( Figure 5E ) , but displayed a similar phosphorylation on Y1045 as compared to EGF stimulation , suggests that rather than Y1045 phosphorylation alone , the formation of EGFR oligomers is a decisive factor for c-Cbl binding and efficient ubiquitination . The distinct molecular states of spontaneous activated versus ligand-activated EGFR that are recognized and processed differently by the endocytic machinery , most likely evolved to provide robustness in ligand responsiveness against the gain in plasticity from its autocatalytic activation mechanism . This feature therefore seems to be common for other receptor tyrosine kinases as we have shown for the cell guidance ephrin receptor type-A2 ( EphA2 ) ( Sabet et al . , 2015 ) . mCitrine-N1 , mCherry-N1 , TagBFP-C1 , and paGFP-N1 were generated by insertion of AgeI/BsrGI PCR fragments of mCitrine , mCherry , TagBFP ( Evrogen , Moscow , Russia ) and paGFP ( gift from J . Lippincott-Schwartz ) cDNA into pEGFP-C1 or -N1 ( Clontech , Mountain View , CA ) . Fluorophore fusions of EGFR , PTB domain PTP1B , PTP1B D181A ( gift from B . Neel , UHN Toronto ) and Rab11a ( Addgene ) and Rab5 ( gift from Y . Wu , CGC Dortmund ) were generated through restriction-ligation of EGFR , PTB domain PTP1B , PTP1B D181A , Rab5 , and Rab11a cDNA into the appropriate vector ( mCherry-C1 , Tagbfp-C1 and mCitrine-C1 ) . Mutants of EGFR were generated by site-directed mutagenesis using the Quickchange Site-Directed-Mutagenesis kit ( Stratagene , Santa Clara , CA ) . To generate EGFR-QG-mCitrine mCitrine flanked with a linker sequence ( LAAAYSSILSSNLSSDS-mCitrine-SDSSLNSSLISSYAAAL ) ( Sabet et al . , 2015 ) was inserted between the amino acids Q958 and G959 of EGFR . LAMP1-TagBFP was generated by excision replacement of mTFP by TagBFP from LAMP1-mTFP ( Allele Biotechnology and Pharmaceuticals , San Diego , CA ) using the restriction sites MfeI and BamHI . Expression construct encoding fluorophore fusions of HA-c-Cbl were obtained by excision replacement of Citrine by mCherry from HA-c-Cbl-Citrine ( gift from I . Dikic , iBC II , Frankfurt am Main ) . HA-Ubiquitin was a gift from I . Dikic . All constructs were sequence verified and tested for correct expression . pCMV-EGFR ( 128* ) -EGFP-TCORS and p4-CMV-U6-PylT were described earlier ( Lang et al . , 2012 ) . Human EGF ( Peprotech , Hamburg , Germany ) was shock frozen at a concentration of 100 μg/ml in PBS + 0 . 1% BSA and stored at -−80°C . Site-specific C-terminal labeling of hEGF-Cys was carried out as described previously ( Sonntag et al . , 2014 ) . In short , the unprotected C-terminal cysteine in hEGF-Cys was labeled with a five times excess of Alexa647-maleimide ( Life Technologies , Darmstadt , Germany ) in aqueous solution ( PBS , pH 7 . 4 ) under argon for 2 hr at 4°C in the dark . Pervanadate was freshly prepared by adding sodium orthovanadate ( S6508 , Sigma Aldrich , St . Louis , MO ) to H202 ( 30% ) according to Huyer et al . ( 1997 ) ) . tet-TAMRA and TCOK were described earlier in Lang et al . ( 2012 ) . COS-7 and HEK293 cells were grown in Dulbecco’s Modified Eagle’s Medium ( DMEM ) supplemented with 10% fetal bovine serum ( FBS ) , 2 mM L-glutamine and 1% non-essential amino acids ( NEAA ) and maintained at 37°C in 5% CO2 . Transfection of COS-7 cells was done using FUGENE6 ( Roche Diagnostics , Mannheim , Germany ) for imaging experiments and for Western blots according to manufacturer’s protocol . Four to twelve hours prior to an experiment , cells were starved in growth medium containing dialyzed 0 . 5% FBS . For live cell microscopy , cells were cultured on 35-mm glass bottom dishes ( MatTek , Ashland , MA ) or 4-well chambered glass slides ( Lab-tek , Thermo Fisher Scientific , Waltham , MA ) . HEK293 cells were plated on 25-mm poly-L-lysine coated coverslips in DMEM with 10% FBS . Transfection was done using lipofectamine 2000 reagent ( Life Technologies , Darmstadt , Germany ) . The cells were incubated at 37°C with 5% CO2 for 5 hr . After this period the media was replaced with 10% FBS DMEM containing 1 mM TCOK . After overnight incubation at 37°C in 5% CO2 , the cells were rinsed thrice in 0 . 5% FBS DMEM and incubated for at least 1 hr in 0 . 5% FBS DMEM . Afterwards , coverslips were inserted into a PFC pro-flow chamber ( Warner instruments , Hamden , CT ) . Medium was flowed in by gravity flow . Labeling was carried out by flow through of 2 ml DMEM containing 400 nM tetrazine-TAMRA conjugate . This was followed by washout in DMEM or in DMEM containing EGF ( 80 ng/ml ) or EGF-Alexa647 ( 20 ng/ml ) , respectively . Transfection of Rab11a siRNA was achieved using siRNA transfection reagent ( sc-29528 , Santa Cruz Biotechnology , Dallas , TX ) in special transfection medium ( sc-36868 , Santa Cruz Biotechnology ) . Both Rab11a siRNA: sc-36340-SH ( Santa Cruz Biotechnology ) or scrambled non-targeting control siRNA: sc-37007 ( Santa Cruz Biotechnology ) was used at a final concentration of 50 nM for 72 hr before validation of knockdown . Transfection of Rab11a/b siRNA was achieved using Lipofectamine 3000 Transfection Reagent ( Thermo Fisher Scientific , Waltham , MA ) . Rab11a and Rab11b isoforms were simultaneously knocked down using double-stranded small interfering RNAs ( siRNAs ) targeted against the following target sequences 5´ ( AATGTCAGACAGACGCGAAAA ) -3´ for Rab11a and 5´ ( AAGCACCTGACCTATGAGAAC ) -3´ for Rab11b at a final concentration of 40 nM for 72 hr before validation of knockdown . Scrambled non-targeting control siRNA ( 1027281 , Qiagen ) was used at a final concentration of 40 nM . Cells were lysed in ready-made Cell Lysis Buffer ( 9803 , Cell Signaling Technology , Danvers , MA ) supplemented with Complete Mini EDTA-free protease inhibitor ( Roche Applied Science , Heidelberg , Germany ) and 100 μl phosphatase inhibitor cocktail 2 and 3 ( P5726 and P0044 , Sigma Aldrich , St . Louis , MO ) . Following lysis , samples were cleared by centrifugation for 10 min , 13 , 000 rpm at 4°C . For ubiquitination IP , cell lysis was performed in modified RIPA ( 50 mM Tris–HCl , 150 mM NaCl , 1 mM EGTA , 1 mM EDTA , 1% Triton X-100 , 1% Sodium deoxycholate , 0 . 2% SDS ) . Following lysis , samples were subjected to sonication for 12 s and then cleared by centrifugation for 15 min , 13 , 000 rpm at 4°C . Equal amounts of protein lysates were incubated with an anti-EGFR antibody over night at 4°C followed by incubation for 2 hr with Protein-G Sepharose® beads . SDS–PAGE was performed using the X-cell II mini electrophoresis apparatus ( Life Technologies , Darmstadt , Germany ) according to the manufacturers instructions . Samples were transferred to preactivated PVDF membranes ( Merck Millipore , Billerica , MA ) and incubated with the respective primary antibodies at 4°C overnight . Detection was performed using species-specific secondary IR-Dye 800 CW and IR-Dye 680 secondary antibodies ( LI-COR Biosciences , Lincoln , NE ) and the Odyssey Infrared Imaging System ( LI-COR Biosciences , Lincoln , NE ) . Cells were fixed with 4% paraformaldehyde in PBS ( pH 7 . 4 ) for 10 min at room temperature ( RT ) . This was followed by permeabilization for 5 min with 0 . 1% Triton X-100 in TBS . Background staining was blocked by incubation with Odyssey® Blocking Buffer ( LI-COR Biosciences , Lincoln , NE ) for 1 hr at RT . Primary and secondary antibodies ( diluted in Odyssey® Blocking Buffer ( LI-COR Biosciences , Lincoln , NE ) were applied for 1 hr 30 min and 1 hr , respectively . Fixed cells were imaged in PBS at 37°C . In background-subtracted fluorescence images , masks for single cells were generated and the mean fluorescence intensity for each channel was measured in ImageJ ( http://imagej . nih . gov/ij/ ) . The relative phosphorylation level ( pYn/EGFR , n=845 ∪ 1045 ∪ 1068 ) per cell was determined and the mean values of pYn/EGFR-mCitrine were calculated . The fluorescence signals emanating from the antibody complexes that were used to monitor phosphorylation of the three tyrosine residues cannot be compared because of the different affinities of the primary antibodies for the epitopes , and distinct fluorescence brightness and binding affinity of the labeled secondary antibodies . In order to be able to compare the level of ligand-independent phosphorylation for the three sites , the measured fraction of auto-phosphorylated EGFR ( mean pYn intensity divided by mean EGFR intensity , where n=845 , 1045 , 1068 ) was normalized to the corresponding fraction after EGF stimulation ( pYn/EGFR-EGF upon 2’ or 5’ EGF stimulation: the maximally attainable phosphorylation at a given EGFR expression level ) . The relative EGFR auto-phosphorylation was computed from immunofluorescence for single cells and western blots as the ratio of the fraction of anti-pYn ( n=845 , 1045 , 1068 ) intensity over EGFR-mCitrine intensity ( for blots this corresponded to the anti-Citrine antibody band intensity ) before and after stimulation with EGF . This 'ratio of fractions' gives a comparative measure for immunofluorescence and Western blots of the auto-phosphorylated fraction with respect to the EGF-induced phosphorylation of each investigated tyrosine residue on EGFR . For Western blots , the relative EGFR expression level was determined by normalizing to the average of the un-/stimulated EGFR-mCitrine bands of the highest levels of expressed EGFR-mCitrine ( 3 µg cDNA ) . Quantification of the spatial distribution of spontaneously phosphorylated Y845 was performed by measuring the mean fluorescence intensity of EGFR-mCitrine and pY845 in a 5-pixel ring at the cell periphery . Mean fluorescence intensities of EGFR-mCitrine and pY845 at the RE were obtained by using binary masks of the RE , generated from thresholded BFP-Rab11a images . The difference between pY845 over EGFR-mCitrine at the PM and RE ( ( pY845/EGFR ) PM - — ( pY845/EGFR ) RE ) was plotted as a function of overall EGFR phosphorylation level in individual cells . Confocal images were recorded using an Olympus Fluoview FV1000 confocal microscope ( Olympus Life Science Europa , Hamburg , Germany ) , a Leica TCS SP5 DMI6000 confocal microscope ( Leica Microsystems , Wetzlar , Germany ) , a Leica SP8 confocal microscope ( Leica Microsystems , Wetzlar , Germany ) and a Zeiss LSM 780 ( Carl Zeiss Jena GmbH , Jena , Germany ) . Time-lapse movies of COS-7 cells expressing BFP-Rab11a , c-Cbl-mCherry and either EGFR-mCitrine wild type or ubiquitination-impaired mutants were obtained through confocal microscopy . Binary masks of the RE were generated from thresholded BFP-Rab11a images and for endosomal EGFR from thresholded EGFR-mCitrine images . To quantify the fraction of endosomal EGFR at the RE , the integrated EGFR-mCitrine fluorescence intensity was determined in the corresponding endosomal masks . The fraction of endosomal EGFR at the RE was determined by dividing the integrated intensity of EGFR at the RE by the integrated intensity of all endosomal EGFR . The initial ratio was normalized to 1 for each cell . The fraction of EGFR at the PM ( PM-EGFR ) was quantified for individual cells as the ratio between the integrated intensity of a 5-pixel ring of the cell periphery and the integrated intensity of the whole cell . The ratio for time point 0min for each cell was normalized to 1 and mean values were calculated for each time point . FRAP experiments were carried out at 37°C on a Leica TCS SP5 DMI6000 confocal microscope equipped with a HCX PL APO 63x/1 . 4 NA oil objective and an environment-controlled chamber maintained at 37°C . The FRAP image acquisition was divided into three steps: ( 1 ) pre-bleach , ( 2 ) bleaching , and ( 3 ) post-bleach . In the pre-bleach step , a total of 10 fluorescence images for EGFR-mCitrine and BFP-Rab11a with a time interval of 10 s were acquired , using the 514 nm Argon laser at 10% power and the 405 nm Cube laser at 2% power , respectively . Bleaching of EGFR-mCitrine was performed with the 514 nm Argon laser at 100% power , and bleaching was restricted to a region of interest ( ROI ) on the RE , identified by expression of BFP-Rab11a . In the post-bleach step , a total of 80 fluorescence images for EGFR-mCitrine and BFP-Rab11a with a time interval of 10 s were acquired , using the 514 nm Argon laser at 10% power and the 405 nm Cube laser at 2% power , respectively . Images were background corrected in ImageJ and the relative fluorescence intensity I ( t ) was computed as follows: ( 1 ) I ( t ) = ( I ( t ) ROII ( t ) TOTAL ) / ( I ( 0 ) ROII ( 0 ) TOTAL ) where I ( t ) ROI is the average fluorescence intensity in the ROI at time t , I ( t ) TOTAL is the average fluorescence intensity of the whole cell at the same time point , I ( 0 ) ROI is the average fluorescence intensity in the ROI of the pre-bleach images , I ( 0 ) TOTAL is the average fluorescence intensity of the whole cell of the pre-bleach images . The relative intensity was then fitted by a single exponential function: ( 2 ) I ( t ) =I0+Aexp-tτ where I ( t ) is the fluorescence intensity ( in arbitrary units ) at time t ( s ) , I0 is the residual intensity after photobleaching in the ROI , τ ( s ) is the exponential recovery time constant and A is the exponential amplitude . FLAP experiments were carried out at 37°C on a Leica TCS SP5 DMI6000 confocal microscope equipped with a HCX PL APO 63x/1 . 4 NA oil objective . The FLAP image acquisition was divided into three steps: ( 1 ) pre-activation , ( 2 ) photoactivation and ( 3 ) post-activation . In the pre-activation step , a total of three fluorescence images for EGFR-paGFP and EGFR-mCherry with a time interval of 40 s were acquired , using the 488nm Argon laser at 10% power and the 561 nm DPSS laser at 11% power , respectively . Photoactivation of EGFR-paGFP was performed with the 405nm Cube laser at 80% power and photoactivation was restricted to a region of interest ( ROI ) on the RE , identified by co-expression of EGFR-mCherry . In the post-activation step , a total of 30 fluorescence images for EGFR-paGFP and EGFR-mCherry with a time interval of 40 s were acquired , using the 488 nm Argon laser at 10% power and the 561 nm DPSS laser at 11% power , respectively . At the end of the experiment an image of BFP-Rab11a was acquired with 5% 405-laser power . Following background correction , fluorescence loss after photoactivation at the RE was quantified as the ratio of local EGFR-paGFP to EGFR-mCherry fluorescence , to account for changes in the structure and intensity in the ROI . Fluorescence gain on the PM was quantified as the ratio of local ( cell periphery ) EGFR-paGFP to EGFR-mCherry fluorescence . Analogously to the FRAP experiment , exponential decay at the RE and increase at the PM were fitted to a mono-exponential function ( Equation 2 ) . Considering the PM and the RE a simple two-compartment system allowed a more detailed interpretation of the retrieved parameters . The loss of EGFR from either compartment was assumed to follow a First-Order Rate Process . This yields the following exponential equation containing association ( kon ) and dissociation ( koff ) rate constants of the fluorescence intensity of EGFR to the RE ( [EGFRRE] ) : ( 3 ) [EGFRRE]=koffkon+koff+konkon+koffe-t ( kon+koff ) The normalized fluorescence decay curves were averaged and fitted to this function ( Equation 3 ) . The residence time of EGFR at the RE was calculated as 1/ koff . Anisotropy microscopy was done at 37°C in vitamin-free imaging medium in COS-7 cells ectopically expressing EGFR-QG-mCitrine and PTB-mCherry . Images were acquired 20–24 hr post-transfection , using an Olympus IX81 inverted microscope ( Olympus , Hamburg , Germany ) equipped with a MT20 illumination system . A linear dichroic polarizer ( Meadowlark optics , Frederick , CO ) was placed in the illumination path of the microscope , and two identical polarizers were placed in an external filter wheel at orientations parallel and perpendicular to the polarization of the excitation light . Fluorescence images were collected via a 20x/0 . 7 NA air objective using an Orca CCD camera ( Hamamatsu Photonics , Hamamatsu City , Japan ) . For each measurement two images were taken , one with the emission polarizer oriented parallel to the excitation polarizer ( I|| ) and one with the emission polarizer oriented perpendicular to the excitation polarizer ( I⊥ ) . Steady state anisotropy ( ri ) was calculated in each pixel i by: ( 4 ) ri=Gi I||i-I⊥iGi I||i-2I⊥i The G-factor ( Gi ) was determined by acquiring the ratio of the intensities at perpendicular and parallel orientations for a fluorophore in solution ( fluorescein ) with a steady-state anisotropy close to zero . The CellR software supplied by the microscope manufacturer ( Olympus , Hamburg , Germany ) controlled data acquisition . Computed anisotropy images were displayed with false color-coding using ImageJ . For analysing the co-localization between EGFR-QG-mCitrine and PTB-mCherry the acquired fluorescence images were background subtracted and masks for individual cells were generated using ImageJ . PTB-mCherry fluorescence intensity was saturated in the nucleus and therefore excluded from the masks of individual cells . The extent of PTB recruitment ( R ) to EGFR was calculated by the following equation: ( 5 ) R=F of EGFRpre*F of PTBpostF of EGFRpost*F of PTBpre From this , a 2D-histogram of R versus EGFR fluorescence intensity per pixel was constructed with a bin size of 50 fluorescence units . FLIM measurements were made on an Olympus FluoView FV1000 laser scanning confocal microscope equipped with a time-correlated single-photon counting module ( LSM Upgrade Kit , Picoquant , Berlin , Germany ) , using 60x/1 . 35 NA oil objective ( Olympus , Hamburg , Germany ) . All pulsed lasers were controlled with the Sepia II software ( PicoQuant GmbH , Berlin , Germany ) at pulse repetition frequency of 40 MHz . The sample was excited using a 507 nm diode laser ( LDH 507 , Picoquant , Berlin , Germany ) . Fluorescence emission was spectrally filtered using a narrow-band emission filter ( HQ 537/26 , Chroma , Olching , Germany ) . Photons were detected using a single-photon counting avalanche photodiode ( PDM Series , MPD , Picoquant , Berlin , Germany ) and timed using a single-photon counting module ( PicoHarp 300 , Picoquant , Berlin , Germany ) . Using the SymPhoTime software V5 . 13 ( Picoquant , Berlin , Germany ) , images were collected after an integration time of ~ 2 min collecting app . ~ 3 . 0–5 . 0 x 106 photons . Data analysis was performed using the global analysis code as described in Grecco et al . ( 2009 ) . To quantify the spatial coincidence between the perinuclear area with high EGFR/PTP1B D181A interaction and the areas in the vicinity of Rab11-positive REs per cell , EGFR-mCtirine fluorescence lifetime and BFP-Rab11a fluorescence images were analyzed as follows: The percentage of interacting EGFR/PTP1B D181A in the perinuclear area that coincided with areas of Rab11-rich endosomes was calculated from 20 cells ( 63 ± 29% , mean ± std . ) . This was compared to the expected overlap percentage , calculated as a ratio of the size of the high Rab11-intensity area to the total size of the cell ( 19 ± 12% ) . Density estimation by a consensus-like algorithm: Given a snapshot image of a cell , from its pixel values we can observe how a quantity of interest ( EGFR-mCitrine lifetime or BFP-Rab11a intensity ) is spatially arranged . Due to fluctuations in the cell in each snapshot a non-smooth spatial arrangement is observed . Nevertheless , the images reveal some patterns , thus it can be presumed that each snapshot is a random realization of an underlying smooth density function . Estimating this function is of interest , as it can be easily compared to density functions of other quantities from the same cell . For each pixel i , we denote with Nithe set of its neighboring pixels from the cell and with |Ni| the size of Ni , that is , the number of neighboring pixels of i . Typically |Ni| is 4 , except for the pixels that lie on the edges of the cell . Each pixel i has an initial value xi ( 0 ) , which depicts the value from the snapshot image . In order to make smoother density estimation , we aim to accommodate the differences between the values of neighboring pixels . Therefore , we assume that each pixel has a cost function of the form ( 6 ) C ( xi , xNi ) =12 ∑j∈Ni ( xi-xj ) 2+12 K ( xi-xj ( 0 ) ) 2 that it tries to minimize . K≥0 is a parameter that represents the 'stubbornness' of every pixel regarding its initial value . Higher values of K lead to larger stubbornness; hence , the final density estimate will be similar to the initial arrangement , thus non-smooth . For K = 0 consensus will be reached among the pixels , and all of them will have the same value in the final estimate . In our analysis , we use K = 0 . 005 to obtain a fairly smooth density estimate . An iterative consensus-like algorithm is applied for minimization of the cost function , similar to the one described in ( Ghaderi and Srikant , 2013 ) . For every pixel i the estimate of the value for the next time step is ( 7 ) xi ( t+1 ) =1|Ni|+K ∑j∈Nixj ( t ) +K|Ni|+Kxi ( 0 ) This algorithm converges exponentially fast to our desired density estimate . All results are expressed as mean ± SEM . If more than two groups were compared a one-way analysis of variance ( ANOVA ) was performed , followed by Tukeys multiple comparision test ( Tukey's HSD ) to estimate significance . For the comparison of the relative phosphorylation of Y845 , Y1045 , and Y1068 in Figure 1B via immunofluorescence a Z-test was performed . In case of comparisons with normalized data , an independent one sample t test was performed . The following set of ordinary differential equations: r1 = kendocyt EGFRPM ( t ) r2 = kendocyt EGFRPMp ( t ) r3 = krecyc EGFRint ( t ) r4 = krecyc EGFRintp ( t ) r5 = kPMPTP EGFRPMp ( t ) r6 = kintPTP EGFRintp ( t ) r7 = EGFRPM ( t ) ( kautonom EGFRPM ( t ) + kautocat EGFRPMp ( t ) ) r8 = EGFRint ( t ) ( kautonom EGFRint ( t ) + kautocat EGFRintp ( t ) ) kautocat = 0 OR kkinase kautonom = 0 . 01 kkinase vesiscular trafficking / phosphorylation d/dt ( EGFRPM ( t ) ) = - r1 + r3 + r5 - r7 d/dt ( EGFRPMp ( t ) ) = - r2 + r4 - r5 + r7 d/dt ( EGFRint ( t ) ) = r1 - r3 + r6 - r8 d/dt ( EGFRintp ( t ) ) = r2 - r4 - r6 + r8 C = EGFRPM ( t ) + EGFRPMp ( t ) + EGFRint ( t ) + EGFRintp ( t ) were numerically integrated with the function 'NDSolve' of Mathematica 10 . 0 . 1 for Mac OS X . The following parameters yield the red curves in Figure 6D: kendocyt=0 . 1/2 , krecyc=0 . 1/1 , kkinase=0 . 06 , kintPTP =0 . 0006*200 , kPMPTP =0 . 0006*15 , C=2 . Variations of these parameters leading to the other curves and images in Figure 6 are indicated in the figure legends and in the main text , e . g . kautocat=0 and kPTP=0 . 0006 for the red curves in Figure 6B , or varying C in the x-axis of Figure 6H . For ligand binding ( Figure 6I ) , the following kinase reaction rates were used: r7 = kligand EGF-EGFRPM ( t ) / ( EGF-EGFRPM ( t ) + Csaturation ) r8 = kligand EGF-EGFRint ( t ) / ( EGF-EGFRint ( t ) + Csaturation ) , with kligand=0 . 06 and Csaturation=0 . 001 . This results in essentially a zeroth order kinase reaction , without the numerical artifacts of a set of stiff ODEs for very small levels of phosphorylation EGF-EGFRp < Csaturation .

In living tissue , the ability of individual cells to grow is influenced by signal molecules in the environment around each cell . For example , after an injury , a molecule called epidermal growth factor can stimulate cells to grow to repair the wound . Epidermal growth factor binds to and activates a receptor protein called EGFR , which faces outwards from the cell surface . However , this signal needs to be switched off again afterwards to prevent the cells from growing too much . Epidermal growth factor activates EGFR by triggering a process called “autophosphorylation” , in which EGFR attaches molecules called phosphates to itself . To quench the signal , EGFRs that are bound to growth factors are removed from the cell surface and taken into the cell in small membrane bubbles called vesicles . Enzymes called phosphatases near the cell nucleus remove the phosphate groups and thereby switch the receptors off , before the receptors are ultimately destroyed . However , EGFR autophosphorylation can also happen spontaneously in the absence of growth factor , so it was not clear how the cell is able to distinguish between this spontaneous activation and a genuine signal . Baumdick , Brüggemann , Schmick , Xouri et al . used biochemical techniques to address this question . The experiments show that EGFRs that have become spontaneously active are also removed from the cell surface in vesicles . However , unlike the EGFRs that are bound to growth factors , the spontaneously active receptors are recycled back to the membrane . On the way , their activity is also switched off by encountering phosphatases so that they are not active when they reach the cell surface again . The experiments also show that EGFRs are targeted for destruction by the presence of a tag called ubiquitin , which is added to the receptor in response to the binding of growth factor . Therefore , Baumdick et al . ’s findings show that epidermal growth factor controls a switch that alters the way active EGFRs are processed in cells . This system acts to suppress the spontaneous activation of EGFRs , whilst maintaining the ability of the cell to respond to epidermal growth factor . The next challenge is to understand how the location of the phosphatases inside the cell influences when and how the EGFRs respond to this external signal .

2015

EGF-dependent re-routing of vesicular recycling switches spontaneous phosphorylation suppression to EGFR signaling

Large Ca transients cause massive endocytosis ( MEND ) in BHK fibroblasts by nonclassical mechanisms . We present evidence that MEND depends on mitochondrial permeability transition pore ( PTP ) openings , followed by coenzyme A ( CoA ) release , acyl CoA synthesis , and membrane protein palmitoylation . MEND is blocked by inhibiting mitochondrial Ca uptake or PTP openings , depleting fatty acids , blocking acyl CoA synthesis , metabolizing CoA , or inhibiting palmitoylation . It is triggered by depolarizing mitochondria or promoting PTP openings . After mitochondrial MEND blockade , MEND is restored by cytoplasmic acyl CoA or CoA . MEND is blocked by siRNA knockdown of the plasmalemmal acyl transferase , DHHC5 . When acyl CoA is abundant , transient H2O2 oxidative stress or PKC activation initiates MEND , but the immediate presence of H2O2 prevents MEND . The PTP inhibitor , NIM811 , significantly increases plasmalemma in normally growing cells . Thus , the MEND pathway may contribute to constitutive as well as pathological plasmalemma turnover in dependence on mitochondrial stress signaling . Endocytic processes that do not use classical endocytic proteins remain poorly understood because they are mechanistically inhomogeneous and difficult to study in isolation ( Mayor and Pagano , 2007; Doherty and McMahon , 2009 ) . Nevertheless , they are involved in a wide range of cellular processes , including cell migration ( Howes et al . , 2010 ) , cell wounding responses ( Tam et al . , 2010 ) , and pathogen internalization ( Vidricaire and Tremblay , 2007 ) , as well as membrane recycling in some neurons ( Gong et al . , 2008 ) and astrocytes ( Jiang and Chen , 2009 ) . Massive endocytic ( MEND ) responses described by us ( Fine et al . , 2011; Hilgemann and Fine , 2011; Lariccia et al . , 2011 ) are the largest endocytic responses ever characterized . Large fractions of the plasmalemma that are internalized bind many amphipathic molecules less well than membrane remaining at the cell surface ( Hilgemann and Fine , 2011 ) . Thus , MEND internalizes preferentially the more ‘ordered’ portions of the surface membrane . Different forms of MEND can be distinguished on the basis of ATP- , polyamine- , and Ca-dependence ( Lariccia et al . , 2011 ) . In addition , MEND can be initiated by rapidly cleaving sphingomyelin in the outer plasmalemma monolayer with bacterial sphingomyelinase C ( Lariccia et al . , 2011 ) . It has been suggested that this form of MEND becomes activated during cell wound responses when native sphingomyelinases are translocated to the cell surface via exocytosis ( Tam et al . , 2010; Corrotte et al . , 2013 ) . We describe here experiments that suggest an entirely different pathway by which MEND occurs in BHK fibroblasts subsequent to large Ca transients . These ATP-dependent endocytic responses are functionally similar to ‘excessive’ endocytosis that occurs after large Ca transients in secretory cells ( Smith and Neher , 1997 ) . Up to 70% of the cell surface of fibroblasts can be internalized , followed by replenishment of the plasmalemma from internal membrane pools over 20 to 40 min ( Lariccia et al . , 2011 ) . To elucidate the cellular pathway by which MEND occurs , we attempted first to resolve how Ca promotes MEND , and second , to understand why more ordered membrane domains would be selectively internalized . As described here , our data support the hypothesis that Ca transients act initially to prime mitochondria to open PTPs ( permeability transition pore ) ( Giorgio et al . , 2013 ) , releasing metabolites to the cytoplasm ( Azzolin et al . , 2010 ) . At the distal end of the pathway , our data suggest that palmitoylation of surface membrane proteins promotes the coalescence of ordered membrane domains by promoting protein clustering , as described biochemically ( Levental et al . , 2010 ) and as occurs in anoxia-related metabolic stress ( Frank et al . , 1980 ) . That mitochondria might regulate surface membrane palmitoylation and thereby endocytosis is a novel hypothesis . Nevertheless , studies that suggest how this might occur have been available for decades . First , it is known that coenzyme A ( CoA ) , which is synthesized on the outer mitochondrial surface , is accumulated into the matrix space by voltage-sensitive transporters ( Tahiliani , 1991; Leonardi et al . , 2005 ) , generating CoA gradients to the cytoplasm of at least 50:1 ( Tahiliani and Neely , 1987 ) . Therefore , transient openings of nonselective mitochondrial pores , PTPs , can potentially release CoA and generate cytoplasmic CoA transients in the range of tens of micromolar without serious consequences for other mitochondrial metabolites . Second , the low micromolar concentration of free CoA in the cytoplasm has long been suggested to limit cytoplasmic acyl CoA synthetase activities ( Idell-Wenger et al . , 1978 ) , whereas free CoA is less likely to limit pyruvate dehydrogenases that have higher CoA affinities ( Bremer , 1969 ) . Third , it is known that surface membrane proteins , visualized as membrane particles in freeze fracture studies , can cluster and decrease in number ( Frank et al . , 1980 ) in a cellular circumstance known to cause PTP openings , namely during anoxia and reoxygenation of cardiac tissue ( Brenner and Moulin , 2012 ) . Therewith , a working hypothesis is suggested that mitochondria might control palmitoylation of membrane proteins in response to metabolic stress by releasing CoA to the cytoplasm . In principle , this release may occur not only via PTP openings , but by reverse CoA transport during mitochondrial depolarization caused by partial PTP openings , and via membrane defects that might occur during mitochondrial fission and fusion events in metabolic stress ( Knott et al . , 2008; Ong et al . , 2010; Frank et al . , 2012 ) . It is now well established that transient PTP openings occur physiologically , causing transient mitochondrial depolarizations accompanied by superoxide flashes ( Hausenloy et al . , 2004; Saotome et al . , 2009; Zhang et al . , 2013 ) . One widely-held view is that low conductance PTP openings cause mitochondrial depolarization without loss of metabolites , and thereby release excess mitochondrial Ca to the cytoplasm ( Huser and Blatter , 1999; Korge et al . , 2011; Zhou and O’Rourke , 2012 ) . Nevertheless , it appears possible for mitochondria to release metabolites by transient PTP openings without loss of mitochondrial membrane integrity or even global mitochondrial depolarization ( Dumas et al . , 2009 ) . Knockout of cyclophilin D , a major PTP regulator ( Basso et al . , 2005 ) , gives rise to intriguing phenotypes: hearts are mildly protected from ischemia/reperfusion damage , but animals are markedly less tolerant of physical exertion ( swimming ) and shift their energy metabolism from fatty acids to glucose ( Elrod et al . , 2010 ) . From the prevailing interpretation of PTP function , these phenotypes will be the result of mitochondria Ca accumulation , driven by the failure of PTPs to open and release Ca when it accumulates in the matrix space . Our working hypothesis predicts that loss of mitochondrial depolarizations will also favor CoA accumulation by mitochondria and cause a decrease of cytoplasmic free CoA . Consequences of a lower cytoplasmic free CoA concentration might include the shift from fatty acid to glucose metabolism , owing to the different CoA affinities of acyl CoA synthetases and pyruvate dehydrogenases noted above . With this background , we describe here multiple lines of evidence that mitochondrial responses to large Ca transients trigger MEND in BHK cells , that CoA and acyl CoAs are intermediates in the pathway , that membrane protein acylation is required , specifically via the acyl transferase , DHHC5 , ( Li et al . , 2010 ) , that PKCs and oxidative stress promote acyl CoA-dependent endocytosis without PTP openings , and finally that the MEND pathway may contribute significantly to constitutive surface membrane turnover in BHK cells . Using BHK fibroblasts , data presented in Figures 1 and 2 lends initial support to the hypothesis that palmitoylation can be initiated by release of CoA from mitochondria where it is accumulated by membrane potential-driven transporters ( Tahiliani , 1991 ) . Plasmalemma surface area is monitored as membrane electrical capacitance ( Cm ) via patch clamp of BHK cells that express cardiac Na/Ca exchangers , NCX1 ( Linck et al . , 1998 ) . Potassium-free solutions are employed to block clathrin-dependent endocytosis ( Altankov and Grinnell , 1995 ) and to minimize membrane conductance . Effects of square wave voltage perturbations ( 0 . 1–0 . 5 kHz ) to determine Cm are removed digitally from current records . 10 . 7554/eLife . 01293 . 003Figure 1 . MEND is initiated by mitochondria . ( A ) Records of membrane current ( above ) and electrical membrane capacitance ( below ) during the Standard MEND protocol in BHK fibroblasts expressing cardiac Na/Ca exchangers ( NCX1 ) . Cells are opened in Ca-free extracellular solution with 0 . 5 mM EGTA , using a cytoplasmic ( pipette ) solution containing 40 mM cytoplasmic Na , 8 mM MgATP and 0 . 2 mM GTP . Thereafter , Ca influx via Na/Ca exchange is activated by extracellular application of 3 mM Ca for 10 s . During Ca influx , outward membrane current reflects 3Na/1 Ca exchange . Membrane area ( i . e . , Cm ) increases by 30% as a result of exocytosis ( i . e . , fusion of vesicles to the cell surface ) . After terminating Ca influx , membrane area is stable for nearly 60 s and then declines over 2 min by 70% as plasmalemma is internalized via endocytosis . ( B ) Composite results implicating a role for mitochondria in the initiation of MEND . From left to right , bar graphs give results for Standard MEND ( CTR , black and white ) , Standard MEND without cytoplasmic Pi ( purple ) , with the mitochondrial Ca uptake inhibitor , RU323 ( 20 μM , blue ) , with the PTP blockers , cyclosporine A ( 5 μM , green ) and NIM811 ( 2 μM , orange ) , after PKC activation by OAG ( 15 μM , yellow ) , and after rapid perfusion of a mitochondrial uncoupler , CCCP ( 20 μM ) with an ATP synthetase inhibitor , oligomycin ( OM , 5 μM , red ) . n > 6 in all panels . MEND was quantified as fractional decrease of Cm from point 2 to point 3 , and stars indicating significance have their usual meanings . DOI: http://dx . doi . org/10 . 7554/eLife . 01293 . 00310 . 7554/eLife . 01293 . 004Figure 2 . Metabolites that promote PTP openings rapidly initiate MEND . Experiments were started with Pi-free cytoplasmic solution . Subsequently , rapid internal perfusion of cells was initiated via a silica mico-capillary tube whose orifice was placed within 50 μm of the patch pipette orifice . KSP solution was prepared by replacing 80 mM NMG and 20 mM TEA in the standard internal solution with 100 mM K . In addition , 5 mM succinate and 1 mM Pi were added , and free Ca was buffered to 0 . 25 μM with 2 mM EGTA . ( A ) The upper record illustrates MEND responses recorded at 37°C upon cytoplasmic perfusion of KSP solution , and the lower record illustrates the failure of MEND to occur at 22°C . ( B ) Composite results for experiments in which the plasmalemma patch within the patch pipette was ruptured by suction ( i . e . , the cell was ‘opened’ ) at 22°C with KSP solution in the pipette , and the Standard MEND protocol was performed 1 to 2 min later at 37°C . n > 6 for all results . DOI: http://dx . doi . org/10 . 7554/eLife . 01293 . 004 The ‘Standard MEND’ protocol is shown in Figure 1A . Patch clamp is established in a Ca-free extracellular solution using a cytoplasmic solution that contains a high concentration of Na ( 40 mM ) , ATP ( 8 mM ) and GTP ( 0 . 2 mM ) . Then , cells are moved into one of multiple parallel flowing solution streams at 37°C , and Ca influx is activated by switching extracellular Ca from 0 to 2 mM Ca for 12 s . Membrane current ( upper trace ) , which reflects 3Na/1Ca exchange ( Yaradanakul et al . , 2008 ) , allows calculation of the absolute quantity of Ca entering cells . Since BHK cells are round after removal from dishes , we can also calculate cell volume from cell diameter ( 25–35 μm ) and determine the absolute change of Ca occurring in cells . In the experiment of Figure 1A , the total cellular Ca concentration increases by 2 . 1 mM , and free Ca typically exceeds 50 μM in these experiments ( Lariccia et al . , 2011 ) . During Ca influx , membrane area ( i . e . , Cm , lower trace ) increases by 35% as a result of the fusion of subplasmalemmal membrane vesicles to the plasmalemma ( exocytosis ) . Then , after a delay of 30–120 s , membrane area begins to decline as a result of endocytosis until more than 50% of the plasmalemma is removed into vesicles just beneath the cell surface ( Lariccia et al . , 2011 ) . As indicated below the experimental traces in Figure 1A , we quantified membrane area in relation to initial area at three points; just before Ca influx ( 1 ) , after exocytosis ( 2 ) , and after the MEND response ( 3 ) . These values are subsequently presented as sets of three bar graphs . MEND responses , defined as the fractional decrease of membrane area between points 2 and 3 , were compared only if Na/Ca exchange currents were not statistically different and exocytic responses in different groups were also similar . The first set of bar graphs in Figure 1B quantifies the Standard MEND response ( Control , CTR; black and white ) . The second bar set ( purple ) shows that removal of inorganic phosphate ( Pi , 1 mM ) from the standard cytoplasmic solution reduces MEND responses to less than 30% of peak membrane area , as expected from the activating effect of Pi on PTPs ( Massari , 1996 ) . The third data set ( blue ) shows that the mitochondrial Ca uniporter inhibitor , RU323 ( Ying et al . , 1991 ) ( 15 μM ) , strongly blocks MEND . Subsequently , it is shown that the nonspecific PTP inhibitor , cyclosporine A ( Crompton et al . , 1988 ) ( CsA , 5 μM in cytoplasmic and extracellular solutions; green ) and the PTP/cyclophilin D-specific cyclosporine , NIM811 ( N-methyl-4-isoleucine cyclosporine ) ( 2 μM; orange ) ( Waldmeier et al . , 2002 ) , also strongly inhibit MEND . Importantly , the calcineurin inhibitor , FK506 , had no effect on MEND , even at a high concentration ( Lariccia et al . , 2011 ) . Activation of protein kinase C epsilon ( PKCε ) is reported to protect cells from oxidative damage by inhibiting PTP openings ( Baines et al . , 2003; Budas and Mochly-Rosen , 2007 ) . Therefore , we incubated cells with a PKC activator , 1-oleoyl-2-acetyl-sn-glycerol ( OAG , 15 μM ) , for 20 min before experiments , and Standard MEND was effectively blocked ( yellow ) . Next , we performed experiments to induce rapid mitochondrial depolarization , which may be expected ‘initially’ to promote PTP openings ( Scorrano et al . , 1997 ) , as well as to cause both reverse Ca transport ( Montero et al . , 2001 ) and reverse CoA transport ( Tahiliani , 1991 ) from mitochondria . To do so , we included the mitochondrial uncoupler , CCCP ( 20 μM ) , and the ATP synthetase inhibitor , oligomycin ( OM , 5 μM ) , in the pipette and opened cells only after placing them in a flowing solution at 37°C . As indicated by the solid red bar graph for data point ‘1’ ( before the Ca transient ) , cells underwent 30% MEND responses within 1 min after rupturing the membrane in the patch pipette ( i . e . , ‘opening’ the cell to the patch pipette ) . Thereafter , Ca influx initiated supernormal exocytic responses , from point ‘1’ ( before ) to ‘2’ ( peak ) , but MEND was fully suppressed . Thus , while rapid mitochondrial depolarization with these agents can activate single MEND responses with very little delay , cells subsequently are refractory to MEND for the duration of patch clamp experiments ( ∼30 min ) . We report lastly in this connection an experiment set in which we tested for inhibition of MEND by an antioxidant that becomes targeted to mitochondria , Mito-Tempo ( Jiang et al . , 2009 ) . When included in the pipette solution at 10 μM , average MEND responses were reduced from 42 ± 7% ( n = 10 ) to 27 ± 6% ( n = 8 ) , but the difference was not significant . Figure 2 describes the initiation of MEND by perfusing the cytoplasm of cells rapidly with metabolic substrates that optimally open PTPs in isolated mitochondria at submicromolar free Ca ( Massari , 1996 ) . To do so , we place the end of a micro-capillary tube within 50 μm of the orifice of the patch pipette opening . Using positive pressure to generate rapid solution flow , diffusible molecules are exchanged between the cytoplasm and the pipette tip within 2–5 s ( Hilgemann and Lu , 1998 ) . As described in Figure 2A , we changed the standard potassium-free cytoplasmic solution as rapidly as possible to one with PTP-promoting constituents ( ‘KSP’ solution ) , namely succinate ( 5 mM ) , Pi ( 1 mM ) , free Ca set to 0 . 25 μM with 2 mM EGTA , and potassium ( 100 mM ) . After a delay of several seconds , MEND responses occurred over a time course of somewhat less that 1 min and amounted to 34 ± 6% of initial cell area at 37°C . As shown by the lower trace in Figure 2A , the same protocol induced no response at 22°C . Nevertheless , Figure 2B shows that Standard MEND responses at 37°C were negligible in cells that were initially opened at 22°C with KSP solution ( potassium/succinate/phosphate-containing solution ) in the pipette . Thus , KSP solution , like mitochondrial depolarization , induces a MEND-refractory state within at most 1 min , even when no MEND occurs at 22°C . Given that cytoplasmic solutes exchange with pipette solutions within a few seconds , this result suggests that metabolites required for MEND might be released and lost into the patch pipette during cytoplasmic application of KSP solution . Initial evidence that MEND requires generation of acyl CoAs and palmitoylation is presented in Figure 3 . The first data set ( black and white ) shows that Standard MEND is normal in cells that were incubated for 1 hr with 1:1 palmitate-loaded albumin ( 50 μM; ‘Alb+FA’ ) . The second data set ( purple ) shows that MEND is suppressed in cells that were incubated with fatty acid ( FA ) -free albumin ( ‘Alb-FA’ , 50 μM ) for 1 hr before experiments to deplete cellular fatty acids . The next data sets show that MEND is inhibited 70% by the palmitoylation inhibitor , 2-bromopalmitate ( BP , 50 μM; blue ) ( Jennings et al . , 2009 ) , is decreased 50% by the acyl CoA synthetase inhibitor , triacsin C ( TrC , 2 μM; green ) ( Omura et al . , 1986 ) when it is applied acutely in both the pipette and extracellular solutions , and MEND is fully blocked by acute application of both bromopalmitate and triascin C ( orange ) . Thus , both synthesis of acyl CoAs and palmitoylation may be required in the reaction pathway leading to MEND . 10 . 7554/eLife . 01293 . 005Figure 3 . MEND is blocked by preventing protein palmitoylation reactions . ( A ) From left to right , composite results for Standard MEND after incubating cells with 1:1 palmitate-loaded albumin ( 50 μM ) for 1 hr ( black ) , after incubating cells with fatty acid-free albumin ( Alb , 50 μM ) for 1 hr ( purple ) , with the palmitoylation inhibitor , bromopalmitate ( BP , 50 μM ) included in all solutions ( blue ) , with the acyl CoA synthetase inhibitor , Triascin C ( TrC , 2 μM ) in all solutions ( green ) , with BP ( 50 μM ) and TrC ( 2 μM ) in all solutions ( orange ) , with cytoplasmic acetate ( Ace , 6 mM; pink ) , with acetate ( 6 mM ) and acetyl CoA synthetase ( ACS , 20 μM; yellow ) , and with a high cytoplasmic CoA concentration ( 3 mM ) to block DHHCs ( red ) . For all results , n > 6 . ( B ) CoA inhibition of DHHC2-mediated palmitoylation of the N-terminus of myristoylated lymphocyte-specific kinase ( myrLckNT ) . For details , see ‘Materials and methods’ . DOI: http://dx . doi . org/10 . 7554/eLife . 01293 . 005 In subsequent experiments , we examined two opposed ways in which CoA can be predicted to modulate the MEND pathway , assuming that CoA is first used to generate acyl CoA from fatty acid and subsequently is released during palmitoylation of membrane proteins . First , we tested whether free CoA is in fact an intermediate in the pathway . To do so , we perfused cells with a high concentration of a CoA-metabolizing enzyme , namely human acetyl CoA synthetase ( ACS , 20 μM; Kd for CoA , 11 μM [Luong et al . , 2000] ) with acetate ( Ace , 6 mM ) to rapidly convert CoA to acetyl CoA . Acetate itself did not affect MEND ( pink ) , but acetate together with acetyl CoA synthetase blocked MEND ( yellow ) . Second , we tested whether a very high concentration of CoA ( 3 mM ) would inhibit MEND , as expected for product inhibition of acyl transferase reaction that releases free CoA . As shown in the last data set , CoA at a concentration of 3 mM strongly inhibited MEND ( red ) , and Figure 3B documents that acyl transferase activity of the DHHC2 transferase is effectively blocked by 3 mM CoA . Inhibition occurs with a Ki of about 0 . 2 mM , and activity is reduced by more than 90% with 3 mM CoA . Figure 4 demonstrates that MEND blockade via the mitochondrial mechanisms just described is entirely overcome by perfusion of myristoyl CoA ( mCoA ) or CoA into cells . We employed mCoA instead of palmitoyl CoA because its lower affinity for membranes is advantageous to achieve rapid cytoplasmic concentration changes via the pipette perfusion technique . In Figure 4A , cyclosporine A ( 3 μM ) was used to block MEND . Ca influx caused a 35% exocytic response , as usual , and membrane area ( Cm ) was then stable until mCoA ( 15 μM ) was perfused into the cell . Within seconds , membrane area began to fall , and within 2 min nearly 70% of the cell surface was removed by endocytosis . Figure 4B shows composite data for mCoA release of MEND block by cyclosporine A ( black and white ) , followed by results for release of MEND block by CCCP and oligomycin ( purple ) , KSP-induced MEND block ( blue ) , and OAG-induced MEND block ( green ) . The last two data sets in Figure 4B show that perfusion of CoA ( 20 μM ) , instead of mCoA , overcomes cyclosporine A MEND block equally well as mCoA ( yellow ) , and finally that MEND block by a high concentration of CoA ( 3 mM ) is immediately relieved by its washout from cells ( red ) , as expected for a simple product inhibition mechanism . 10 . 7554/eLife . 01293 . 006Figure 4 . Cytoplasmic perfusion of acyl CoA or CoA circumvents four mitochondrial MEND blocks . ( A ) Typical experiment in the presence of cyclosporine A ( CsA , 3 μM ) to block MEND . Ca influx cause a 40% increase of Cm via exocytosis , and Cm remains stable thereafter for minutes . Cytoplasmic perfusion of myristoyl CoA ( mCoA , 15 μM ) via the micro-capillary within the patch pipette causes a MEND response that begins within 10 s and internalizes 70% of the plasmalemma within 2 . 5 min . ( B ) Composite results quantifying MEND that occurs when mCoA or CoA is perfused into the cytoplasm of cells in which MEND has been blocked by interventions acting on mitochondria . From left to right , bar graphs present results for pipette perfusion of mCoA ( 15 μM ) in cells in which MEND was blocked by cyclosporine A ( CsA , 3 μM; black and white ) , by opening cells at 22°C with CCCP ( 20 μM ) and oligomycin ( 5 μM; purple ) , by opening cells at 22°C with KSP ( blue ) , and by pretreatment of cells with OAG ( 15 μM ) for 30 min ( green ) . The penultimate results quantify MEND caused by pipette perfusion of CoA ( 20 μM ) into cyclosporine A-blocked cells ( yellow ) . The final data set shows results for cells in which MEND was blocked by a high cytoplasmic CoA concentration ( 3 mM ) . Ca influx caused on average 38% exocytic responses , and Cm was then stable . When CoA was perfused out to the cytoplasm ( ‘wash-out’ ) , endocytosis started within 15 s and amounted to 56% of the plasmalemma on average . For all results , n > 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 01293 . 006 We next address the role of DHHC acyl transferases that can mediate surface membrane protein palmitoylation in MEND . From more than 20 DHHC transferases ( Mitchell et al . , 2006 ) only DHHC2 and DHHC5 are known at this time to traffic to and become active at the plasmalemma ( Greaves et al . , 2011; Li et al . , 2011; Thomas et al . , 2012 ) , and DHHC5 may be more ubiquitous ( Li et al . , 2011 ) . Using siRNA for DHHC5 with lipofectamine2000 transfection , Figure 5A documents in Western blots that DHHC5 expression was decreased by >80% in BHK cells . However , the magnitudes of Standard MEND responses became variable after transfection protocols using either lipofectamine2000 or RNAmax . We were able to overcome this variability and generate even larger MEND responses by incubating BHK cells during the entire siRNA protocol with a low concentration of a PKC inhibitor , staurosporine ( STS , 0 . 1 μM ) , that prevents PKC-induced inhibition of PTPs ( Ytrehus et al . , 1994 ) . Cells were returned to STS-free solutions 5 min before experiments . As shown in Figure 5B using LF2000 to transfect cells , exocytic responses in control transfected cells were larger than normal , and subsequent MEND responses removed fully 75% of the cell surface . In cells transfected with siRNA for DHHC5 ( right data set in Figure 5B ) , these MEND responses were decreased on average to 25% of the cell surface , corresponding to a 66% decrease of the MEND response . As shown in Figure 5C , siRNA of DHHC5 also strongly blocked MEND responses induced by pipette perfusion of mCoA into cells that were MEND-blocked by cyclosporine A ( 3 μM ) . Figure 5D quantifies KSP-induced MEND in BHK cells that were not STS-treated . Accordingly , KSP-induced MEND in cells transfected with scrambled DHHC5 siRNA amounts to only 14% . Nevertheless , the responses were reproducible , and they were highly significantly reduced to 2 . 5% by DHHC5 siRNA transfection . 10 . 7554/eLife . 01293 . 007Figure 5 . The acyl tranferase , DHHC5 , is required for MEND in BHK cells . ( A ) DHHC5 knockdown by siRNA to ZDHHC5 . BHK cells were transfected with control siRNA or ZDHHC5-specific siRNA at indicated concentrations using Lipofectamine 2000 . 72 hrs after transfection , cells were harvested and cell lysates were processed for Western blot analysis using 20 µg protein per lane . A human non-small lung carcinoma cell line , H1299 , was used as a positive control for anti-ZDHHC5 antibody . ( B ) MEND in BHK cells amounts to 75% of the cell surface after growing cells with staurosporin ( 0 . 1 μM ) , and DHHC5 siRNA decreases MEND by 63% ( p<0 . 01 ) . ( C ) mCoA-relief of MEND block by cyclosporine A ( 3 μM ) is reduced by 75% ( p<0 . 05 ) in BHK cells transfected with siRNA for DHHC5 vs scrambled siRNA . ( D ) KSP-induced MEND in BHK cells is reduced from 14 to 2 . 5% by DHHC5 siRNA transfection using RNAmax ( p<0 . 01 ) . For all results , n > 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 01293 . 007 As established in Figures 3 and 4 , the generation of acyl CoA is required for the occurrence of MEND in these experiments . As shown in Figure 6A , however , the introduction of acyl CoA ( mCoA ) into cells is not sufficient to initiate MEND . Under standard conditions , cytoplasmic perfusion of mCoA ( 15 μM ) causes only 5 to , at most , 10% endocytic responses over 10 min ( n > 20; green square in Figure 6A ) . Therefore , Ca must be acting by at least one additional mechanism for MEND to occur . From many possibilities , we describe in Figure 6 that two common Ca-dependent cell signals , activation of PKCs and generation of reactive oxygen ( ROS ) , both can promote MEND in the presence of acyl CoA and in the absence of Ca transients . 10 . 7554/eLife . 01293 . 008Figure 6 . Activation of acyl CoA-dependent MEND without Ca transients . ( A ) Pipette perfusion of mCoA ( 15 μM ) without previously activating a Ca transient causes little or no endocytosis over 5 min ( green square ) . Extracellular application of OAG ( 15 μM ) for 1 to 2 min also has little or no effect , even after 6 min . With mCoA ( 15 μM ) in the cytoplsamic solution , OAG initiates a decrease of membrane area that continues for several minutes after OAG is removed , amounting to 24% on average after 6 min . ( B ) Application of H2O2 ( 80 μM ) for 4 min has little or no effect over 15 min in the absence of mCoA . With mCoA ( 15 μM ) in the cytoplasmic solution , however , a large decrease of membrane area occurs when H2O2 is removed , amounting to 26% on average after 6 min . ( C ) MEND responses quantified as percent decrease of membrane area per min over 4 min . From left to right , the bar graphs quantify MEND caused by OAG , inhibition of OAG-induced MEND by PKC peptide 19–36 ( 1 μM in the pipette ) , MEND caused by H2O2 , lack of effect of PKC peptide ( 19–35 ) on H2O2 MEND , and average MEND responses to H2O2 and OAG applied sequentially . ( D ) MEND in T-Rex-293 cells with inducible PLM expression . The bar graphs quantify the percent decrease of membrane area over 4 min after applying OAG . OAG-activated MEND is increased nearly three-fold when PLM expression has been induced for 24 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 01293 . 008 Figure 6A presents results for applying the same PKC activator , OAG , used to inactivate MEND ‘prior’ to experiments in Figure 1 . As shown by the dashed line , application of OAG ( 15 μM ) on its own causes very little or no endocytosis . However , when OAG ( 15 μM ) is applied for 1 . 5 min in the presence of mCoA ( 15 μM ) in the pipette , membrane area begins to decrease with a delay of about 30 s and continues to decrease for several minutes after removal of OAG . As shown in Figure 6B , application of H2O2 ( hydrogen peroxide ) ( 80 μM ) for 4 min has little or no effect in the absence of mCoA in the cytoplasm . In the presence of mCoA ( 15 μM ) , application of H2O2 causes little or no change initially , but upon its removal membrane area begins to decline robustly within 1 min and on average more than 25% of the cell surface is lost over 10 min . This pattern suggests that ROS have an acute inhibitory effect on MEND , which is rapidly relieved when H2O2 is removed . The stimulatory effect of ROS on MEND , however , is a long-lived effect . The results suggest that transient generation of ROS during the MEND protocol promotes the final steps of MEND , subsequent to generation of acyl CoA . In Figure 6C we provide further details about the PKC and H2O2 effects , quantifying the occurrence of MEND as the average rate of membrane loss over 4 min . MEND responses promoted by OAG are strongly inhibited by a peptide substrate inhibitor of conventional PKCs ( PKC peptide ( 19–36 ) , 1 μM ) , confirming that conventional PKCs are involved ( House and Kemp , 1987 ) . By contrast , the induction of MEND by H2O2 is not inhibited by the PKC peptide , indicating that H2O2 is not acting through the PKCs that can promote MEND . Finally , we show that MEND activation by H2O2 and OAG are not additive; the activation of MEND by one reagent precludes a further effect by the other . It is beyond the scope of this article to resolve how PKCs and H2O2 promote MEND . However , one possibility raised by the literature is that phosphorylation of some DHHC protein substrates by PKCs can be permissive for their subsequent palmitoylation . This is the case for the dually palmitoylated , Na/K pump regulatory protein , phospholemman ( PLM ) ( Tulloch et al . , 2011 ) , which is expressed preferentially in muscle ( Geering , 2006 ) . With this background , we next generated a T-REx-293 cell line in which expression of PLM can be activated in a tetracycline-dependent manner . As shown in Figure 6D , we quantified the rate at which MEND occurred over 4 min after a 1 min activation of PKCs by OAG ( 15 μM ) in the presence of cytoplasmic mCoA ( 15 μM ) . The overexpression of PLM caused a three-fold increase in the rate of PKC/mCoA-dependent MEND . This result clearly supports the idea that MEND can be promoted by the presence of proteins that can be palmitoylated and therefore is a ‘cargo-dependent’ form of endocytosis . At this time , we can only speculate that long-term protein modifications caused by transient oxidative stress also promote the availability of membrane protein palmitoylation sites . Given that MEND can be activated without large Ca transients , the question is raised whether the MEND pathway may be constitutively active at the cytoplasmic acyl CoA concentrations occurring normally in cells . We describe in Figure 7 experiments that lend initial support for this possibility by analyzing surface membrane changes in response to treating BHK cells with the putatively specific cyclosporine D/PTP inhibitor , NIM811 ( 2 μM ) . As shown in Figure 7A , BHK cells are essentially round after removal from dishes . From micrographs , we measured the diameter of each cell in both the ‘X’ and ‘Y’ directions , calculated the spherical area of each cell , and determined the Cm of each cell after opening it via patch clamp . In the first group of experiments , BHK cells were removed from dishes and were incubated for 1 or 3 hr , either with or without NIM811 at 37°C . Results for control cells were pooled , giving three groups of measurements . The average diameters of cells employed ( 28 . 2 ± 0 . 9 , 27 . 8 ± 1 . 0 μm , and 29 . 9 ± 1 . 4 μm ) were very similar . Assuming a specific membrane capacitance of 1 pF per 100 μm2 , we calculated for each cell the ratio of surface membrane area to the area of a sphere with the cell diameter . For control cells , a ratio of 2 . 2 ± 0 . 1 was obtained , indicating that surface membrane undulations account for one-half of the cell surface . After 1 hr of NIM811 treatment this ratio was not changed , but the ratio increased significantly to 2 . 6 ± 0 . 1 after 3 hr of incubation . In the second set of experiments , we grew BHK cells normally for 28 hr in the presence of NIM811 ( 2 μM ) and carried out equivalent measurements after removing cells from dishes . Membrane area in cells of equal size was increased on average by 27% ( p<0 . 01 ) . In summary , treatment of BHK cells with NIM811 causes a significant increase of membrane area after a delay of 1 hr , reaching a plateau within about 3 hr , indicating ‘either’ that membrane insertion at the cell surface is increased ‘or’ that membrane removal from the cell surface is decreased , as expected for partial inhibition of endocytosis . 10 . 7554/eLife . 01293 . 009Figure 7 . NIM811 increases surface area of cultured BHK cells . ( A ) Micrograph of BHK cell during patch clamp . 25× LWD lens . Cell diameters were calculated as the average of a horizontal and vertical line transecting each cell , as indicated . The spherical area of each cell was calculated , and the ratio of the electrically determined area to the spherical area was calculated assuming 1pF to be 100 μm2 . ( B ) First three bar graphs give results for BHK cells removed from dishes and incubated at 37°C for 1 or 3 hr without ( CTR ) and with NIM811 ( 2 μM ) . Results for control cells were not significantly different and were pooled . The fourth and fifth bar graphs give results for control cells ( CTR ) and cells grown with NIM811 ( 2 μM ) for 30 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 01293 . 009 That mitochondria are involved in the initiation of MEND is supported by results for six interventions expected to act via mitochondria ( Figures 1 and 2 ) . The following results support the idea that PTP openings are a prerequisite for Ca-activated MEND during the standard protocol: ( 1 ) reproducible MEND , constituting 50% of the plasmalemma , requires cytoplasmic Pi , an activator of PTPs ( Massari , 1996 ) . ( 2 ) MEND is blocked by inhibiting Ca uptake by mitochondria , a mechanism that will stop both mitochondrial depolarization and PTP openings . ( 3 ) MEND is blocked by two cyclosporines , one considered a specific cyclophilin D/PTP inhibitor ( Waldmeier et al . , 2002 ) . Although cyclosporine A also inhibits calcineurin , we determined previously that calcineurin inhibition by FK506 does not inhibit MEND ( Lariccia et al . , 2011 ) . ( 4 ) MEND can be blocked by prior PKC activation , consistent with suggestions that PKCs protect cells from reperfusion injury by inhibiting PTP openings ( Ytrehus et al . , 1994 ) . ( 5 ) MEND occurs when cells are internally perfused with mitochondrial substrates that support PTP openings , namely succinate with Pi in potassium-rich cytoplasmic solution at an optimal free Ca concentration ( 0 . 2 μM , [Massari , 1996] ) . These responses ( Figure 2 ) occur without an initial exocytic phase , indicating that Ca release from mitochondria , if it occurs , does not cause large Ca transients in these experiments . ( 6 ) Finally , we bring to bear original descriptions of PTP openings , monitored via mitochondrial light scattering ( Haworth and Hunter , 1979 ) . Under the conditions of those experiments , the rate at which PTPs open increases over the Ca concentration range of 10–500 μM , but maximal responses are still obtained with 30 μM free Ca in less than 10 s . Free cytoplasmic Ca concentrations in our experiments definitively exceed 50 μM ( Lariccia et al . , 2011 ) , and we activate Ca influx for 10 to 14 s in the Standard MEND protocol . Therefore , it is very likely that full PTP pore openings occur during these experiments . Rapid cytoplasmic application of KSP solution or CCCP with oligomycin in BHK cells induces a MEND-refractory state ( Figure 2 and Figure 1B ) , which we interpret to be the consequence of CoA release via PTP openings with subsequent loss of CoA into the pipette . We admit in this regard that CCCP and oligomycin have not previously been shown to cause transient PTP openings when rapidly introduced into cells . However , depolarization definitively promotes PTP openings ( Scorrano et al . , 1997 ) . In fact , more specific explanations may emerge because recent work indicates that PTPS are formed by dimers of the target of oligomycin , the ATP synthetase ( Giorgio et al . , 2013 ) . Related to these experiments , it is an important question whether mitochondrial depolarization per se causes substantial CoA release via reverse CoA transport . Studies of cardiac mitochondria indicate that reverse CoA transport can decrease matrix CoA with a half-life on the order of 15 min when mitochondria are depolarized with valinomycin ( Tahiliani , 1991 ) . Accordingly , reverse transport of CoA by depolarized mitochondria would release only about 10% of the mitochondrial CoA content within the delay of 1–2 min that precedes the onset of MEND after a Ca transient . This may well be significant in intact cells . However , the fact that KSP solution effectively causes a MEND-refractory state within 1 min at room temperature ( Figure 2 ) is much more consistent with rapid CoA loss via diffusion through a pore than via the reverse transport process . That the pathway to Ca-activated MEND in BHK cells requires generation of acyl CoAs and palmitoylation of proteins at the cell surface is supported by the following evidence: ( 1 ) fatty acid depletion of cells for 1 hr , a specific inhibitor of acyl CoA synthetase activity ( Omura et al . , 1986 ) , and a nonspecific inhibitor of palmitoylation , bromopalmitate , all reduce MEND , and triascin C with bromopalmitate causes complete MEND block . ( 2 ) An acetyl CoA synthetase , when perfused into cells together with acetate to deplete CoA , also fully blocks Ca-activated MEND . ( 3 ) A high concentration of CoA , which fully blocks acylation activity of a DHHC acyl transferase , fully blocks Ca-activated MEND . ( 4 ) When MEND is blocked by four different interventions that probably act through mitochondria , introduction of acyl CoA into the cytoplasm fully restores MEND . ( 5 ) Low concentrations of CoA are equally effective as acyl CoA in promoting MEND in these protocols . ( 6 ) High CoA concentrations are demonstrated to block DHHC acyl transferase activity and to strongly block MEND , and this block is rapidly reversed by perfusion CoA out of the cytoplasm . ( 7 ) Ca-activated MEND , MEND induced by acyl CoA in cyclosporine A-blocked cells , and MEND induced by PTP-promoting substrates ( KSP solution ) are all selectively inhibited by knockdown of the DHHC5 acyl transferase , which is known to be active at the surface membrane ( Li et al . , 2011 ) . As with evidence for mitochondrial involvement , multiple explanations are possible for individual outcomes . However , the results together provide clear evidence that Ca-activated MEND requires acyl CoA synthesis and acyl transferase activity , that is protein palmitoylations , within the time frame in which MEND occurs . While palmitoylation within the secretory pathway clearly can target proteins to the plasmalemma ( Greaves et al . , 2009 ) , the present data sets clearly suggest that progressively greater palmitoylation of surface membrane proteins can promote removal of proteins from the cell surface by endocytosis . Whether or not MEND requires palmitoylation of specific proteins is not addressed by the present data sets . However , we have demonstrated in Figure 6D that the presence of a membrane protein that can be dually palmitoylated in a PKC-dependent manner ( Tulloch et al . , 2011 ) facilitates the occurrence of MEND . This clearly suggests that MEND is a ‘cargo-dependent’ endocytic process . It seems likely that Ca-activated MEND in BHK cells is mechanistically related to endocytosis that occurs during cell wounding ( Tam et al . , 2010 ) . As noted in the Introduction , it is proposed that exocytosis during cell wounding translocates sphingomyelinases to the cell surface , where they subsequently generate ceramide and promote endocytosis ( Tam et al . , 2010 ) . Further , it is proposed that caveolae are internalized in a specific manner ( Corrotte et al . , 2013 ) . From our perspective , it is certain that several different mechanisms can cause MEND ( Lariccia et al . , 2011 ) , and extracellular application of sphingomyelinase C does so very effectively ( Lariccia et al . , 2011 ) when about 50% of cellular sphingomyelin is cleaved ( Van Tiel et al . , 2000 ) . Our tests for involvement of native sphingomyelinases in BHK MEND responses were entirely negative , and we were able to dissociate exocytosis from MEND ( Lariccia et al . , 2011 ) . While caveolae might well become involved in MEND , the loss of 70% of the cell surface during large MEND is inconsistent with the much smaller membrane area represented by caveolae in almost all cells ( Bretscher and Whytock , 1977; Parton and del Pozo , 2013 ) . In our view , therefore , MEND involves the internalization of ordered membrane domains that can form via phase separation from bulk membrane as the result of multiple synergistic cellular processes , one of them being progressively more extensive palmitoylation of membrane proteins that appears to occur in Ca-activated MEND in BHK cells . Constitutive endocytosis in BHK cells has been analyzed rather extensively ( Marsh and Helenius , 1980; Griffiths et al . , 1989 ) . Between one and two percent of the surface membrane is internalized per min , mostly via well-defined coated pits ( Marsh and Helenius , 1980 ) . Nevertheless , the precise quantitation of different endocytic forms does not seem secure . Certainly , endocytic mechanisms that internalize ‘lipid raft’ proteins , in particular GPI-anchored proteins ( Lakhan et al . , 2009 ) , are also well characterized in BHK cells ( Fivaz et al . , 2002; Refaei et al . , 2011 ) . It therefore cannot be discounted that the mechanisms underlying MEND are constitutively active , possibly in dependence on spontaneous mitochondrial depolarizations and superoxide flashes ( Elrod et al . , 2010; Zhang et al . , 2013 ) and consistent with the increase of sarcolemma area caused by NIM811 ( Figure 7 ) . MEND can internalize surface membrane at a rate of 50% per min ( Figure 1 ) , at least 25-times greater than constitutive membrane turnover . Therefore , a constitutive activity of the MEND pathway equal to even one percent of its maximal activity would be a substantial endocytic flux in BHK cells . Independent of mitochondrial involvement , our experiments raise a question whether the distal part of the MEND pathway , that is acyl CoA/palmitoylation-dependent endocytosis , might be a form of ‘lipid raft’-dependent endocytosis that is physiologically activated by PKCs and/or oxidative stress . Clearly , DHHC5 is constitutively functional in cells ( Li et al . , 2011; Thomas et al . , 2012 ) , and both PKC activation and H2O2 promote acyl CoA-dependent endocytosis without the activation of Ca transients ( Figure 6 ) . Endocytosis following PKC activation might in principle reflect the endocyotic process that normally inactivates PKCs by removing PKCs from the cell surface ( Carmena and Sardini , 2007 ) . Although generally ascribed to conventional , ubiquitin-dependent endocytosis , a ‘caveolar’ or ‘lipid raft’ pathway that is blocked by nystatin has also been shown to participate in PKC inactivation ( Leontieva and Black , 2004 ) . Our work suggests that the dually palmitoylated membrane protein , PLM , can be used as a simple model to study PKC-dependent endocytosis in future work , independent of mitochondrial involvement . PLM appears to be palmitoylated in dependence on its phosphorylation by PKCs ( Tulloch et al . , 2011 ) , and expression of PLM promotes acyl CoA/PKC-dependent endocytosis ( Figure 6 ) . Therefore , mutations of PLM to modulate selectively its palmitoylation and phosphorylation status should allow insightful experiments . It appears more challenging to elucidate how transient H2O2 oxidative stress activates endocytosis , although an obvious starting point will be to determine how the palmitoylation status of membrane proteins is changed by transient oxidative stress . In summary , our work lends support to a working hypothesis that mitochondrial signaling impacts the plasmalemma by controlling and/or modulating surface membrane protein palmitoylation and subsequently palmitoylation-dependent endocytosis . One trigger appears to be release of CoA from mitochondria during mitochondrial depolarizations that involve transient PTP openings ( Huser and Blatter , 1999; Korge et al . , 2011 ) . Activation of PKCs prior to a Ca transient blocks MEND , while both PKCs and transient generation of ROS support MEND progression at the cell surface . Accordingly , a wave of PKC activation during MEND will tend to restabilize mitochondria but promote ongoing endocytic events . How MEND is related to different categories of ‘lipid raft’-dependent endocytosis delineated to date ( Lajoie and Nabi , 2007; Mayor and Pagano , 2007 ) remains to be established . It will be of great interest to establish how and if the MEND pathway is related to Ca-activated excessive endocytosis ( Smith and Neher , 1997 ) and bulk endocytosis ( Cousin , 2009 ) in secretory cells . Patch clamp ( Yaradanakul et al . , 2008 ) and cell cultures ( Linck et al . , 1998 ) were as described ( Lariccia et al . , 2011 ) . Effects of voltage pulses ( 0 . 1–0 . 5 kHz ) to determine Cm are filtered out of records . We employed relatively large BHK cells ( 40–120 pF ) because MEND was more reliable in that subpopulation . A T-REx-293 cell line with inducible PLM expression was prepared as follows: PLM was excised from pAdTrack-CMV-PLM ( a gift from Lois Carl , Pennsylvania State University ) and then inserted into pcDNA5/FRT/TO vector ( Life technologies ) using EcoRV and HindIII restriction sites . Flp-In T-REx 293 cells were transfected with pcDNA5/FRT/TO-PLM and pOG44 and then selected with hygromycin to generate cells expressing PLM in the tetracycline inducible system . Standard MEND solutions minimize all currents other than NCX1 current . Extracellular solution contained in mM: 120 n-methyl-d-glucamine ( NMG ) , 4 MgCl2 ± 2 CaCl2 , 0 . 5 EGTA , 20 TEA-OH , 10 HEPES , pH 7 . 0 with aspartate . Cytoplasmic solution contained in mM: 75 NMG , 20 TEA-OH , 15 HEPES , 40 NaOH , 0 . 5 MgCl2 , 0 . 8 EGTA , 0 . 25 CaCl2 , 1 Pi set to pH 7 . 0 with aspartate . Unless stated otherwise , 8 mM MgATP , 2 mM TrisATP , and 0 . 2 mM GTP were employed in cytoplasmic solutions with a free Mg of 0 . 5 mM . KSP cytoplasmic solution contained 110 KOH , 40 NaOH , 10 histidine , 2 . 0 EGTA , 0 . 45 CaCl2 , nucleotides as just given , and pH 7 . 0 with aspartate . Bath solution in NIC recording and FITC-dextan uptake experiments contained in mM: 120 NaCl , 5 KCl , 0 . 5 NaHPO4 , 0 . 5 MgCl2 , 1 . 5 CaCl2 , 15 histidine , and 15 glucose . During anoxia , 5 mM deoxyglucose was substituted for glucose , 35 mM KCl was added to stop spontaneous activity , 50 μM FA-free albumin with 50 μM myristate was added , and solution was degassed by stirring under vacuum . Unless specified otherwise , reagents were from Sigma-Aldrich . NIM811 was a gift of Novartis Pharmaceutical , Basal . Unless stated otherwise , error bars represent standard error of 6 and usually 8 or more observations . Significance was assessed by Student’s t-test or , in rare cases of unequal variance , by the Mann–Whitney Rank Sum test . In all figures , ‘*’ denotes p<0 . 05 , ‘**’ denotes p<0 . 01 , and ‘***’ denotes p<0 . 001 . Lipofectamine 2000 or RNAiMax transfection reagents were used according to the manufacturer’s instructions . Briefly , cells were plated 1 day before transfection in 12-well plates at 2–4 × 105 cells per well without antibiotics . A non-targeting siRNA ( # D-001206-14-05; Thermo Scientific , Waltham , MA ) was used as the negative control . A set of three Sigma pre-designed Mission siRNAs were used for ZDHHC5 gene silencing: #1 , sense , CGUUACACAGGGUUGCGAAdTdT , anti-sense , UUCGCAACCCUGUGUAACGdTdT; #2 , sense , GAUAGUAGCUUAUUGGCCAdTdT , anti-sense , UGGCCAAUAAGCUACUAUCdTdT; #3 , sense , GUUUCAGAUGGGCAGAUAAdTdT , anti-sense , UUAUCUGCCCAUCUGAAACdTdT . Control or specific siRNAs were diluted in Opti-MEM ( Invitrogen; Grand Island , NY ) for 5 min , and Lipofectamine 2000 ( Invitrogen ) was also diluted in Opti-MEM at room at 23°C for 5 min . Then the diluted siRNA and Lipofectamin 2000 were mixed and incubated for 30 min at 23°C . The siRNA-Lipofectamine 2000 complexes were added to the cells and incubated for 48–72 hr . When employed , STS ( 0 . 1 μM; Santa Cruz Biotechechnology , Dallas , TX , # sc-3510 ) was present throughout the incubation but not during experiments . Effectiveness of siRNA was assessed after 72 hr as demonstrated in Figure 5A . Purification of reagents and methods were as described ( Jennings et al . , 2009 ) . Briefly , CoA solutions were diluted into reactions to the indicated final concentrations of CoA along with 1 µM myrLckNT , 0 . 85 µM [3H]-palmitoyl-CoA , and 10 nM DHHC2 . Reactions were incubated 5 min at 25°C before stopping with 5X sample buffer . Reactions were resolved by SDS-PAGE and either the myrLckNT bands were excised from gels and treated for scintillation counting ( Figure 3B , above ) or gels were treated for fluorography and exposed to film ( Figure 3B , below ) . CoA inhibited incorporation of [3H]-palmitate into both DHHC2 and myrLckNT .

Cells use a process called endocytosis to absorb proteins and other molecules . There are many forms of endocytosis , but they usually involve the molecule of interest becoming tucked into a bud that forms in the cell membrane . This bud is then pinched off to leave the molecule inside a vesicle that is inside the cell . In general endocytosis is triggered by ‘caging’ proteins such as clathrin , but other forms are also possible . These “non-classical” forms of endocytosis are involved in processes as diverse as the internalization of pathogens and the response of cells to wounding , and sometimes they involve large fractions of the cell membrane being pinched off . Several different forms of “massive endocytosis” have been observed , but they have remained enigmatic in comparison to the classical forms of endocytosis . Now Hilgemann et al . report a new pathway for massive endocytosis that is triggered by a sudden influx of calcium ions into Baby Hamster Kidney ( BHK ) fibroblasts . Up to 70% of the cell membrane can be pinched off during this process , especially areas of the membrane in which lipids and proteins are arranged in a more ordered pattern than in the average cell membrane . After massive endocytosis , it takes BHK cells about 30 minutes to replace the regions that were lost during the endocytosis . Hilgemann et al . find that calcium ions exert their influence via mitochondria , which are the primary source of energy for most cells . In contrast to all other cell organelles , the mitochondria are surrounded by two concentric membranes . The influx of calcium ions causes pores in the inner membrane of the mitochondria , called permeability transition pores , to open so that coenzyme A , a small molecule that is required for fatty acid metabolism , is released into the cytoplasm of the cell . This is followed by the condensation of coenzyme A with a fatty acid and the attachment of fatty acids to the surface membrane proteins . The attachment of these fatty acids ( a process known as palmitoylation ) evidently promotes ordered regions of the cell surface to coalesce and be pinched off into the cytoplasm as membrane vesicles . A key unanswered question is whether the release of coenzyme A by mitochondria regulates biochemical processes in addition to endocytosis .

2013

Massive endocytosis triggered by surface membrane palmitoylation under mitochondrial control in BHK fibroblasts

The receptor tyrosine kinase ( RTK ) AXL is induced in response to type I interferons ( IFNs ) and limits their production through a negative feedback loop . Enhanced production of type I IFNs in Axl-/- dendritic cells ( DCs ) in vitro have led to speculation that inhibition of AXL would promote antiviral responses . Notwithstanding , type I IFNs also exert potent immunosuppressive functions . Here we demonstrate that ablation of AXL enhances the susceptibility to infection by influenza A virus and West Nile virus . The increased type I IFN response in Axl-/- mice was associated with diminished DC maturation , reduced production of IL-1β , and defective antiviral T cell immunity . Blockade of type I IFN receptor or administration of IL-1β to Axl-/- mice restored the antiviral adaptive response and control of infection . Our results demonstrate that AXL is essential for limiting the immunosuppressive effects of type I IFNs and enabling the induction of protective antiviral adaptive immunity . AXL is a member of the TAM ( TYRO3 , AXL , and MERTK ) subfamily of RTK that potently inhibits the production of type I IFNs ( Bhattacharyya et al . , 2013; Rothlin et al . , 2015; Rothlin et al . , 2007; Zagorska et al . , 2014 ) . In DCs , AXL is an IFN-stimulated gene ( ISG ) and hijacks molecular components of type I IFN signaling to induce the expression of Suppressor of Cytokine Signaling ( SOCS ) 1 and SOCS3 ( Rothlin et al . , 2007 ) . SOCS1 and SOCS3 , in turn , downregulate type I IFN signaling . Therefore , AXL is a key component of a homeostatic mechanism that controls type I IFN levels . Recent studies using an array of enveloped viruses have identified AXL as an enhancer of infection in vitro , including in DCs ( Bhattacharyya et al . , 2013; Meertens et al . , 2012; Morizono et al . , 2011; Shimojima et al . , 2007; 2012 ) . Enveloped viruses exploit apoptotic mimicry by exposing phosphatidylserine on their lipid envelopes . Binding of phosphatidylserine to the AXL agonist growth arrest-specific 6 ( GAS6 ) protein leads to the activation of AXL on host cells ( Bhattacharyya et al . , 2013; Lew et al . , 2014 ) . Activation of AXL through viral apoptotic mimicry leads to the induction of the Socs genes and the suppression of type I IFN production and signaling ( Bhattacharyya et al . , 2013 ) . It was also shown that the non-enveloped virus SV40 can engage AXL directly by structural mimicry to facilitate infection ( Drayman et al . , 2013 ) . Type I IFNs were identified based on their ability to inhibit the propagation of viruses ( Isaacs and Lindenmann , 1957; Taniguchi et al . , 1980 ) . Accordingly , genetic ablation of Axl resulted in an enhanced production and signaling of type I IFN during viral infection of cells in vitro and increased the resistance of DCs to the virus ( Bhattacharyya et al . , 2013 ) . These studies have speculated that disabling AXL RTK function might have potent antiviral activity in vivo ( Bhattacharyya et al . , 2013; Meertens et al . , 2012; Morizono et al . , 2011; Shimojima et al . , 2007; 2012 ) . Type I IFNs also mediate a vast array of immunoregulatory functions ( McNab et al . , 2015 ) . For example , sustained production of type I IFNs during chronic lymphocytic choriomeningitis ( LCMV ) infection inhibited the generation of virus-specific T cells and prevented viral clearance ( Teijaro et al . , 2013; Wilson et al . , 2013 ) . Similar detrimental effects of type I IFNs have been described during bacterial infections . In particular , type I IFNs inhibit protective cell-intrinsic responses against intracellular bacteria , including Mycobacterium tuberculosis ( Mayer-Barber et al . , 2010; 2011 ) . Additionally , immunosuppressive effects of type I IFNs may underlie their pharmacological efficacy in the treatment of multiple sclerosis ( Prinz et al . , 2008 ) . Given the contrasting immunosuppressive and antiviral functions of type I IFNs , we sought to directly test whether disabling AXL RTK signaling indeed translates into increased resistance to viral infection in vivo . Unexpectedly , Axl-/-mice were more susceptible than WT mice to influenza A virus ( IAV ) infection . This enhanced susceptibility correlated with reduced maturation of DCs and deficient induction of antiviral T cell responses . A similar impairment in inducing an effective adaptive T cell response in Axl-/- mice was detected during infection with the unrelated neurotropic West Nile virus ( WNV ) . The failure to engage antiviral adaptive immunity could be ascribed to increased type I IFN and the associated reduction in IL-1β production in infected Axl-/-mice . Neutralization of type I IFN function restored the production of IL-1β in infected Axl-/- DCs and rescued the capacity of Axl-/- mice to induce the protective antiviral adaptive immune response and resist IAV infection . Similarly , delivery of IL-1β restored antiviral adaptive immunity in Axl-/- mice and survival to IAV infection . In summary , our studies underscore the function of AXL in calibrating the antiviral versus the immunosuppressive functions of type I IFNs during viral infection . To better understand the function of AXL during the course of IAV infection in vivo , mice were challenged with 10 PFU of A/Puerto Rico/8/1934 ( H1N1 ) ( PR8 ) and monitored for clinical signs of disease . By 11 days after intranasal administration of PR8 , significantly more Axl-/- mice than WT mice succumbed to the infection ( Figure 1A ) . This result is in agreement with a recent report by Fujimori et al ( Fujimori et al . , 2015 ) . The increased susceptibility of Axl-/- mice to IAV infection correlated with higher viral titers in the bronchoalveolar lavage ( BAL ) fluid than in WT mice 7 and 9 days post-infection ( Figure 1B ) , corresponding to when the CD8+ T cell response is critical in viral clearance . However , no significant differences in viral loads were detected during the early phase of the infection between WT and Axl-/- mice ( day 3 post-infection , Figure 1B ) . 10 . 7554/eLife . 12414 . 003Figure 1 . Loss of Axl increases susceptibility to influenza A virus infection in vivo . ( A ) Kaplan-Meier survival curves for wild-type ( WT ) and Axl-/- mice infected with 10 PFU of A/PR8 virus , 8–11 mice of each genotype and representative of 5 independent experiments . ( B ) Viral titers in the bronchoalveolar lavage ( BAL ) of WT and Axl-/- mice on days 3 , 7 , and 9 post infection with 10 PFU of PR8 , as determined by qPCR of PR8 polymerase acidic protein ( PA ) RNA . PFU = plaque forming units . 6–12 mice were used per condition . ns , non-significant; *p<0 . 05; **p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 12414 . 003 In contrast to these in vivo observations , previous studies have reported increased resistance to infection by other viruses in AXL-deficient DCs in vitro ( Bhattacharyya et al . , 2013; Meertens et al . , 2012; Morizono et al . , 2011; Shimojima et al . , 2007; 2012 ) . Therefore , we tested whether Axl-/- DCs were more or less susceptible to IAV in vivo by using a recombinant strain of PR8 carrying a GFP reporter gene in the NS segment ( PR8-GFP ) ( Manicassamy et al . , 2010 ) and analyzing percentages of GFP+ lung DCs 3 days post-infection . Two subsets of pulmonary DCs , CD11c+CD11b+CD103- and CD11c+CD11b-CD103+ , have been identified as responsible for presenting and cross-presenting IAV antigens ( Ballesteros-Tato et al . , 2010; Helft et al . , 2012; Kim et al . , 2014 ) . Flow cytometry analyses revealed that AXL is expressed in both of these DC subsets during influenza infection ( Figure 2—figure supplement 1 ) . Importantly , AXL ablation did not affect the total number of CD11c+CD11b+CD103- and CD11c+CD11b-CD103+ DCs in the lung ( Figure 2—figure supplement 2 ) . When PR8-GFP infected Axl-/- mice were compared to WT mice , we detected significantly fewer infected CD11c+CD11b+CD103- DCs ( Figure 2A , Figure 2—figure supplement 3 ) . Likewise , fewer GFP+ CD11c+CD11b-CD103+ DCs were identified in Axl-/- mice ( Figure 2A , Figure 2—figure supplement 3 ) . 10 . 7554/eLife . 12414 . 004Figure 2 . Genetic ablation of Axl confers resistance to IAV infection in dendritic cells in vivo and in vitro . WT and Axl-/- mice were infected with 3x106 PFU of PR8-GFP for 72 hr and lung DCs were identified by flow cytometry . ( A ) Top , representative flow cytometry plots ( left ) and percentage of GFP+CD11c+MHCII+CD11b+ DCs ( right ) in infected WT and Axl-/- mice . n = 9 for each genotype , representing 3 independent experiments . Bottom , representative plots ( left ) and percentage of GFP+CD11c+MHCII+CD103+ DCs ( right ) in infected WT and Axl-/- mice . n = 4 for each genotype , representative of 3 independent experiments . ( B ) Representative flow cytometry plots ( left ) and percentage of GFP+ alveolar macrophages ( right ) in infected WT and Axl-/- mice . 14–16 mice per genotype , 3 independent experiments . ( C ) WT and Axl-/- BMDCs were infected with PR8-GFP with indicated multiplicities of infection ( MOIs ) for 12 hr . Representative flow cytometry plots ( left ) and percentage of GFP+ BMDCs ( right ) are shown . ( D ) Abundance of PR8 PA RNA normalized to Gapdh in WT and Axl-/- BMDCs after 12 hr of infection with 0 . 25 MOI of PR8-GFP , as determined by qPCR . ( E ) WT and Axl-/- BMDCs were infected as in ( C ) . Representative plots ( left ) and percentage of IAV M2 ion channel+ BMDCs ( right ) are shown . For ( C ) and ( E ) , 5–9 samples were tested in each condition . Data are shown as representative or as the mean ± SEM of at least 4 independent samples per group representative of 4 independent experiments . ns , non-significant; *p<0 . 05; **p<0 . 01; ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 12414 . 00410 . 7554/eLife . 12414 . 005Figure 2—figure supplement 1 . AXL and MERTK expression in naive lung dendritic cells and alveolar macrophages . Representative histograms showing AXL ( top ) and MERTK ( bottom ) expression in ( A ) CD11c+MHCII+CD11b+CD103- lung DCs , ( B ) CD11c+MHCII+CD11b-CD103+ lung DCs , and ( C ) CD11chighMHCIIint alveolar macrophages . DOI: http://dx . doi . org/10 . 7554/eLife . 12414 . 00510 . 7554/eLife . 12414 . 006Figure 2—figure supplement 2 . Total number of CD11c+MHCII+CD11b+CD103- and CD11c+MHCII+CD11b-CD103+ cells in the lung 72 hr post infection with 3x106 PFU A/PR8 NS1-GFP . Data are shown as the mean ± SEM of independent experiments , n = 8–9 of each genotype . ns , non-significant . DOI: http://dx . doi . org/10 . 7554/eLife . 12414 . 00610 . 7554/eLife . 12414 . 007Figure 2—figure supplement 3 . Axl-/- mice have fewer IAV-infected lung DCs than WT mice . Total number of GFP+CD11c+MHCII+CD11b+CD103- and GFP+CD11c+MHCII+CD11b-CD103+ cells in the lung 72 hr post infection with 3x106 PFU A/PR8 NS1-GFP . Data are shown as the mean ± SEM , n = 8–9 of each genotype . *p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 12414 . 007 Another important cell type in the anti-IAV response is the alveolar macrophage ( Iwasaki and Pillai , 2014 ) . Axl-/- and WT alveolar macrophages were equally susceptible to infection by PR8-GFP ( Figure 2B ) . Alveolar macrophages express both AXL and the related receptor MERTK , while only AXL but not MERTK was detected on lung DCs ( Fujimori et al . , 2015 ) and Figure 2—figure supplement 1 ) . Thus , it is possible that MERTK compensates for the loss of AXL in alveolar macrophages , that MERTK is the relevant TAM receptor or that neither AXL nor MERTK regulate the susceptibility of alveolar macrophages to PR8-GFP . Bone marrow-derived ( BM ) -DCs also express AXL ( Rothlin et al . , 2007 ) . These cells were infected in vitro with a range of multiplicities of infection ( MOI ) of PR8-GFP and the degree of infection was measured as the percentage of GFP+ BMDCs . Similar to a previous report using pseudotyped HIV-1 and WNV ( Bhattacharyya et al . , 2013 ) , Axl-/- BMDCs were significantly more resistant to infection by PR8-GFP than WT BMDCs ( Figure 2C ) . Additionally , the abundance of transcript of PR8 polymerase acidic protein ( PA ) in Axl-/- BMDCs was lower than in WT BMDC cultures ( Figure 2D ) . AXL is a well-established phagocytic receptor that mediates the engulfment of apoptotic cells ( Rothlin et al . , 2015; Zagorska et al . , 2014 ) . The AXL agonist , GAS6 , can bind to phosphatidylserine exposed on the surface of apoptotic membranes and thus bridge apoptotic cells to AXL-expressing phagocytes . To rule out the possibility that the difference in GFP signal between WT and Axl-/- BMDCs was due to uptake of infected GFP+ apoptotic cells , we measured the expression of IAV M2 ion channel on the cell surface . Newly synthesized IAV M2 channel is transported to the plasma membrane of infected cells for incorporation into the envelope of budding virions . Therefore , membrane-associated M2 is a marker of active infection . We detected reduced percentages of M2+Axl-/- BMDCs in comparison to M2+ WT BMDCs throughout the range of tested MOIs of PR8-GFP ( Figure 2E ) . Collectively , our results recapitulate the previously described resistance of Axl-/-DCs to viral infection , but do not translate into improved antiviral response during infection in vivo . The induction of protective antiviral CD4+ and CD8+ T cell responses to IAV requires antigen presentation by DCs on MHC-II and MHC-I , respectively . In agreement with the increased resistance of lung DC subsets to IAV infection in Axl-/- mice , we detected a reduced maturation of these cells in the mediastinal lymph nodes ( MLNs ) . Significantly lower amounts of MHC-I and MHC-II were measured on CD11c+CD11b+CD103- DCs in the draining MLN in Axl-/- mice 72 hr post-infection with IAV ( Figure 3A ) . The reduced expression of MHC-I and MHC-II was also observed in Axl-/- CD11c+CD11b-CD103+ cells ( Figure 3B ) . IL-1β has been shown to be required for effective activation of lung dendritic cells and induction of adaptive immunity during IAV infection ( Pang et al . , 2013 ) . We found significantly fewer IL-1β-producing CD11c+CD11b+CD103- and CD11c+CD11b-CD103+ DCs in the lung of Axl-/- mice 72 hr post-infection in comparison to WT mice ( Figure 3C and D ) . In contrast , alveolar macrophages from both infected WT and Axl-/- mice produced equal amounts of IL-1β ( Figure 3E ) . 10 . 7554/eLife . 12414 . 008Figure 3 . DCs in Axl-/- mice are less activated and produce less IL-1β than WT mice during IAV infection . ( A ) Expression of MHC-I and MHC-II molecules on CD11c+MHCII+CD11b+CD103- mediastinal lymph node ( MLN ) DCs after 72 hr of infection with 3x106 PFU of PR8-GFP IAV as detected by flow cytometry . ( B ) Expression of MHC-I and MHC-II molecules on CD11c+MHCII+CD11b-CD103+ MLN DCs in mice infected as in ( A ) . ( C ) Intracellular staining of IL-1β in lung CD11b+ DCs 72 hr post infection with 3x106 PFU of PR8-GFP . ( D ) Intracellular staining of IL-1β in lung CD103+ DCs infected as in ( C ) . Data are presented as the mean ± SEM of 4–6 mice per condition , representative of 2–4 independent experiments . ns , non-significant; *p<0 . 05; **p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 12414 . 008 AXL expression is not limited to DCs and macrophages—it is also detected on mature NK cells during viral infection ( Figure 4—figure supplement 1 ) and non-hematopoietic cells ( Rothlin et al . , 2015 ) . To test whether the loss of AXL expression on myeloid cells was sufficient to lead to increased susceptibility to IAV infection , we generated Axlfl/fl mice ( Figure 4A ) and crossed them to Cd11c-Cre ( Caton et al . , 2007 ) . CD11c is a classic marker of DCs ( Caton et al . , 2007 ) and it is also expressed by alveolar macrophages . Ablation of AXL was confirmed in lung DCs and alveolar macrophages in Cd11c-Cre+ Axlfl/fl mice , while its expression remained intact in NK cells ( Figure 4—figure supplement 2 ) . BMDCs derived from Cd11c-Cre+ Axlfl/fl recapitulated the increased resistance to PR8-GFP infection characteristic of Axl-/- BMDCs , confirming the functional ablation of AXL in this line ( Figure 4—figure supplement 3 ) . Axlfl/fl mice were also crossed to Csf1r-Cre mice ( Deng et al . , 2010 ) . CSF1 receptor is preferentially expressed in macrophages , although it can also be detected in CD11b+ but not CD103+ lung DCs ( Ginhoux et al . , 2009 ) . In agreement with this reported expression pattern , Csf1r-Cre+ Axlfl/fl featured preferential ablation of AXL in alveolar macrophages than lung DCs ( Figure 4—figure supplement 2 ) . 10 . 7554/eLife . 12414 . 009Figure 4 . Cd11c-Cre+Axlfl/fl mice but not Csf1r-Cre+Axlfl/fl mice succumb to IAV infection . ( A ) Cloning strategy for the generation of Axl-floxed mice . Axlfl/fl mice were subsequently crossed with Cd11c-Cre or Csf1r-Cre mice . ( B ) Kaplan-Meier survival curves for Cre-Axlfl/fl , Cd11c-Cre+Axlfl/fl , and Csf1r-Cre+Axlfl/fl mice infected with 10 PFU of A/PR8 virus , 7–18 mice per group and representative of 2 independent experiments . ns , non-significant; *p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 12414 . 00910 . 7554/eLife . 12414 . 010Figure 4—figure supplement 1 . AXL expression by immune cells in the lung during IAV infection . Representative histograms of AXL expression in CD11c+MHCII+CD11b-CD103+ DCs , CD11c+MHCII+CD11b+CD103- DCs , CD11c-Ly6g+ neutrophils , CD11c-Ly6g-NK1 . 1+DX5+ NK cells , CD11c-CD4+ T cells , CD11c-CD8+ T cells , CD11c-CD11b-B220+ B cells , and CD11chighMHCIIint alveolar macrophages . Samples were collected from naïve mice or those infected with 10 PFU of PR8 for 3 days or 7 days , as indicated . Histograms are representative of 6–20 mice from 2–5 independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 12414 . 01010 . 7554/eLife . 12414 . 011Figure 4—figure supplement 2 . AXL is selectively ablated in Cd11c-Cre+Axlfl/fl and Csf1r-Cre+Axlfl/fl mice . ( A ) AXL expression in lung C CD11chigh MHCIIint alveolar macrophages , CD11c+MHCII+CD11b-CD103+ DCs , CD11c+MHCII+CD11b+CD103- DCs , and CD11c-Ly6g-NK1 . 1+DX5+ NK cells from WT , Axl-/- , Cd11c-Cre-Axlfl/fl and Cd11c-Cre+Axlfl/fl mice . Histograms are representative of 4–5 mice per genotype from 2–5 independent experiments . DC and alveolar macrophage AXL expression is shown from naive mice and NK cell AXL expression is represented from mice infected with 10 PFU of PR8 for 7 days . ( B ) AXL expression in lung CD45+CD11c+CD115+Siglec F+CD11blo alveolar macrophages and CD45+CD11c+CD115-Siglec F- DCs from Csf1r-Cre-Axlfl/fl and Csf1r-Cre+Axlfl/fl mice . Histograms are representative of 4–5 mice per genotype from 2 independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 12414 . 01110 . 7554/eLife . 12414 . 012Figure 4—figure supplement 3 . Cd11c-Cre+Axlfl/fl BMDCs are resistant to IAV infection . WT , Axl-/- , Cd11c-Cre-Axlfl/fl , and Cd11c-Cre+Axlfl/fl BMDCs were infected with 0 . 05 MOI of PR8-GFP for 12 hr . Representative flow cytometry plots ( left ) and percentage of GFP+ BMDCs ( right ) are shown . Data are shown as representative or as the mean ± SEM of 4 independent samples per genotype from 2 independent experiments . *p<0 . 05; **p<0 . 01DOI: http://dx . doi . org/10 . 7554/eLife . 12414 . 01210 . 7554/eLife . 12414 . 013Figure 4—figure supplement 4 . Cre-Axlfl/fl and Cd11c-Cre+Axlwt/wt mice clear IAV infection . Viral titers in the bronchoalveolar lavage ( BAL ) of WT , Cre-Axlfl/fl , Cd11c-Cre+Axlwt/wt , and Axl-/- mice 9 days post infection with 10 PFU of PR8 . Data represents 5 mice per condition . ns , non-significant; *p<0 . 05; **p<0 . 01DOI: http://dx . doi . org/10 . 7554/eLife . 12414 . 013 Next , we challenged Cd11c-Cre+Axlfl/fl , Csf1r-Cre+Axlfl/fl and respective controls to IAV infection . Analogous to that seen in Axl-/- mice , Cd11c-Cre+Axlfl/fl mice succumbed at a higher frequency to PR8 than Cd11c-Cre-Axlfl/fl control mice ( Figure 4B ) . The sensitivity of Cd11c-Cre+Axlfl/fl to PR8 infection was not due to Cre expression , as Cd11c-Cre+Axlwt/wt mice were not more susceptible to infection than WT mice ( Figure 4—figure supplement 4 ) . In contrast , Csf1r-Cre+ Axlfl/fl were as resistant to IAV infection as control mice ( Figure 4B ) . Take together , these results indicate that the ablation of AXL in myeloid cells is sufficient to confer susceptibility to IAV infection and that preserved expression in DCs appears to be required to resist the infection . Clearance of IAV depends on the optimal activation of the adaptive immune response ( Iwasaki and Pillai , 2014; Strutt et al . , 2013; Sun and Braciale , 2013 ) . We therefore investigated the induction of adaptive antiviral immunity in the absence of AXL . Expression of CD69 , an early T cell activation marker , was significantly lower in draining MLN CD8+ T cells in Axl-/- mice in comparison to WT controls 3 days post-infection with PR8 ( Figure 5—figure supplement 1 ) . Lung CD8+ T cells also showed a diminished production of IFN-γ 9 days post infection ( Figure 5A ) . The number of IFN-γ+ antigen-restricted CD8+ T cells specific for IAV PA amino acids 224–233 was similarly reduced in the lung of Axl-/- mice ( Figure 5A ) . Furthermore , as an additional marker of activation , the number of CD8+CD44+ cells in the MLN as well as the expression level of CD44 on CD8+ T cells in the lung was less in Axl-/- mice compared to WT mice ( Figure 5—figure supplement 2 ) . Similarly , CD4+ T cells in Axl-/- mice were less activated as evidenced by their reduced expression of IFN-γ and CD44 ( Figure 5B and Figure 5—figure supplement 2 ) . 10 . 7554/eLife . 12414 . 014Figure 5 . Axl-/- mice and Cd11c-Cre+Axlfl/fl mice mount impaired T cell responses to IAV infection . ( A ) Representative plots ( left ) and percentage ( middle ) of CD8+IFN-γ+ T cells in the lung of WT and Axl-/- mice after 9 days of infection with 10 PFU of PR8 . 4–5 mice per genotype , representative of 4 independent experiments . Right , quantification of IFN-γ-producing H-2Db-restricted CD8+ T cells specific for IAV PA amino acids 224–233 in the lung 9 days post infection with 10 PFU of PR8 . 7–8 mice per genotype , 2 independent experiments . ( B ) Representative plots ( left ) and percentage ( right ) of CD4+IFN-γ+ T cells in the lung of WT and Axl-/- mice infected as in ( A ) . ( C ) Number of CD8+IFN-γ+ T cells ( left ) , CD8+PA+IFN-γ+ T cells ( middle ) , and CD4+IFN-γ+ ( right ) in the lung 9 days post-infection with 10 PFU of PR8 in Cre-Axlfl/fl , Cd11c-Cre+Axlfl/fl , and Csf1r-Cre+Axlfl/fl mice , as indicated . 5–10 mice per genotype , representative of 2–3 independent experiments . ns , non-significant; *p<0 . 05; **p<0 . 01; ***p<0 . 001DOI: http://dx . doi . org/10 . 7554/eLife . 12414 . 01410 . 7554/eLife . 12414 . 015Figure 5—figure supplement 1 . Axl-/- mice display an early defect in CD8+ T cell activation during PR8 infection . Representative histogram ( left ) and MFI ( right ) for CD69 expression on CD8+ T cells in the MLN of WT and Axl-/- mice 3 days post infection with 10 PFU of PR8 . n = 4 of each genotype , representative of 2 independent experiments . **p<0 . 01DOI: http://dx . doi . org/10 . 7554/eLife . 12414 . 01510 . 7554/eLife . 12414 . 016Figure 5—figure supplement 2 . T cells of Axl-/- and Cd11c-Cre+Axlfl/fl mice have reduced CD44 expression during IAV infection . ( A ) Number of CD8+CD44+ T cells in the MLN 9 days post-infection with 10 PFU of PR8 ( left ) . n = 5 of each genotype , representative of 4 independent experiments . Representative histogram and MFI ( right ) for CD44 expression on CD8+ T cells in the lung of WT and Axl-/- mice after 9 days of infection with 10 PFU of PR8 . n = 6 of each genotype , representative of 4 independent experiments . ( B ) Number of CD4+CD44+ T cells in the MLN infected as in ( A ) ( left ) . n = 5 of each genotype , representative of 4 independent experiments . Representative histogram and MFI ( right ) for CD44 expression on CD4+ T cells in the lung of WT and Axl-/- mice after 9 days of infection with 10 PFU of PR8 . n = 6 of each genotype , representative of 4 independent experiments . ( C ) Quantification of CD8+CD44+ T cells ( left ) and CD4+CD44+ T cells ( right ) in the MLN 9 days post-infection with 10 PFU of PR8 in Cd11c-Cre-Axlfl/fl and Cd11c-Cre+Axlfl/fl mice . 5 mice per genotype , representative of 2 independent experiments . *p<0 . 05; **p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 12414 . 016 As was seen in Axl-/- mice , Cd11c-Cre+Axlfl/fl but not Cd11c-Cre-Axlfl/fl had significantly fewer number of IFN-γ producing CD8+ T cells ( Figure 5C ) . This reduction was in part due to lower numbers of PA-specific CD8+ T cells ( Figure 5C ) . In contrast , Csf1r-Cre+Axlfl/fl mice had preserved CD8+ T cell responses . ( Figure 5C ) . Similarly , Cd11c-Cre+Axlfl/fl mice developed fewer IFN-γ producing CD4+ T cells in comparison to respective controls , while this response was conserved in Csf1r-Cre+Axlfl/fl mice ( Figure 5C ) . The number of CD8+CD44+ and CD4+CD44+ T cells was also significantly reduced in the lung of Cd11c-Cre+Axlfl/fl infected mice compared to control mice ( Figure 5—figure supplement 2 ) . These results indicate that myeloid-specific ablation of AXL in Cd11c-Cre+Axlfl/fl mice is sufficient to account for the impaired T cell activation phenotype seen in complete Axl-/- mice . Furthermore , the ability of Csf1r-Cre+Axlfl/fl infected mice to mount protective adaptive antiviral responses is consistent with the preserved expression of AXL in lung DCs and the lack of increased susceptibility to infection in this conditional knock out line . To corroborate the findings in the context of an unrelated virus , Axl-/- mice were infected subcutaneously with WNV and spleens were collected 8 days after infection . As seen during IAV infection , Axl-/- mice developed deficient CD8+ T cell responses to WNV . The percentage and number of NS4B tetramer+ cells ( Figure 6A ) and IFN-γ-producing CD8+ T cells ( Figure 6B ) were reduced in Axl-/- mice in comparison to WT mice . Similarly , levels of intracellular granzyme B were reduced in both NS4B tetramer+Axl-/- CD8+ T cells and total Axl-/- CD8+ T cell populations ( Figure 6C ) . These results are in agreement with the increased susceptibility of Axl-/- mice to WNV infection ( Miner et al . , 2015 ) . Collectively , these results show that loss of AXL signaling leads to a defect in priming the adaptive antiviral T cell responses after IAV and WNV infections . 10 . 7554/eLife . 12414 . 017Figure 6 . Axl-/- mice mount a deficient CD8+ T cell response to WNV infection . WT and Axl-/- mice were infected subcutaneously with 102 PFU of WNV , and spleens were harvested 8 days post infection after extensive cardiac perfusion with PBS . ( A ) Representative flow cytometry plots ( left ) and percentage and number ( right ) of NS4B tetramer+ CD8+ T cells . ( B ) Representative flow cytometry plots ( left ) and percentage and number ( right ) of CD8+IFN-γ+ T cells . ( C ) Representative flow cytometry plots ( left ) and percentage and number ( right ) of CD8+ T cells stained for NS4B tetramer and granzyme B . Data are presented as the mean ± SEM of 6–7 mice per genotype . Data are pooled from two independent experiments . p<0 . 05; **p<0 . 01; ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 12414 . 017 AXL is a negative regulator of type I IFN production and genetic ablation of Axl has been shown to lead to increased production of type I IFNs upon viral infection ( Bhattacharyya et al . , 2013; Rothlin et al . , 2007 ) . We detected increased production of IFN-β in IAV-infected Axl-/- versus WT BMDC cultures ( Figure 7A ) . Furthermore , neutralization of type I IFN signaling by MAR1-5A3 anti-IFNAR antibody restored the susceptibility of Axl-/- BMDCs to PR8 infection ( Figure 7B ) . Next , we tested whether the enhanced type I IFN response in Axl-/- mice accounted for their increased susceptibility to IAV infection . Axl-/- and WT mice were injected with MAR1-5A3 or the respective isotype control one day prior to infection with PR8 . We detected a significant increase in the survival of IAV-infected Axl-/- mice treated with MAR1-5A3 ( Figure 7C ) . This correlated with a restoration of IFN-γ+ PA-restricted CD8+ T cells and IFN-γ producing CD4+ T cells ( Figure 7D ) . Similarly , the number of CD8+CD44+ and CD4+CD44+ T cells in infected Axl-/- mice treated with the anti-type I IFN receptor antibody were restored to that of control mice ( Figure 7D ) . 10 . 7554/eLife . 12414 . 018Figure 7 . Blockade of IFNAR signaling protects Axl-/- mice to IAV infection and rescues T cell activation . ( A ) IFN-β in the supernatant of WT and Axl-/- BMDCs after 12 hr of infection with 0 . 25 MOI of PR8 , as determined by ELISA from 4 independent experiments . ( B ) Percentage of GFP+ WT and Axl-/- BMDCs infected with 0 . 05 MOI PR8 for 12 hr treated with 10 μg/ml of IgG1 isotype control or α-IFNAR MAR1-5A3 antibody . Data is compiled from 3 independent experiments . ( C ) Kaplan-Meier survival curves for WT and Axl-/- mice given α-IFNAR MAR1-5A3 antibody or isotype control by IP injection one day prior to infection with 10 PFU of A/PR8 virus , 8–10 mice per group , 2 independent experiments . ( D ) WT and Axl-/- mice were treated with antibody and infected as in ( C ) . Number of IFN-γ-producing PA tetramer+ CD8+ T cells ( left ) and IFN-γ+ CD4+ T cells ( middle ) in the lung 7 days post infection with 10 PFU of PR8 . 4–5 mice in each group , representative of 2 independent experiments . Number of CD8+CD44+ T cells ( middle ) and CD4+CD44+ T cells ( right ) in the MLN 9 days post-infection with 10 PFU of PR8 . 8–10 mice per group , representing 2 independent experiments . Data are shown as the mean ± SEM . *p<0 . 05; **p<0 . 01; ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 12414 . 018 The immunosuppressive properties of type I IFNs are mediated , in part , by their ability to block the production of IL-1β ( Guarda et al . , 2011; Mayer-Barber et al . , 2010; 2011 ) . This is particularly relevant in the context of IAV infection , as IL-1β is required for effective priming of antiviral T cells and antibody responses ( Ichinohe et al . , 2009; Pang et al . , 2013; Schmitz et al . , 2005 ) . The increased production of IFN-β in IAV-infected Axl-/- BMDCs correlated with a concomitant reduction in the production of IL-1β ( Figure 8A ) . This is in agreement with the decreased production of IL-1β found in lung DCs of infected Axl-/- mice ( Figure 3C and D ) . Furthermore , neutralization of type I IFN signaling in infected Axl-/- BMDCs rescued IL-1β production by these cells ( Figure 8B ) . To test if the diminished production of IL-1β in infected Axl-/- mice was causal for their increased susceptibility to IAV , we administered recombinant IL-1β intranasally to WT and Axl-/- mice on days 1 , 2 , and 3 post-infection with 10 PFU of PR8 . Axl-/- mice that were given IL-1β were completely protected and survived the infection ( Figure 8C ) . Consistent with this protection , viral titers in the BALF of Axl-/- mice given IL-1β were controlled to the level of WT controls ( Figure 8D ) . Furthermore , the administration of IL-1β to Axl-/- mice restored the antiviral adaptive immune response as measured by the number of IFN-γ+ PA-restricted CD8+ T cells and IFN-γ producing CD4+ T cells ( Figure 8E ) as well as CD8+CD44+ and CD4+CD44+ T cells ( Figure 8—figure supplement 1 ) . In summary , these results demonstrate that the genetic ablation of Axl leads to enhanced production of type I IFNs and decreased production of IL-1β resulting in impaired induction of antiviral adaptive immunity and clearance of virus . 10 . 7554/eLife . 12414 . 019Figure 8 . Intranasal administration of IL-1β rescues Axl-/- T cell activation and confers protection to IAV infection . ( A ) IL-1β levels in supernatant of WT and Axl-/- BMDCs after 12 hr of infection with 0 . 25 MOI of PR8 , as determined by ELISA from 4 independent experiments . ( B ) IL-1β in supernatant of WT and Axl-/- BMDCs infected with 0 . 05 MOI of PR8-GFP for 12 hr treated with 10 μg/ml of isotype control or α-IFNAR MAR1-5A3 antibody , as determined by ELISA from 3 independent experiments . ( C-E ) WT and Axl-/- mice were intranasally administered PBS or 20 ng of recombinant IL-1β on days 0 , 1 , 2 , and 3 post infection with 10 PFU of PR8 . ( C ) Kaplan-Meier survival curves for mice treated as indicated with 5 mice per group , representative of 4 independent experiments . **Axl-/- mice given PBS succumbed to infection significantly more than the other experimental groups . ( D ) Viral titers in the bronchoalveolar lavage ( BAL ) collected 9 days post infection determined by qPCR of PR8 PA RNA . 6–10 mice per group , representing 3 independent experiments . ( E ) Number of IFN-γ-producing PA tetramer+ CD8+ T cells and IFN-γ+ CD4+ T cells in the lung 7 days post infection with PR8 . 4–5 mice in each group , representative of 2 independent experiments . Data are shown as the mean ± SEM . *p<0 . 05; **p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 12414 . 01910 . 7554/eLife . 12414 . 020Figure 8—figure supplement 1 . Intranasal IL-1β delivery rescues Axl-/- T cell CD44 expression during IAV infection . WT and Axl-/- mice were intranasally administered PBS or 20 ng of recombinant IL-1β on days 0 , 1 , 2 , and 3 post infection with 10 PFU of PR8 . ( A ) CD44 expression on CD8+ and CD4+ T cells was assessed by flow cytometry . ( B ) Number of CD8+CD44+ T cells and CD4+CD44+ T in the lung with 4–5 mice in each group , representative of 3 independent experiments . Data are shown as the mean ± SEM . *p<0 . 05; **p<0 . 01; ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 12414 . 020 In vitro experiments have led to the speculation that AXL promotes the infection of several enveloped viruses including vaccinia ( Morizono et al . , 2011 ) , Lassa ( Shimojima et al . , 2012 ) , dengue ( Meertens et al . , 2012 ) , and WNV ( Bhattacharyya et al . , 2013 ) . Since loss of AXL function in DCs protected the cells from viral infection , these results suggested that AXL inhibition might lead to improved antiviral response in infected hosts . This idea is of particular importance given that small molecule inhibitors against AXL are currently in development and one of them is in Phase 1 clinical trial ( Graham et al . , 2014 ) . In direct opposition to this hypothesis , our experimental evidence demonstrates that mice featuring the genetic ablation of Axl , even after selective deletion in myeloid cells , are more susceptible to viral infection than WT mice . We report that loss of Axl leads to a reduced ability to mount an adequate adaptive antiviral response , as exemplified by deficient priming of T cells after IAV and WNV infection . This phenotype is consistent with previous findings on the susceptibility to IAV infection of mice lacking CD8+ T cells ( Bender et al . , 1992; Wu et al . , 2010 ) . How does AXL signaling protect the host against viral infection ? Albeit paradoxical , the ability of AXL to inhibit type I IFN production appears to be important for induction of antiviral adaptive immunity . While type I IFNs are considered classical inducers of DC maturation ( Gallucci et al . , 1999; Montoya et al . , 2002 ) , the immunosuppressive effects of type I IFNs are also well known ( McNab et al . , 2015 ) . Indeed , the increased production of type I IFNs in response to infection in Axl-/- cells limited the production of IL-1β , a cytokine required for the effective priming of antiviral T cells ( Ichinohe et al . , 2009; Pang et al . , 2013; Schmitz et al . , 2005 ) . Activation of the NLRP3 inflammasome and production of IL-1β are important components of the antiviral immune response to a variety of RNA viruses , including IAV , WNV , Sendai virus , adenovirus , and vaccinia virus ( Kanneganti , 2010; Ramos et al . , 2012 ) . Similar to our findings in Axl-/- mice , hosts deficient in inflammasome signaling experience heightened mortality to IAV or WNV infection ( Allen et al . , 2009; Ichinohe et al . , 2009; Ramos et al . , 2012; Thomas et al . , 2009 ) . Thus , our studies highlight the central role of AXL in the protection of the host against viral infections through the tightly regulated production of type I IFNs . AXL is expressed not only in myeloid cells , but also in NK cells and non-hematopoietic cells ( Rothlin et al . , 2015 ) . We generated an Axlfl/fl mouse line and demonstrated that ablation of Axl in DCs and alveolar macrophages was sufficient to result in deficient T cell responses and confer susceptibility to IAV . While AXL function in DCs appears to be required for the appropriate priming of adaptive immunity , our approach did not ablate Axl in exclusively these cells . The development of more selective approaches to induce the ablation of genes in distinct DC populations will provide a better understanding . In contrast , mice featuring preferential ablation in macrophages did not succumb to the same infection . AXL and the related RTK MERTK are expressed in alveolar macrophages . These receptors are not only inhibitors of the immune response but also mediate the phagocytosis of apoptotic cells ( Fujimori et al . , 2015; Rothlin et al . , 2015 ) . Thus it is possible that AXL and MERTK function in alveolar macrophages is dispensable for the regulation of the priming of adaptive immunity , but participates in the clearance of apoptotic debris during the resolution phase . Human myeloid cells , including DCs , express AXL ( Scutera et al . , 2009 ) . Our study suggests that disabling this receptor function using small molecule inhibitors or blocking antibodies could lead to increased susceptibility to viral infections in humans rather than the desired increased resistance predicted by in vitro studies ( Bhattacharyya et al . , 2013; Meertens et al . , 2012; Morizono et al . , 2011; Shimojima et al . , 2007 ) . The prevalence of this AXL function in distinct viral infections should be carefully considered in the development of pharmacological tools that inhibit this RTK . Bone marrow-derived dendritic cells ( BMDCs ) were generated from bone marrow progenitor cells flushed from mouse femurs and tibias of gender- and age-matched donors . 2x105 progenitor cells/ml were incubated at 37°C on 24-well cell culture plates in complete media containing RPMI 1640 , 10% fetal bovine serum ( FBS ) , 1% penicillin and streptomycin , and supplemented with granulocyte macrophage colony-stimulating factor ( GM-CSF ) at a concentration of 20 ng/ml ( PeproTech , Rocky Hill , NJ ) . Fresh enriched media was added on days 3 and 6 of differentiation . BMDCs were subsequently used on day 7 . A/PR8 ( H1N1 ) and A/PR8 NS1-GFP ( gift of Adolfo García-Sastre ( Manicassamy et al . , 2010 ) ) were propagated for 2 days at 35°C in the allantoic cavities of 10- to 11-day old fertilized chicken eggs . BAL fluid from mice was collected for the measurement of viral titers at the indicated days post-infection by washing the trachea and lungs with 3 ml of PBS containing 0 . 1% BSA . Viral titers were quantified by standard viral plaque assay using Madin-Darby canine kidney ( MDCK ) cells or by qPCR quantification comparing samples to a standard curve generated from cDNA of RNA isolated from MDCK-titered stock . Viral stocks and BAL fluid samples were stored at -80°C . The WNV strain ( 3000 . 0259 ) was isolated in New York in 2000 and passaged once in C6/36 Aedes albopictus cells . Prior to infection of BMDCs , media was aspirated and wells were washed once with PBS . Viral stock was diluted in 0 . 1% BSA in PBS to the indicated multiplicities of infection ( MOIs ) . 100 μl of diluted virus was added to each well of BMDCs on the 24-well plates and placed in a 37°C incubator . Plates were lightly tapped every 20 min during a 1 hr incubation process to keep cells evenly covered by liquid . After 1 hr , virus was aspirated and wells were washed once with PBS . 1 ml of complete RPMI ( RPMI 1640 , 10% FBS , 1% penicillin and streptomycin ) was added to each well and cultures were incubated at 37°C for 12 hr . Anti-mouse IFN-α/β receptor 1 ( IFNAR1 ) antibody ( MAR1-5A3 , Leinco Technologies , Fenton , MO ) or IgG1 control was used at a concentration of 10 μg/ml and added to the complete media for the 12-hr incubation following infection in the indicated experiments . After infection , supernatants were stored at -80°C , and cells were collected by washing with ice-cold PBS . Mice were anesthetized by intraperitoneal injection of ketamine and xylazine . 10 PFU PR8 or 3x106 PFU PR8-GFP was suspended in 20 μl and was intranasally administered . Weight change and overall appearance of health was monitored daily . Mice were sacrificed by CO2 asphyxiation at the indicated time points or euthanized upon falling below 80% initial starting weight . For WNV studies , mice ( 8- to 10-week old , both sexes ) were inoculated subcutaneously via footpad injection with 102 PFU of WNV . In experiments where mice were treated with recombinant IL-1β , 20 ng of IL-1β ( eBioscience , San Diego , CA ) or PBS vehicle control were administered intranasally in 20 μl immediately following inoculation with virus . The same dose of IL-1β or PBS control was subsequently given days 1 , 2 , and 3 post-infection while mice were anesthetized by isolfurane . Mice treated with α-IFNAR MAR1-5A3 ( Leinco Technologies ) or IgG1 isotype control ( 600 μg/mouse ) were injected intraperitoneally one day prior to infection with PR8 . To stain for flow cytometry , collected cells were washed once with PBS and then incubated with α-CD16/32 Fc-block clone 93 ( Biolegend , San Diego , CA ) in PBS for 10 min at room temperature . After washing again with PBS , cells were incubated with their respective antibody cocktail for 30 min at 4°C . Following subsequent washes , prepared cells were fixed with 1% paraformaldehyde . Anti-mouse antibodies used in the study were purchased from BioLegend , BD-Biosciences , eBioscience , Santa Cruz Biotechnology ( Dallas , TX ) , Invitrogen ( Carlsbad , CA ) , Becton Dickinson ( Franklin Lakes , NJ ) , or R&D Systems ( Minneapolis , MN ) . Conjugated antibodies to FITC , PE , PE-Cy7 , PerCP , PerCP-Cy5 . 5 , APC , APC-Cy7 , or Pacific Blue were used for flow cytometry to target CD45 ( 30-f11 ) , CD11c ( N418 ) , CD11b ( M1/70 ) , H-2kb MHC-I ( AF6-88 . 5 . 5 . 3 ) , MHC-II I-A/I-E ( M5/114 . 15 . 12 ) , CD103 ( 2E7 ) , CD3ε ( 145-2C11 ) , CD4 ( GK1 . 5 , RM4-5 ) , CD8α ( 53–6 . 7 ) , CD8β ( YT5156 . 7 . 7 ) , CD69 ( H1 . 2F3 ) , CD44 ( IM7 ) , MERTK ( 108928 ) , IFN-γ ( XMG1 . 2 ) , and granzyme B ( GB12 ) . Allophyocyanin-labeled H-2Db MHCI class I tetramers for IAV viral acid polymerase amino acids 224–233 ( SSLENFRAYV ) and WNV NS4B ( 2488–2496 , SSVWNATTA ) were obtained from the National Institutes of Health Tetramer Core Facility . Unconjugated antibodies used included IAV M2 ion channel clone 14C2 ( Novus Biologicals , Littleton , CO ) and IFNAR1 ( Leinco Technologies , MAR1-5A3 ) . When detecting AXL expression , αAXL AF854 ( R&D systems ) was used with DCs and when testing AXL expression across immune cells during the course of IAV infection while αAXL C-20 ( Santa Cruz Biotechnology ) was used with naive alveolar macrophages . Secondary antibodies used in the study were donkey anti-mouse IgG-PE ( Santa Cruz Biotechnology , clone 3744 ) and chicken anti-goat IgG-AF647 ( Invitrogen , clone A21469 ) . Flow cytometry data were analyzed using FlowJo software . MLN were removed and homogenized by plunger against 40 μm strainers while suspended in complete RPMI . Lungs were minced with razor blades and placed in HBSS containing 2 . 5mM HEPES buffer and 1 . 3 mM EDTA for 37°C for 30 min while shaking . Samples were then transferred into RPMI 1640 containing 5% FBS , 2 . 5 mM HEPES buffer , and 0 . 5 mg/ml collagenase D ( Roche , Indianapolis , IN ) and incubated for 37°C for 60 min while shaking . Lung samples were homogenized and passed through a 40 μm cell strainer . Single cell suspensions from lung or MLN were treated with ACK lysis buffer before staining for flow cytometry . 2x106 single-cell suspensions of lung cells were stimulated for 6 hr with phorbol 12-myristate 13-acetate ( PMA , 20 ng/ml ) and ionomycin ( 1 μg/ml ) with protein secretion blocked with GolgiStop ( BD Biosciences ) when testing intracellular staining for IFN-γ . Staining was performed after fixing and permeabilizing cells using the BD Cytofix/Cytoperm kit ( BD Biosciences ) following the manufacturer’s protocol . Spleens of WT and Axl-/- mice were harvested 8 days post-infection after extensive cardiac perfusion with PBS . Splenocytes were dispersed into single cell suspensions with a cell strainer . Intracellular IFN-γ or TNF-α staining was performed after ex vivo restimulation with a Db-restricted NS4B immunodominant peptide using 1 μM of peptide and 5 μg/ml of brefeldin A ( Sigma , St Louis , MO ) . Intracellular granzyme B staining was performed in separate samples that also were stained with the APC-conjugated immunodominant NS4B tetramer . RNA was collected at the indicated time points and according to RNeasy mini kit ( QIAGEN , Valencia , CA ) manufacturer’s instructions . iScript cDNA Synthesis kit ( BIO-RAD , Hercules , CA ) was used for reverse transcription and KAPA SYBR Fast qPCR kit ( KAPA BIOSYSTEMS , Wilmington , MA ) was used for qPCR reactions . Dissociation curves were used to assess specificity of products . Primers used in this study are listed in Supplementary file 1 . IFN-β was measured using VeriKine Mouse Interferon Beta ELISA kits ( Pestka Biomedical Labs , Piscataway , NJ ) . IL-1β was detected by ELISA Ready-Set-Go ( eBioscience , USA ) . Data are represented as mean ± SEM . Differences between the means of experimental groups were analyzed with two-tailed Student’s t-test ( GraphPad Software Inc . , La Jolla , CA ) . Survival was calculated using Kaplan-Meier plot , and survival curves were compared by the Mantel-Cox log-rank test ( GraphPad Software Inc . ) . p values ≤ 0 . 05 were considered significant .

The immune system must be ever vigilant to ward off infections . Immune cells called T-cells can identify and eliminate microbes , but if they are too aggressive , they can damage the body . To prevent this , the body has systems that control immune responses . For example , another type of immune cell called a dendritic cell produces proteins known as type 1 interferons , which help to fight viral infections while limiting other immune responses . An enzyme called AXL blocks the production of type 1 interferons . Many scientists believe that this activity reduces the ability of individual cells in the body to defend themselves against attacking viruses . In fact , experiments with cells grown in the laboratory have shown that some viruses activate the AXL enzyme to help them infect . Similar studies have also shown that inhibiting AXL and related enzymes can make cells more able to fight off certain types of viral infection . These and other studies suggested that some drugs that block AXL might be useful treatments for viral infections , however it was not clear if this was the case for all viruses . Now , Schmid et al . show that the loss of AXL actually makes mice more prone to infections by the influenza virus and West Nile Virus . In the experiments , mice genetically engineered to lack AXL were more likely than normal mice to become ill after exposure to one of the viruses . Furthermore , fewer T cells matured to the stage where they could attack the virus in these mice . Next , Schmid et al . show that blocking the production of type 1 interferons in the mice that lack AXL restores their ability to fight off these viral infections . This is because type 1 interferons limit the production of a protein that helps the dendritic cells to mature . Therefore , Schmid et al . ’s findings show that AXL is vital for mice to fight off viral infections because it helps to balance the antiviral and immune suppressing activities of type 1 interferons . The findings also suggest that using drugs that block AXL to treat infections with certain viruses , including influenza and West Nile Virus , might do more harm than good .

2016

AXL receptor tyrosine kinase is required for T cell priming and antiviral immunity

Malaria has been a major driving force in the evolution of the human genome . In sub-Saharan African populations , two neighbouring polymorphisms in the Complement Receptor One ( CR1 ) gene , named Sl2 and McCb , occur at high frequencies , consistent with selection by malaria . Previous studies have been inconclusive . Using a large case-control study of severe malaria in Kenyan children and statistical models adjusted for confounders , we estimate the relationship between Sl2 and McCb and malaria phenotypes , and find they have opposing associations . The Sl2 polymorphism is associated with markedly reduced odds of cerebral malaria and death , while the McCb polymorphism is associated with increased odds of cerebral malaria . We also identify an apparent interaction between Sl2 and α+thalassaemia , with the protective association of Sl2 greatest in children with normal α-globin . The complex relationship between these three mutations may explain previous conflicting findings , highlighting the importance of considering genetic interactions in disease-association studies . Complement Receptor One ( CR1 ) plays a key role in the control of complement activation and the immune clearance of C3b/C4b-coated immune complexes ( Krych-Goldberg and Atkinson , 2001 ) . CR1 is expressed on a range of cells including red blood cells ( RBCs ) , leucocytes and glomerular podocytes ( Krych-Goldberg and Atkinson , 2001 ) . A number of CR1 polymorphisms have been described , including four molecular weight variants and variation in the number of CR1 molecules expressed on the surface of RBCs ( reviewed by [Krych-Goldberg et al . , 2002; Schmidt et al . , 2015] ) . Missense mutations of CR1 form the basis of the Knops blood group system of antigens , that includes the antithetical antigen pairs of Swain-Langley 1 and 2 ( Sl1 and Sl2 ) and McCoy a and b ( McCa and McCb ) ( Moulds , 2010 ) . The non-synonymous single nucleotide polymorphisms ( SNPs ) A4828G ( rs17047661 ) and A4795G ( rs17047660 ) within exon 29 of the CR1 gene give rise to the Sl1/Sl2 and McCa/McCb alleles , encoding R1601G and K1590E , respectively ( Moulds et al . , 2001 ) ( Figure 1 ) . CR1 has been implicated in the pathogenesis of multiple diseases , with epidemiological and in vitro data suggesting a role in malaria ( Schmidt et al . , 2015 ) . The Sl2 and McCb alleles occur at high frequencies only in populations of African origin ( Figure 2 ) ( Thathy et al . , 2005; Zimmerman et al . , 2003; Moulds et al . , 2004; Noumsi et al . , 2011; Fitness et al . , 2004; Covas et al . , 2007; Gandhi et al . , 2009; Yoon et al . , 2013; Hansson et al . , 2013; Kariuki et al . , 2013; Eid et al . , 2010 ) , which , given the historical prevalence of the malaria-causing parasite Plasmodium falciparum in sub-Saharan Africa , might suggest a possible survival advantage against malaria ( Rowe et al . , 1997; Rowe et al . , 2000 ) . CR1 is a receptor for the invasion of RBCs by Plasmodium falciparum merozoites ( Spadafora et al . , 2010; Tham et al . , 2010 ) and for the formation of clusters of P . falciparum-infected RBCs ( iRBCs ) and uninfected RBCs , known as rosettes ( Rowe et al . , 1997 ) . The rosetting phenotype is associated with severe malaria in sub-Saharan Africa ( Doumbo et al . , 2009 ) , with pathological effects likely due to the obstruction of microcirculatory blood flow ( Kaul et al . , 1991 ) . RBCs from donors with the high-frequency African CR1 Knops mutations bind poorly to the parasite ligand P . falciparum erythrocyte membrane protein-1 ( PfEMP1 ) that mediates rosetting by iRBCs , potentially protecting against severe malaria by reducing rosetting ( Rowe et al . , 1997 ) . Nevertheless , epidemiological data supporting this possibility are contradictory , with some studies showing an association between Sl and McC genotypes and severe malaria ( Thathy et al . , 2005; Kariuki et al . , 2013; Tettey et al . , 2015 ) and others finding none ( Zimmerman et al . , 2003; Hansson et al . , 2013; Jallow et al . , 2009; Manjurano et al . , 2012; Toure et al . , 2012; Rockett et al . , 2014 ) . Some previous studies have not considered Sl and McC genotypes together in the same statistical model , despite their physical adjacency in the CR1 molecule , nor taken into account potential interactions with other malaria resistance genes . Given the important biological role of CR1 in malaria host-parasite interactions , we aimed to clarify the relationship between the Sl and McC alleles and severe malaria in a case-control study of Kenyan children . These investigations were supplemented with a separate longitudinal cohort study of Kenyan children , examining the associations of these alleles with uncomplicated malaria and other common childhood illnesses . Finally , we also investigated the influence of these alleles on the formation of P . falciparum rosettes , as a potential functional explanation for these results through ex vivo laboratory studies . Data were obtained from 5545 children enrolled in a case-control study of severe malaria ( Figure 3 ) . The general characteristics of the cases and controls are shown in Supplementary file 1A , and the characteristics of the dataset by Sl and McC genotype are shown in Supplementary file 1B . The Sl2 and McCb allele frequencies ( 0 . 68 and 0 . 16 respectively ) were comparable to other African populations ( Figure 2 ) . There was no significant deviation from Hardy-Weinberg equilibrium for the Sl or McC genotypes among controls ( Supplementary file 1C ) . Using a simple logistic regression model containing only Sl and McC genotypes ( referred to as the unadjusted analysis below ) , we found a non-significant association between the Sl2 allele and severe malaria overall , with the Sl2/Sl2 genotype being associated with an OR for severe malaria of 0 . 90 ( 95% CI 0 . 79–1 . 01; p=0 . 07 ) ( Supplementary file 1D ) . We attempted to refine this signal by fitting a more complete model to the data , including the potential confounding factors of ethnicity , location , sickle cell trait , ABO blood group and α+thalassaemia genotype , as well as considering possible first-order interactions between terms ( referred to as the full adjusted analysis below ) . A significant protective association was observed for Sl2 in the recessive form ( adjusted Odds Ratio ( aOR ) 0 . 78; 95% CI 0 . 64–0 . 95; p=0 . 011 ) , which was most marked for cerebral malaria ( aOR 0 . 67; 0 . 52–0 . 87; p=0 . 006 ) ( Figure 4 and Table 1 ) . The Sl2/Sl2 genotype was also associated with significant protection against death from severe malaria ( aOR 0 . 50; 0 . 30–0 . 80; p=0 . 002 ) , and death among children admitted with a specific diagnosis of cerebral malaria in the full adjusted analysis ( aOR 0 . 44; 0 . 23–0 . 78; p=0 . 007 ) ( Figure 4 and Table 1 ) . Unexpectedly , we observed a significant interaction between Sl2 and α+thalassaemia genotype , such that the protective associations of Sl2 were only seen in individuals of normal α-globin genotype ( Figure 5 ) . We found no evidence for an association between Sl2 and any other clinical form of severe malaria ( Table 1 ) , or with P . falciparum parasite density ( Figure 6 ) . The unadjusted analysis showed a borderline significant association between McCb and increased susceptibility to severe malaria overall ( OR 1 . 17; 1 . 00–1 . 25; p=0 . 056 , Supplementary file 1D ) , and significant associations with increased risk of cerebral malaria ( OR 1 . 21; 1 . 05–1 . 39; p=0 . 008 ) and death ( OR 1 . 34; 1 . 00–1 . 77; p=0 . 046 , Supplementary file 1D ) . Similar associations were seen in the full adjusted analysis , although this only reached statistical significance for cerebral malaria ( aOR 1 . 19; 1 . 10–1 . 38; p=0 . 025 ( additive model ) , Figure 4 and Table 1 ) . We found no association between McCb and any other clinical form of severe malaria ( Table 1 and Supplementary file 1D ) or with P . falciparum parasite density ( Figure 6 ) . We considered whether the observed results for Sl and McC could be consistent with the effect of a single haplotype spanning Sl and McC , or with the effect of a specific genotype combination . Sl and McC are 33 bp apart and are in linkage disequilibrium , with only three of four possible haplotypes observed in our data . We therefore reanalyzed the data under a haplotype model in which the per-individual count of each of the three observed haplotypes was included as a predictor along with the potential confounding factors , as well as under a genotypic model in which the count of each of the six possible Sl/McC genotype combinations was included as a predictor ( Appendix 2 ) . These analyses suggest an additive protective association with the Sl2/McCa haplotype ( aOR = 0 . 85; 0 . 75–0 . 96; p=0 . 007 ) , with broadly consistent results observed for analysis of genotype combinations ( Supplementary file 1E and 1F ) . Thus , the opposing effects of Sl2 and McCb observed above could plausibly result from the protective association of a single haplotype at the locus , although this is difficult to distinguish from the individuals SNPs acting independently and additively based on the statistical evidence alone . We next examined the association between Sl2 and McCb alleles and uncomplicated malaria in a longitudinal prospective study of 208 Kenyan children . General characteristics of the cohort study population by Sl and McC genotypes are shown in Supplementary file 1G . After adjusting for variables known to influence malaria susceptibility , the Sl2 allele was associated with a >50% reduction in the incidence of uncomplicated malaria ( additive model ) ( Table 2; the number of episodes , incidence and unadjusted Incidence Rate Ratios for the diseases studied in the longitudinal cohort are shown in Supplementary file 1H , I and J ) . Once again , a significant interaction was seen with α+thalassaemia , such that the protective association of Sl2 was only demonstrated in children of normal α-globin genotype ( Table 3 ) . We found no significant association between the McCb allele and uncomplicated malaria ( Table 2 ) . The data shown above are incompatible with malaria being the selective pressure for McCb in the Kenyan population , and suggest that other life-threatening childhood diseases may have been responsible for selection of McCb . We therefore used the same longitudinal cohort study to investigate whether the McCb and Sl2 alleles influence the risk of other childhood diseases . McCb was associated with borderline significant protection against several common infectious diseases including LRTIs , URTIs and gastroenteritis ( Table 2 ) . Sl2 was associated with a borderline reduced incidence of gastroenteritis ( Table 2 ) . The association of McCb with gastroenteritis was predominantly seen in children of normal α-globin genotype , echoing the interaction seen with Sl2 and malaria . A previous in vitro study based on a culture-adapted P . falciparum parasite line suggested that RBC from Sl2 genotype donors had a reduced ability to form rosettes , providing a possible mechanism for protection against severe malaria ( Rowe et al . , 1997 ) . P . falciparum clinical isolates were not available from the Kenyan case-control study to investigate this potential mechanism in that population . However , the association of Sl and McC genotypes with ex vivo P . falciparum rosette frequency could be examined using 167 parasite isolates from a case-control study of children with clinical malaria in Mali ( Doumbo et al . , 2009 ) . Analysis of this small case-control study suggested a protective association between the Sl2/Sl2 genotype and cerebral malaria ( aOR 0 . 35 , 95% CI 0 . 12–0 . 89 , p=0 . 024 ) and the Sl2/Sl2-McCa/McCa genotype combination was associated with protection against cerebral malaria ( aOR 0 . 14 , 95% CI 0 . 02–0 . 84 , p=0 . 031 , Appendix 1 ) . As such , we considered samples from this population to be appropriate for testing rosetting as a potential mechanism of action . The median rosette frequency ( percentage of iRBC that form rosettes ) was significantly lower in P . falciparum isolates from malaria patients with one or more Sl2 alleles than in isolates from Sl1/Sl1 donors ( Figure 7 ) , whereas McC genotype had no significant associations with P . falciparum rosette frequency ( Figure 7 ) . The data presented here provide epidemiological evidence supporting a role for CR1 in the pathogenesis of cerebral malaria . Two neighboring CR1 polymorphisms belonging to the Knops blood group system of antigens had opposing associations on risk of cerebral malaria . The Sl2/Sl2 genotype was associated with protection against cerebral malaria and death , while the McCb allele was associated with increased susceptibility ( Figure 4 and Table 1 ) . The Sl2 allele was also associated with significant protection against uncomplicated malaria , whereas the McCb allele was associated with borderline protection against several common infections in Kenyan children ( Table 2 ) . The protective association of Sl2 against cerebral malaria , death and uncomplicated malaria was influenced by α+thalassaemia , being most evident in children of normal α-globin genotype . The protective association between Sl2 and cerebral malaria was first reported in a small case-control study from western Kenya ( Thathy et al . , 2005 ) , but has remained controversial , especially as most prior studies have been underpowered . Hence , our study is the first adequately powered independent sample set that replicates the protective association between Sl2 and cerebral malaria . Other studies found no consistent significant associations between Sl genotypes and severe malaria ( Zimmerman et al . , 2003; Hansson et al . , 2013; Jallow et al . , 2009; Manjurano et al . , 2012; Toure et al . , 2012; Rockett et al . , 2014 ) , including a recent multi-centre candidate gene study that included the sample set analysed here ( Rockett et al . , 2014 ) . A weak association between McCb and an increased odds ratio for cerebral malaria was shown in the multi-centre study ( Rockett et al . , 2014 ) . The complex interactions between Sl2 , McCb and α+thalassaemia revealed by our study provide possible reasons for the previous inconsistent findings . Although Sl2 was associated with protection against cerebral malaria in our study , McCb and α+thalassaemia both counteracted this effect . The protective association of Sl2 was observed most clearly when both McCb and α+thalassaemia genotypes were included in the statistical model , something that has not been considered in previous studies . It is possible that some of the other discrepant genetic associations with severe malaria ( Rockett et al . , 2014 ) might result from interactions between multiple loci that vary across populations and may not be revealed by standard analyses . Biologically , it makes sense to account for McC genotype when investigating associations with Sl2 and vice versa , as the two polymorphisms encode changes only 11 amino acids apart in the CR1 molecule ( Figure 1 ) . The possibility that the observed association might be due to a haplotype rather than independent effects of Sl and Mc cannot be discounted . The interaction we describe here between Sl2 and α+thalassaemia is reminiscent of the epistatic interactions that have been observed between α+thalassaemia and other malaria-protective polymorphisms including sickle cell trait ( HbAS ) ( Williams et al . , 2005a ) and haptoglobin ( Atkinson et al . , 2014 ) . It is possible , therefore , that α+thalassaemia has a broad effect on multiple malaria-protective polymorphisms , influencing their restricted global frequencies ( Penman et al . , 2009 ) , and contributing to the discrepant outcomes of previous association studies . Recent large genetic association studies on malaria do not include data on α+thalassaemia , because the causal deletions are not typed on automated platforms ( Rockett et al . , 2014 ) , instead requiring manual genotyping using labour-intensive PCR-based methods ( Chong et al . , 2000 ) . Replication of the Sl2/α+thalassaemia interaction will be required , and we suggest that α+thalassaemia genotype should be included as an important confounding variable in future malaria epidemiological studies and that efforts should continue to discover the mechanism of protection afforded by α+thalassaemia , which remains controversial ( Carlson et al . , 1994; Fowkes et al . , 2008; Krause et al . , 2012; Opi et al . , 2014; Opi et al . , 2016 ) . We examined one possible biological mechanism by which the Sl2 allele might influence cerebral malaria by studying P . falciparum rosetting , a parasite virulence factor associated with severe malaria in African children ( Doumbo et al . , 2009 ) . Previous in vitro experiments showed that CR1 is a receptor for P . falciparum rosetting on uninfected RBCs , and that RBCs serologically typed as negative for the Sl1 antigen ( likely to be from donors with Sl1/Sl2 or Sl2/Sl2 genotypes ) ( Moulds et al . , 2001 ) show reduced binding to the parasite rosetting ligand PfEMP1 ( Rowe et al . , 1997 ) . In this study , we found a significantly lower median rosette frequency in P . falciparum parasite isolates from Malian patients with Sl2 genotypes compared to Sl1/Sl1 controls ( Figure 4 ) . Therefore , similar to HbC ( Fairhurst et al . , 2005 ) , blood group O ( Rowe et al . , 2007 ) and RBC CR1 deficiency ( Cockburn et al . , 2004 ) , it is possible that reduced rosetting and subsequent reduced microvascular obstruction ( Kaul et al . , 1991 ) may in part explain the protective association of Sl2 against cerebral malaria . However , given the protective association of Sl2 with uncomplicated malaria , and the possible associations of Sl2 and McCb with other common childhood infections , it seems likely that the Knops polymorphisms may be associated with broader effects , for example on the complement regulatory functions of CR1 . Previously , we have shown that neither cofactor activity for the breakdown of C3b and C4b nor binding to C1q are influenced by the Sl2 and McCb mutations ( Tetteh-Quarcoo et al . , 2012 ) . In addition , we can find no association between Knops genotype and CR1 clustering on erythrocytes ( Paccaud et al . , 1988; Swann et al . , 2017 ) . However , other potential effects such as altered immune complex binding and processing or activation of the complement lectin pathway via mannose-binding lectin ( Ghiran et al . , 2000 ) have not yet been investigated . Our studies have several limitations: McCb homozygotes are relatively infrequent in Kenya , which limited our power to detect associations with McCb in the homozygous state . Our longitudinal cohort study generated several values of borderline statistical significance for the McCb allele which are inconclusive . Studies with larger sample sizes will be needed to examine the specific associations of McCb on assorted childhood diseases . Another limitation is that our functional ( Mali ) and epidemiological ( Kenya ) studies were conducted in different populations . The mechanisms of rosetting and associations with malaria severity are thought to be similar across sub-Saharan Africa ( Rowe et al . , 2009 ) , suggesting that data collected in either location are likely to be comparable . Furthermore , examination of a small set of cerebral malaria cases and controls from Mali suggests a protective association between Sl2/Sl2 genotype and cerebral malaria also occurs in this setting ( Appendix 1 ) . Ideally , future epidemiological and functional studies of specific polymorphisms on malaria should be conducted within a single population , although this remains logistically challenging . In conclusion , we show that two high frequency CR1 polymorphisms have opposing associations with cerebral malaria and death in Kenyan children . While the Sl2 allele may have reached high frequency in African populations by conferring a protective advantage against cerebral malaria , our data suggest that McCb arose due to a survival advantage afforded against other non-malarial infections ( Noumsi et al . , 2011; Fitness et al . , 2004 ) . Sl2 may in part protect against cerebral malaria by reducing rosetting , but additional effects seem likely . Further work is needed to examine both the epidemiological effects of the Knops polymorphisms on diverse childhood diseases , and the biological effects of the Sl2 and McCb polymorphisms on CR1 function . Future epidemiological studies should account for the effect of α+thalassaemia on the associations between Sl2 and McCb on malaria and other infectious diseases . This study uses data from a Kenyan case-control study of severe malaria , with samples collected between 2001 and 2010 , a Kenyan longitudinal cohort study , with samples collected between 1998 and 2001 and a Malian case-control study performed between July 2000 and December 2001 . Historic datasets ( i . e . >10 years old ) are widely used in genetic epidemiological studies of malaria due to the logistical challenges of sample collection in malaria endemic countries and the changing epidemiological patterns of disease . All epidemiological and clinical studies in Kenya were carried out in the area defined by the Kilifi Health and Demographic Surveillance System ( KHDSS ) , with Kilifi County Hospital ( KCH ) serving as the primary point of care ( Scott et al . , 2012 ) . Malaria transmission is seasonal in this region following the long and short rains . An Entomological Inoculation Rate ( EIR ) of up to 50 infective bites per person per year was measured in the late 1990s ( Mbogo et al . , 2003 ) , but transmission has since declined ( O'Meara et al . , 2008 ) . Between January 2001 and January 2008 , children aged <14 years who were admitted to KCH with severe malaria were recruited as cases , as described previously ( Rockett et al . , 2014 ) , except that children who were resident outside the KHDSS were excluded ( Figure 3 ) . Severe malaria was defined as the presence of blood-film positive P . falciparum infection complicated by one or more of the following features: cerebral malaria ( CM ) ( a Blantyre coma score ( BCS ) of <3 ) n = 943; severe malarial anaemia ( SMA ) ( hemoglobin concentration of <5 g/dl ) n = 483; respiratory distress ( RD ) ( abnormally deep breathing ) n = 522 or ‘other severe malaria’ ( no CM , SMA or RD but other features including prostration ( BCS 3 or 4 ) , hypoglycemia and hyperparasitemia ) n = 318 . Controls ( n = 3829 ) consisted of children 3–12 months of age who were born consecutively within the KHDSS study area between August 2006 and September 2010 and were recruited to an ongoing genetic cohort study ( Williams et al . , 2009 ) . As such , controls were representative of the general population in terms of ethnicity and residence but not of age . The use of controls who are considerably younger than cases differs from the classical structure of a case-control study . However , this method ( using cord blood or infant samples as controls ) has been widely used in African genetic association studies ( e . g . [Band et al . , 2013; Busby et al . , 2016; Clarke et al . , 2017] ) and is the most logistically feasible way of collecting sufficiently large numbers of control samples in many sub-Saharan African settings . The Sl and McC polymorphisms were originally typed as part of a larger study by Rockett et al . , 2014 , which included case-control data from 12 global sites . In Kenya , 0 . 5 ml blood samples were collected into EDTA tubes and DNA extracted using Qiagen DNeasy blood kits ( Qiagen , Crawley , UK ) . DNA was stored at −20°C and shipped frozen to Oxford . Sample processing is described in detail in the supplementary methods of Rockett et al . , 2014 . Briefly , samples underwent a whole-genome amplification step using Primer-Extension Pre-Amplification . Genotyping was performed using SEQUENOM iPLEX Gold with 384 samples processed per chip . In Rockett et al . ’s study , samples were typed for 73 SNPs; 55 of these SNPs were chosen on the basis of a known association with severe malaria , 3 SNPs were used to confirm gender and the remaining 15 SNPs to aid quality control . Samples were excluded if they did not have clinical data for gender or if genotypic gender of the sample did not match clinical gender . Samples were included if they were successfully genotyped for more than 90% of 65 ‘analysis’ SNPs . The Kenyan samples studied by Rockett et al . originally comprised 2741 cases of severe malaria and 4183 controls . After the quality control of both phenotypic and genotypic data described above , 2268 cases and 3949 controls were analysed by Rockett et al . , 2014 . The 2268 Kenyan cases and 3949 controls that were analyzed by Rockett et al . , 2014 were the starting point for our study . Children living outside the KHDSS were excluded , because this allowed us to use ‘location’ as a random effect in the final statistical model , which greatly improved model fit . Children with missing genotypes ( Sl , McC , sickle cell , α+thalassaemia or ABO blood group ) were also excluded ( Figure 3 ) . After applying these exclusion criteria , 1716 severe malaria cases and 3829 community controls were available for analysis . Hence , the number of severe malaria cases differs between our study and Rockett et al . , 2014 due to differing exclusion criteria . The inclusion of the severe malaria cases who lived outside the KHDSS into our statistical models did not alter the findings of our analysis ( Supplementary file 1K ) . In both our study and Rockett et al . , 2014 , the control samples were identical and all came from within the KHDSS . Our study has 120 fewer controls than Rockett et al . , 2014 due to missing genotypes , because we only used controls for whom full Sl , McC , sickle cell genotype , α+thalassaemia genotype and ABO blood group data were available . Our analytical methods differed from Rockett et al . , 2014 , in that we included both Sl and McC in the same statistical model and adjusted for confounders , whereas Rockett et al . examined each SNP independently . This study has been described in detail previously ( Nyakeriga et al . , 2004 ) . Briefly , this study was established with the aim of investigating the immuno-epidemiology of uncomplicated clinical malaria and other common childhood diseases in the northern part of the KHDSS study area , approximately 15 km from KCH ( Williams et al . , 2005b ) . The study was carried out between August 1998 and August 2001 involving children aged 0–10 years recruited either at the start of the study or at birth when born into study households during the study period . They were actively followed up on a once-weekly basis for both malaria and non-malaria related clinical events . In addition , on presentation with illnesses , cohort members were referred to a dedicated outpatient clinic for more detailed diagnostic tests . The cohort was monitored for the prevalence of asymptomatic P . falciparum infection through four cross-sectional surveys carried out in March , July and October 2000 and June 2001 . Exclusion criteria included migration from the study area for more than 2 months , the withdrawal of consent and death . Uncomplicated clinical malaria was defined as fever ( axillary temperature of > 37 . 5°C ) in association with a P . falciparum positive slide at any density . The most common non-malaria-related clinical events reported during the study period included upper respiratory tract infections ( URTIs ) , lower respiratory tract infections ( LRTIs ) , gastroenteritis , helminth infections and skin infections , as defined in detail previously ( Williams et al . , 2005b ) . Malaria negative fever was defined as an axillary temperature of > 37 . 5°C in association with a slide negative for P . falciparum . This analysis includes 208 children aged < 10 years for whom full Sl , McC , sickle cell genotype , α+thalassaemia genotype and ABO blood group data were available . This study has been described in detail previously ( Lyke et al . , 2003 ) . Briefly , between July 2000 and December 2001 , children ranging from 1 month to 14 years of age were recruited into a case-control study in the Bandiagara region in East Central Mali , an area of intense and seasonal P . falciparum malaria infection . In order to address the specific question of whether the Sl2/Sl2 genotype is associated with protection against cerebral malaria in Mali , only the subset of children suffering strictly defined cerebral malaria ( a BCS of <3 , with other obvious causes of coma excluded , n = 34 ) or uncomplicated malaria ( n = 184 , symptomatic children with P . falciparum parasitemia and an axillary temperature ≥37 . 5°C , in the absence of other clear cause of fever ) , and for whom Sl and McC genotyping was available were analyzed . The rosette frequency ( percentage of mature infected erythrocytes forming rosettes with two or more uninfected erythrocytes ) of P . falciparum isolates from patients recruited into the Mali case-control study was determined by microscopy after short term culture ( 18–36 hr ) , as described in detail previously ( Doumbo et al . , 2009 ) . Of the 209 isolates studied previously ( Doumbo et al . , 2009 ) , 167 were successfully genotyped for the Sl and McC alleles and are analysed here . The rosetting assays were performed before we genotyped the study participants , excluding observer bias . The rosette frequency of parasites from hosts with differing Sl and McC genotypes were compared by a Kruskal-Wallis test with Dunn’s multiple comparisons ( Prism v6 . 0 , Graphpad Inc , San Diego , CA ) . DNA was extracted either from fresh or frozen whole blood by proprietary methods using either the semi-automated ABI PRISM 6100 Nucleic acid prep station ( Applied Biosystems , Foster City , CA ) or using QIAamp DNA Blood Mini Kits ( Qiagen , West Sussex , UK ) . SNPs giving rise to the Sl and McC alleles were genotyped using either the SEQUENOM iPLEX Gold multiplex system ( Agena Biosciences , Hamburg , Germany ) ( Kenyan study ) ( Rockett et al . , 2014 ) or by an established PCR-RFLP method as described previously ( Malian study ) ( Moulds et al . , 2004 ) . Genotyping for sickle cell trait ( HbAS ) and the common African α+thalassaemia variant caused by a 3 . 7 kb deletion in the HBA gene were performed by PCR as described in detail elsewhere ( Chong et al . , 2000; Waterfall and Cobb , 2001 ) . The effects of the Sl and McC alleles were examined in genotypic , dominant , recessive and additive models of inheritance , with the best fitting model selected based on Akaike information criterion ( AIC ) . Analyses for the Kilifi case-control study were performed in R ( R Foundation for Statistical Computing , Vienna , Austria ) ( R Development Core Team , 2010 ) using the ‘ggplot2’ , ‘lme4’ , and ‘HardyWeinberg’ packages ( Wickham , 2009; Bates et al . , 2015; Graffelman and Camarena , 2008 ) , while analyses for the longitudinal study were performed in Stata v11 . 2 ( StataCorp , Texas , USA ) . In both studies , a p value of < 0 . 05 was considered statistically significant . Graphs were generated using R or Prism v6 . 0 ( Graphpad Inc , San Diego , CA ) . For the Kenyan case-control study , Sl and McC genotype were included together in a statistical model to examine their associations with malaria susceptibility . Odds Ratios ( ORs ) and 95% Confidence Intervals ( CI ) were generated using mixed effect logistic regression analysis both with and without adjustment for ethnicity and location of residence as random effects , and sickle cell genotype , α+thalassaemia genotype , and ABO blood group ( O or non-O ) as fixed effects ( variables which have been associated with malaria susceptibility in multiple previous studies in this population ) ( Jallow et al . , 2009; Rockett et al . , 2014; Williams et al . , 2005a; Atkinson et al . , 2014; Rowe et al . , 2007; Williams et al . , 2005b; Fry et al . , 2008; Malaria Genomic Epidemiology Network et al . , 2015 ) . The ethnicity variable was compressed from 28 categories to four; Giriama ( n = 2728 ) , Chonyi ( n = 1800 ) , Kauma ( n = 588 ) and other ( n = 429 ) . Binary parameterization of the α+thalassaemia variable was used , that is , comparing those children with no α+thalassaemia alleles against those with one or more α+thalassaemia alleles . This division was chosen in accordance with a previous report showing that both heterozygous and homozygous α+thalassaemia genotypes are associated with protection against severe malaria and death in the Kilifi area ( Williams et al . , 2005c ) . 2000 bootstrapped iterations were run to give 95% CIs and p values . For the Kenyan longitudinal cohort study , Incidence Rate Ratios ( IRRs ) and 95% CIs were generated using a random effects Poisson regression model that took into account within-person clustering . Data were examined with and without adjustment for confounding by McC genotype ( for Sl analyses ) , Sl genotype ( for McC analyses ) sickle cell genotype , α+thalassaemia genotype , ABO blood group , ethnic group , season ( defined as 3 monthly blocks ) , and age in months as a continuous variable . For the Malian case-control study , ORs and 95% CIs were computed using mixed effect logistic regression analysis with adjustment for location of residence as a random effect and age , ABO blood group ( O or non-O ) and ethnicity ( Dogon or non-Dogon ) as fixed effects . α+thalassaemia genotyping was not available for the Malian study and sickle cell trait is extremely uncommon in this population , therefore neither variable was included in the model . 2000 bootstrapped iterations were run to give adjusted ORs . Corrections for multiple comparisons were not performed , instead all adjusted odds ratios , confidence intervals and p values have been clearly reported . This approach has been repeatedly advocated , particularly when dealing with biological data ( Rothman , 1990; Perneger , 1998; Nakagawa , 2004; Fiedler et al . , 2012; Rothman , 2014 ) . A detailed description of the Malian dataset is given in Appendix 1 , and a detailed description of the statistical model fitting for the Kenyan studies is given in Appendix 2 .

Malaria kills more than half a million children in Africa every year . The disease is caused by the Plasmodium falciparum parasite , and mosquitos infected with the parasites spread them to humans when they bite . Once inside a human , the parasites infect the red blood cells . In severe cases , these red blood cells can stick to the walls of small blood vessels that supply the brain and so hinder the flow of oxygen , causing a coma . This is called cerebral malaria . Malaria can also result in the destruction of many oxygen-carrying red blood cells , which causes severe anemia . Both cerebral malaria and severe anemia can lead to death . Small changes ( called mutations ) in certain human genes can protect against malaria . Over time , mutations that protect people living in Africa from dying from malaria have been passed down through generations . A good example is the sickle cell mutation , which causes red blood cells to be of an unusual shape , but also affects the ability of malaria parasites to grow normally within red cells . Finding new mutations that protect against malaria may help scientists understand how severe malaria happens and eventually develop new drugs and vaccines against the disease . Some studies have found that mutations in a gene called complement receptor 1 ( CR1 ) may be protective , although others have disagreed . Now , Opi , Swann et al . show that children with one of the CR1 mutations were one-third less likely to get cerebral malaria and half as likely to die as children without the mutation . In the study , genetic and health information on more than 5 , 500 children in Kenya were analyzed to see if the severity of malaria differed depending on whether they had a CR1 mutation . They also found that the CR1 mutation is only protective against severe malaria when the child does not have another malaria- protective mutation called α-thalassemia . In children with α-thalassemia , the CR1 mutation does not make a difference . The interaction between the CR1 mutation and α-thalassemia may explain why some studies did not show a benefit of CR1 . If the researchers did not include α-thalassemia in their assessment , they could not have seen the whole picture . Future studies showing how the CR1 mutation protects against cerebral malaria could help identify new treatments that prevent severe disease or death . More study of interactions between genes that play a role in malaria may also be helpful .

2018

Two complement receptor one alleles have opposing associations with cerebral malaria and interact with α+thalassaemia

Cdk5 is a post-mitotic kinase with complex roles in maintaining neuronal health . The various mechanisms by which Cdk5 inhibits and promotes neurodegeneration are still poorly understood . Here , we show that in Drosophila melanogaster Cdk5 regulates basal autophagy , a key mechanism suppressing neurodegeneration . In a targeted screen , Cdk5 genetically interacted with Acinus ( Acn ) , a primarily nuclear protein , which promotes starvation-independent , basal autophagy . Loss of Cdk5 , or its required cofactor p35 , reduces S437-Acn phosphorylation , whereas Cdk5 gain-of-function increases pS437-Acn levels . The phospho-mimetic S437D mutation stabilizes Acn and promotes basal autophagy . In p35 mutants , basal autophagy and lifespan are reduced , but restored to near wild-type levels in the presence of stabilized AcnS437D . Expression of aggregation-prone polyQ-containing proteins or the Amyloid-β42 peptide , but not alpha-Synuclein , enhances Cdk5-dependent phosphorylation of S437-Acn . Our data indicate that Cdk5 is required to maintain the protective role of basal autophagy in the initial responses to a subset of neurodegenerative challenges . Cdk5 shares strong homology with other members of the family of cyclin-dependent kinases ( Cdks ) , but it is distinct in its modes of regulation and function ( Dhavan and Tsai , 2001; Pozo and Bibb , 2016 ) . Unlike other Cdks , Cdk5 is best known for its function in post-mitotic cells rather than cell cycle regulation ( Dhariwala and Rajadhyaksha , 2008 ) . In post-mitotic cells , Cdk5 is regulated by binding to its obligatory membrane-associated p35 or p39 co-activators ( Tsai et al . , 1994; Tang et al . , 1995 ) . These co-activators are highly expressed in the brain and loss of Cdk5 activity in mice or flies has been associated with defects in neurite outgrowth ( Su and Tsai , 2011; Trunova et al . , 2011 ) , neuronal migration ( Nishimura et al . , 2014 ) , pre- and post-synaptic functions ( Bibb et al . , 1999; Li et al . , 2001 ) , the maintenance of synaptic plasticity ( Hawasli et al . , 2007 ) , retinal degeneration ( Kang et al . , 2012 ) , and neurodegeneration in aging brains ( Trunova and Giniger , 2012; Shah and Lahiri , 2017 ) . Accordingly , dysregulation of Cdk5 has been observed in numerous brain diseases , including schizophrenia and epilepsy ( Patel et al . , 2004; Engmann et al . , 2011 ) and neurodegenerative disorders , including Huntington’s disease , Alzheimer’s disease and Amyotrophic Lateral Sclerosis ( ALS ) ( Cheung and Ip , 2012 ) . Cdk5 substrates regulating microtubule-based transport ( Klinman and Holzbaur , 2015 ) and synaptic function ( Tan et al . , 2003; Lai and Ip , 2015 ) , have been identified , but the long-term neuroprotective function of Cdk5 activity remains poorly understood ( McLinden et al . , 2012; Meyer et al . , 2014 ) . Autophagy , here short for macroautophagy , is a key cellular process for maintaining the health of neurons ( Menzies et al . , 2015 ) . Cytoprotective functions of autophagosomes in neurons include the engulfment and lysosomal delivery of aggregates of misfolded proteins and the disposal of dysfunctional mitochondria ( Green and Levine , 2014 ) . This protective role of basal autophagy was demonstrated by neuron-specific mutations in autophagy core components: neuronal loss of Atg5 or Atg7 yields rapid neurodegeneration ( Hara et al . , 2006; Komatsu et al . , 2006; Juhász et al . , 2007 ) . These findings are consistent with results in mouse and Drosophila models of Huntington’s disease or spinocerebellar ataxia type 3 ( SCA3 ) in which elevated autophagy corresponded to reduced loads of aggregated Huntingtin ( Htt ) protein and ameliorated neuronal phenotypes ( Ravikumar et al . , 2004; Bilen and Bonini , 2007; Sarkar et al . , 2007; Zheng et al . , 2010; Jaiswal et al . , 2012; Menzies et al . , 2015 ) . Although the rapid induction of autophagy in response to glucose or amino-acid deprivation is well described ( Galluzzi et al . , 2014 ) , little is known about the modulation of basal levels of autophagy in response to stress caused by protein aggregates ( Ashkenazi et al . , 2017 ) . We have previously identified Acinus ( Acn ) as a novel regulator of basal autophagy in Drosophila ( Haberman et al . , 2010; Nandi et al . , 2014 ) . Mammalian Acn had originally been identified as a caspase target aiding in chromatin modifications in apoptotic cells ( Sahara et al . , 1999; Joselin et al . , 2006 ) . In mammalian and Drosophila cells , Acn is highly enriched in the nucleus where , together with its binding partners Sap18 and RNPS1 , it forms the ASAP complex ( Schwerk et al . , 2003; Murachelli et al . , 2012 ) . ASAP interacts with the exon junction complex ( Tange et al . , 2005 ) and participates in the regulation of alternative splicing ( Schwerk et al . , 2003; Jang et al . , 2008; Hayashi et al . , 2014; Malone et al . , 2014 ) . Antagonistic activities of Akt1 kinase and caspase-3 homologs regulate Acn levels ( Hu et al . , 2005 ) and genetic manipulations that prevent the caspase-mediated cleavage of endogenous Acn in Drosophila can elevate its levels in a cell type-specific manner ( Nandi et al . , 2014 ) . An unexpected consequence of such elevated Acn was an increase in basal , starvation-independent autophagy . High-level Acn overexpression by the Gal4/UAS system triggers autophagy-dependent death in Drosophila ( Haberman et al . , 2010 ) . By contrast , Acn levels were modestly increased by the AcnD527A mutation that interferes with its Caspase-mediated cleavage or by phospho-mimetic mutations in two AKT1 target sites that reduce this cleavage ( Nandi et al . , 2014 ) . Such mildly elevated Acn levels yielded elevated basal autophagy with beneficial outcomes including enhanced starvation resistance , prolonged life span and reduced loads of polyQ aggregates in a Drosophila model of Huntington’s disease ( Nandi et al . , 2014 ) . Here , we show that similar benefits are gained by phosphorylation of Acn at serine 437 . We identify Cdk5 as the kinase which mediates this phosphorylation and show that Cdk5 activity is enhanced in the presence of multiple aggregation-prone proteins , including Huntingtin-Q93 ( Htt . Q93 ) , SCA3 . Q78 , and Amyloid-beta peptide 42 ( Aβ42 ) . These findings offer new insights into the complex mechanisms balancing the effects of loss and gain-of-function of Cdk5 on neurodegenerative diseases . Phosphorylation of conserved C-terminal Akt1-target sites ( Figure 1A ) regulates Acn levels in flies and mammalian cells ( Hu et al . , 2005; Nandi et al . , 2014 ) . This motivated us to investigate the functional consequences of phosphorylation of another highly conserved residue of Acn , namely serine 437 ( Figure 1B ) . Phosphorylated S437 has been detected by phosphoproteomics approaches in mammalian cancer cells ( Patwa et al . , 2008; Francavilla et al . , 2017 ) and Drosophila ( Bodenmiller et al . , 2007 ) . To explore phosphorylation of this residue in vivo , we raised an antibody that specifically recognizes Acn phosphorylated at serine 437 . To assess its specificity for S437-phosphorylated Acn ( pS437-Acn ) , we generated flies in which endogenous Acn was replaced by N-terminally Myc-tagged AcnWT or AcnS437A ( Figure 1A ) . Expression of these transgenes was under control of the endogenous acn promoter and enhancers within a 4 kb acn genomic region ( Nandi et al . , 2014 ) . All genomic acn transgenes were inserted at 96F3 to avoid insertion site-specific differences in expression levels ( Figure 1K ) . The previously observed lethality , developmental defects and endocytic trafficking defects of acn-null alleles were rescued by both transgenes and also a corresponding phospho-mimetic AcnS437D ( Figure 1—figure supplement 1 ) . From here on , we will refer to these rescued flies that , in an acn-null background exclusively express transgenic forms of Acn as AcnWT , AcnS437D or AcnS437A flies . Full genotypes for each experiment are listed in Supplementary file 3 . Staining of AcnWT larval eye discs with the phospho-specific pS437-Acn antibody revealed a dynamic expression pattern consistent with staining for the Myc epitope tag ( Figure 1C ) and the previously described distribution of Acn in retinal cells ( Haberman et al . , 2010; Nandi et al . , 2014 ) . Two approaches were used to assess specificity for pS437-Acn . First , staining with pS437-Acn antibody of AcnS437A eye discs was reduced to background level ( Figure 1D ) . Second , treatment of AcnWT eye discs with Calf Intestinal Phosphatase ( CIP ) reduced staining for pS437-Acn but not the Myc epitope ( Figure 1E ) , demonstrating the specificity of the pS437-Acn antibody and the high level of Acn phosphorylation at S437 in developing eye discs . To test the effect of S437 phosphorylation on Acn levels , we stained the larval eye disc with Acn antibody . Acn levels were slightly reduced in AcnS437A eye discs compared to AcnWT ( Figure 1F , G ) . The small difference between AcnWT and AcnS437A levels in western blots of larval lysates ( Figure 1I ) may reflect the contribution of other tissues with low levels of S437 phosphorylation of AcnWT or with AcnS437A stabilized by alternative mechanisms . By contrast , Acn levels were further enhanced in flies expressing only the phospho-mimetic AcnS437D ( Figure 1H ) . This was consistent with changes in Acn levels when compared by western blot analysis of larval lysates: AcnS437D levels were elevated compared to AcnWT or phospho-inert AcnS437A ( Figure 1I , J ) . Moreover , we detected a smaller form of Acn that lost the N-terminal Myc-epitope ( arrow in Figure 1I ) . This likely represents a cleaved form of Acn that was stabilized for phospho-mimetic AcnS437D but degraded for AcnWT or AcnS437A ( Figure 1I ) . Proteolytic cleavage of Acn close to the S437 residue has previously been observed for Drosophila and mammalian Acn proteins ( Sahara et al . , 1999; Hu et al . , 2005; Nandi et al . , 2014 ) . Together , these data indicate that Acn S437 is phosphorylated in developing tissues and plays a critical role in regulating Acn levels . As altering Acn levels can modulate the level of basal autophagy ( Nandi et al . , 2014 ) , we tested whether the stabilized AcnS437D mutant exhibited elevated levels of autophagy . First , we analyzed endogenous Atg8a in eye discs of fed wandering third instar larvae . Atg8a punctae mark autophagosomes and early autolysosomal structures ( Klionsky et al . , 2016 ) . The number of Atg8a-positive punctae was higher for phosho-mimetic AcnS437D eye discs compared to AcnWT or phospho-inert AcnS437A , indicating elevated levels of autophagy ( Figure 2A–D ) . Another tissue with highly regulated autophagy are larval fat bodies ( Rusten et al . , 2004; Scott et al . , 2004 ) . When we examined fat bodies of fed 96 hr larvae , Atg8a punctae were rare in AcnWT or AcnS437A larvae , but numerous and brightly stained in fed AcnS437D larval fat bodies ( Figure 2E–H , Figure 2—figure supplement 1A ) . Importantly , a 4-hr amino acid starvation further increased the accumulation of Atg8a punctae in all three genotypes ( Figure 2I–L , Figure 2—figure supplement 1 ) . We used three approaches to distinguish whether the elevated levels of Atg8a punctae in fed AcnS437D larvae represent an accumulation of stalled autophagosomes , or elevated flux through the pathway . First , we inhibited lysosomal acidification and degradation with chloroquine to reveal Atg8a that has been delivered to autophagosomes and otherwise would be degraded ( Lőw et al . , 2013; Mauvezin et al . , 2014 ) . For AcnWT , AcnS437A and AcnS437D fat bodies , chloroquine treatment resulted in further elevation of ATG8a staining after starvation , consistent with elevated flux in these starved tissues ( Figure 2I’–L , Figure 2—figure supplement 1 ) . Chloroquine treatment also significantly enhanced ATG8a staining in fed AcnS437D fat bodies consistent with elevated autophagic flux ( Figure 2G’ , H , Figure 2—figure supplement 1 ) . Second , we used LysoTracker to evaluate acidification of autolysosomes and lysosomes ( DeVorkin and Gorski , 2014a ) . Under fed conditions , few LysoTracker-positive punctae were detected in AcnWT or AcnS437A fat bodies , and their number increased upon amino acid starvation ( Figure 2M–P ) , consistent with previous reports of starvation-induced autophagic flux in larval fat bodies ( Rusten et al . , 2004; Scott et al . , 2004 ) . AcnS437D larval fat bodies , however , displayed numerous LysoTracker-positive structures even in fed 96 hr larvae ( Figure 2O , P ) and their number further significantly increased after a 4 hr amino acid starvation ( Figure 2O’ , P ) . Third , we analyzed these changes on the ultrastructural level using transmission electron microscopy ( TEM ) . Previous work has shown that fat bodies from fed wild-type larvae contain predominately lysosomes smaller than 400 µm , while starved fat bodies contain lysosomes and autolysosomes larger than 1 µm ( e . g . Scott et al . , 2004; Takáts et al . , 2013; Takáts et al . , 2014 ) . When we analyzed fat bodies of fed 96 hr AcnS437D larvae , we observed significantly larger and more abundant lysosomes and autolysosomes compared to AcnWT ( Figure 2Q , R ) . The mean diameter of lysosomal/autolysosomal structures is increased more than eight-fold from 250 ± 20 nm in AcnWT to 2044 +/−135 nm in AcnS437D larvae ( Figure 2—figure supplement 2S ) and the mean area they occupied was increased almost 65-fold ( Figure 2T ) . We previously observed that increased levels of Acn exert a concentration-dependent physiological response . Mild elevation , for example by preventing caspases-mediated cleavage of Acn , enhanced starvation resistance and extended life span of well-fed flies ( Nandi et al . , 2014 ) . By contrast , da-Gal4-driven expression at higher levels caused autophagy-mediated lethality ( Haberman et al . , 2010 ) . Therefore , we explored the physiological consequences of elevated autophagy in AcnS437D flies . When challenged with starvation stress , AcnS437D flies survived significantly longer than AcnWT or AcnS437A flies ( Figure 3A ) . Furthermore , AcnS437D life span under well-fed conditions was significantly extended , whereas the phospho-inert AcnS437A mutants died somewhat faster compared to AcnWT ( Figure 3B ) . The median life expectancy for the stabilized AcnS437D mutant was extended by about 50% to 57 days compared to 38 days for AcnWT . Together , our data indicate that stabilized phospho-mimetic AcnS437D elevates basal autophagy leading to beneficial outcomes under standard growth conditions . To identify the kinase responsible for phosphorylating Acn at serine 437 , we performed a targeted RNAi screen . We screened a pre-selected subset of kinases based on hits in software packages [GPS3 . 0; http://gps . biocuckoo . org ( Xue et al . , 2008 ) and NetPhosK http://www . cbs . dtu . dk/services/NetPhosK/ ( Miller and Blom , 2009 ) . To test the ability of these kinases to modify Acn function , we used a sensitized genetic system . Eye-specific GMR-Gal4-driven expression of UAS-AcnWT at 28°C yields a rough-eye phenotype that is susceptible to enhancement or suppression by modifiers of Acn levels ( Nandi et al . , 2014 ) . We reasoned that reduced activity of any kinase responsible for phosphorylating and stabilizing Acn should at least partially suppress the roughness induced by UAS-AcnWT . Among the kinases tested , RNAi lines targeting Cdk5 and the MAP kinase p38b exhibited more than 50% suppression of Acn-induced eye roughness ( Figure 4 , Figure 4—figure supplement 1 , Supplementary file 1 ) . Moreover , expression of UAS-Cdk5-RNAi and UAS-p38b RNAi transgenes by themselves did not result in visible phenotypes in the eye ( Figure 4F , M and Supplementary file 1 ) . We further investigated these two hits from the RNAi screen by examining interactions of gain-of-function and loss-of-function mutants with Acn . GMR-Gal4-driven co-expression of AcnWT with the dominant-negative kinases p38b MAPKK53R or Cdk5K33A effectively suppressed the rough-eye phenotype ( Figure 4C , J , Figure 4—figure supplement 2 and Supplementary file 2 ) . Furthermore , co-expression of AcnWT with p38b MAPK , Cdk5 or its coactivator p35 ( Tsai et al . , 1994; Connell-Crowley et al . , 2000 ) further enhanced eye roughness ( Figure 4D , K , L and Supplementary file 2 ) . By contrast , expression of these indicated Cdk5/p35 and p38b MAPK transgenes by themselves yielded little to no visible eye phenotypes ( Figure 4G , H , N–P and Supplementary file 2 ) . These strong genetic interactions with Acn point to Cdk5 and p38b MAPK as two candidate kinases that may phosphorylate Acn and thereby enhance its activity . To test whether genetic interactions reflect modification of Serine 437 , we used the phospho-specific pS437-Acn antibody to stain eye tissue from wandering third instar larvae . Larvae with Cdk5 or p35 knocked down , as well as Cdk5 or p35 mutant larvae exhibited a dramatic reduction of Acn phosphorylation at serine 437 compared to wild-type controls ( Figure 5A–I ) . Some brightly pS437-Acn-positive cells remained , however , close to the morphogenetic furrow of Cdk5 or p35 loss-of-function eye discs . In wild-type ( Figure 5B , C ) and Cdk5 loss-of-function eye discs ( Figure 5E , F ) , the apical position of these pS437-Acn-positive cells and their shape and DNA distribution identified them as mitotic cells ( Figure 5C , F ) . In dividing cells , Acn may be phosphorylated , possibly by mitotically active kinases of the Cdk family with target recognition sequences similar to Cdk5 ( Malumbres and Barbacid , 2009 ) . Interestingly , Cdk1 knockdown enhanced the Acn overexpression phenotype in eyes ( Supplementary file 1 ) . As AcnS437A or AcnS437D flies are viable and without obvious mitotic defects , we did not further analyze the significance of elevated pS437-Acn levels in mitotic cells . Importantly , in Cdk5 mutant larvae , Acn phosphorylation at serine 437 was restored to wild-type levels with a genomic Cdk5 transgene ( Figure 5J ) . Furthermore , overexpression of p35 drastically enhanced Acn-S437 phosphorylation ( Figure 5K ) . By contrast , in p38b mutant larvae pS437-Acn levels were not altered compared to wild type ( Figure 5A , L ) . These data suggest that although both Cdk5 and p38b MAPK genetically interact with Acn , only Cdk5/p35 mediates phosphorylation of Acn-S437 . To further test the in vivo importance of S437 phosphorylation by Cdk5 , we compared GMR-Gal4-driven co-expression of UAS-p35 or UAS-p38b MAPK with UAS-AcnWT or UAS-AcnS437A , respectively . At 25°C , UAS-AcnWT or UAS-AcnS437A expression yielded mildly rough eyes ( Figure 6A , B ) , and expression of only p35 or p38b MAPK did not alter eye morphology ( Figure 6C , D ) . Notably , co-expression of p35 enhanced UAS-AcnWT-induced roughness significantly more than that of AcnS437A ( Figure 6E , F , I; p<0 . 0001; Chi-square test ) . By contrast , both UAS-AcnWT and UAS-AcnS437A rough-eye phenotypes were enhanced by p38b MAPK co-expression ( Figure 6G , H , I ) . Furthermore , Acinus S437 phosphorylation by Cdk5 was also observed in vitro , when purified Acinus proteins , either bacterially-expressed GST-Acn402-527 fusion proteins ( Figure 6J ) or S2 cell-expressed full-length Acinus proteins ( Figure 6K ) were exposed to purified Cdk5/p35 kinase complex . Together these findings indicate that , unlike for p38b MAPK , Acn S437 is the physiologically relevant residue targeted by the Cdk5/p35 complex . Flies mutant for the Cdk5 co-activator p35 display adult onset neurodegeneration and reduced lifespan ( Connell-Crowley et al . , 2007; Trunova and Giniger , 2012 ) . To test whether altered basal autophagy contributes to these phenotypes , we first examined the distribution of Atg8a in eye discs from fed 96-hr old larvae . Compared to wild type ( Figure 7A ) , p35 mutant eye discs displayed a significantly reduced number of Atg8a-positive punctae ( Figure 7B , K ) . Basal autophagy in p35 eye discs was restored to wild-type level by expression of phospho-mimetic AcnS437D , but not AcnWT or AcnS437A , under control of the endogenous acn promoter ( Figure 7C–E , K ) . Cellular levels of the autophagy receptor p62 depend on autophagic flux and thus are elevated in cells with impaired basal autophagy ( Pircs et al . , 2012; DeVorkin and Gorski , 2014b ) . By immunostaining , p62 levels were significantly elevated in p35 eye discs compared to wild type ( Figure 7F , G , L ) , consistent with impaired basal autophagy . Accumulation of p62 in p35 eye discs was not prevented by expression of AcnWT or AcnS437A ( Figure 7H , I , L ) , whereas AcnS437D expression restored wild-type p62 levels ( Figure 7J , L ) . Next , we explored the physiological consequences of manipulating levels of basal autophagy in p35 mutants . Consistent with previous reports ( Connell-Crowley et al . , 2007 ) , p35 mutants had reduced life expectancy compared to wild type ( Figure 7M ) . Near wild-type life span was restored in p35 mutants that express phospho-mimetic AcnS437D , but not AcnS437A or AcnWT ( Figure 7M ) . Together , these data indicate that Acn-S437 is a physiologically relevant target for Cdk5/p35-mediated phosphorylation . To maintain normal life span , autophagy is required in neurons for the suppression of neurodegeneration . To test a possible role of Cdk5-mediated Acn phosphorylation in this context , we used two different Drosophila Huntington’s disease models expressing huntingtin-polyQ polypeptides in the eye , either through GMR-Gal4-driven expression of UAS-Htt . Q93 ( Figure 8B , C ) or through a direct fusion of Htt-Q120 with the GMR enhancer/promoter region ( GMR-Htt . Q120 , Figure 8D , E ) . Compared to wild-type eye discs ( Figure 8A ) , Acn-S437 phosphorylation was elevated in posterior cells of Htt-polyQ expressing eye discs ( Figure 8B , D , F ) . This activation of Acn was dependent on Cdk5/p35 as it was suppressed in eye discs with p35 knocked down ( Figure 8C , F ) or mutant for p35 ( Figure 8E , F ) . To examine the consequence of Acn activation for polyQ accumulation , we stained eye discs for polyQ proteins . GMR-Htt . Q120 expression resulted in its accumulation a few rows posterior to the furrow in eye discs expressing AcnWT or AcnS437A under control of the acn promoter ( Figure 8G–I , L ) . By contrast , expression of the stabilized phospho-mimetic AcnS437D protein yielded an initial reduction in polyQ accumulation just posterior to the furrow ( Figure 8J , L ) . This is consistent with the data above that show elevated autophagy in AcnS437D eye discs ( Figure 2A , C ) and with the known role of autophagy in the clearance of protein aggregates ( Menzies et al . , 2017 ) . In p35 mutant eye discs , polyQ protein levels were further elevated ( Figure 8K , L ) , which resulted in the accumulation polyQ aggregates in adults as detected by a filter retardation assay ( Figure 8M ) . Acn activation was not specific for Htt-polyQ peptides , as overexpression of other neurodegeneration linked polyQ proteins , specifically SCA3 . Q78 ( Figure 9B , G ) and ATX1 . Q82 ( Figure 9C , G ) yielded similar increases in Acn-S437 phosphorylation . Furthermore , expression of human Aβ42 peptide resulted in elevated levels of S437 phosphorylated Acn ( Figure 9D , G ) . By contrast , overexpression of Parkinson’s disease-linked alpha-Synuclein ( Figure 9E , G ) or Amyotrophic Lateral Sclerosis ( ALS ) -linked human SOD1 ( Figure 9F , G ) in larval eye disc failed to elevate Acn S437 phosphorylation compared to wild type ( Figure 9A ) . Numerous studies have implicated dysregulation of Cdk5 activity in neurodegenerative diseases due to its role in regulating cytoarchitecture , axonal transport , and synaptic activity ( Su and Tsai , 2011; Cheung and Ip , 2012; Shah and Lahiri , 2017 ) . Here , we show that reduced Cdk5 activity can also reduce neuronal fitness by compromising basal autophagy . We find that the effect of the Cdk5/p35 complex on autophagy depends on its role in phosphorylating the conserved S437 in Acn . Interfering with Acn-S437 phosphorylation , either by loss of Cdk5/p35 function , or by mutation of its target site in Acinus , reduces the level of basal , starvation-independent autophagy and shortens life span . Importantly , these phenotypes were reversed by the phospho-mimetic AcnS437D mutation . The beneficial outcomes that result from Acinus stabilization , including the extended lifespan and suppression of polyQ protein accumulation ( our data and Nandi et al . , 2014 ) , argue that increases in autophagic flux and autolysosome size in stabilized phospho-mimetic AcnS437D mutants are not a response to a proteotoxic stress or reflective of a defect in lysosomal function due to AcnS437D expression , but reflect a beneficial activation of autophagy as previously observed in multiple systems ( e . g . Sarkar et al . , 2007; Simonsen et al . , 2008; Menzies et al . , 2015; Gelino et al . , 2016 ) . Therefore , these findings indicate the importance of Cdk5-mediated Acn-S437 phosphorylation for maintaining neuronal health . How does phosphorylation of S437 stabilize Acinus and boost its function ? One possibility is a reduction in caspase-mediated cleavage of Acinus ( Sahara et al . , 1999 ) . Inhibition of this cleavage has previously been shown as a consequence of Akt1-mediated Acinus phosphorylation in apoptotic ( Hu et al . , 2005 ) and also in non-apoptotic cells ( Nandi et al . , 2014 ) and furthermore , upon binding of the anti-apoptotic protein AAC11 ( Rigou et al . , 2009 ) . Our data , however , argue against this possibility . We have previously shown that preventing its caspase-mediated cleavage in the AcnD527A mutant stabilized Acn only in a subset of photoreceptor cells ( predominantly R3 and R4 , Nandi et al . , 2014 ) . By contrast , S437 phosphorylation elevates Acinus levels in the majority of photoreceptor cells ( Figure 1 ) . We thus favor the notion of a different Acn cleavage , closer to the S437 residue . Such a cleavage has been reported for mammalian Acinus ( Sahara et al . , 1999 ) at a residue corresponding to A423 in Drosophila Acn . It will be important to identify the protease responsible for this cleavage and test its impact on Acn function . In the context of neurodegenerative diseases , much interest has been focused on the aberrant activation of Cdk5 . Elevated Cdk5 activity can be induced by multiple stressors such as inflammation , ischemia , or mitochondrial dysfunction ( Su and Tsai , 2011 ) . These stressors can trigger pathological activation of Cdk5 by calpain-mediated cleavage of its p35 or p39 co-activators ( Lee et al . , 2000 ) . The cleavage product p25 lacks the myristoylation-tag that anchors activated Cdk5 in complexes with p35 or p39 to membranes and escapes the inhibitory auto-phosphorylation of p35/39 that limits their life time in active complexes ( Patrick et al . , 1999 ) . The resulting unrestrained phosphorylation of proteins involved in microtubule-based axonal transport and synaptic proteins contributes to the progression of ALS , Alzheimer’s diseases and other neurological diseases ( McLinden et al . , 2012; Klinman and Holzbaur , 2015 ) . In Drosophila , we observe Cdk5-depedent elevated Acn-S437 phosphorylation in response to the expression of polyQ proteins , but we do not know yet whether this increase reflects elevated Cdk5/p35 activity or pathological Cdk5/p25 activity following calpain-mediated cleavage of p35 . Alternatively , the phosphatases that remove the phosphate group from pS437- Acn may be inhibited . Acn-S437 phosphorylation is highly dynamic in developing eyes ( Figure 1 ) , but phosphatases acting on pS437-Acn have not yet been identified . Interestingly , in addition to multiple polyQ proteins , Aβ42 expression is another Drosophila neurodegenerative disease model that triggered elevated phosphorylation of Acn-S437 . This is consistent with increased Cdk5 activity reported for multiple Alzheimer’s disease models ( Otth et al . , 2002; Shah and Lahiri , 2017 ) which in turn may contribute to the Cdk5-dependent phosphorylation of Tau as a possible contribution to the progression of Alzheimer’s disease ( Cruz et al . , 2003 ) . Not all neurodegeneration models induced Cdk5 activity as visualized by Acn-S437 phosphorylation . Pathological Cdk5 activation is believed to contribute to Parkinson’s disease ( Smith et al . , 2006 ) . Nevertheless , overexpression of alpha-Synuclein , which mimics some aspects of Parkinson’s disease in Drosophila ( Kontopoulos et al . , 2006 ) , is not sufficient to increase pS437-Acn levels . Similarly , Acn phosphorylation was unchanged in response to hSOD1 over-expression in a Drosophila model of ALS ( Watson et al . , 2008 ) . Both of these models trigger substantial cellular stress and rapid neurodegeneration ( Jaiswal et al . , 2012 ) , indicating that Acn phosphorylation is not a generic response to cellular stress , but more likely involves specific activation of Cdk5 . An example of such a specific interaction was recently elucidated in the context of polyQ proteins . Ashkenazi et al . ( 2017 ) showed that the levels of the autophagy regulator Beclin1 are maintained by its Ataxin3-mediated de-ubiquitination . Their interaction is mediated by the polyQ tract in wild-type Ataxin3 and inhibited by the presence of pathological polyQ tracts known to trigger Huntington’s disease or SCA3 ( Ashkenazi et al . , 2017 ) . Furthermore , an increasing number of autophagy receptors are being identified that drive selective autophagy of specific cargoes ( Rogov et al . , 2014; Khaminets et al . , 2016 ) . Interestingly , several of these receptors are activated by modifications such as phosphorylation or ubiquitination in response to specific stressors , such as polyQ proteins ( Deng et al . , 2017 ) . To which extend such modifications of autophagy receptors contribute to the induction of basal autophagy by Acn remains to be tested . Intriguingly , Cdk5-mediated phosphorylation stabilizes Huntingtin ( Luo et al . , 2005 ) and thereby may promote its function as a scaffold for selective autophagy ( Ochaba et al . , 2014; Rui et al . , 2015 ) . Cdk5 may thus act through different effectors in different diseases . Endophilin B1 was identified as a Cdk5 substrate , the phosphorylation of which is required for starvation-induced autophagy ( Wong et al . , 2011 ) . Importantly , Cdk5-mediated phosphorylation of Endophilin B1 appeared necessary for the elimination of dopaminergic neurons in an MPTP mouse model of Parkinson's disease ( Wong et al . , 2011 ) . Furthermore , MEKK1 is a key target for Cdk5 in a Drosophila model of retinitis pigmentosa ( Kang et al . , 2012 ) . Such diversity of targets may not only apply to different diseases , but also to different stages of a given disease . Although the pathological activation of Cdk5 activity appears to be a contributing factor to the progression of some neurodegenerative diseases , we show that wild-type levels of Cdk5/p35 activity are necessary to support basal autophagy in the clearance of protein aggregates . We show that this effect , at least in part , is due to Cdk5-mediated phosphorylation of Acn as phospho-mimetic AcnS437D reverses the reduced basal autophagy and the shortened life span observed in p35 mutants due their adult-onset , progressive neurodegeneration ( Connell-Crowley et al . , 2007; Trunova and Giniger , 2012 ) . How does the Cdk5-mediated phosphorylation of Acn-S437 elevate the level of basal autophagy ? We previously have shown that levels of Acn are critical for regulating basal autophagy in an Atg1-dependent manner ( Nandi et al . , 2014 ) . High Acn levels can drive excessive autophagy , even in the presence of activated mTor , and cause autophagy-dependent lethality ( Haberman et al . , 2010 ) . More subtle effects resulted when inhibiting caspase-mediated cleavage or phospho-mimetic mutations elevated levels of Acinus expressed from its own promoter ( Nandi et al . , 2014 , Figure 1 ) . For example , the phosphorylation of two conserved Akt1-target sites has been shown to elevate Acn levels ( Hu et al . , 2005; Nandi et al . , 2014 ) . In Drosophila , this stabilized Acn promotes starvation-independent basal autophagy ( Nandi et al . , 2014 ) . Similarly , we find that Cdk5-mediated phosphorylation stabilizes Acn protein and promotes autophagy ( Figure 7 ) , as does the phospho-mimetic AcnS437D mutation ( Figure 2 ) . Nevertheless , stabilized Acinus proteins do not reach the levels necessary to cause the developmental defects that can be triggered by overexpression through the Gal4/UAS system ( Haberman et al . , 2010; Nandi et al . , 2014 ) . Nuclear localization of Acn as a Cdk5 substrate seems to be in conflict with the localization of the activated Cdk5/p35 complex to the plasma membrane due the myristoylation of p35 ( Patrick et al . , 1999 ) . However , non-myristoylated p35 and p39 preferentially accumulate in the nucleus and can bind and activate nuclear Cdk5 ( Asada et al . , 2008 ) . How nuclear or possibly cytoplasmic Acn induces autophagy remains unclear . Phosphorylated and unphosphorylated Acn proteins are primarily nuclear and neither phosphorylation at S437 nor the two Akt1 target sites S641 and S731 are necessary for starvation-induced autophagy , although they enhance basal autophagy ( Figure 2 and Nandi et al . , 2014 ) . Acn is a required component of the nuclear ASAP complex ( Schwerk et al . , 2003; Murachelli et al . , 2012 ) which participates in the regulation of alternative splicing ( Hayashi et al . , 2014; Malone et al . , 2014 ) . Future work will therefore focus on attempts to identify specific Acn-dependent transcripts that may play a role in the regulation of autophagy or identifying alternative mechanisms for its role in the regulation of basal , starvation-independent autophagy . Flies were maintained using standard conditions . Bloomington Stock Center provided Da-Gal4 , Arm-Gal4 , GMR-Gal4 driver lines , w1118 , the RNAi lines and human neurodegenerative disease model lines ( BS lines: 33769 , 8141 , 33818 , 51376 , 33606 , 8534 ) . Other fly strains used were p38bΔ45 , a null generated by transposon excision and removing most of the p38b coding region , UAS-p38b , UAS-p38bK53R ( Vrailas-Mortimer et al . , 2011 ) ; p3520C , which deletes ~90% of the p35 coding region , including all sequences required for binding to and activating Cdk5; ( Connell-Crowley et al . , 2007 ) , UAS-p35 , UAS-Cdk5 , UAS-Cdk5K33A; ( Connell-Crowley et al . , 2000 ) . A Cdk5 null allele and genomic rescue transgene ( Kissler et al . , 2009 ) were gifts from Edward Giniger , National Institute of Neurological Disorders and Stroke , Bethesda , Maryland . UAS-Htt-exon1-Q93 , ( Steffan et al . , 2001 ) , abbreviated UAS-Htt . Q93 , was a gift from Robin Hiesinger , Free University Berlin , Berlin , Germany . Transgenic flies were generated by BestGene , Inc . DNA constructs related to genomic acn were generated by standard mutagenesis of a 4 kb Acn DNA fragment sufficient for genomic rescue ( Haberman et al . , 2010 ) , confirmed by sequencing , cloned into an Attb vector , and inserted into the 96F3 AttP landing site ( Venken et al . , 2006 ) . Similarly , UAS-controlled wild-type and mutant Acn transgenes were generated by standard mutagenesis from full-length Acn cDNA , confirmed by sequencing , and inserted into pUAS vectors modified by addition of an AttB site ( Nandi et al . , 2014 ) . Experiments with UAS-RNAi transgenes were performed at 28°C to maximize knockdown efficiency . Starvation resistance and life span were analyzed as described previously ( Nandi et al . , 2014 ) . Briefly , for starvation resistance 4- to 5-day-old virgins were kept in vials containing 1% agarose in 1X PBS at 25°C and dead flies are counted every 6 hr intervals . To measure life spans , males that emerged within a 2-day period were aged for an additional 3 days , kept in demographic cages and their survival at 25°C was recorded every other day . Antibodies against pS437-Acn was raised in rabbits by Genemed Synthesis against the Acn peptide H I V R D P- S ( p ) -P A R N R A S and double-affinity purified . For Western blots , five 96 hr larvae were crushed in 300 µl lysis buffer ( 10% SDS , 6 M urea , and 50 mM Tris-HCl , pH 6 . 8 ) at 95°C , boiled for 2 min , and spun for 10 min at 20 , 000xg . 20 µl lysate from larvae were separated by SDS-PAGE , transferred to nitrocellulose membranes , blocked in 3% non-fat dried milk and probed with mouse antibodies against Actin ( JLA20 ) or Myc ( 9E10; both at 1:2 , 000; Developmental Studies Hybridoma Bank ) , guinea pig anti-Acn ( aa 423–599 , 1:3 , 000 , Haberman et al . , 2010 ) . For Western blots with pS437-Acn antibodies , 1% BSA was used as blocking reagent . Using IR-dye labeled secondary antibodies and the Odyssey scanner ( LI-COR Biosciences ) bound antibodies were detected and quantified by comparison to Actin . Prestained molecular weight markers ( HX Stable ) were obtained from UBP-Bio . Non-radioactive Cdk5 in vitro kinase assays were performed essentially as described ( Hong and Guan , 2017 ) . In brief , GST-Acn402-527 fusion proteins ( WT or S437A mutant ) were expressed in a 30-ml culture of BL21 bacteria and immobilized and purified on 10 µl glutathione sepharose beads using standard procedures ( Hong and Guan , 2017 ) . Alternatively , Streptag-2xMyc-tagged full-length Acinus proteins ( WT or S437A ) were expressed in S2 cells as described ( Stenesen et al . , 2015 ) , immobilized and purified on Strep-Tactin magnetic beads ( IBA ) according to the manufacturer’s instructions . Immobilized GST-Acn fusion proteins or Streptag-Myc-tagged Acn proteins were exposed to 10 or 30 ng Cdk5/p35 complex ( Upstate ( Sigma ) 14–477 ) in 30 µl kinase assay buffer ( 25 mM MOPS , pH 7 . 2 , 12 . 5 mM glycerol 2-phos-phate , 25 mM MgCl2 , 5 mM EGTA , 2 mM EDTA , 0 . 25 mM DTT , and 0 . 5 mM ATP ) for 20 min at 30°C . Immobilized Acn proteins were washed twice with PBS containing 0 . 1% Triton-X100 , eluted with 20 mM glutathione and analyzed by western blots using antibodies against Acinus ( 1:2000 , Haberman et al . , 2010 ) or pS437-Acn ( 1:2000 ) . Quantitative RT-PCR was used to measure transcript levels of Myc-tagged Acn transgenes and knockdown efficiencies as previously described ( Akbar et al . , 2011 ) . In short , RNA was isolated using TRIZOL ( Ambion ) according to the manufacturer’s instructions . 2 µg RNA was reverse transcribed using High-Capacity cDNA Reverse Transcription kit ( Applied Biosystems ) using random hexamer primers . Quantitative PCR was performed using the Fast SYBR Green Master Mix in a real-time PCR system ( Fast 7500; Applied Biosystems ) . Each data point was repeated three times and normalized for the message for ribosomal protein 49 ( RP49 ) . Primers were: Myc-left , 5′-CTGGAGGAGCAGAAGCTGAT-3′ , within the Myc and Acn_right , 5′-GGAGTCTCGACCTCGGTCTT-3′ , within the Acn coding regions , and RP49_left , 5′-ATCGGTTACGGATCGAACAA-3′ , and RP49_right , 5′-GACAATCTCCTTGCGCTTCT-3′ . Cdk5_left: 5’-AATGGAGAAGATCGGGGAGG-3’ Cdk5_right: 5’-GGGAGATCTGCC TGCTGA-3’ p38b_left: 5’-GACGCCGATCTGAACAACAT-3’ p38b_right: 5’-ATCCTGGATTTCGGTTTGGC-3’ p35_left: 5’-TGTTCTTGCACTGTCGTTGT-3’ p35_right: 5’-TCAGCGGAGAAGAGAGCAAG-3’ SEMs of fly eyes were obtained as previously described ( Wolff , 2011 ) . Briefly , eyes were fixed in 2% paraformaldehyde , 2% glutaraldehyde , 0 . 2% Tween 20 , and 0 . 1 M cacodylate buffer , pH 7 . 4 , for 2 hr . Fixed samples were washed for 12 hr each in a series of four washes with increasing ethanol ( 25–100% ) . This is followed by a series of hexamethyldisilazane washes ( 25–100% in ethanol ) for 1 hr each . Flies air dried overnight were mounted on SEM stubs and coated in fast-drying silver paint on their bodies only . Flies were sputter coated with a gold/pallidum mixture for 90 s and imaged at 1000 × magnification , with extra high tension set at 3 . 0 kV on a scanning electron microscope ( SIGMA; Carl Zeiss ) . The microscope was equipped with the InLens detector ( Carl Zeiss ) . For TEM , size-matched 96 hr fed larvae were dissected and processed as described earlier in Nandi et al . ( 2014 ) . In short , dissected larvae were fixed in 2% glutaraldehyde in 0 . 1 M cacodylate buffer , pH 7 . 2 and postfixed with 2% OsO4 and 1 . 5% KFeCN in the same buffer . Samples were embedded in epoxy resin , sectioned . Sections were stained with uranyl acetate and lead citrate to enhance contrast , examined with a transmission electron microscope ( 120 kV; Tecnai G2 Spirit BioTWIN; FEI ) , and images were captured with an 11-megapixel camera ( Morada; Olympus ) . From TEMs , measurements of autolysosomal diameters and areas were obtained using Macnification software ( Orbicule ) . Whole-mount tissues were prepared for immunofluorescence staining as previously described ( Akbar et al . , 2011 ) . Briefly , dissected samples were fixed in periodate-lysine-paraformaldehyde , washed in PBS , permeabilized with 0 . 3% saponin in PBS ( PBSS ) , blocked with 5% goat serum in PBSS , and stained with the indicated primary antibodies: guinea pig anti-Acn ( 1:1000 , Haberman et al . , 2010 ) , mouse anti-Myc ( 9E10 , 1:1000 ) , mouse anti-FLAG ( 1:1000; Sigma ) , rabbit anti-p62 ( 1:2000 , Pircs et al . , 2012 ) , a gift from G . Juhàsz ( Eötvös Loránd University , Budapest , Hungary ) , rabbit anti-GABARAP ( 1:200; Abcam , ab109364 ) , which detects endogenous Atg8a ( Kim et al . , 2015 ) , mouse-anti 1C2 ( 1:1 , 000; MAB1574; EMD Millipore ) , rabbit anti-Boss ( 1:2000 , Krämer et al . , 1991 ) , and secondary antibodies were labeled with Alexa Fluor 488 , 568 , or 647 ( 1:500; Molecular Probes ) and mounted in Vectashield containing DAPI ( Vector Laboratories ) . Fluorescence images were captured with 63× , NA 1 . 4 or 40× , NA1 . 3 Plan Apochromat lenses on an inverted confocal microscope ( LSM 510 Meta or LSM 710; Carl Zeiss Jena ) . Confocal Z-stacks of eye discs were obtained at 1 µm step size . For phosphatase treatment , dissected third instar larval carcasses were fixed in periodate-lysine-paraformaldehyde and treated with 130 U/ml Calf Intestinal Phosphatase ( New England Biolabs , Inc . ) for 3 hr at 37°C with 1X protease inhibitor tablet ( Roche ) dissolved in 1X PBS , pH 7 . 5 . Subsequently , tissues were processed and stained , and eye disc mounted as described above . For autophagy flux experiments , 72-hr-old larvae were transferred to fresh medium containing 3 mg/ml chloroquine ( Sigma ) as described ( Lőw et al . , 2013 ) . LysoTracker staining ( GFP-Certified Lyso-ID red lysosomal detection kit; Enzo Life Sciences ) of size-matched 90–96 hr fat bodies from fed and starved larvae was performed as previously described ( Rusten et al . , 2004; Scott et al . , 2004 ) . In brief , larvae were dissected in Schneider’s Drosophila media ( Gibco ) , inverted to expose fat bodies , and incubated in 100 µM LysoTracker Red DND-99 for 1 min . Inverted carcasses were then washed in 1X PBS and fat bodies were mounted onto a droplet of Vectashield ( Vector Laboratories ) . Samples were imaged immediately on an inverted confocal microscope ( LSM 510 Meta using 63× , NA 1 . 4 Plan Apochromat lens ) . Z-projections of three optical sections of fat body tissue , each 1 µm apart were used to quantify LysoTracker and Atg8a punctae in fat bodies using Imaris software ( Bitplane ) . The number of punctate was quantified per fat body cell . Digital images for display were imported into Photoshop ( Adobe ) and adjusted for gain , contrast , and gamma settings . Integrated intensities of Atg8a punctae in fat bodies were determined using Image J and normalized to AcnWT . Integrated densities for pS437-Acinus and polyQ in eye discs were quantified using Image J software . Identical areas posterior to the morphogenetic furrow were quantified , thereby excluding dividing cells close to the furrow that were strongly stained for pS437-Acinus . All immunofluorescence experiments were repeated at least three times with at least three samples each . Polyglutamine aggregates were detected with a modified filter assay ( Scherzinger et al . , 1997 ) . Briefly , 25 heads from 2 week-old flies were homogenized in 200 µl Cytoplasmic Extraction Reagent I buffer and fractionated using NE-PER Nuclear and Cytoplasmic Extraction Reagents following the manufacturer’s protocol ( Thermo Fisher Scientific ) . Cytosolic fractions were adjusted to 1% SDS , incubated at room temperature for 15 min , denatured at 95°C for 5 min , and filtered through a 0 . 2 µm cellulose acetate membrane ( Sterlitech Corporation ) preequilibrated with 1% SDS . Membrane was washed twice with 0 . 2% SDS and blocked in TBS ( 100 mM Tris-HCl , pH 7 . 4 , and 150 mM NaCl ) containing 3% nonfat dried milk and probed with a mouse anti-Htt antibody ( 1:1 , 000; MAB5490; EMD Millipore ) . The bound antibodies are detected and quantified using anti-IR dye conjugated secondary antibodies and Odyssey scanner and software ( LI-COR Biosciences ) . Statistical significance was determined in Prism using log-rank for survival assays , chi square analysis for eye roughness frequencies , and one-way analysis of variance for multiple comparisons , followed by Tukey’s test . To separate effects of treatment and genetic background we used two-way analysis of variance for multiple comparisons , followed by Bonferroni’s test for individual comparisons . All bar graphs resulting from these analysis show means ±SD . For quantifications of fluorescence images and Western blots , at least three independent experiments were used . P values smaller than 0 . 05 are considered significant , and values are indicated with one ( <0 . 05 ) , two ( <0 . 01 ) , three ( <0 . 001 ) , or four ( <0 . 0001 ) asterisks .

Cells have a problem that we recognize from our own homes: if nobody cleans up , garbage accumulates . Unwanted material in cells can include proteins that clump together and can no longer carry out their normal tasks . If left to build up , these protein aggregates can damage the cell and even kill it . Many neurodegenerative disorders , including Huntington’s disease and Alzheimer's disease , arise when such faulty proteins accumulate inside brain cells . Autophagy is a process that can destroy protein aggregates and other defective material to keep cells healthy . Understanding how cells regulate autophagy is thus of great interest to scientists . A protein called Acinus promotes autophagy and is found in many organisms including fruit flies and humans . All Acinus proteins share a common feature; they contain a site called Serine437 that can be modified by the attachment of a phosphate group , in a process known as phosphorylation . However , the significance of this modification was not clear . Nandi et al . have now asked if this phosphorylation event is important for the role of Acinus in autophagy . The experiments were carried out in the fruit fly , Drosophila melanogaster . Flies were engineered such that the normal Acinus protein was replaced with a mutant version that mimics the phosphorylation at Serine437 . These mutant flies had higher levels of Acinus , showed more autophagy , and lived longer when compared to normal flies . Further work identified a protein called Cdk5 as being responsible for attaching phosphate to Acinus at Serine437 . Making Cdk5 inactive using experimental tools led to lower levels of autophagy in brain cells and shortened the flies’ life span . Moreover , some aggregation-prone proteins linked to neurodegenerative diseases can enhance the activity of Cdk5 towards Acinus , thereby reducing their own accumulation through elevated autophagy . Together these findings show that phosphorylation of Acinus by Cdk5 maintains healthy brain cells and improves life span by enhancing autophagy . The next step is to understand how phosphorylation at Serine437 stabilizes Acinus to boost autophagy . This may lead to new ways to adjust the levels of autophagy to benefit different organisms .

2017

Stress-induced Cdk5 activity enhances cytoprotective basal autophagy in Drosophila melanogaster by phosphorylating acinus at serine437

An important challenge of crop improvement strategies is assigning function to paralogs in polyploid crops . Here we describe the circadian transcriptome in the polyploid crop Brassica rapa . Strikingly , almost three-quarters of the expressed genes exhibited circadian rhythmicity . Genetic redundancy resulting from whole genome duplication is thought to facilitate evolutionary change through sub- and neo-functionalization among paralogous gene pairs . We observed genome-wide expansion of the circadian expression phase among retained paralogous pairs . Using gene regulatory network models , we compared transcription factor targets between B . rapa and Arabidopsis circadian networks to reveal evidence for divergence between B . rapa paralogs that may be driven in part by variation in conserved non-coding sequences ( CNS ) . Additionally , differential drought response among retained paralogous pairs suggests further functional diversification . These findings support the rapid expansion and divergence of the transcriptional network in a polyploid crop and offer a new approach for assessing paralog activity at the transcript level . The transition from basic research in Arabidopsis to new model systems for monocot and dicot crops has focused attention on the implications of polyploidy for current models of genetic processes developed in Arabidopsis . The expansion of gene content through whole genome duplication ( WGD ) , tandem duplication , or transposed duplicates has been predicted to account for the evolution of morphological complexity ( Freeling and Thomas , 2006 ) . Improving crop yield in rapidly changing climates depends on our ability to integrate these gene content expansions into functional classifications of physiological importance . This will rely on the growing collection of sequenced genomes , not just across crop species but of ecotypes within species , including complementary genomic datasets such as transcriptomes , methylomes , chromatin accessibility profiles , and metabolomes . One difficulty in assigning new or overlapping functions among paralogs arises from heterogeneity in transcript abundance datasets generated under various environmental conditions , from various tissue types , and at distinct times of the day . Many studies have explored the potential for functional divergence of duplicated genes by comparing the expression levels normalized across a collection of expression studies ( Blanc and Wolfe , 2004; Ganko et al . , 2007; Schnable et al . , 2011; Woodhouse et al . , 2014 ) which limits the search to genes showing very dramatic differences in transcript abundance at a single time point . The importance of daily rhythms was recognized with the 2017 Nobel prize in physiology or medicine to Jeffrey Hall , Michael Rosbash , and Mike Young for their discoveries of the molecular mechanisms generating circadian rhythms in Drosophila ( Rosbash , 2017; Young , 2018 ) . The conservation of circadian oscillators across the animal and plant lineages supports a role for these rhythms in maintaining fitness and evolving new regulatory pathways to fulfill that role ( Bell-Pedersen et al . , 2005 ) . Many lines of evidence support the importance of circadian rhythms to plant biology , including photosynthesis , starch metabolism , biomass accumulation , and reproduction ( Greenham and McClung , 2015; Millar , 2016 ) . The circadian clock responds to environmental conditions to set these circadian rhythms to local time ( Greenham and McClung , 2015 ) . As a consequence , circadian rhythms and thus much of plant biology are likely to be influenced by climate change . Examples of natural variation in plant circadian function are accumulating , as is evidence that many domestication traits that facilitated the geographic expansion of crops are due to alterations in circadian clock genes ( Nakamichi , 2015; Müller et al . , 2016; Müller et al . , 2018 ) . This supports the utility in targeting circadian clock processes as a means of trait improvement without disrupting critical pathways required for growth and yield . Plant circadian biologists have focused primarily on Arabidopsis as a model for defining circadian clock components and function in plants ( Creux and Harmer , 2019 ) . Transcriptome studies have revealed extensive circadian control of transcript abundance resulting in time of day changes in expression ( Covington and Harmer , 2007; Mockler et al . , 2007; Michael et al . , 2008 ) . These rhythmic changes in transcript abundance are not unexpected given the daily changes in light , temperature , and precipitation that affect physiological processes such as photosynthesis . Dynamic changes in metabolism and physiology must be driven by dynamic changes in gene expression and ultimately protein regulation and activity . To identify candidate circadian regulators for trait improvement in crops , a more detailed time-course resolution of transcript abundance levels is needed to confirm whether the diel and circadian patterns observed in Arabidopsis are maintained in polyploid crops . The crop plant Brassica rapa offers an excellent model system for studies in crops . It is a member of the Brassicaceae and a close relative of Arabidopsis making comparative studies feasible . The morphological diversity in B . rapa with turnip , Chinese cabbage , pak choi , leafy and oil-type varieties provides a wealth of phenotypic traits to study in one species allowing for broad applicability to other crops . Preliminary studies have shown diversity in circadian clock parameters among morphotypes that correlate with various physiology measures suggesting that circadian clock variation has contributed to B . rapa diversification ( Yarkhunova et al . , 2016 ) . Examination of the orthologs of known circadian clock genes in Arabidopsis revealed preferential retention of these genes in B . rapa following the triplication and extensive fractionation of the genome after the divergence of Brassica and Arabidopsis from their common ancestor around 24 million years ago ( MYA; Lou et al . , 2012 ) . The preferential retention of clock genes is consistent with their involvement in protein complexes and regulation of critical pathways making them sensitive to dosage effects . The gene dosage balance hypothesis proposes that duplication of the entire genome is favored over single or chromosome level duplications because it maintains the appropriate concentration of gene products ( Conant et al . , 2014 ) . This is supported by studies in yeast where genes of protein complexes tend to be lost simultaneously with their interacting proteins ( Pires and Conant , 2016 ) . The increase in expression of one duplicate could lead to or permit the loss of the other duplicate or neo-functionalization ( Pires and Conant , 2016 ) . To assess the functional significance of the retention of circadian clock genes in B . rapa and look for possible examples of neo-functionalization , we performed two high-resolution circadian transcriptome experiments to characterize the circadian network . To compare the expression dynamics of paralogous genes , we developed a novel method for identifying and classifying changes in expression patterns , an R package called DiPALM ( Differential Pattern Analysis via Linear Models ) . DiPALM facilitated a comparison of paralog expression patterns , revealing the genome-wide expansion of phase domains among paralogs providing novel insight into the rapid divergence of the transcriptional network in this crop . To identify transcriptional regulators exhibiting functional divergence , we compared the circadian gene regulatory networks ( GRNs ) between B . rapa and Arabidopsis . Using previously generated circadian microarray data in Arabidopsis , we compared GRNs to identify the more Arabidopsis-like versus the more divergent ( less Arabidopsis-like ) among members of B . rapa paralogous transcription factors ( TF ) gene pairs based on conservation of connected targets in the network . The identification of the more Arabidopsis-like TF ortholog was supported by the presence of conserved noncoding sequences ( CNSs ) surrounding TF target genes , reinforcing the importance of these CNSs for regulating gene expression . To associate these diverging patterns with a biological pathway , we applied DiPALM to our recent drought time-course experiment in B . rapa ( Greenham et al . , 2017 ) and discovered differential responses to mild drought stress among paralogs suggesting neo- and sub-functionalization . Studies in Arabidopsis have demonstrated widespread circadian regulation of the transcriptome with distinct and overlapping genes responding to various entraining conditions ( Covington et al . , 2008; Michael et al . , 2008 ) . As noted above , genes contributing to circadian clock function have been retained in multiple copies following a WGD in B . rapa ( Lou et al . , 2012 ) . We wondered whether there have been concomitant effects on the extent of circadian regulation of the B . rapa subsp . trilocularis ( Yellow Sarson ) R500 ( henceforth called B . rapa R500 ) transcriptome and the expression patterns of circadian regulated paralogs . We conducted two RNA-seq experiments designed to capture the genes under circadian regulation entrained by light and temperature zeitgebers ( German for ‘time givers’ ) . We entrained plants to light-dark ( LD ) cycles at constant temperature or thermocycles ( HC ) in constant light before transfer to constant light at 20°C ( LLHH; see Materials and methods ) . Following 24 hr in constant conditions leaf tissue from the youngest leaf was collected and flash-frozen in liquid nitrogen every 2 hr for 48 hr . All samples were harvested from biological replicates of plants from the LD and HC entrainment that had been transferred to constant conditions of LLHH . To identify the circadian transcriptome , we analyzed the LD and HC datasets and ran the circadian analysis program RAIN ( Thaben and Westermark , 2014 ) , a nonparametric method for the detection of rhythms from a variety of waveforms that are typical of transcript abundance datasets . The 2 hr sampling regimen provided the resolution to capture more cycling genes than possible with typical 4 hr sampling ( Hughes et al . , 2017 ) . Using a Benjamini-Hochberg corrected p-value of 0 . 01 , we identified 16 , 447 high confidence circadian regulated genes from the two datasets . Of the 22 , 204 genes that were expressed in the RNA-seq datasets , 74% of them passed our cutoff for cycling in one or both conditions , indicating retention of circadian regulation of the transcriptome following WGD in B . rapa . To assign cycling genes to specific phase bins based on the timing of peak expression , we generated co-expression networks for each dataset using the weighted gene correlation network analysis ( WGCNA ) package in R ( Langfelder and Horvath , 2008 ) , which has proven to be an effective method for grouping similarly phased genes based on their expression pattern ( Greenham et al . , 2017 ) . This resulted in 14 modules in the LD dataset and 10 modules in the HC dataset . To demonstrate the uniformity of the genes within each module , a heatmap was generated with the log2 transformed expression data of each gene for all modules numbered based on their phase ( Figure 1 ) . Each module reflects a similarly phased set of genes that collectively are phased throughout the day with the LD_01 module showing peak expression at ZT24 ( subjective dawn; ZT refers to Zeitgeber Time and indicates the number of hours since the last dark-to-light or cold-to-warm transition ) and LD_09 showing peak expression at ZT36 ( subjective dusk ) . The similarity in patterns seen in the LD and HC heatmaps in terms of phasing and distribution of genes within those phase bins suggests that there may be considerable overlap in gene phasing after LD and HC entrainment . To quantify the overlap , we matched the genes across the two networks and compared the correlation of eigengenes ( first principle component of the expression matrix for each module ) between LD and HC modules . The circos graph in Figure 1 depicts the overlap between the two networks where the width of the ribbon represents the number of genes in common between the two modules and the color signifies the Pearson correlation coefficient between the eigengenes of the two datasets with dark orange being a correlation coefficient of 1 . Because the modules are numbered based on their phase , similarly phased modules are arranged in the same order in the circos plot and the significant overlap and expression pattern between these modules is evident . Figure 1—figure supplement 1 shows the distributions of phase difference between LD and HC for all 16 , 447 genes that cycle in both conditions . 13 , 850 genes ( ~84% ) have a phase difference of less than or equal to 4 hr . These comparisons demonstrate that most genes have the same or similar phasing when entrained by either photocycles or thermocycles . To further assess the similarity between these two datasets , GO ontologies were compared for each module from the LD and HC experiments ( Supplementary file 1 ) . This revealed similar biological processes enriched in the modules with a high correlation in expression patterns . For example , genes in LD_03 and HC_02 were both significantly enriched ( FDR adjusted p-value<0 . 01 ) for photosynthetic processes and response to abiotic stress , consistent with their morning phased expression . Interestingly , genes in LD_07 and HC_07 were significantly enriched for protein phosphorylation suggesting a time-of-day regulation of this process , as in Arabidopsis ( e . g . Choudhary et al . , 2015 ) . The correlation between module membership is not a rigorous test of differential transcript abundance and many genes have low correlations with their module eigengene that are not reflected in the analysis in Figure 1 . To the best of our knowledge , there was not a rigorous test available for identifying significantly different gene expression patterns that would classify the change based on phase , amplitude , or a combination of both . Rather than looking at differentially expressed genes at any given time point , we felt it was more important to classify a pattern change that encapsulated the entire time course . Therefore , we developed an R package , DiPALM , which takes advantage of the network analysis that assigns an eigengene to each module and thereby produces a minimal set of patterns representing the entire dataset . The expression correlation of a given gene to any module’s eigengene defines the module membership ( kME ) of that gene to the module . The combination of kMEs for a gene across all modules can be used to encode its expression pattern numerically and allows for quantitative comparisons between any two gene’s expression patterns . This allowed us to run a set of linear model contrasts ( one for each eigengene ) that is analogous to running a contrast of gene expression data between time points or treatments except in this case the kME value represents the entire expression profile across the time course . We first tested for differential expression patterns between the LD and HC datasets for the set of 15 , 101 genes that cycle after entrainment to either LD and HC . Both datasets were collected in constant conditions of light and temperature ( LLHH ) following different entrainment regimens as specified by their names . To generate a significance cutoff , we also ran the analysis on a permuted gene expression set where gene accessions were randomly re-assigned to expression patterns . P-values were then calculated using this permuted set . Using a p-value cutoff of 0 . 01 , we identified just 1713 genes , or 11% of all cycling genes , that have altered patterns following LD versus HC entrainment . To quantify overall expression level variation , we ran a similar linear model analysis on the median expression level for all genes and identified 3465 ( 23% ) of genes , only 448 of which overlapped with the pattern change list ( Supplementary file 2 ) . The 11% of cycling genes with entrainment-dependent cycling patterns are interesting but not within the scope of this manuscript . A functional enrichment analysis of these genes revealed translation initiation factors and ncRNA metabolic process among the significant ( FDR adjusted p-value<0 . 01 ) functional categories ( Supplementary file 1 , ‘Differential_Pattern’ Tab ) . For the remainder of analyses with these datasets , we have combined LD and HC to increase our statistical power by having four replicates per time point rather than two . One significant advantage of a linear model-based framework is the ability to account for any identified effect . We make use of this feature by modeling any LD versus HC entrainment effects as a covariate . In other words , DiPALM gives us the ability to combine these datasets while still accounting for any differences between LD and HC entrainment . This combined dataset ( referred to as LDHC ) provides more statistical power for further analysis , particularly for the 89% of genes that show no significant change between LD and HC . For the other 11% that do have differences , these differences are taken into account and will not adversely affect the results . The network analysis revealed that a large portion of the transcriptome exhibits rhythmic expression patterns . This implies that multi-copy paralogs have retained their rhythmic expression patterns and circadian regulation consistent with the preferential retention of circadian clock orthologs in B . rapa ( Lou et al . , 2012 ) . Based on the gene dosage model , the balance in expression among different subunits of protein complexes must be maintained resulting in the proper adjustment of paralog expression level , in some cases resulting in one copy maintaining high expression while the other is repressed ( Conant et al . , 2014 ) . To evaluate this model , we calculated the mean expression level for all cycling genes across the 48 hr time course . The mean expression level for the set of retained multi-copy paralogs was significantly higher ( p-value=0 . 0 based on one-way ANOVA with Tukey’s test ) than for genes retained in a single copy ( Figure 2A ) . It is possible that one of the retained copies is expressed at a much higher level than the average single-copy gene as well as its paralog . To test whether this is the case , we separated all the cycling two- and three-copy paralogs into the highest and lowest expressed copies based on the average expression level across all time points . In the case of three-copy paralogs , we only included the highest and lowest expressed genes in the analysis . We compare this to randomly paired single-copy gene pairs that were also separated into high and low expression groups ( Figure 2B ) . Surprisingly , the difference is observed in the low expressed paralog where these duplicated genes have significantly higher ( p-value=0 . 0 based on one-way ANOVA with Tukey’s test ) average expression levels compared to the genes retained in a single copy , suggesting that although the duplicate paralogs do appear to exhibit gene dosage , their overall expression is retained at a higher level than the expression of single-copy genes . The retention of multi-copy genes that are under circadian regulation and maintained at a relatively high expression level led us to explore whether there is evidence for divergence in expression pattern that would support neo- or sub-functionalization among paralogs . To associate similar patterns , we applied the same WGCNA method to the combined LDHC dataset as was done for the individual analyses shown in Figure 1 . This resulted in 12 modules with distinct phasing throughout the day that is visible when the eigengene expression for each module is presented as a heatmap ( Figure 2C ) . Next , we wondered whether there was any association between the phase of expression and retained copies that may suggest certain biological processes that are phased to specific times of the day and may preferentially retain multi-copy genes . Based on the number of genes within each module , we ran a hypergeometric test to look for over- and under-enrichment of multi-copy genes within the modules . Surprisingly , we found that modules with phasing from morning to midday tend to be enriched for multi-copy genes . By contrast , evening and night phased modules were depleted for multi-copy genes ( Figure 2D ) . These trends were not associated with the number of genes within the module as can be seen with two of the largest modules , LDHC_02 and LDHC_09 , being over- and under-enriched , respectively . GO enrichment was carried out on a combined group of all multi-copy genes from the five morning modules with significant enrichment for multi-copy genes ( p-value<0 . 05 ) . The same was done for the group of all multi-copy genes from all four evening modules with significant depletion in multi-copy genes . Both of these sets appear to be representative of the whole modules from which they came with the morning-phase copied genes being significantly enriched ( FDR adjusted p-value<0 . 01 ) for photosynthesis , translation , and response to abiotic stimulus genes . The evening-phase multi-copy genes were significantly enriched for protein phosphorylation and glycosinolate biosynthesis genes ( Figure 2D , Supplementary file 3 ) . To look for signs of possible neo- or sub-functionalization among paralogs , we compared expression profiles to identify paralogs with significantly different expression patterns . For three-copy paralogs , where all three copies ( a , b , and c ) were expressed , these sets were converted into three , two-member pairs ( ab , bc , and ac ) . We applied a similar linear modeling test using DiPALM that we ran on the LD and HC comparison but including the covariate in the model to account for differences following LD versus HC entrainment . We ran this analysis on 4433 pairs where both genes are expressed . We found 3743 ( 84% ) pairs with differential median expression and 1883 ( 42% ) pairs exhibiting differential expression patterns ( p-value<0 . 01 ) , the vast majority ( 1607/1883; 85% ) of which showed both differential median expression and pattern ( Supplementary file 4 ) . However , this does not describe how the patterns differ . As with standard differential expression tests , it is critical to associate a direction of change in expression to know how a gene transcript is affected by treatment or condition . To isolate the type of pattern change for the paralogous pairs exhibiting significantly different patterns , we performed clustering on the vector of expression values across the combined LDHC data for each gene whose expression differed significantly from its paralog . This clustering had the effect of grouping genes based on their phase . Similar clustering was done for the significantly different median expression set . This clustering component is also part of the DiPALM package . A detailed description and example dataset of the analysis pipeline is provided with the package on CRAN ( R Development Core Team , 2018 ) . To visualize the degree of pattern change , we generated a heatmap of each of the clustering methods with the paralogous gene pairs in the same row for comparison . As expected , the pattern clustering uncovered the changes in rhythmic patterns between pairs ( Figure 3A ) while the median expression clustering uncovered overall changes in transcript abundance ( Figure 3—figure supplement 1 ) . From the heatmap visualization , it is clear that the majority of the pattern change among paralogous pairs is the result of an altered phase of peak expression . In some cases ( Figure 3A and B , clusters 4 and 12 ) the pairs are completely antiphase . This suggests a genome-wide expansion of phase domains among retained paralogs . The divergence in expression pattern among retained paralogs led us to ask how diverged the retained pairs are with respect to their Arabidopsis ortholog ( Figure 4A ) and whether one copy exhibits an Arabidopsis-like expression pattern while the other copy exhibits a different expression pattern . One method of comparing the orthologs between B . rapa and Arabidopsis is to compare the phase of expression . However , assigning an accurate phase to circadian data from two cycles is challenging and often gene expression patterns show very broad peaks in abundance that can be difficult to classify , especially with the resolution of only 4 hr that is available for Arabidopsis . Also , because we are comparing two species , we have to consider the properties of the clock in each species . The B . rapa R500 clock has a slightly shorter period than Arabidopsis Col-0 that results in altered phasing among genes that is not indicative of divergence in function but arises merely from the different paces of the Arabidopsis and B . rapa oscillators . For example , if we plot the expression of single-copy circadian clock genes from B . rapa and their corresponding orthologs in Arabidopsis we see a leading phase in B . rapa ( Figure 4—figure supplement 1 ) . To avoid these phase complications , we chose a gene regulatory network ( GRN ) approach using GENIE3 ( Huynh-Thu et al . , 2010 ) that would provide additional statistical robustness by first predicting transcription factor ( TF ) targets based on expression dynamics within each species followed by a comparison of network connections between the species ( Figure 4B ) . We obtained previously published circadian microarray data from Arabidopsis that were generated under similar conditions with LD and HC entrainment ( LL_LDHC , LL_LLHC , LL12_LDHH , and LL23_LDHH from Mockler et al . , 2007 ) . We selected Arabidopsis TFs from the Arabidopsis TF database ( https://agris-knowledgebase . org/AtTFDB/ ) and B . rapa TFs from the Mapman annotation ‘RNA regulation of transcription’ in addition to known circadian clock TFs . This resulted in a list of 612 Arabidopsis and 2147 B . rapa TFs that were expressed in their respective datasets . For the target set , we included 9201 Arabidopsis expressed genes and the corresponding 14 , 541 B . rapa expressed orthologs . Separate GRNs were generated for Arabidopsis and B . rapa . To identify the significance of TF-target edges , we generated a permuted network by using the same TFs but shuffling the expression values for the target genes . These permuted network edges were used as a null distribution to select edges from the actual networks with a 5% FDR , resulting in an Arabidopsis GRN with 71 , 216 edges and a B . rapa GRN with 947 , 062 edges . A gene was said to be a target of a TF if one of these significant edges existed between them . We hypothesized that the paralogous TF in B . rapa that displayed more of the Arabidopsis orthologous function would have a greater overlap in targets in the network compared to the more divergent pair member . In other words , using TFs as features to describe the targets , how well can the expression of the target genes be explained by the expression of that TF ( Figure 4B ) ? A set of 256 TFs exists where one Arabidopsis TF can be associated with two B . rapa paralogs and all three of these genes had target groups defined by their respective GRNs . For each of these 256 sets , we examined the significance of the overlap between the Arabidopsis TF target group versus the corresponding B . rapa orthologous TF target groups . This resulted in two p-values for each group indicating how similar each B . rapa TF is to its orthologous Arabidopsis TF in terms of target gene overlap ( Supplementary file 5 ) . Next , we wanted to determine if the difference in these two p-values was significant; that is , does one of the B . rapa TFs show more conservation of target gene overlap with its Arabidopsis ortholog than the B . rapa paralogous TF ? This was accomplished with another permutation-based test where genes were randomly sampled to create target groups of the same sizes . The two B . rapa versus Arabidopsis p-values were calculated and the difference was taken . This was repeated 10 , 000 times for each of the 256 TF sets . 49 TF pairs exhibited significant enrichment ( p-value<0 . 05 ) for one B . rapa TF ( assigned Br1 ) being more similar to Arabidopsis than its paralog ( Supplementary file 5 ) , suggesting possible divergence in function between these TFs ( Figure 4C and D ) . It is worth noting that the size of the target group is not driving the enrichment as we see a broad distribution in target size and significance ( Figure 4C ) . Among the list of 49 TFs , six are part of the core circadian clock ( ELF3 , ELF4 , PRR9 , PRR7 , PRR5 , and TOC1 ) , and a seventh , RVE1 , integrates the circadian clock and auxin pathways ( Rawat et al . , 2009; McClung , 2019 ) . Based on the predicted targets of these TFs in the GENIE3 model , there are 11 , 559 B . rapa and 3387 Arabidopsis genes regulated by these 49 TFs , providing further support for an expansion of the circadian network in B . rapa . The divergence in TF target genes indicates several possible changes have occurred; these could include modifications to regulatory elements of the target genes , mutations that alter the TF protein binding efficiencies for motifs or interacting partners , or a combination of both . Alterations to regulatory elements associated with core TFs and/or target genes can lead to whole pathway-level restructuring . One possible mechanism for altered expression regulation is the distribution of conserved noncoding sequences ( CNSs ) . A set of CNSs was identified across the Brassicaceae that show signs of selection ( Haudry et al . , 2013 ) . To associate CNSs with the B . rapa R500 genome , we performed a BLAST analysis with a collection of ~63 , 000 CNSs against the B . rapa R500 genome . Provided the alignment met our BLAST filters , we allowed each CNS to have a maximum of three targets ( see Materials and methods ) . We repeated the BLAST with the Arabidopsis genome but restricted each CNS to one target gene . To test for altered regulatory element occurrences between target genes of the identified diverged TFs , we asked whether variation in CNS retention followed the same pattern as the observed gene expression changes . Do we see a similar divergence in B . rapa paralogous TF enrichment with Arabidopsis in the GENIE3 network if we replace the target set gene ortholog data with CNSs ? With the list of genes and associated CNSs , we replaced the target genes in the GENIE3 networks with CNSs resulting in a network with TFs targeting a group of CNSs rather than genes . We performed the same permutation tests to assign significant enrichment to the groups to ask whether we could identify a B . rapa R500 TF ortholog that was more Arabidopsis-like than its paralog . We identified 68 significant TFs ( p-value<0 . 05 ) in the CNS network ( Figure 4E and F , Supplementary file 5 ) , 35 of which overlapped with the 49 TFs identified based on target gene overlap ( Figure 4G ) . The agreement between these two approaches is apparent when the corresponding p-value distributions from the overlap of targets are plotted ( Figure 4D and F ) . In these boxplots , the Br1 enrichment for At target gene overlap is shown with the paralogous pairs connected by the red lines . Not only does the agreement between the target gene and CNS overlap further strengthen the support for those 35 TF pairs showing signs of divergence but suggests that the CNS distribution is associated with gene expression patterns and is a good predictor of expression variation . With this set of high confidence diverged TFs from the overlap group , we wondered whether changes in TF amino acid sequence contribute to the divergence between paralogous pairs in which case we would expect the Arabidopsis-like B . rapa TF to be more similar than its paralog . To test this , we ran a protein BLAST using the B . rapa TFs against the Arabidopsis genome and examined the distribution of blast scores for the more and less Arabidopsis-like TF . Results from the BLAST suggest very little association between amino acid sequence and TF divergence ( Figure 4I ) suggesting that changes to regulatory regions associated with target genes are likely to be a major driver of TF divergence . TF expression pattern changes are likely contributing to the divergence in regulation . To test this further , we conducted a similar analysis to the BLAST comparison where we looked at the expression correlation of B . rapa TFs versus the orthologous Arabidopsis TF . In general , the more Arabidopsis-like B . rapa TF was more likely to maintain a higher expression correlation to the Arabidopsis ortholog but the effect is not significant ( Figure 4H , p-value 0 . 057 ) and several of the 35 TF sets tested do not show this result . However , due to the period difference between the Arabidopsis and B . rapa datasets ( longer period confers delayed phase ) , it is difficult to compare expression patterns directly ( Figure 4—figure supplement 1 ) . While paralogous TFs likely bind to the same motifs , a divergence in expression may result in the loss of transcriptional coactivators or corepressors required for gene activation or repression due to temporal separation in expression pattern resulting from the shift in phase of that TF . Similarly , an altered phase of expression might allow interaction with new TF interacting factors to provide a new target affinity for that TF resulting in a new target set . If the CNSs are driving the target gene expression differences we would expect them to be enriched for TF binding motifs ( Haudry et al . , 2013 ) . To test for motif enrichment , we selected 12 of the TF groups where Br1 was significantly more Arabidopsis-like ( see Supplementary file 5 , third tab ) . For each TF ( B . rapa paralogs and Arabidopsis ortholog ) , we took the collection of CNSs represented by their target genes in the GRNs and ran them against the HOMER motif analysis algorithm ( Heinz et al . , 2010 ) . Because the CNSs are located throughout the gene ( promoter , 5’UTR , introns , 3’UTR ) we selected the sequence from 2 kb upstream of the start codon to the 3’UTR for each target gene for comparison . We also included just the 2 kb upstream sequence to compare to standard motif search parameters . For all 12 TFs tested , we found three– to fivefold greater enrichment for motifs in the CNS elements compared to the full length and 2 kb promoter background sets ( Figure 4—figure supplement 2 ) . This is consistent with the results from the GRN analysis showing that CNSs are as predictive as expression dynamics and contain important regulatory elements . Further studies are needed to look for associations between groups of CNSs with their corresponding binding motifs and specific gene expression patterns . Since the GENIE3 algorithm associates target genes based on a TF being an activator or repressor , the target genes typically have several major expression patterns ( Figure 4B ) . Isolating distinct patterns and analyzing CNS variation between the target groups may reveal new motif groupings or novel motifs . The gene balance hypothesis posits that multi-subunit complexes are sensitive to variations in stoichiometry resulting in dosage compensation to produce the same amount of product ( Birchler and Veitia , 2014 ) . Clustering based on paralog median expression levels did reveal a consistent trend with one paralog having significantly higher median expression levels compared to the other retained paralog of that pair ( Figure 3—figure supplement 1 ) . However , as previously demonstrated , the overall expression levels for multi-copy genes is higher than single-copy genes ( Figure 2A and B ) . In addition , the rhythmicity in the paralogous pairs is still apparent , providing further support for focusing on the importance of the pattern of expression rather than simply the overall expression levels . This led us to wonder whether there is any indication that these pattern changes may contribute to new temporal responses to environmental stimuli such as abiotic stress . Gated stress response has been characterized in several plant species including Arabidopsis , poplar , rice , and B . rapa ( Fowler et al . , 2005; Wilkins et al . , 2009; Wilkins et al . , 2010; Greenham et al . , 2017; Grinevich et al . , 2019 ) . If a time-of-day dependent stress-responsive gene in Arabidopsis is now present in two copies in B . rapa with altered expression patterns does this confer an expanded stress response window or does one copy exhibit stress response while the other does not ? To look for indications of divergence in the function , we used a mild drought time-course RNA-seq dataset ( Greenham et al . , 2017 ) to test for altered responses to drought among pairs of paralogs . We first used the well-watered control samples to identify the paralogous pairs with altered patterns . Consistent with the divergence in pattern change under circadian conditions , the same trend is apparent under the diel ( LD ) conditions of our drought time course . Out of 4664 total pairs where two copies show detectable expression levels , 3259 pairs had significantly different patterns under control conditions ( p-value<0 . 01 ) . In the circadian dataset , we observed just 42% of genes with altered pattern but these diel data reveal 70% of pairs with altered pattern . Similarly , 77% of pairs ( 3602 ) had significantly different median expression levels ( Supplementary file 6 ) . Of the total pairs , 50% had both genes identified as circadian-regulated and an additional 35% had one circadian-regulated gene , emphasizing the importance of the circadian clock in diel conditions . For the pairs with altered patterns under the well-watered conditions , 93% were tested ( both genes were expressed ) under circadian conditions and 52% showed significant pattern changes . These results point to even greater divergence of paralog expression patterns under real-world , diel conditions compared to circadian conditions . To identify differential response to drought among paralogous pairs , we first ran DiPALM on all ( 23 , 248 ) expressed genes and tested each gene for pattern and expression changes in response to drought . We identified 3891 with a significant pattern change and only 327 with median expression changes ( p-value<0 . 01; Supplementary file 7 ) . Unlike the circadian dataset , the largest source of variation in expression is due to a phase change consistent with the importance of time-of-day gating of the stress response ( Wilkins et al . , 2010; Greenham et al . , 2017 ) . This provides further support for the dynamic nature of expression regulation and the limitations of simply quantifying transcript abundance differences at single time points . It should also be noted that this drought treatment captured the early signs of drought perception with very subtle expression changes during the first 24 hr and more evident changes in the subsequent 24 hr ( Greenham et al . , 2017 ) . The ability to detect these distinct patterns using DiPALM highlights the effectiveness of the network pattern approach for capturing unique and unpredicted patterns . To test whether the paralogous pairs exhibiting differential expression patterns under well-watered conditions are enriched for drought-responsive genes , we performed a permutation test . We randomly sampled the same number of pairs from the full set of pairs ( 3259 out of 4664 ) for 10 , 000 permutations to identify the likelihood of selecting drought-responsive genes within the 3891 sampled set . As a result , the differentially patterned copies were enriched for genes with drought responsive patterns ( p-value 0 . 0005; Figure 5A ) , whereas copies with differential median expression level were not ( p-value 0 . 6457; Figure 5A ) . With enrichment of drought-responsive genes among these pairs exhibiting different patterns , we wondered whether these pairs are more or less likely to have one or both copies responding to drought . Using the same 10 , 000 randomly sampled sets of 3891 pairs , we estimated a null distribution of the expected number of pairs with one and two drought-responsive genes . Results from the permutations indicated significant enrichment for pairs in which one member is drought responsive ( p-value<0 . 0001 ) and no enrichment for both copies being drought responsive ( p-value 0 . 1852; Figure 5B and C ) . Thus , we observed enrichment for only one but not both paralogs responding to drought stress in B . rapa . This suggests that the genome-wide expansion of expression domains among paralogs is biologically meaningful , in this case for drought stress response . More broadly , these results have important implications for how we capture and characterize transcriptomic responses or ‘states’ when making predictions about paralog function . Temporal , spatial , and conditional regulation can reveal new expression dynamics . An assessment of functional comparisons among paralog expression levels defines the gene with the highest expression in one or multiple tissue samples as ‘winning’ over the other ( Woodhouse et al . , 2014 ) . This classification is often referred to when looking for signs of subgenome dominance within polyploid species . With our set of pairs with one drought-responsive paralog , we wondered whether the responsive member of a pair had a higher median expression level under control conditions . Of the 764 pairs with a drought-responsive paralog , the drought responsive gene was the lower expressing member in 420 pairs and the higher expressing member in 344 pairs in the control conditions . Thus , we conclude that transcript abundance , whether at a single time point or combined across a time series , is not a reliable predictor of a gene’s functional importance . As validation of this , we ran a standard linear model test at each time point comparing well-watered and drought treatments and did not detect significant transcript abundance differences for any genes . To detect the initial transcriptional response to drought perception , we had to incorporate the complete transcript profile into the differential expression test . What appears to be very subtle changes in the abundance of a subset of transcripts are contributing to the measured temporal physiological changes in Fv’/Fm’ and stomatal conductance also occurring at specific time points ( Greenham et al . , 2017 ) . The ability to capture these early transcriptomic responses to the onset of drought offers new insight into how responsive the network is to slight adjustments in the temporal regulation of expression . The next challenge is to capture this fine-scale resolution across genotypes with diverse physiological responses to stress and identify the associated patterns . The circadian regulation of the transcriptome leads to time of day changes in gene expression that coordinate physiological responses to environmental conditions . This study emphasizes the power of B . rapa as a model system for investigating the consequences of polyploidy on transcriptional network dynamics . The close relationship of B . rapa to Arabidopsis facilitates comparative studies and guides gene function hypotheses due to the wealth of genomic and molecular resources developed in Arabidopsis . B . rapa , a morphologically diverse crop , has undergone whole-genome triplication since diverging from its common ancestor with Arabidopsis resulting in an expansion of gene copy number . To examine how this expansion has influenced the circadian transcriptome , we developed a new R package ( DiPALM ) to compare gene expression in time-course experiments . DiPALM enables a new line of inquiry into how temporal regulation of paralogs influence GRNs in B . rapa in which single time point comparisons of differential expression are replaced with temporal pattern analysis to provide a more complete view of the transcriptional network and the pervasiveness of rhythmic gene expression . In particular , this method facilitated the discovery of genes with altered expression patterns independent of expression levels . DiPALM was benchmarked using a publicly available simulated dataset ( Spies et al . , 2019 ) . DiPALM was designed specifically for large time-course datasets in multicellular species that span daylength-scale experiments . This benchmark data consists of simulations that mimic short time-frame cell culture treatments with as little as four timepoints . Despite this , DiPALM performed on par with the top three methods evaluated by Spies et al . ( Supplementary file 8 ) and DiPALM remains the only method to use a linear model-based framework that allows for complex contrasts between any number of different treatments . This framework has become the standard for differential expression analysis with packages such as edgeR ( McCarthy et al . , 2012 ) and DEseq2 ( Love et al . , 2014 ) . In this simulated dataset , a simple pairwise comparison between corresponding timepoints using established differential expression methods was shown to outperform all other methods tested . However , we show that when this ‘pairwise’ approach is applied to a real dataset involving mild drought response , it yields no significantly differentially expressed genes , whereas DiPALM detects 3891 genes with differential patterning . Our data support extensive circadian and diel regulation of the B . rapa transcriptome , as documented in Arabidopsis and a few other plant species ( Covington et al . , 2008; Wilkins et al . , 2009; Wilkins et al . , 2010; Li and Zhang , 2015; Oakenfull and Davis , 2017 ) . Our circadian time-course experiments with 2 hr sampling density provided sufficient resolution to reliably assess rhythmicity for all expressed genes resulting in roughly 74% showing circadian clock regulation in B . rapa , consistent with a pervasive role of the circadian clock in regulating diverse aspects of plant physiology . Interestingly , we found that genes retained in multi-copies in B . rapa are enriched in network modules that are phased during the day whereas evening and night phased modules are depleted for multi-copy genes . This difference might be related to the dosage sensitivity of processes occurring during the day . Gene ontology enrichment processes for the genes in these daytime phased modules included photosynthesis and abiotic stress response . Given the importance of proper transcriptional regulation of photosynthetic processes and the balance of enzyme components , it is not surprising that there is higher retention of multi-copy genes within these pathways . However , whether these genes are performing similar functions to their orthologous counterparts in Arabidopsis or have acquired new functions that could lead to additional regulation of the photosynthetic process is one exciting avenue of future study . However , it should be noted that in general , evening phased genes are likely to be greatly understudied because most experiments are performed during the day ( Grinevich et al . , 2019 ) , which could artifactually reduce the identification of abiotic stress ( and other ) genes in our evening modules . The retention of multi-copy genes that are circadian regulated provided an opportunity to assess the level of retention of transcript abundance patterns among paralogs to look for signs of possible neo- or sub-functionalization . Applying DiPALM to our list of circadian regulated paralogs uncovered evidence for extensive rearrangement of the transcriptional network through the divergence in expression pattern among retained paralogs in B . rapa that , in extreme examples , results in paralogs being expressed in antiphase to one another . This demonstrates the genome-wide expansion of phase domains among retained paralogs . The expansion of expression domains is reminiscent of the PSEUDO-RESPONSE REGULATOR ( PRR ) and REVEILLE ( RVE ) families of circadian clock genes that were retained following WGD as well as tandem duplication events ( Linde et al . , 2017 ) . The PRR genes in Arabidopsis have a temporally sequential expression pattern with PRR9 expressed just after dawn followed by PRR7 , PRR5 , PRR3 , and finally PRR1/TOC1 in the evening ( Matsushika et al . , 2000 ) . The PRR proteins appear to retain some common functions ( e . g . repression of expression of CCA1 , LHY , and TOC1 ) but their diverged expression patterns also allow differential contributions to the circadian network ( Nakamichi , 2011; Nakamichi et al . , 2012; Liu et al . , 2013; Liu et al . , 2016 ) . This also emphasizes the benefit of applying a network framework to data that associate similarly regulated genes as observed for paralogs with similar expression patterns having their paralogous pairs diverging in expression together . These changes in phasing among paralogs occur in similar network modules where groups of genes are classified with a particular phase while their paralogous pair-mates exhibit a similar phase difference indicative of common regulatory control . The expansion of expression domains among paralogous pairs provides ample opportunity for neo- and sub-functionalization through new network connections and novel interacting targets now expressed in phase with the pair member with the altered phase of expression . To test this hypothesis and predict the possible network rearrangements that have occurred in B . rapa since diverging from its common ancestor with Arabidopsis , we took a GRN approach to model the relationships between TFs and gene expression patterns . The identification of a set of 49 TF B . rapa paralogs where one paralog showed significantly more overlap with the Arabidopsis ortholog supports the divergence in network regulation between retained paralogous TFs in B . rapa . This is further supported by the overlap of CNSs among target genes in the GRNs . By replacing genes with CNSs in the GRN we were able to identify the more Arabidopsis-like paralog with strong consensus with the original GRN . Among the 35 TFs that overlapped between the CNS and gene expression networks are several genes involved in light signaling ( CRY2 , PHYA , ELF4 , HYH , ZFN1 ) , photosynthesis ( GLK2 ) , flowering time ( COL2 , COL4 , CDF1 , CDF2 ) , abiotic/biotic stress ( RAP2 . 4 , STO , IBH1 , TIP ) and the circadian clock ( CDF1 , CDF2 , ELF4 , PRR5 , PRR9 , RVE1 , TIC , TOC1 ) . The divergence in TF among light signaling , photosynthesis , and abiotic stress pathways is consistent with the GO enrichment observed for the daytime phase modules that are enriched for retained copies . Each of these pathway lists , other than photosynthesis , contains one or more TFs with zinc finger domains . A worthwhile future study would be to investigate whether certain features of TFs make them more likely to neo-functionalize within certain biological pathways . The success of the CNS network approach supports a predictive role of these CNSs for expression dynamics and provides a refined nucleotide sequence space to explore in future studies to associate regulatory elements with specific expression patterns . Associating CNSs with specific gene pattern responses may uncover new regulatory elements or novel combinations of regulatory elements that contribute to the differential regulation of paralogs and their targets , for example in response to drought stress . In the case of mild drought response , we find that two-copy B . rapa paralog pairs with significantly different expression patterns are enriched for pairs in which only one of the copies is drought responsive . This highlights the improved sensitivity gained with time series resolution and the ability to capture critical diel responses to environmental conditions that would otherwise be missed or assumed for all paralogs based solely on homology to Arabidopsis . The significant enrichment of one rather than both members of a paralogous pair being drought responsive also supports neo- or sub-functionalization . This is consistent with observations in Arabidopsis , rice , and poplar , that homeologous genes undergo expression partitioning among tissues ( De Smet and Van de Peer , 2012; Langfelder and Horvath , 2012 ) . This is also seen in the polyploids , Gossypium hirsitum and Tragopogon mirus ( Adams et al . , 2003; Buggs et al . , 2010 ) . In Arabidopsis , the majority of duplicated genes show divergent expression ( Blanc and Wolfe , 2004; Haberer et al . , 2004 ) , and this is particularly evident among abiotic and biotic stress-responsive genes ( Casneuf et al . , 2006; Ha et al . , 2007 ) . That significant amino acid sequence variation apparently does not contribute to the divergence between B . rapa paralogous TFs reinforces the importance of regulatory element variation in target genes . This raises the question of how two paralogous TFs with the same motif binding affinities can have such diverse targets in the GRN . One possibility is that the presence of new interacting partners at the novel phase of expression could modify binding affinity to either enhance or prevent binding to certain motifs or affect the consequences of TF binding ( activation versus repression ) . Conversely , the lack of critical interacting partners due to a mismatch in phasing could affect target binding and/or regulatory consequence . Further study into the temporal regulation of known binding partners for the divergent TFs is needed . These findings bring up several questions surrounding the importance of the variation in paralog expression pattern . Are these differential patterns maintained across B . rapa morphotypes or is there additional within-species variation ? Did these phase differences arise post-genome triplication or were they present in the diploid progenitors that gave rise to B . rapa ? Different B . rapa morphotypes ( e . g . leafy vegetable , turnip , and oilseed ) exhibit differential circadian clock parameters , as assessed by leaf movement analysis ( Yarkhunova et al . , 2016 ) , strongly supporting additional within-species variation in the transcriptomic network . An examination of the circadian networks across B . rapa morphotypes is needed to characterize these differences and begin to associate network plasticity with morphotype-specific traits . Our analysis reveals divergence in drought response among retained paralogs; how do these responses differ in more or less drought-tolerant genotypes ? Applying these pattern analysis approaches to pan-transcriptome time-course studies has the potential to identify regulatory elements that contribute to transcriptional network architecture and the evolution of new forms of transcriptional control in polyploids . Seeds of Brassica rapa subsp . trilocularis ( Yellow Sarson ) R500 were planted in ( 3 . 25’ x 3 . 625’ ) pots with a soil mixture of two parts Metro-Mix PX1 + one part Pro-Mix amended with 0 . 5 mL of Osmocote 18-6-12 fertilizer ( Scotts , Marysville , OH ) . The first photocycle experiment ( LD ) involved entraining plants under 12 hr light/12 hr dark and constant 20°C ( LDHH ) for 15 days after sowing ( DAS ) before transfer to constant light at 20°C ( LLHH ) . The second thermocycle experiment ( HC ) involved entraining B . rapa R500 plants under 24 hr light with 12 hr 20°C and 12 hr 10°C temperature cycles ( LLHC ) until 15 DAS before transfer to LLHH . Following 24 hr in constant conditions leaf tissue from the youngest leaf was collected and flash-frozen in liquid nitrogen every 2 hr for 48 hr . Lights in the chamber at plant height were ~130 µmol photons m−2 s−1 . Plants were shifted to constant light and temperature ( LLHH ) for 24 hr before starting the leaf tissue sampling at ZT24 . Leaf tissue ( ~100 mg ) from the youngest fully developed leaf was harvested and frozen in liquid nitrogen every 2 hr for 48 hr ( ZT24 – ZT72 ) . At each time point , leaf tissue from 10 plants was collected . Leaf tissue was ground to a fine powder using a Retsch Mixer Mill MM 400 ( Vendor Scientific , Newtown , PA ) . The mRNA extraction was performed according to Greenham et al . , 2017 and the strand-specific libraries according to Wang et al . , 2011 . For each leaf sample ( ~100 mg ) , 1 mL lysis binding buffer ( LBB ) was used to resuspend ground tissue . For each of two biological replicates , 200 µL aliquots of LBB lysate from each of five plants were pooled before mRNA isolation . Library size and quality was verified using a 2100-bioanalyzer ( Agilent Technologies , Santa Clara , CA ) . Libraries were indexed and pooled into 12 sample sets and sequenced as 101 bp paired-end reads using Illumina HiSeq2500 ( Illumina , San Diego , CA ) . Raw data have been submitted to GEO ( http://ncbi . nlm . nih . gov/geo ) under accession number GSE123654 ( https://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE123654 ) . The raw fasta reads were filtered using trimmomatic ( Bolger et al . , 2014 ) with mostly default settings ( ILLUMINACLIP: . /Tru-Seq3-PE . fa:2:30:10 LEADING:3 TRAILING:3 SLIDINGWINDOW:4:25 MINLEN:50 ) . Reads were aligned to the B . rapa R500 genome Brapa_R500_V1 . 2 . fasta ( https://genomevolution . org/CoGe/GenomeInfo . pl ? gid=52010 ) using tophat2 ( Kim et al . , 2013; https://ccb . jhu . edu/software/tophat/index . shtml ) with the following options: --library-type fr-firststrand -I 12000 G R500_v1 . 6 . gff -M --max-segment-intron 12000 --max-coverage-intron 12000 . Sample LD_ZT62_rep2 was identified as an outlier and removed , to avoid over-weighting rep1 the rep2 values were imputed by averaging the values of ZT60_rep1 and ZT64_rep1 . Raw counts were generated in Subread version 1 . 6 . 3 ( Liao et al . , 2019; http://subread . sourceforge . net/ ) with the following options: -F SAF -M -T 6 --fraction -s 2 p -B -C . Count data was normalized using edgeR ( Robinson et al . , 2010; McCarthy et al . , 2012; https://bioconductor . org/packages/release/bioc/html/edgeR . html ) version 3 . 22 . 1 using 'calcNormFactors' and log2 FPKM values were calculated using 'rpkm' with log = TRUE , prior . count = 0 . 1 . We created an R package that takes a raw count table of RNAseq data and runs differential pattern analysis from time series gene expression data . DiPALM is available through the Comprehensive R Archive Network ( CRAN; https://cran . r-project . org/package=DiPALM ) or via the Greenham Lab Github page ( https://github . com/GreenhamLab/Brapa_R500_Circadian_Transcriptome; Greenham , 2020; copy archived at swh:1:rev:9d59bbc84659cbbdffa96413694e01298c9868bb ) . A sample dataset is provided with the package along with a detailed vignette and manual that describes the analysis pipeline . A published set of simulated time-course expression datasets ( Spies et al . , 2019 ) was used for benchmarking . Data consists of eight separate time courses meant to simulate cell culture treatments with samples taken at various times after treatment along with untreated controls at each timepoint . In each dataset , 1200 genes were intentionally perturbed at various intensities using a set of predefined patterns of perturbation . These were taken to be the ‘differentially expressed’ genes or ‘positives’ in the test . This benchmarking test was run through DiPALM as well as various other methods tested by Spies et al . Table S1 shows the area under the receiver operating characteristic ( AUROC ) curve for all tests . The three top-performing methods found by Spies et al . are shown for comparison: splineTC ( Michna et al . , 2016 ) , maSigPro ( Nueda et al . , 2014 ) , and impulseDE2 ( Fischer et al . , 2018 ) . Both the differential pattern ( kME ) and differential expression ( median ) detection parts of DiPALM were tested along with a combined score which simply takes the sum of the scores from both parts ( combined ) . The method of running a traditional differential expression analysis on each timepoint separately and then taking the most significant timepoint for each gene was also tested ( pairwise ) . The entire analysis pipeline , starting with raw count data , was carried out using the R Statistical Programing Language ( R Development Core Team , 2018 ) along with the Rstudio integrated development environment ( R Studio Team , 2015 ) . A comprehensive R markdown file is available through the Greenham Lab Github page ( Greenham , 2020; https://github . com/GreenhamLab/Brapa_R500_Circadian_Transcriptome ) . This analysis script includes all data processing , statistical analysis and plotting that was used for this publication . Additional R packages were used in this analysis , including ‘edgeR’ ( Robinson et al . , 2010; McCarthy et al . , 2012 ) , ‘stringr’ ( Wickham , 2019 ) , ‘ggplot2’ ( Wickham , 2016 ) , ‘rain’ ( Thaben and Westermark , 2014 ) , ‘WGCNA’ ( Langfelder and Horvath , 2008; Langfelder and Horvath , 2012 ) , ‘circlize’ ( Gu et al . , 2014 ) , and ‘pheatmap’ ( Kolde , 2019 ) . For paralog expression comparisons we converted all three-copy paralogs into three two-way comparisons ( copy 1 versus copy 2 , copy 1 versus copy 3 , copy 2 versus copy 3 ) in order to perform one analysis with the two-copy tests and enable consistency in the interpretation of the results ( see line 606 of the markdown file ) . For the heatmap presented in Figure 3 , Br1 and Br2 were assigned based on the sign of the summed t-statistics from the limma output ( see line 709 of the markdown file ) . For the drought RNAseq analysis we used our previous dataset ( Greenham et al . , 2017 ) and aligned the data to our new B . rapa R500 genome assembly ( https://genomevolution . org/CoGe/OrganismView . pl ? org_name=Brassica%20rapa ) using the same pipeline described for the circadian datasets . These raw counts are available in a file called ‘DroughtTimeCourse_CountTable . csv’ on the Greenham Lab Github page ( Greenham , 2020; https://github . com/GreenhamLab/Brapa_R500_Circadian_Transcriptome ) . Using a list of canonical CNS sequences derived from Haudry et al . , 2013 , the B . rapa R500 and TAIR10 genomes were annotated for those CNSs using NCBI BLAST+ . A local BLAST database for each reference genome was created using the command: To get an initial set of CNS alignments , the following command was run for each reference genome: Filtering was disabled in favor of a different scheme also used in Yocca et al . , 2019 . All alignments with a bitscore of 28 . 2 were dropped , and alignments with a smaller than 60% coverage of the CNS sequence ( BLAST+’s ‘qcovs’ value ) were also dropped . Finally , to ensure the CNS alignments are reasonably unique , all of the alignments for a particular CNS sequence were discarded if they appeared more than once in the TAIR10 reference , or more than three times in the B . rapa R500 reference . Since the B . rapa genome has undergone a genome triplication event relative to A . thaliana , three occurrences were seen as the maximum reasonable amount . Two BED files were generated for each genome containing the coordinates of each resulting alignment . To associate the resulting CNS alignments with genes in the references , BEDtools was used to find the closest gene to each CNS location: bedtools closest -s -t all -D a -a < cns . bed> -b < reference . bed>>CNS_prox_genes . txt cns . bed is one of the two BED files generated in the previous section , and reference . bed is a gene annotation for the respective reference genome . The -s option constraints reported associations to be only on the same strand -- that is , CNS alignments and genes must appear on the same strand . -D tells BEDtools to report distances and ensures that the reported distances are signed ( negative for occurring before the gene , positive for occurring after , and 0 for being intragenic ) . Motif analysis was performed using HOMER -- specifically , findMotifsGenome . pl . This program requires that two sets of sequences be provided: a set of target sequences to be searched for motifs , and a set of background sequences for comparison to the target sequences to be compared to . Motif analyses were performed on three different target groups . The first analysis consisted of searching the CNSs against an ‘extended’ promoter background . For each transcription factor group , the CNSs corresponding to all of the target genes in Ath , Br1 , and Br2 separately were pulled and placed into three BED files containing the coordinates of the CNSs in TAIR10 and B . rapa R500 , respectively . A background set of sequences was then generated for every gene in the TAIR10 and B . rapa R500 genomes using BEDtools: bedtools slop -s -i < reference . bed> -g < reference . genome> -l 2000 r 0 > 2 kb_and_gene . bed Reference . bed is a gene annotation for the reference genome , and reference . genome is a text file containing the lengths of each chromosome that BEDtools uses to ensure that the coordinates it outputs are valid . This outputs a BED file that annotates a background consisting of a 2 kb promoter region before each gene , as well as the gene itself , to the end of the 3’ UTR . This larger background sequence was selected rather than the 2 kb promoter alone since many CNSs occurred in the UTRs and introns , and so using only 2 kb promoters would leave out background sequence relevant to many of the CNSs , potentially skewing the results . The second analysis searched the ‘extended’ promoter sequences against themselves . The target set consisted of ‘extended’ promoters corresponding to target genes in each transcription factor group , and the background consisted of a total list of sequences , including the target group , as per HOMER’s recommendations . The third analysis consisted of a more traditional motif search of promoter regions against promoter-only background . In this case , neither the target nor background sequences contain CDS , introns , or UTRs as with the ‘extended’ regions defined in the previous two analyses . These promoters were pulled using BEDtools: bedtools flank -s -i < reference . bed> -g < reference . genome> -l 2000 r 0 > 2 kb_promoters . bed Everything is identical as above , except that bedtools flank does not include the genes themselves in the output , and only generates locations for promoter regions . As before , promoters corresponding to TF target groups were selected and then analyzed against the entire set of promoters . HOMER was run using its included plant motif database on all three datasets ( Supplementary files 9–10 ) . Default parameters were used . Motif analyses were performed separately for the Ath , Br1 , and Br2 target groups . Given that three analyses were performed for each of these target groups , a total of nine motif analyses were performed for each TF group for a total of 108 motif analyses . Only the ‘knownMotifs’ output of HOMER was considered , which consists of a database search of target and background sequences against known plant motifs with a hypergeometric test to quantify significance . The de novo results were not used . For each of the 108 analyses , the outputs of the ‘knownMotifs’ analysis were simplified by grouping together found motifs that correspond to the same DNA-binding protein domain . Of each of these domain groups , the best p-value out of all the motifs found for that domain was selected as representative for the entire group . For each TF group , significant ( p<0 . 01 ) domains were counted for the Ath , Br1 , and Br2 target groups , for the CNS , ‘extended’ promoter , and promoter-only motif analyses . The B . rapa R500 genome was used for all analyses in this study . It is available through CoGe under the Genome ID: 52010 ( https://genomevolution . org/CoGe/GenomeInfo . pl ? gid=52010 ) . Circadian time-course RNA-seq data for B . rapa entrained in light cycles ( LDHH ) and temperature cycle ( LLHC ) are available through NCBI GEO under accession number GSE123654 ( https://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE123654 ) . Diel time-course RNA-seq under mild drought stress and well-watered conditions is from Greenham et al . , 2017 . Raw data are available through NCBI GEO under accession number GSE90841 ( https://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE90841 ) . For this study , the data were re-mapped to the B . rapa R500 genome and raw count data are available through a Greenham Lab Github repository ( Greenham , 2020; https://github . com/GreenhamLab/Brapa_R500_Circadian_Transcriptome ) . The file is named ‘DroughtTimeCourse_CountTable . csv’ . Full details of the data analysis including explanations , code and additional data files can be found in the above Github repository . The analysis is laid out in an R markdown file named ‘Brapa_CircadianTranscriptome_Markdown . Rmd’ .

Like animals , plants have internal biological clocks that allow them to adapt to daily and yearly changes , such as day-night cycles or seasons turning . Unlike animals , however , plants cannot move when their environment becomes different , so they need to be able to weather these changes by adjusting which genes they switch on and off . To do this , plants keep track of how long days are using external cues such as light or temperature . One of the effects of climate change is that these cues become less reliable , making it harder for plants to adapt to their environment and survive . This is a potential problem for crop species , like Brassica rapa . This plant has many edible forms , including Chinese cabbage , oilseed , pak choi , and turnip . It is also a close relative of the well-studied model plant , Arabidopsis . Since evolving away from Arabidopsis , the genome of B . rapa tripled , meaning it has one , two , or three copies of each gene . This has allowed the extra gene copies to mutate and adapt to different purposes . The question is , what impact has this genome expansion had on the plant's biological clock ? One way to find out is to perform RNA-sequencing experiments , which record the genes a plant is using at any one time . Here , Greenham , Sartor et al . report the results of a series of RNA-sequencing experiments performed every two hours across two days . Plants were first exposed to light-dark or temperature cycles and then samples were taken when the plants were in constant light and temperature . This revealed which genes B . rapa turned on and off in response to signals from the internal biological clock . It turns out that the biological clock of B . rapa controls close to three quarters of its genes . These genes showed distinct phases , increasing or decreasing in regular patterns . But the different copies of duplicated and triplicated genes did not necessarily all behave in the same way . Many of the copies had different rhythms , and some increased and decreased in patterns totally opposite to their counterparts . Not only did the daily patterns differ , but responses to stressors like drought were also altered . Comparing these patterns to the patterns seen in Arabidopsis revealed that often , one B . rapa gene behaved just like its Arabidopsis equivalent , while its copies had evolved new behaviors . The different behaviors of the copies of each gene in B . rapa relative to its biological clock allow this plant to grow in different environments with varying temperatures and day lengths . Understanding how these adaptations work opens new avenues of research into how plants detect and respond to environmental signals . This could help to guide future work into targeting genes to improve crop growth and stress resilience .

2020

Expansion of the circadian transcriptome in Brassica rapa and genome-wide diversification of paralog expression patterns

Intrinsically disordered proteins/regions ( IDPs/IDRs ) are proteins or peptide segments that fail to form stable 3-dimensional structures in the absence of partner proteins . They are abundant in eukaryotic proteomes and are often associated with human diseases , but their biological functions have been elusive to study . In this study , we report the identification of a tin ( IV ) oxochloride-derived cluster that binds an evolutionarily conserved IDR within the metazoan TFIID transcription complex . Binding arrests an isomerization of promoter-bound TFIID that is required for the engagement of Pol II during the first ( de novo ) round of transcription initiation . However , the specific chemical probe does not affect reinitiation , which requires the re-entry of Pol II , thus , mechanistically distinguishing these two modes of transcription initiation . This work also suggests a new avenue for targeting the elusive IDRs by harnessing certain features of metal-based complexes for mechanistic studies , and for the development of novel pharmaceutical interventions . Intrinsically disordered proteins/regions ( IDPs/IDRs ) constitute a significant fraction of the metazoan proteome ( Liu et al . , 2006; Uversky , 2013 ) . By virtue of their structural malleability and propensity to interact with multiple-binding partners , these peptide stretches of ∼30 or more amino acid residues have become increasingly recognized for their pivotal and prevalent role in cellular functions including many implicated in human disease pathogenesis ( Babu et al . , 2011 ) . IDRs usually function through transient and weak interactions , rendering them difficult subjects for mechanistic studies . IDRs are typically composed of low-complexity sequences and are often rich in polar amino acid residues , making them challenging targets for intervention by conventional small-molecule inhibitors , which often require stable hydrophobic-binding pockets ( Metallo , 2010 ) . However , conceptually , this feature of IDRs may be suitable for interactions with the hydrophilic and periodic metal–oxygen backbones found in a group of metal clusters known as polyoxometalates ( POMs ) . POMs , primarily composed of transition metals and oxygen , have been previously reported to exhibit potent biological activities ( Rhule et al . , 1998 ) , although these activities are rather promiscuous due to POMs' limited structural diversity , large size , and dependence on charge-based interactions in protein binding ( Judd et al . , 2001; Li et al . , 2009; Geng et al . , 2011; Narasimhan et al . , 2011 ) . Efforts have been made to improve the specificity and potency of POMs ( Flutsch et al . , 2011; Gao et al . , 2014 ) , and it seems that biological application of this metal complex-based protein-targeting strategy could benefit from new chemistry beyond conventional POMs to gain greater chemical space . Transcription initiation by eukaryotic RNA polymerase II ( Pol II ) is a highly regulated process requiring the coordinated actions of Pol II and a group of general transcription factors ( GTFs , i . e . , TFIIA , TFIIB , TFIID , TFIIE , TFIIF , and TFIIH ) ( Roeder , 1996 ) . Of all the GTFs , TFIID is the one with intrinsic DNA sequence specificity and is responsible for nucleating the assembly of a preinitiation complex ( PIC ) at core promoters ( Roeder , 1996 ) . TFIID-promoter binding , which can be further stabilized by TFIIA , creates a platform for TFIIB loading , which in turn allows the engagement of TFIIF and Pol II . TFIIE and TFIIH are the last to join the PIC , after which promoter melting and RNA synthesis can occur . After Pol II leaves the promoter during the initial ( de novo ) round of transcription , a reinitiation scaffold containing TFIID has been proposed to remain on the DNA template and facilitate subsequent rounds of transcription ( reinitiation ) in reconstituted mammalian systems ( Hawley and Roeder , 1987; Zawel et al . , 1995 ) . TFIID is a protein complex composed of the TATA-binding protein ( TBP ) and ∼14 TBP-associated factors ( TAFs ) ( Albright and Tjian , 2000; Matangkasombut et al . , 2004 ) . TBP recognizes and binds to the TATA box , while TAF1 and TAF2 recognize the Initiator element ( Inr ) , and the TAF6-TAF9 module recognizes the downstream core promoter element ( DPE ) ( Juven-Gershon and Kadonaga , 2010 ) . It is generally accepted that higher TFIID-promoter affinity leads to more robust transcription , and indeed it is thought that a primary role of sequence-specific activators is to recruit TFIID to promoters ( Roeder , 1996; Albright and Tjian , 2000; Matangkasombut et al . , 2004; Juven-Gershon et al . , 2006 ) . However , transcription initiation is a dynamic process and the release of at least a portion of the TFIID-promoter DNA contacts has been shown to be a critical step for productive initiation ( Yakovchuk et al . , 2010 ) . In addition , TFIID may assume diverse structures and its recognition of core promoter elements can be modulated by activators and post translational modifications ( Lewis et al . , 2005; Juven-Gershon et al . , 2008; Liu et al . , 2009; Cianfrocco et al . , 2013 ) . These distinct functional states of TFIID and their transitions are likely critical for gene-specific transcription regulation , but they are difficult to probe by conventional biochemistry and genetic analysis . In a search for small molecule compounds to selectively perturb TFIID function , we identified a tin-based metal cluster as a specific binder and modulator targeting an IDR within the TAF2 subunit of metazoan TFIID . By virtue of its specificity for interfering with the first round of transcription initiation , this metal cluster compound serves as a useful tool for studying the role of TFIID in both transcription initiation and reinitiation . This non-POM metal cluster revealed a novel mode of interaction with a low-complexity protein domain , demonstrating the feasibility of using metal-based compounds to selectively target IDRs . We screened a library of ∼10 , 000 organic compounds for binders to metazoan TFIID using a small-molecule microarray platform ( Casalena et al . , 2012 ) . In this screen , chemicals of diverse structures were printed on a functionalized glass surface and the binding of TFIID was detected by specific antibodies ( Figure 1A ) . We identified one compound ( 1 , ChemDiv 7241-4207 ) that reproducibly and selectively bound to both Drosophila and human TFIID ( Figure 1B , C ) . As controls , no binding to the antibodies or two other multi-subunit complexes of the human Pol II core transcription machinery , TFIIH and Pol II , was observed in counter screenings ( Figure 1B ) . 10 . 7554/eLife . 07777 . 003Figure 1 . Small-molecule microarray screening for TFIID-specific modulators . ( A ) Screening schematic: chemicals of diverse structure were covalently attached to a functionalized glass surface , and incubated with protein of interest , the binding of which was indicated by primary antibodies recognized by a specific fluorescently ( Cy5 , red ) labeled secondary body . ( B ) Representative images of an area of arrays probed with bovine serum albumin ( BSA , control ) , Drosophila ( d ) TFIID , human ( h ) TFIID , TFIIH ( control ) , and Pol II ( control ) in combination with specified primary antibodies . Yellow arrows denote the lead compound ( 1 , ChemDiv 7241-4207 ) . The images were scanned at 532 nm ( green , for reference spots ) and 635 nm ( red , for antibody signal ) . ( C ) Background subtracted average signal in arbitrary unit ( A . U . ) in the Cy5 fluorescent channel picked up by Drosophila ( left ) or human ( right ) TFIID is plotted against their respective BSA controls . Yellow circles depict the data points of the lead compound . DOI: http://dx . doi . org/10 . 7554/eLife . 07777 . 003 To assess the effect of compound 1 on transcription , we developed an integrated functional assay consisting of a reconstituted human cell-free transcription system ( Figure 2A ) . In this assay , a complete set of highly purified GTFs ( TFIIB , TFIID , TFIIE , TFIIF , and TFIIH; TFIIA is not required ) plus Pol II was incubated with the lead compound first , followed by incubation with a promoter-containing DNA template for transcription . As a control , TBP was used in place of TFIID to support a ‘basal’ transcription that also requires the rest of the protein factors . We found that the commercially supplied compound 1 ( ChemDiv 7241-4207 ) inhibited both Drosophila and human TFIID-directed transcription , but not transcription directed by TBP ( Figure 2B and Figure 2—figure supplement 1A , B ) , suggesting a TAF-specific mechanism of inhibition . Further characterization indicated that this inhibition ( i ) is sensitive to the dose of TFIID used in the reaction ( Figure 2—figure supplement 1B ) , ( ii ) can be alleviated by the addition of more TFIID after chemical treatment , but not by the addition of any other protein factors ( Figure 2C ) , confirming that TFIID is the most likely target of inhibition in the reaction , and ( iii ) is insensitive to various mutations in core promoter elements ( Figure 2—figure supplement 1C ) . Taken together , our transcription results suggest that the inhibitory activity specifically targets an evolutionarily conserved TAF subunit of TFIID that is required for a basic function of TFIID during Pol II transcription initiation in vitro . 10 . 7554/eLife . 07777 . 004Figure 2 . TFIID-specific transcription inhibition in a reconstituted system . ( A ) Cartoon illustration of TFIID- or TATA-binding protein ( TBP ) -directed transcription assays . Highly purified protein factors were mixed with the chemical and incubated before the addition of DNA templates for preinitiation complex ( PIC ) assembly . The DNA template contains the synthetic super core promoter ( SCP1 ) ( Juven-Gershon et al . , 2006 ) . ( B ) Dose-dependent inhibition of hTFIID-directed transcription , but not TBP-directed transcription , by the originally purchased lead compound ( 1 , ChemDiv 7241-4207 ) . The images were the primer extension products of the synthesized RNA and their signals were quantified and normalized to the respective controls ( first lane from left , DMSO vector only ) . ( C ) Transcription rescue with individual protein factors supplemented in twofold excess ( relative to the default dosage of each factor ) immediately after chemical treatment ( ChemDiv 7241-4207 at 5 µg/ml ) and before the addition of the DNA template . Fold of inhibition was calculated for each reaction pair . DOI: http://dx . doi . org/10 . 7554/eLife . 07777 . 00410 . 7554/eLife . 07777 . 005Figure 2—figure supplement 1 . TFIID dependency of the in vitro transcription assay and controls for the TFIID-specific inhibition . ( A ) Titration of hTFIID in the reconstituted transcription reaction . The reaction contains all other protein factors specified in Figure 2 , except for the lead compound . ( B ) Titration of hTFIID and dTFIID and their response to the lead compound 1 ChemDiv 7241-4207 . Asterisk ( * ) denotes quantifications affected by cross-over signal from a neighbor . ( C ) TFIID-specific transcription inhibition on mutant SCP1s ( Juven-Gershon et al . , 2006 ) . These images were from a same gel with the same display setting as those shown in Figure 2B . Note that although the absolute signal from TFIID-directed transcription may be core promoter element-dependent , ∼1 . 5 µg/ml of lead compound 1 is always sufficient to cause ∼50% transcription inhibition . In contrast , TBP-directed transcription is rather constant . DOI: http://dx . doi . org/10 . 7554/eLife . 07777 . 005 In an effort to perform a structure-activity relationship analysis , we resynthesized compound 1 in-house and were surprised to find that the resynthesized compound was completely inactive in the transcription assay ( Figure 3A ) . By comparing three batches of an analog compound ( 2 ) with varying levels of inhibitory activity ( Figure 3—figure supplement 1 ) , we found that the inhibitory activity correlated with levels of a tin-containing material detected by elemental analysis ( Figure 3B ) . This material is likely derived from tin ( II ) chloride ( SnCl2 ) added as an anti-oxidant in the final recrystallization step in a subset of the commercially supplied samples ( Figure 3—figure supplement 2 ) . After excluding most common tin-containing compounds as candidates , we found that tin ( IV ) oxochloride , prepared by any of several established routes ( Dehnicke , 1961; Messin and Janierdubry , 1979; Sakurada et al . , 2000 ) , consistently reproduced the specific inhibition of TFIID-directed transcription ( Figure 3C and Figure 3—figure supplement 3 ) . Dose-response titration revealed a Hill coefficient of ∼1 , suggesting a non-cooperative binding of this chemical to its biological target ( Figure 3D ) . This compound , which consists of tin , bridging oxygen , and chlorine ligands , may form ladder-like clustered structures and coordinate to atoms with lone electron pairs , such as the nitrogen in pyridine ( Dehnicke , 1961; Messin and Janierdubry , 1979; Holmes et al . , 1987; Sakurada et al . , 2000 ) , or as is perhaps more functionally relevant , the imidazole groups of histidine residues in proteins ( Figure 3—figure supplement 3E ) . We concluded that the tin ( IV ) oxochloride-derived cluster is the ingredient within the active commercial supplies responsible for the TFIID-specific transcriptional inhibitory activity . 10 . 7554/eLife . 07777 . 006Figure 3 . Identification of a tin ( IV ) oxochloride-derived cluster as the TFIID-specific transcription inhibitor . ( A ) Inhibition of TFIID-directed transcription by the originally purchased lead compound ( 1 , ChemDiv 7241-4207 ) , but not the in-house resynthesized one . ( B ) Elemental analysis of three batches of analog compound ( 2 ) with varying levels of inhibitory activity ( see Figure 3—figure supplement 1B , C ) . ( C ) Inhibition of TFIID-directed transcription by tin ( IV ) oxochloride synthesized using different methods . From left to right: DMSO control , SnCl2 refluxed in isopropanol , SnCl2 oxygenation with H2O2 or O2 , and SnCl4 oxygenation with bis ( trimethylsilyl ) peroxide ( BTSP ) , and SnOCl2 in complex with pyridine . ( D ) Dose response titration ( left ) and Hill Plot ( right ) of SnOCl2·pyridine inhibiting hTFIID-directed transcription . Three independent replicates were used for plotting . The Hill coefficient was 1 . 062 . DOI: http://dx . doi . org/10 . 7554/eLife . 07777 . 00610 . 7554/eLife . 07777 . 007Figure 3—figure supplement 1 . Discrepancy between organic compound structures and the transcription inhibitor activity of commercial compounds . Transcription inhibition by lead compound 1 ( A ) and its analog 2 ( B ) from different sources . Because the actual inhibitory activity was found to be from minor contaminations unrelated to the assigned chemical structure , the reagents are named by an abbreviation of the vender name followed by the catalog number ( or as ‘in-house’ preparation ) . ChemDiv 7241-4207 ( original purchase , lead compound 1 ) was also used in ( B ) as a positive control for transcription inhibition . By analytical organic chemistry standards ( nuclear magnetic resonance [NMR] and mass spectrometry ) , all these compounds are >95% ‘pure’ . The three supplies of analog compound 2 subjected to elemental analysis ( Figure 3B ) are highlighted with colored stars ( * ) in ( B ) , and their high-resolution NMR analysis results are shown in panel C ( left ) . ( C ) ( Right ) is a hTFIID-directed transcription assay showing that the minor dimethylformamide ( DMF ) solvent contamination detected in the active materials by NMR as two chemical shifts between 2 . 5 and 3 . 0 ppm is irrelevant to the inhibitory activity . DOI: http://dx . doi . org/10 . 7554/eLife . 07777 . 00710 . 7554/eLife . 07777 . 008Figure 3—figure supplement 2 . Tracking of the TFIID inhibitory activity to a tin-containing complex . ( A ) Scheme of the recrystallization procedure provided by the vender for the resynthesis of Princeton OSSK_462080 . ( B ) Detection of robust TFIID inhibitory activity within the recrystallization leftover . For comparison , the corresponding ‘pure’ chemical ( Princeton OSSK_462080 large-scale resynthesis ) only inhibited TFIID-directed transcription by twofold at 10 µM ( 4 . 6 µg/ml; 10-fold at 100 µM ) ( Figure 3—figure supplement 1B ) . ( C ) Recreation of the TFIID inhibitory activity by mimicking the recrystallization procedure . Top , the scheme . Inactive compound 2 ( LifeChem F1566-0338 large-scale resynthesis ) , tin ( II ) chloride , and acetate acid were dissolved in isopropanol in a sealed container with atmosphere air and incubated at 83°C . Samples were taken at different time point and assayed for activity ( with a final concentration equivalent to 10 µM elemental tin ) , as shown on the bottom . ( D ) Comparing different tin compounds in recreating the inhibitory activity , in the presence ( left ) or absence ( right ) of the inactive mother molecule ( compound 2 ) . The reaction conditions are as described in ( C ) . Tin ( II or IV ) oxides were used as suspension in corresponding reactions . DOI: http://dx . doi . org/10 . 7554/eLife . 07777 . 00810 . 7554/eLife . 07777 . 009Figure 3—figure supplement 3 . Tin ( IV ) oxochloride-derived cluster identified as the TFIID-specific transcription inhibitor . ( A ) Simple inorganic and organic tin-containing complexes tested for inhibition of TFIID-directed transcription . Nano particles ( ‘nano’ ) or regular tin oxide compounds were used as suspension ( 13 . 5 and 15 . 1 µg/ml for SnO and SnO2 , respectively ) . Concentrations for the rest of the chemicals were expressed in µM . SnCl2 oxygenated by refluxing in air or H2O2 treatment ( concentration expressed as elemental tin ) , or ChemDiv 7241-4207 ( original purchase ) was used as positive controls . ( B ) Oxygenation converts SnCl2 into the active species . Left: generation of the TFIID-specific inhibitory activity by refluxing SnCl2 in isopropanol under air atmosphere . Right: TFIID-inhibitory activity generated by other oxygenation methods ( H2O2: hydrogen peroxide; TBHP: tert-butyl hydroperoxide; mCPBA: m-chloroperbenzoic acid; NaBO3 , sodium perborate ) . ( C ) In-house X-ray photoinduced spectroscopy ( XPS ) elemental analysis of tin ( IV ) oxochloride prepared by refluxing in isopropanol . The atomic ratio of Sn:O:Cl:C of ∼1:2:2:3 is consistent with SnOCl2 coordinating with isopropanol at a ratio of 1:1 . ( D ) Infrared spectrum of tin ( IV ) oxochloride in complex with pyridine . Marked peaks match those reported previously ( Dehnicke , 1961 ) . ( E ) Hypothetical structure of a tin oxochloride tetramer coordinated with pyridines , explaining the proposed 2:3 coordination ratio ( Dehnicke , 1961 ) . Similar ladder-like tin ( IV ) -oxo backbone structures have been reported for tin ( IV ) compounds ( Holmes et al . , 1987 ) . Multiple histidine residues , when presented in close vicinity from a surface of a protein , are expected to replace these individual pyridines , as driven by entropy . DOI: http://dx . doi . org/10 . 7554/eLife . 07777 . 009 Identification of tin ( IV ) oxochloride led us to consider presumptive targets that should be histidine-rich domains of phylogenetically conserved TAF subunits . Testing this hypothesis , we found that a GST fusion of the histidine-rich Drosophila TAF2 ( dTAF2 ) C-terminal fragment ( residues [1125–1221] ) bound selectively to the original tin-oxochloride containing sample in the arrayed library ( Figure 4A ) . The targeted polypeptide fragment is part of a conserved IDR with adjacent low-complexity poly ( K ) , poly ( KH ) , and poly ( KD/E ) motifs found in both Drosophila and human TAF2 ( Verrijzer et al . , 1994; Kaufmann et al . , 1998 ) ( Figure 4B and Figure 4—figure supplement 1 ) . The tin ( IV ) oxochloride cluster , with its hydrophilic , periodic surface features , presents a likely complementary ligand for these polar , repetitive , and histidine-rich IDRs . 10 . 7554/eLife . 07777 . 010Figure 4 . The inhibitor targets an IDR of TAF2 through histidines . ( A ) Binding of GST-dTAF2 ( 1125–1221 ) to the lead compound 1 ( ChemDiv 7241-4207 ) ( yellow arrow head ) in the microarray . GST antibody was used for detection . Recombinant GST protein was the control . ( B ) Alignment of dTAF2 and hTAF2 C-terminal IDRs , with the conserved repetitive motifs highlighted . ( C–F ) Surface plasmon resonance sensorgrams of synthetic tin ( IV ) oxochloride cluster binding to GST-dTAF2 ( 1125–1221 ) ( C , D ) or Halo-hTAF2 ( 990–1199 ) ( E , F ) fragment , under varying imidazole concentrations ( C , E ) or pH ( D , F ) ( R . U . : resonance unit ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07777 . 01010 . 7554/eLife . 07777 . 011Figure 4—figure supplement 1 . The intrinsically disordered nature of metazoan TAF2 C-terminus and the stringent conservation of and around the low-complexity sequences . Full-length TAF2 proteins from Drosophila ( A ) or human ( B ) were analyzed for protein-disordered region ( http://bioinf . cs . ucl . ac . uk/psipred/ ? disopred=1 ) . The predicted α-helix and β-sheet structures are highlighted in purple and yellow , respectively , and the C-terminus-disordered regions are underscored . ( C ) Alignment of the very C-terminus of TAF2 proteins from multiple vertebrates . The low-complexity sequences and other well-conserved residues were highlighted in distinct colors . DOI: http://dx . doi . org/10 . 7554/eLife . 07777 . 011 To further validate tin ( IV ) oxochloride cluster as responsible for binding to the GST-dTAF2 ( 1125–1221 ) fragment , we carried out an independent surface plasmon resonance assay ( Figure 4C , D ) . In this assay , the GST fusion protein was immobilized on a functionalized surface and an aqueous solution prepared with pure tin ( IV ) oxochloride chemicals was injected . We detected binding and dissociation of the chemical to the fusion protein upon chemical injection and buffer washing , respectively . In addition , we found that the binding is sensitive to imidazole and to pH values of ≤5 . 8 , consistent with an essential role of histidine residues , which are protonated at lower pH and therefore incapable of coordinate bonding with the inhibitor . Similar results were observed in a parallel assay using a human TAF2 C-terminus ( 990–1199 ) fusion protein ( Figure 4E , F ) , validating the conservation of the inhibitor target sites between Drosophila and human TAF2 proteins . Previous reports indicated that TAF2 can directly interact with DNA and is required for the recognition of the Inr element by TFIID ( Verrijzer et al . , 1994; Kaufmann et al . , 1998 ) . In addition , the inhibitor-targeted dTAF2 IDRs are rich in lysine residues that may non-specifically interact with negatively charged DNA backbone . We therefore examined whether GST fusions containing dTAF2 ( 1125–1221 ) or its sub-regions could retain DNA in a GST-pull down assay using DNA fragments corresponding to different regions of the promoter used in the transcription assay ( Figure 5A ) . We found that indeed , this fragment of TAF2 can directly interact with DNA , primarily through the poly ( K ) and poly ( KH ) motifs , and more weakly through the adjacent poly ( KD/E ) motif . In addition , this interaction seems to be independent of the DNA sequence , which is not surprising given the low complexity of this region ( Figure 5A ) . This suggests that this newly identified DNA interaction surface of TAF2 may be quite distinct from the previously reported Inr-selective DNA-binding activity of TAF2 ( Verrijzer et al . , 1994 ) . 10 . 7554/eLife . 07777 . 012Figure 5 . The inhibitor and DNA synergistically bind to the TAF2 IDR . ( A ) GST pull-down of double-stranded ( ds ) DNA oligonucleotides by GST-dTAF2 fragments . Top , the GST fusions ( G: GST alone ) . Middle , the four DNA fragments tested , which are parts of the SCP1 ( Juven-Gershon et al . , 2006 ) used in this study ( sequence shown , with the TATA , Inr , and DPE elements , from left to right , underscored ) . Bottom , DNA staining raw images ( left ) and bar representation of the bound/unbound DNA signals ( right ) . ( B ) Binding of nuclease-treated GST-dTAF2 ( 1125–1221 ) to the lead compound 1 ( ChemDiv 7241-4207 ) printed in quadruplicates in the microarray ( yellow rectangle ) , and its rescue by a double-stranded ( ds ) DNA oligonucleotide ( 1 µg/ml ) or heparin ( 2 µg/ml ) . GST antibody was used for detection . Recombinant GST-dTAF5 was used as a negative control . ( C ) DNase I footprinting assay on TFIID-promoter binding , in the presence of the lead compound ( 1 , ChemDiv 7241-4207 ) , a structural analog ( 2 , Princeton OSSK_462080 ) , or an unrelated , non-specific ( NS ) inhibitor ( Maybridge BTB08547 , see Figure 5—figure supplement 1B ) . Shown is the digestion product of the end-labeled DNA template separated by gel electrophoresis . The DNA template contains the SCP1 . Black boxes depict the positions of the TATA , Inr , and DPE elements , respectively . Blue bracket indicates the ‘footprint’ of TFIID . For simplicity , only two bands ( denoted by arrowheads ) , which were protected ( A ) or intensified ( B ) upon TFIID binding , were selected for quantification ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07777 . 01210 . 7554/eLife . 07777 . 013Figure 5—figure supplement 1 . ( A ) DNase I footprinting assay of TFIID-promoter binding and its enhancement by the specific inhibitor at different salt concentrations . ( B ) A non-specific inhibitor ( Maybridge BTB08574 ) : its structure ( top ) and inhibition of both TFIID- and TBP-directed transcription ( bottom ) . The lead compound used here was ChemDiv 7241-4207 . DOI: http://dx . doi . org/10 . 7554/eLife . 07777 . 013 These results led us to re-examine the finding in Figure 4A , where the printed lead compound was bound by a bacterially expressed recombinant GST fusion protein ( which likely contained co-purifying contaminating nucleic acids ) . We indeed found that interaction of GST-dTAF2 ( 1125–1221 ) with the tin-complex in the arrayed library was dependent on nucleic acids , as binding was ablated by nuclease pre-treatment and restored by addition of double-stranded DNA oligonucleotides ( Figure 5B ) . Heparin , a poly-anionic mimic of DNA , did not rescue the binding , indicating that structural features of DNA were required , not simply its high density of negative charges ( Figure 5B ) . These data suggested that DNA may stabilize an otherwise transient structural state of this IDR domain that then becomes susceptible to targeted binding by the tin compound . We thus hypothesized that there is some sort of synergy between DNA and the inhibitor in binding to the TAF2-disordered region , and further examined this interaction in the context of holo-TFIID complex using a DNase I footprinting assay . We found the holo-TFIID protects an extended area of the promoter DNA from DNase I digestion , as previously reported on the super core promoter ( SCP1 ) ( Juven-Gershon et al . , 2006 ) ( Figure 5C ) . The signatures of this footprint include the protection of some hypersensitive sites ( such as the band A at the upstream edge of the Inr element , and many bands covered by the bracket ) , and the exposure of new hypersensitive sites ( such as band B at the downstream edge of the Inr element ) ( Figure 5C , lane 1 and 2 ) . The TATA box is not protected because the engagement of TBP in the context of holo-TFIID complex at this promoter requires TFIIA ( Cianfrocco et al . , 2013 ) . Interestingly , the inhibitor and its analog specifically enhanced features corresponding to promoter DNA binding by TFIID ( lanes 3–8 ) ( exemplified by the enhanced protection of band A and sensitization of band B , quantified by image analysis ) . This observed enhancement was subtle but highly reproducible using the various commercially supplied compounds and likely to be a specific consequence because a non-specific inhibitor picked up in our initial screen clearly reduced rather than enhanced TFIID binding ( Figure 5C lanes 9–10 and Figure 5—figure supplement 1 ) . Therefore , we speculate that the specific tin ( IV ) oxochloride chemical inhibitor may stabilize TFIID-promoter interactions under certain conditions via the TAF2 IDR . It has generally been accepted that TFIID-promoter binding is a rate-limiting step in transcription initiation , and overall affinity of TFIID for promoter DNA correlates with transcription activity ( Juven-Gershon et al . , 2006 ) . It is important to note that the inhibitor doesn't block TFIID-promoter binding , thus , it is likely to affect some downstream events . Individual TFIID-promoter contacts are likely dynamic , and a rate-limiting isomerization of TFIID could influence the transition from an initially bound Pol II to a productively engaged Pol II in the assembly of a functional PIC ( Yakovchuk et al . , 2010 ) . We therefore hypothesized that an unnatural stabilization of the TAF2-DNA interaction by the inhibitor might specifically interfere with the presumptive conformational rearrangement required for the transition to a productively engaged Pol II . To test this hypothesis , we first examined whether the inhibitor blocks transcription at the initiation stage . The primer extension assay used for transcription detection requires the synthesis of transcripts of 155 nucleosides , which could be limited by elongation . To identify the step ( s ) during transcription targeted by the inhibitor , we examined the synthesis of the very first dinucleotide during initiation , using only adenosine triphossphate ( ATP , the first nucleotide ) and alpha-32P labeled guanosine triphosphate ( GTP , the second nucleotide ) in the reaction . In this assay , a productive PIC would lead to the synthesis an ApG dinucleotide that is detected by autoradiography . Indeed , we observed inhibition of ApG synthesis ( Figure 6A ) , suggesting the inhibitor acts very early during a stage at or before the synthesis of the first phosphodiester bond , and thus , very likely some step during PIC assembly . 10 . 7554/eLife . 07777 . 014Figure 6 . Tin ( IV ) oxochloride cluster specifically blocks de novo transcription initiation at the step of Pol II engagement . ( A ) Analysis of the first dinucleotide synthesis . Left , the scheme of the experiment . Green ‘A’ is where transcription starts . ATP and GTP are sufficient for the formation of the first phosphodiester bond at the SCP1 . Right , the autoradiography image of the dinucleotide products . ( B ) Scheme of the step-wise PIC assembly perturbation experiment . The cartoon illustrates the question under investigation—which step of PIC assembly is inhibited . In box is the flow chart of the experiment . Inhibitor was added to the reaction after a subset of GTF ( set 1 ) was incubated with the template DNA , followed by the addition of the rest of the protein factors ( GTF set 2 ) for PIC assembly . All four nucleoside triphosphates ( NTPs ) were added in the end to allow RNA synthesis . ( C ) Gel image and quantification of transcription inhibition by treatment at different stages of PIC assembly . D , B , E , F , H , and P represent TFIID , TFIIB , TFIIE , TFIIF , TFIIH , and Pol II , respectively . ‘/’ indicated the lack of any protein factor in GTF set 1 or 2 . On the left part ( lanes 1–14 ) , GTF set 1 contains protein factors added incrementally one by one ( from none to the complete set ) following the order of in PIC assembly ( Roeder , 1996 ) . On the right part ( lanes 11–14 ) , individual factors were omitted in GTF set 1 . SnOCl2·pyridine was used at 5 µg/ml ( A ) or 2 . 5 µg/ml ( B ) as the inhibitor . DOI: http://dx . doi . org/10 . 7554/eLife . 07777 . 01410 . 7554/eLife . 07777 . 015Figure 6—figure supplement 1 . The originally purchased lead compound specifically blocks de novo transcription initiation at the step of Pol II engagement . Refer to Figure 6 legends for more details . The inhibitor used here was ChemDiv 7241-4207 ( original purchase ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07777 . 015 To test which step during PIC assembly is affected by the inhibitor , we next performed a series of experiments in which DNA templates were pre-incubated with only a subset of protein factors ( ‘GTF set 1’ , with factors incrementally added according to the order of PIC assembly [Roeder , 1996] ) to maintain PIC assembly at specific stages prior to the addition of the inhibitor . After inhibitor treatment for 15 min , the missing factors were added ( as ‘GTF set 2’ ) to complete PIC assembly and initiate transcription ( Figure 6B , and Figure 6C lanes 1–14 ) . As controls , we had either no factor or all the requisite protein factors in GTF set 1 that would be equivalent to inhibitor treatment before the beginning or after the completion of PIC assembly ( lanes 1–2 and 13–14 ) . As expected , inhibitor treatment before PIC assembly efficiently blocked transcription ( by 6 . 5-fold at 2 . 5 µg/ml , comparing lanes 1 and 2 ) . In contrast , the inhibitor failed to interfere with transcription if added after PIC assembly ( compare lanes 13 and 14 ) , indicating some step ( s ) during initial PIC assembly is sensitive to the inhibition . Interestingly , pre-incubation of TFIID or TFIID together with TFIIB with the DNA template led to a more severe inhibition ( 16-fold , Figure 6C , lanes 4 and 6 ) when compared to adding all the protein factors after the inhibitor ( 6 . 5-fold , lane 2 ) . This suggests that a TFIID-TFIIB-DNA sub-complex is likely the most susceptible substrate for inhibition , while the presence of other core promoter factors may facilitate the progression of PIC assembly and reduce the time interval of this vulnerable stage . Addition of TFIIF partially alleviated the inhibition ( comparing lanes 6 and 8 ) , suggesting that TFIIF may play a necessary but insufficient role in driving TFIID into an inhibitor-resistant state . Upon further addition of Pol II to the assembly , the system became resistant to the inhibitor to a level comparable to that of preformed PICs ( lanes 10 and 14 ) . In contrast to the essential role of TFIID , TFIIB , TFIIF , and Pol II ( the same set of factors required for Pol II engagement ) in forming the minimal PIC intermediate resistant to inhibition , adding TFIIE and TFIIH had no effect on inhibition ( comparing lane 10 , 12 , and 14 ) , even though these two factors are required for robust transcription . Very similar results were observed when individual protein factors were left out one by one from GTF set 1 as an alternative strategy to arrest PIC assembly at specific stages ( Figure 6C lanes 11–24 for the pure tin ( IV ) oxochloride compound; and Figure 6—figure supplement 1 for the original lead compound ChemDiv 7241-4207 ) . Taken together , these results strongly suggest that the step most sensitive to inhibition involves the initial binding of Pol II to a DNA-TFIID complex that must then transition into a conformation compatible with productive Pol II engagement . Importantly , after Pol II engagement takes place during de novo PIC assembly , the system becomes resistant to the inhibitor . To better understand how Pol II engagement might be inhibited by the chemical , we performed DNase I footprinting assays to directly examine the potential conformational isomerization events during PIC assembly . In this assay , various PIC components were incubated with the template DNA under conditions that would lead to optimal transcription output if all the other components were included . We found that , as expected , TFIID alone caused a footprint covering the Inr , DPE , and extending downstream ( to ∼+55 ) ( Figure 7A , lane 1 and 2 ) . The addition of TFIIB enhanced protection over the upstream region ( from the TATA box to the Inr ) , and this protection became further enhanced by the addition of TFIIF ( lanes 3 and 4 ) , consistent with the synergy between TBP and TFIIB binding to promoter DNA ( Tsai and Sigler , 2000 ) , and the stabilization of TFIIB binding by TFIIF ( Luse , 2012 ) . Interestingly , we observed significant changes in the footprint pattern upon the addition of Pol II ( lane 5 ) . These changes include ( i ) the exposure of some hypersensitive sites protected by the initial TFIID binding ( such as position ∼+16 , and the sites flanking the DPE ) , suggesting the release or unmasking of some DNA from the bound TFIID; ( ii ) emergence of new hypersensitive sites ( such as position ∼+14 ) , suggesting structural changes in the DNA trajectory itself caused by Pol II binding; ( iii ) reduction of the hypersensitive site induced by TFIID binding at the edge of the Inr , consistent with the release of some DNA from TFIID and/or the association of Pol II; and ( iv ) a strong and extended protection covering the upstream of the TATA box ( ∼−37 ) to the Inr , consistent with the engagement of Pol II and other PIC components with this region of the promoter . These results are also consistent with a previous report using a different promoter ( Yakovchuk et al . , 2010 ) that suggested some kind of a conformational isomerization at the promoter associated with Pol II engagement . 10 . 7554/eLife . 07777 . 016Figure 7 . DNase I footprinting assays monitoring PIC assembly and structural isomerization . ( A ) Early steps of PIC assembly . Specified GTFs were incubated with the end-labeled DNA template , followed by DNase I digestion . Blue bracket highlights the TFIID footprint . The numbers are relative to the transcription start site ( +1 ) . ( B ) Arresting of the conformational isomerization . Top , the scheme . The inhibitor used here was SnOCl2·pyridine . Lanes 1 and 9 are 10 bp DNA ladder . Lanes 2 and 10 are digestion of naked DNA . The lower case ‘d’ and ‘f’ in lanes #11–14 reflect the use of less TFIID and TFIIF ( together with the omission of spermidine and carrier nucleic acid in the reaction—see ‘Material and methods’ for detail ) . Letter abbreviations are explained in Figure 6 legend . DOI: http://dx . doi . org/10 . 7554/eLife . 07777 . 016 To examine the effect of the chemical inhibitor on this isomerization step , we used suboptimal levels of TFIID to first assemble a TFIID-TFIIB sub-complex at the promoter . This sub-complex gives a weak footprint ( that is expected to be more sensitive to perturbation than the footprint generated by optimal levels of TFIID ) ( Figure 7B , lanes 2 and 3 ) . Despite using suboptimal levels of TFIID , no significant change in the footprint protection was observed upon inhibitor treatment ( lane 4 ) . This is consistent with the hypothesis that the inhibitor likely acts at a stage after TFIID and TFIIB binding . Because Pol II and TFIIF are known to be tightly associated to each other physically and functionally ( Roeder , 1996 ) and they both rescue PIC assembly from the inhibition ( TFIIF partially and Pol II completely ) , we next added them together to the reaction and observed , as expected , the characteristic footprint changes associated with Pol II engagement ( lane5 ) . Importantly , in the presence of the inhibitor , the Pol II-induced footprint changes over the TATA box , the Inr , and in between were largely abolished , while the protection downstream of the Inr were partially altered , suggesting that the full engagement of Pol II is significantly disrupted by the tin compound along a major portion of the promoter ( lane 6 ) . In control experiments where the inhibitor was added after TFIID , TFIIB , TFIIF , and Pol II , it had almost no detectable effect on the footprint corresponding to the post-isomerization state of the partial PIC ( lanes 7 and 8 ) . This is consistent with the transcription profile where addition of Pol II renders the system resistant to the inhibitor ( Figure 6C ) . To better focus on the changes over the downstream promoter region , we also performed the experiment under conditions that generate optimal TFIID footprints . We verified that the inhibitor only partially compromised the downstream changes associated with Pol II engagement , although it almost completely blocked the changes covering the TATA and Inr elements ( lanes 9–14 ) . The changes of DNase I protection patterns observed in the downstream region suggest that the inhibitor may still allow Pol II to partially interact with this part of the promoter ( either directly or indirectly via TFIID ) , and that a simple complete exclusion of Pol II from accessing promoter DNA is unlikely to be the primary inhibition mechanism . These results are consistent with the possibility that the inhibitor directly stabilizes TFIID–Inr interactions and prevents the progression of the conformational isomerization . We conclude that transcription inhibition is most likely a consequence of incomplete Pol II engagement due to arrested TFIID isomerization . Once PIC assembly passes through this Pol II-dependent isomerization stage , the inhibitor apparently can no longer interfere with transcription . The tight correlation between Pol II engagement and inhibitor sensitivity does not rule out the possibility of a mutual steric exclusion between inhibitor binding to the TAF2 IDR and Pol II engagement at the promoter , which may be independent of the conformational isomerization . Because Pol II and TFIIF are thought to leave the promoter during the elongation phase after de novo transcription initiation , while TFIID may be retained at the promoter as part of a reinitiation scaffold ( Zawel et al . , 1995 ) , we reasoned that examination of transcription reinitiation in our system may help to further dissect the inhibition mechanism . A specialized transcription reinitiation process that is distinct from the process of de novo PIC assembly pathway has been reported for in vitro transcription systems supported by human and yeast nuclear extracts or purified factors ( Hawley and Roeder , 1987; Zawel et al . , 1995; Yudkovsky et al . , 2000 ) . Although TFIID is thought to be part of the reinitiation scaffold , if and how the retained TFIID might differ from complexes assembled during de novo initiation has remained unclear . In addition , using Drosophila embryo nuclear extracts and a dual template assay system , Kadonaga has demonstrated that each round of transcription requires complete assembly and disassembly of the PIC . Thus , at least in vitro , under certain transcription conditions , there is no commitment of any limiting transcription factor to the initial template to form a pre-licensed reinitiation scaffold ( Kadonaga , 1990 ) . This observed difference is likely a result of various regulatory protein factors and/or promoter elements present in different reaction systems . We therefore decided to further investigate whether TFIID is committed to an initiating DNA template and becomes pre-licensed for reinitiation in our completely defined highly purified human Pol II transcription system . We first performed transcription reactions with two DNA templates that produce transcripts of distinct lengths ( Figure 8A ) . We incubated the first DNA template ( DNA 1 ) with a complete set of protein factors to form a stable PIC , then added the second DNA template ( DNA 2 ) to see whether the essential factors are still available . To constrain transcription to a single round so that we can compare de novo PIC assembly on the two templates , we added 0 . 1% Sarkosyl immediately after the addition of nucleoside triphosphates ( NTPs ) for RNA synthesis following previously reported procedures using crude nuclear extracts ( Hawley and Roeder , 1985; Kadonaga , 1990 ) , which is also validated in our highly purified system ( Figure 8—figure supplement 1A ) . We found that the presence and pre-incubation of DNA 1 severely compromised ( by 11–18-fold ) transcription from DNA 2 ( Figure 8A , compare lanes 2–4 with 7 for the short transcript , and lanes 9–11 with 14 for the long transcript ) , indicating that some limiting factors become stably committed to the first DNA template during initial PIC assembly . Adding more fresh TFIID ( 1 × equivalent to the initial dosage ) immediately after DNA 2 restored transcription activity from DNA 2 by ∼fourfold ( comparing lanes 4 with 5 , and 11 with 12 ) , suggesting that TFIID is likely one of the limiting and DNA template committed factors during de novo PIC assembly . 10 . 7554/eLife . 07777 . 017Figure 8 . Template commitment of TFIID and its resistant to tin ( IV ) oxochloride inhibition during reinitiation . ( A ) Two-template assay to test the template commitment of GTFs . The cartoon illustrates the question . Bottom left is the experimental scheme . After PIC assembled on the first DNA template ( DNA 1 ) , the second DNA template ( DNA 2 ) was added ( as an option , onefold extra TFIID can also be added immediately following DNA 2 ) and incubated for specified time , followed by the addition of nucleoside triphosphates ( NTPs ) for RNA synthesis . Sarkosyl was added to a final concentration of 0 . 1% ( within 30 s after the addition of NTPs ) as an option to restrict transcription to a single-round . The two DNA templates both contain a SCP1 , but lead to primer extension products of different length ( L: long , 192 bases; S: short , 155 bases ) . Bottom right is the results , sub-divided into four groups ( dashed boxes ) . The transcription signals were normalized within each group for each ( L or S ) specific primer extension product ( shown in blue immediately under the specific bands ) . ( B ) Comparison of single-round transcription vs multiple-round transcription . The cartoon illustrates the question to address: whether TFIID in the reinitiation scaffold is sensitive to inhibition or not . Blue ‘x’ indicates the position of the first G residue where the first Pol II will be stalled in the absence of GTP . Bottom left is the scheme . ‘DNA buffer’ contained no template DNA . SnOCl2·pyridine was used as the inhibitor . Bottom right is the result . Black arrows indicate the bands corresponding to the first ( 1 ) , second ( 2 ) , and third ( 3 ) transcript synthesized from the DNA template . Blue numbers are the normalized quantification of the first transcript from each lane . All three band intensity is plotted as ‘Total signal’ ( the first transcript of the first lane from the left was arbitrarily set as 100 ) . Reinitiation rate was plotted by setting the first transcript of each lane as 100 ( to calculate the chance of the second and third round of transcription to occur in each reaction ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07777 . 01710 . 7554/eLife . 07777 . 018Figure 8—figure supplement 1 . Controls for reinitiation experiments . ( A ) Sarkosyl titration . Sarkosyl was added at the beginning of PIC assembly ( a ) , immediately ( within 30 s ) before ( b ) or after ( c ) the addition of NTPs ( top scheme ) , and the effect on transcription was detected by primer extension ( bottom images ) . In blue are the normalized transcription signals . ( B ) Comparison of single-round transcription vs multiple-round transcription using Sarkosyl treatment . Left , the scheme . Right , the image of the primer extension products . The transcription signals were quantified , normalized , and fold of inhibition under each condition were calculated . 0 . 1% Sarkosyl was added within 30 s after NTPs to restrict transcription to a single round . ( C ) Sarkosyl control for the G-less cassette transcription . 0 . 02% Sarkosyl was added right before the addition of NTPs to prevent reinitiation . Black arrowheads points to the bands corresponding to the first , second , and third transcript synthesized from the same DNA templates , and their signals were quantified , normalized ( the blue numbers ) , and plotted below the image . DOI: http://dx . doi . org/10 . 7554/eLife . 07777 . 018 Omitting Sarkosyl treatment so that transcription can go through multiple rounds led to an increase in transcription signal by ∼threefold ( comparing lanes 1 , 4 , and 6 for the long transcript , and lanes 8 , 11 , and 13 for the short transcript ) , indicating the detection of one de novo round of transcription initiation plus ∼2 rounds of reinitiation . In the absence of Sarkosyl treatment , the presence and pre-incubation of DNA 1 with the protein factors still severely compromised ( by 12–16-folds ) transcription from DNA 2 ( comparing lanes 15 and 17 for the short transcript , and lanes 18 and 20 for the long transcript ) . As expected , this template commitment under multi-round transcription conditions can also be partially alleviated by the addition of extra TFIID ( by ∼sixfold ) ( lanes 16 and 19 ) . These findings suggest that commitment of the limiting factor ( including TFIID ) to DNA 1 during de novo transcription initiation mostly persists through multiple rounds of reinitiation . Therefore , as reported with other human transcription systems , our highly purified system also involves a stable reinitiation scaffold of which TFIID is likely a critical component . To probe the functional states of TFIID ( particularly that of the TAF2 IDR ) during transcription reinitiation , we treated the reaction with the inhibitor either before or after PIC assembly , either under the condition that would allow multiple rounds of transcription or under Sarkosyl treatment condition limiting transcription to the first , de novo round ( Figure 8—figure supplement 1B ) . As expected , tin-compound treatment before PIC assembly efficiently inhibited TFIID-directed transcription ( comparing lanes 3 , 4 , 7 , and 8 from left ) under both conditions . On the other hand , when the inhibitor was added after PIC assembly ( before the addition of NTPs , which in turn is before the formation of the reinitiation scaffold ) , there was no inhibition of transcription under either the single ( de novo initiation ) or multiple ( de novo plus reinitiation ) round conditions , indicating that reinitiation and the re-entry of Pol II to the reinitiation scaffold is largely refractory to inhibition by the tin-oxochloride compound . To further independently validate the resistance of the reinitiation scaffold to inhibition , we used a colliding polymerase assay ( Szentirmay and Sawadogo , 1994 ) . In this assay , a sequence containing no G residues was inserted after the adenovirus major late promoter ( for this ‘G-less’ cassette-based assay , the SCP1 is not suitable because it contains multiple G residues in the initially transcribed sequence ) . Transcription in the absence of GTP ( but in the presence of all the other three required nucleoside triphosphates ) would cause the first Pol II molecule to stall at the position of the first G-residue after the G-less cassette , and to stall the second and third Pol II molecules at positions further upstream . Using alpha-32P labeled CTP as a substrate , this system will allow the first Pol II molecule to synthesize a nascent transcript corresponding to the length of the G-less cassette , followed by progressively shorter transcripts that result from reinitiation ( Szentirmay and Sawadogo , 1994 ) . Using our highly purified transcription factors , we observed just such a ladder of nascent transcripts of expected sizes as previously reported ( Szentirmay and Sawadogo , 1994 ) . Importantly , addition of 0 . 02% Sarkosyl just before the NTP substrates only weakly diminished the synthesis of the longest transcript , but almost completely abolished the shorter ones ( Figure 8—figure supplement 1C ) , mirroring the effect of 0 . 02% Sarkosyl in mildly affecting preassembled PIC for the first round of transcription while completely blocking reinitiation ( Figure 8—figure supplement 1A ) . Thus , the longest transcript likely represents the products of de novo ( first round ) transcription , while the short ones represent reinitiation products in our system . Quantification of the longest transcription confirmed robust ( ninefold ) inhibition of de novo transcription initiation when the chemical was added before PIC assembly , and this inhibition became highly attenuated ( twofold ) when the chemical was added after PIC assembly . When we normalized the signal from each round of transcription in each reaction to the first round of transcription , we found the ratio of the second and third transcripts to the first one ( ‘reinitiation rate’ ) to be highly consistent across all the samples . In particular , inhibitor added after PIC assembly ( but before NTPs allow the first Pol II to escape the promoter ) had no effect on reinitiation rate , confirming that the chance of a reinitiation scaffold to support more transcription events ( after the first Pol II escapes the promoter ) is not sensitive to the inhibitor . Taken together , our studies on reinitiation suggest that after Pol II leaves the promoter , the template-committed TFIID complex remains in an inhibitor-resistant state , distinct from that of the TFIID-TFIIB-DNA intermediate during the early steps of de novo initiation that appears to be the target of the inhibitor . Therefore , the inhibitor-resistance caused by Pol II engagement during de novo PIC assembly is unlikely simple steric masking , but rather a result of rearrangement of TFIID–DNA interactions . This difference in TFIID conformation and functional state between de novo initiation and reinitiation also suggests how TFIID and the pre-licensed scaffold might bypass certain stages of de novo PIC assembly to facilitate reinitiation . Accordingly , a model is proposed summarizing the mechanistic insights revealed by the inhibitor ( Figure 9 ) . 10 . 7554/eLife . 07777 . 019Figure 9 . A model of inhibition that mechanistically distinguishes the two modes of transcription initiation . ( A ) Initially , TFIID forms multiple contacts with an extended promoter DNA region that is stabilized by TFIIA and TFIIB . TFIIA doesn't affect transcription at this promoter with purified factors , but it does facilitate the TATA box protection by TFIID alone ( Cianfrocco et al . , 2013 ) or by TFIID together with TFIIB ( ZZ and RT unpublished ) . We propose a critical isomerization step during de novo PIC assembly involving a TFIID conformational change ( i . e . , release of at least part of the promoter DNA , illustrated by the change in the shape of TFIID ) to allow entry and engagement of Pol II . Once Pol II becomes engaged and further stabilized by other factors ( TFIIE , TFIIF , etc ) transcription can proceed . ( B ) The inhibitor , by binding and interfering with the TAF2 IDR , arrests TFIID isomerization and Pol II engagement , thus , blocking the assembly of a functional PIC . DNase I footprint assay reveals that Pol II molecules can still partially interact with the downstream portion of promoter DNA in the presence of the inhibitor . ( C ) Once the first round of Pol II engagement is accomplished and isomerization has occurred , the PIC intermediate establishes a state resistant to inhibition . After Pol II enters the elongation phase , TFIID remains at the isomerized state as part of a reinitiation scaffold . This reinitiation complex bypasses the initial stages of de novo PIC assembly where TFIID contacts an extended DNA region and thus is resistant to the inhibition by the tin ( IV ) oxochloride cluster . In addition , this shortcut may be a mechanism for the reinitiation scaffold to facilitate reloading of more Pol II molecules . DOI: http://dx . doi . org/10 . 7554/eLife . 07777 . 019 In this study , we identified an unconventional chemical inhibitor of Pol II transcription and used it to probe the function of an IDR within the TAF2 subunit of metazoan TFIID . Although IDRs are abundant , mechanistic studies of their functions have been challenging . The TAF2 IDR targeted by the tin ( IV ) oxochloride-derived inhibitor turned out to be a newly identified TAF2 DNA-binding domain that likely becomes structured upon DNA binding , perhaps reminiscent of basic domains in the leucine zipper family of activators ( Shuman et al . , 1990; Weiss et al . , 1990 ) . On the other hand , the TAF2 IDR domain , with its low-complexity sequences , does not appear to be responsible for the sequence-specific recognition of the Inr element by TAF2 ( Verrijzer et al . , 1994 ) . Instead , this IDR domain apparently binds DNA non-specifically , consistent with our finding that template DNA bearing Inr mutations still responded to the inhibitor in transcription reactions . One possible explanation for the inhibition mechanism is an over-stabilization of TAF2 IDR-DNA interaction by the inhibitor . In this model , the inhibitor-stabilized TAF2 IDR is an impediment to full Pol II engagement . We postulate that this Pol II-dependent isomerization is a multi-step process that engages an extended region of the promoter DNA that gets blocked at the Inr by the inhibitor . This model is consistent with our DNase I footprinting results . Alternatively , if the TAF2 IDR is an active helper for Pol II to overcome the barrier posed by other bound factors ( like TFIID ) to interact with the upstream DNA and the associated protein factors ( such as TFIIB ) , interference of TAF2 IDR function could also lead to a partially impeded Pol II engagement . This is reasonable because IDRs are known to interact with multiple partners with diverse functions ( Uversky , 2013 ) . Based on the different levels of sensitivity to inhibition , Pol II engagement can be dissected into at least two sections: the release of the downstream elements from TFIID ( which can partially occur in the presence of the inhibitor ) and the binding and reorganization of the TATA-Inr region ( which is almost completely inhibited ) . Further study using the TFIID inhibitor and advanced techniques , such as single-molecule imaging suitable to capture fast dynamics in complex biochemical reactions ( Revyakin et al . , 2012 ) , could be useful to reveal additional mechanistic aspects of this elusive part of PIC assembly . Given that IDRs are highly sensitive to their local molecular environment ( Uversky , 2013 ) , it is possible that the TAF2 IDR may be selectively targeted by some cellular factors to modulate transcription initiation in a manner analogous to the action of the tin-based inhibitor described here . Interestingly , although generally considered a coactivator ( Albright and Tjian , 2000 ) , TFIID has also been found at promoters of silenced/inactive genes ( Breiling et al . , 2001; Tatarakis et al . , 2008 ) and may even function as a ‘checkpoint’ during activation ( Marr et al . , 2006 ) . It remains to be seen whether the TAF2 IDR is required for such reported checkpoint functions . Our findings regarding the TAF2 IDR are reminiscent of regulatory functions mediated by the AT-hook domains of TAF1 . Much like TAF2 , the TAF1 subunit of TFIID also plays critical roles in core promoter recognition ( Juven-Gershon and Kadonaga , 2010 ) . The two AT-hook DNA-binding domains of TAF1 preferentially associate with the narrower minor grooves associated with AT-rich sequences ( Metcalf and Wassarman , 2006 ) . Most interestingly , AT-hooks are also intrinsically disordered ( Liu et al . , 2006 ) , and their phosphorylation by casein kinase II was proposed to regulate core promoter selectivity by altering the conformation of TFIID ( Lewis et al . , 2005 ) . Therefore , the various intrinsically disordered DNA-binding domains of TFIID subunits may underscore previously underappreciated versatility of this complex in recognizing core promoter elements and responding to activators . The tin-based chemical reported here is the first known reagent capable of directly and mechanistically distinguishing the functional states of TFIID during initiation and reinitiation and discriminating between these two modes of transcription . Reinitiation is likely an important aspect of transcription regulation in vivo , possibly related to ‘bursting’ ( Raj et al . , 2006 ) , but its mechanism is poorly understood . We found that although the engagement of Pol II during de novo PIC assembly is sensitive to the inhibitor , re-entry of Pol II during reinitiation is inhibitor resistant . The mechanism behind this difference is linked to the different functional states of TFIID and the TAF2 IDR targeted by the inhibitor . As a reagent for mechanistic studies , the tin ( IV ) oxochloride cluster is distinct from Sarkosyl , which has traditionally been used to study reinitiation ( Hawley and Roeder , 1987 ) . Sarkosyl in fact does not discriminate between initiation and reinitiation , even in cases where specialized reinitiation scaffolds are involved . Proper targeting of reinitiation by Sarkosyl relies on the timing of addition ( after de novo PIC assembly but before reinitiation in synchronized reactions ) . In addition , the actual target of Sarkosyl has remained largely a mystery . In contrast , the tin ( IV ) oxochloride cluster identified here has a clear and specific target and obviates the necessity for timed addition by directly blocking initiation while leaving reinitiation and on-going transcription intact . We should point out that de novo initiation and reinitiation may not necessarily represent mechanistically distinct pathways under certain conditions in vitro . Numerous possibilities have been proposed by researchers for the apparent differences in observing disengagement of PIC components at each round ( Kadonaga , 1990 ) vs retention of ‘reinitiation scaffolds’ for reinitiation through pathways distinct from de novo transcription initiation ( Hawley and Roeder , 1987; Zawel et al . , 1995; Yudkovsky et al . , 2000 ) . For example , the role of activators and/or specific promoters in retaining TFIID , Mot1 proteins in removing and recycling TFIID , and other limiting factors , etc has all been proposed to influence these two potential mechanistically distinct pathways to achieve multiple rounds of transcription . It is also possible that the reported discrepancies can be a reflection of the complexity and diversity of regulatory factors assembled as part of the core transcription machinery responsible for promoter recognition in different organisms and possibly even distinct cell types . Although the newly identified compound may present complications for direct in vivo applications , this reagent revealed an intriguing possibility of specifically blocking de novo transcription initiation without affecting reinitiation via targeting of TFIID . Interestingly , the canonical TFIID complex has been reported in several cases to be dispensable for ongoing transcription in terminally differentiated , non-dividing cells ( Cler et al . , 2009; Goodrich and Tjian , 2010; Muller et al . , 2010 ) . Accordingly , these cells are likely to be resistant to perturbation by chemicals that specifically target the canonical TFIID complex , particularly the function of TFIID in de novo transcription initiation . In contrast , biological processes stringently dependent on de novo transcription initiation , such as the onset of viral transcription after infection or re-establishment of mRNA synthesis in rapidly dividing cancer cells , should be more sensitive to this kind of chemical perturbation . Therefore , with specific functions required for de novo transcription initiation but not reinitiation , the canonical TFIID complex has the potential to be even more selective than TFIIH , another component of the Pol II core transcription machinery that was recently found to be a promising target for cancer therapy development ( Titov et al . , 2011; Chipumuro et al . , 2014; Kwiatkowski et al . , 2014 ) . Our unbiased de novo identification of a tin-based metal cluster as a selective transcription inhibitor provides an example of a potential category of metal-based biologically active compounds that is distinct from the conventional transition metal-based POMs ( Rhule et al . , 1998 ) . The critical role of histidine coordination demonstrated here is a direct expansion of a recent report on a nickel/cobalt substitution in potentiating POM compounds to prevent amyloid β peptide aggregation involved in Alzheimer's diseases ( Gao et al . , 2014 ) . With locally enriched charged residues , the TAF2 IDR binding to the metal complex may involve electrostatic interactions as well . Histidine coordination and salt bridges by themselves are unlikely to be sufficient for high-specificity interactions . However , the specific biochemical activity and the synergy with DNA ( but not heparin ) binding suggest a reasonable level of structural selectivity . This is consistent with our finding that a tris-Ni-nitrilotriacetic acid derivative capable of coordinating multiple histidines ( Lata et al . , 2005 ) can bind TAF2 in the context of human TFIID but fails to inhibit transcription under relevant concentrations ( data not shown ) . We speculate this selectivity may be partially ascribed to the periodic nature of the rigid metal–oxygen backbone of our tin compound , which can at least discriminate secondary structures ( α helices , β sheets , and random coils ) of the protein target ( presenting the repetitive residues with distinct periodicity ) . This discrimination may be sufficient to shift the equilibrium between the functional states of the dynamic IDR , as exemplified recently by a phosphorylation-induced β-barrel formation and its regulation of a disorder-to-helix transition within an IDR of the 4E-BP2 translational regulator ( Bah et al . , 2015 ) . We thus propose that certain features of metal complexes can be harnessed in the design of new reagents targeting polar and repetitive IDRs that have largely evaded intervention by traditional organic small-molecule modulators . TFIID complexes were affinity purified from fractionated nuclear extracts with homemade monoclonal antibodies . In brief , for dTFIID , nuclear extract from 0 to ∼12 hr Drosophila melanogaster embryos was prepared as described ( Biggin and Tjian , 1988 ) , fractionated by SP-sepharose resin ( GE Healthcare , Pittsburgh , PA ) , and elution at 1 M KCl salt concentration from the SP–column was reloaded to a Q-sepharose column ( GE Healthcare ) . A fraction eluted at 0 . 3 M KCl salt concentration from the Q-column was collected , immune-precipitated by a monoclonal antibody ( 2B2 ) raised against the TAF1 subunit of dTFIID , and eluted with a specific epitope peptide . The purity and integrity of the complex was verified by silver staining of sodium dodecyl sulfate-polyacrylamide gel electrophoresis ( SDS-PAGE ) -separated samples and protein mass spectrometry . For hTFIID , nuclear extract from HeLa cells was fractionated by P11 phosphate-cellulose resin ( Whatman , Maidstone , United Kingdom ) and collected at 1 M KCl salt elution , then affinity purified as described ( Liu et al . , 2009 ) . Human TFIIH and Pol II complexes were purified from HeLa nuclear extract as described ( Revyakin et al . , 2012 ) . Recombinant proteins were expressed in Escherichia coli and purified according to manufacturer suggested methods . GST-fusion proteins were purified by Glutathione-sepharose 4B resin ( GE Healthcare ) , and ( His ) 6-Halo fusion proteins was purified by Ni-NTA agarose resin ( Qiagen , Venlo , Netherlands ) to ∼>90% purity in SDS-PAGE Coomassie brilliant blue-stained gels . Nuclease treatment of recombinant protein for small-molecule microarray binding was carried out on beads before elution with 0 . 5 U/µl DNase I ( Roche , Basel , Switzerland ) and 10 ng/µl RNase ( Sigma–Aldrich , Milwaukee , WI ) at room temperature with mixing for 1 hr in a buffer containing 50 mM Tris pH 8 . 0 , 5 mM MgCl2 , 2 . 5 mM CaCl2 , 5% glycerol , and 1 mM dithiothreitol ( DTT ) . All the SCP1-related constructs , except for the ‘L’ DNA used in Figure 7A , were constructed by inserting the SCP1 and mutants as described ( Juven-Gershon et al . , 2006 ) in place of the adenovirus E1a promoter ( TATA box ) of the pG3-BCAT plasmid ( Ryu et al . , 1999 ) upstream to a chloramphenicol acetyl transferase ( CAT ) reporter gene . The ‘L’ DNA template had an extra 37 base-pair insertion after the promoter . The G-less cassette DNA template containing the adenovirus major late promoter is the pML ( C2AT ) 19Δ-50 as reported ( Szentirmay and Sawadogo , 1994 ) ( a gift from Dr Manabu Mizuguchi ) . For the primer extension assay used to quantify transcription products , a reverse primer ( 5′ GCCATTGGGATATATCAACGGTGG 3′ ) starting at position +155 ( +192 for the ‘L’ DNA template ) relative to the transcription start site was used . For the DNase I footprinting assay , the reverse primer was used together with a forward primer ( 5′ CATAACCTTATGTATCATACACATACG 3′ ) , starting at position −153 , to generate a DNA template by PCR . DNA oligonucleotide primers were custom synthesized ( Integrated DNA Technology , Coralville , IA ) . Both transcription and footprinting assays used wild-type SCP1-containing DNA unless otherwise specified . Small-molecule microarray manufacture and screening were carried out as described ( Casalena et al . , 2012 ) . In brief , the slides were blocked in 3% ( wt/vol ) bovine serum albumin ( BSA ) ( Sigma–Aldrich , Milwaukee , WI ) in phosphate-buffered saline supplemented with 0 . 1% Tween 20 ( PBST ) for 1 hr at room temperature with gentle shaking , rinsed with 0 . 04% BSA in PBST , then incubated at 4°C for 1 hr , sequentially , with protein of interest ( concentrations optimized for the best signal/noise ratio , usually 1–10 µg/ml ) , primary antibody ( at optimized dilutions ) and secondary antibody ( 1/1000 ) diluted in PBST ( with or without 0 . 04% BSA ) and washed 5 min in between with the same buffer . For the rescue of nuclease-treated protein , heparin ( Sigma–Aldrich ) ( 2 µg/ml ) or a dsDNA oligonucleotide ( 5′ GCTTGCATGCGTACTTATATAAGGGGGTGGGGGCGCGTT 3′ ) ( 1 µg/ml ) was incubated together with the recombinant protein ( 0 . 5 µg/ml ) . The slides were further rinsed with PBST and water briefly , spun dry , immediately scanned at 532 nm and 635 nm using a GenePix 4000B ( Molecular Devices , Sunnyvale , CA ) slide scanner and the images were analyzed by GenePix Pro6 software ( Molecular Devices ) . The primary , monoclonal antibodies are either in-house raised and Protein G affinity purified from hybridoma culture supernatants ( anti-dTAF4 , 3E12; anti-hXPB , 30C1-1; anti-Rpb1: 8WG16 ) or commercially available ( anti-hTAF4: BD Biosciences ( Sao Paulo , Brazil ) Cat #612054; Anti-GST: Sigma–Aldrich Cat #G-1160_0 . 5ML ) . Cy5-labeled secondary antibody against mouse IgG was purchased from GE Healthcare Life Sciences ( Cat #PA45002 ) . For the plots in Figure 1C , the strongest signals from reference dyes were removed . Note: the retention of SnOCl2 cluster on the microarray is likely mediated through coordinating interactions and/or local polymerization of the material , which will preserve similar surface features as proposed in Figure 3—figure supplement 3E for protein interactions . The surface plasmon resonance experiments were performed using a BIACORE T100 ( GE Healthcare ) . GST and GST tagged dTAF2 ( 1125–1221 ) were captured on reference cell and active cell , respectively , through GST capturing kit ( GE Healthcare BR100223 ) on CM5 sensor chip ( GE Healthcare BR100012 ) . For the capturing of hTAF2 ( 990–1199 ) , a HaloTag Amine ( O4 ) Ligand ( Promega , Fitchburg , WI , Cat #P6741 ) was immobilized on all flow cells using amine-coupling chemistry on CM5 sensor chip . A ( His ) 6-Halo-tagged reference protein ( PP7 bacteriophage coat protein [Chao et al . , 2008] ) and a ( His ) 6-Halo-tagged hTAF2 ( 990–1199 ) were immobilized through Halo tag on the reference and active cell , respectively ( no evidence for direct interaction between the ( His ) 6 tag and the inhibitor was observed ) . The inhibitor binding assay was performed in running buffer ( 20 mM HEPES pH 7 . 5 , 100 mM KCl , 0 . 5 mM EDTA , 0 . 005% surfactant Polysorbate 20 , and 2% DMSO ) ( HEPES: 4- ( 2-hydroxyethyl ) -1-piperazineethanesulfonic acid; EDTA: ethylenediaminetetraacetic acid; DMSO: dimethyl sulfoxide ) at a flow rate of 30 µl/min with 120 s of association and dissociation at 25°C . SnOCl2 or SnOCl2·pyridine was prepared at concentration of 5 µg/ml in different buffers adjusted with citrate for lower pH or supplemented with imidazole at specified concentrations . For TAF2 IDR-DNA binding , GST fusions immobilized on glutathione sepharose 4B beads ( ∼2 mg/ml ) were briefly pretreated with 100 µg/ml RNase A in PBST in the presence of complete protease inhibitor ( Roche ) to remove endogenous nucleic acid ( predominantly RNA ) from E . coli . 10 µl beads were incubated in a 384-well plate with 20 ng/µl double-stranded DNA oligonucleotides suspended in 40 µl PBST for 1 hr with gentle shaking at 4°C . 20 µl of the unbound fraction was saved and the beads were washed and re-suspended in 20 µl PBST . Both the bound and unbound were supplemented with 10 µl , 4 µg/ml ethidium bromide and scanned by Typhoon scanner ( GE Healthcare ) and the signals were quantified by ImageQuant TL software ( GE Healthcare ) . The procedure was as described ( Revyakin et al . , 2012 ) with minor modifications . The GTFs included in a standard 25 µl reaction are ∼10 ng ( 0 . 1 µl ) affinity purified h/dTFIID , 5 ng hTFIIB , 20 ng hTFIIE-α , 20 ng hTFIIE-β , 20 ng hTFIIF , ∼5 ng TFIIH , and ∼20 ng Pol II . As controls for TFIID specificity , 5 ng recombinant dTBP ( his-tagged ) was used . These factors were pooled and diluted in a GTF buffer ( 10% glycerol , 25 mM HEPES pH 7 . 9 , 12 . 5 mM MgCl2 , 100 mM KCl , 0 . 1 mM EDTA , 0 . 01% NP40 , 0 . 02% Tween 20 , 1 mM DTT , and 100 µg/ml BSA ) . The chemicals or the DMSO control was dissolved with the same GTF buffer ( with a final DMSO concentration of 2% ) , kept on ice for a few hours , spun at 15 , 000 rpm for 15 min , before they were used to treat the protein factors ( total volume 12 . 5 µl ) at specified concentrations at room temperature ( 23°C ) for 15 min . Next , 100 ng DNA in 12 . 5 µl DNA buffer ( 0 . 2 µl RNasin ( Promega ) and 8 mM spermidine in water ) was added and incubated at 30°C for 30 min for PIC assembly . NTPs were then added to 0 . 5 mM each to allow RNA synthesis at 30°C for 30 min . RNA products were analyzed by primer extension and 6% urea-polyacrylamide gel electrophoresis , scanned by Typhoon scanner ( GE Healthcare ) , and quantified by ImageQuant TL software ( GE Healthcare ) . For the dinucleotide synthesis monitoring abortive initiation , following modifications were made according to a previous report ( Orphanides et al . , 1998 ) : ( 1 ) the NTP substrates only included 1 mM ATP and 1 . 3 µM GTP ( including 0 . 3 µM from α-32P labeled GTP , 3000 Ci/mmole , 10 mCi/ml , Perkin Elmer , Waltham , MA ) ; ( 2 ) the transcription was stopped by incubation at 65°C for 30 min , followed by 4 U shrimp alkaline phosphatase ( New England Biolabs ( NEB ) , Ipswich , MA ) treatment at 37°C for 1 hr; and ( 3 ) the end product was separated by 15% TBE-urea gel ( Thermo Fisher Scientific , Waltham , MA ) . For the step-wise PIC assembly test shown in Figure 6 , GTF set 1 ( 6 µl ) was incubated with template DNA ( 6 µl ) for 30 min at 30°C before the addition of inhibitor/DMSO ( 8 µl , in GTF buffer ) to specified concentrations . After inhibitor treatment at specified concentration for 15 min at 23°C , GTF set 2 ( 3 µl ) was added together with 11 µl DNA buffer , and PIC assembly was finished by another incubation at 30°C for 30 min . For the results shown in Figure 6—figure supplement 1 , 6 µl of inhibitor/DMSO was used to get the specified concentrations , GTF set 2 was in 1 µl , and DNA or DNA buffer was 7 µl . For the comparison of inhibitor effect before or after PIC assembly ( Figure 8 and Figure 8—figure supplement 1B ) , the complete GTF mix , DNA template ( 100 ng DNA in DNA buffer ) , inhibitor/DMSO diluted in GTF buffer , and DNA buffer alone are all 6 µl each . The colliding polymerase assay using G-less cassette DNA template has the following modifications based on a previous report ( Szentirmay and Sawadogo , 1994 ) : ( 1 ) the template DNA was pML ( C2AT ) 19Δ-50 as reported ( Szentirmay and Sawadogo , 1994 ) ; ( 2 ) ATP and UTP were 0 . 6 mM; CTP was 25 μM ( including 0 . 3 μM diluted from α-32P labeled CTP , 3000 Ci/mmole , 10 mCi/ml , Perkin Elmer ) ; ( 3 ) no GTP was included in the reaction; ( 4 ) the reaction was stopped with 100 µl of Stop Solution ( 20 mM EDTA , 1% SDS , 0 . 2 M NaCl , 0 . 15 μg/μl glycogen , and 0 . 1 μg/μl Proteinase K ) , further incubated at 55°C for 10 min , then extracted by phenol:chloroform:isoamylalcohol ( 50:49:1 ) and precipitated with ethanol; ( 5 ) the precipitation pellet was re-suspended in 4 µl buffer ( 50 mM Tris pH 7 . 5 and 2 mM EDTA ) containing 10 U/µl RNase T1 ( Thermo Fisher ) , digested at 37°C for 10 min , then separated by 6% polyacrylamide-urea denaturing gel . Sarkosyl was tested in our system following previously a reported procedure ( Hawley and Roeder , 1985 ) . To restrict transcription to a single round , Sarkosyl was added either to a final concentration of 0 . 02% just before the addition of NTPs or to 0 . 1% immediately ( within ∼30 s ) after the addition of NTPs . For the two-DNA , template-commitment assay , DNA 1 ( or DNA buffer only ) was incubated with the complete set of protein factors in GTF buffer for 30 min at 30°C in a total volume of 12 µl , followed by an addition of 6 µl GTF buffer ( with extra TFIID as an option ) and 6 µl DNA 2 . After incubation at 30°C for a specified period of time , NTPs were added and the incubation continued at 30°C for another 30 min for RNA synthesis . Sarkosyl was added to 0 . 1% immediately ( within ∼30 s ) after the addition of NTPs as an option to restrict transcription to a single , first round . The assay is essentially as described ( Cianfrocco et al . , 2013 ) with modifications . In brief , a DNA fragment was generated by PCR using a primer starting at −153 ( from the transcription start site ) of the non-template strand , and a radioactively ( 32P ) labeled primer starting at the +155 of the template strand . The TFIID-promoter binding ( Figure 5C ) was carried out in 10 µl buffer containing ( 5% glycerol , 12 . 5 mM HEPES pH 7 . 9 , 6 mM MgCl2 , 50 mM KCl , 50 μM EDTA , 0 . 05% NP40 , 0 . 01% Tween 20 , 50 µg/ml BSA , 1% DMSO , 0 . 5 mM DTT , ∼240 ng hTFIID , and ∼0 . 3 nM template DNA ) , with or without the chemical at specified concentrations , at 30°C for 30 min . For the high-salt challenge experiments in Figure 5—figure supplement 1 , extra KCl was added to bring to the specified final concentrations and the 30°C incubation was extended by another 5 min . The reaction was brought to 100 µl ( with 90 µl: 5% glycerol , 12 . 5 mM HEPES pH 7 . 9 , 6 mM MgCl2 , 50 mM KCl , 50 μM EDTA , 0 . 5 mM DTT , and 2 . 5 mM CaCl2 ) then immediately digested by pre-diluted DNase I ( Worthington , Lakewood , NJ ) at 30°C for 60 s . The DNase I digestion was stopped with 100 µl of Stop Solution ( 20 mM EDTA , 1% SDS , 0 . 2 M NaCl , 0 . 15 μg/μl glycogen , and 0 . 1 μg/μl Proteinase K ) . The digestion products were purified , separated by 6% polyacrylamide-urea denaturing gel , and the image further processed as in the transcription assay . The DNase I footprinting assay monitoring conformational isomerization during PIC assembly ( Figure 7A ) was carried out under conditions very similar to the transcription assay . Following are the modifications from a typical 25 µl transcription assay ( 1 ) the DNA template was 0 . 3 nM of the radioactively labeled PCR product ( a 308-bp fragment , instead of the ∼4 kb plasmid ) ; ( 2 ) 100 ng purified yeast tRNA and 3 ng poly ( dG:dC ) were included in each reaction as carriers , and ( 3 ) ∼40 ng TFIID was used; ( 4 ) at the end of PIC assembly , 2 µl 40 mU/µl DNase I ( NEB ) in a buffer ( 50% glycerol , 12 . 5 mM HEPES pH 7 . 9 , 6 . 25 mM MgCl2 , 50 mM KCl , 0 . 05 mM EDTA , 0 . 005% NP40 , 0 . 5 mM DTT , and 5 mM CaCl2 ) was added , and after 30 s , the digestion was stopped and further processed as described above . To monitor the inhibition of the conformational isomerization during PIC assembly ( Figure 7B , lanes 1–8 ) , GTF set 1 ( 30 ng hTFIID plus 5 ng hTFIIB , with or without 20 ng TFIIF plus 20 ng Pol II ) in 8 µl GTF buffer was mixed with 0 . 6 nM radioactively labeled DNA template in 8 µl DNA buffer ( 0 . 2 µl RNasin and 8 mM spermidine ) supplemented with 64 ng purified yeast tRNA and 2 ng poly ( dG:dC ) , incubated at 30°C for 30 min . Then , 4 µl SnOCl2·pyridine solution ( 15 µg/ml in GTF buffer with 2% DMSO ) or the control solution was added and incubated at room temperature for 15 min . Next , GTF set 2 ( 0 . 5 µl , with 20 ng TFIIF and 20 ng Pol II , or the GTF buffer only ) was supplemented , together with 4 . 5 µl DNA buffer ( supplemented with 36 ng purified yeast tRNA and 1 ng poly ( dG:dC ) ) and the incubation was continued for another 30 min at 30°C . The final reaction mixture was subjected to digestion by 2 µl 40 mU/µl DNase I ( NEB ) and the end digestion product was analyzed by gel electrophoresis . The experiment in Figure 7B lanes 9–14 was performed slightly differently in that: ( 1 ) the TFIID amount was 10 ng and TFIIF was 2 ng; ( 2 ) the DNA buffer was replaced with ( 10 mM Tris pH 8 . 0 , and 0 . 1 mM EDTA ) , and the carrier nucleic acids were eliminated; and ( 3 ) 2 µl 20 mU/µl DNase I ( NEB ) was used for the digestion . These two sets of conditions used in Figure 7B , when supplemented with other factors and reagents , both lead to optimal transcription output , and indistinguishable response to the tin ( IV ) oxochloride inhibitor . The commercially supplied lead compound and its analogs were named by an abbreviated vender name followed by specific catalog number . More specifically , ‘ChemDiv’ stands for ChemDiv Inc . ( San Diego , CA ) ; ‘Otava’ stands for Otava Ltd ( Vaughan , Canada ) ; ‘Princeton’ stands for Princeton Biomolecular Research Inc . ( Monmouth Junction , NJ ) ; and ‘LifeChem’ stands for Life Chemicals Inc ( Niagara-on-the-Lake , Canada ) . Nuclear magnetic resonance ( NMR ) and mass spectrometry were used by these commercial suppliers for quality control , which all indicated ‘>95%’ purity of the intended organic compounds ( note: those assays are not sensitive to inorganic metal compounds ) . Commercial proton-induced X-ray emission elemental analysis ( Figure 3B ) was carried out by Elemental Analysis Inc . ( Lexington , KY ) with 5 mg of each material and the abundance of detected elements was plotted . Details of in-house synthesis and analysis of the original lead ( 1 ) and analog ( 2 ) compounds are available upon request . Detailed procedure for resynthesis and purification of analog compound 2 was also purchased from Princeton Biomolecular Research Inc . for the track of the tin-related activity ( available upon request ) and used to draw the scheme in Figure 3—figure supplement 2A . In-house high-resolution NMR spectroscopy was carried out on Varian 500 MHz instrument ( Figure 3—figure supplement 1C ) . In-house recreation of the TFIID inhibitory activity was initially carried out in 2 ml glass vials with 1 ml isopropanol and 5 mM SnCl2 , sealed with a Teflon-lined screw cap and incubated at 83°C in a heat block ( Figure 3—figure supplement 2C , D ) . Samples were taken at different time points for activity analysis ( dried under vacuum then re-dissolved with DMSO ) . Acetic acid and the original lead/analog organic molecules are not required for generating the active species . Other organic solvents like methanol and DMSO can also be used to generate the inhibitory activity . Synthesis of the organic solvent coordinated SnCl4 compounds ( Figure 3—figure supplement 3A ) was accomplished by drop wise addition of neat SnCl4 to stirring liquid-coordinating compounds ( DMSO , n-propanol , iso-propanol , sec-butanol , iso-butanol , acetylacetone ) . The obtained precipitates were vacuum filtered and quickly transferred into sealed flasks and dried under high vacuum . The rest of the simple organic and inorganic tin compounds were purchased from Sigma–Aldrich . Oxygenation of SnCl2 could be achieved by refluxing in iso-propanol under air atmosphere for several hours , which led to potent TFIID-specific transcription inhibitory activity ( Figure 3C , and Figure 3—figure supplement 3B ) . In-house X-ray photoelectron spectroscopy measurement ( Figure 3—figure supplement 3C ) was completed on the Surface Science , model SSX-100 at Harvard's Center for Nanoscale Systems . The probe for the measurement was monochromatic aluminum K-alpha X-ray line with energy at 1 . 4866 keV . Flood gun was used throughout the entire experiment for sample surface charge compensation . The survey spectrum ( 0–1000 eV ) scan was completed by taking the average of 4 scans with X-ray spot size at 800 µm and passing energy at 150 eV . Data were analyzed using CasaXPS program , and ratios of integrated peak areas for each individual element were used for quantification . Alternatively , oxygenation of SnCl2 could be achieved by several oxidizing agents yielding active material ( Figure 3C and Figure 3—figure supplement 3B ) : ( a ) bubbling oxygen gas through a sintered glass tube into a solution of SnCl2 ( 1 g ) in tetrahydrofuran ( 26 ml ) , or water ( 26 ml ) at room temperature for 2 hr; ( b ) addition of H2O2 ( 30% aqueous , 6 ml ) to solution of SnCl2 ( 2 g ) in water ( 52 . 7 ml ) ; ( c ) addition of sodium perborate ( NaBO3 ) to aqueous solution of SnCl2; ( d ) addition of m-chloroperbenzoic acid ( mCPBA ) ( 58 mg ) to a solution of SnCl2 ( 64 mg ) in water ( 1 . 8 ml ) ; ( e ) addition of tert-butyl hydroperoxide ( TBHP ) to solution of SnCl2 in dichloromethane . The oxygenation of SnCl2 in tetrahydrofuran by oxygen to produce ( SnOCl2 ) X was reported previously ( Messin and Janierdubry , 1979 ) . The other oxygenation methods were our development . Over drying may compromise the biological activity , which can be restored by adding miniscule amount of water ( 1 µl per 1 mg of dry material ) prior to DMSO dissolution . Independently of oxygenation of SnCl2 , SnOCl2 cluster was also prepared from tin ( IV ) chloride ( SnCl4 ) ( Figure 3C ) as previously reported ( Sakurada et al . , 2000 ) : to a cooled ( −20°C ) 1 M solution of SnCl4 in dichloromethane ( 10 ml , 10 mmol , Sigma–Aldrich ) , bis ( trimethylsilyl ) peroxide ( BTSP ) ( 1 . 784 g , 10 ml , Gelest ) was added drop wise , and the reaction was allowed to proceed for 1 hr . The volatiles were removed in vacuum and the residue was used for the activity assay , or to complex with pyridine . SnOCl2·pyridine ( Figure 3C , D , Figure 3—figure supplement 3D , E , Figure 4C–F , Figures 6–8 ) was prepared by dissolving the residue obtained above in ethyl acetate ( 10 ml ) , followed by adding pyridine until heavy white precipitate kept forming . The precipitate was filtered , washed with excess ethyl acetate , and dried under high vacuum . Infrared spectrum for thus prepared compound was recorded on Bruker ALPHA FT-IR instrument and the resonances matched those published previously ( Dehnicke , 1961; Sakurada et al . , 2000 ) .

DNA contains instructions to make all the proteins and other molecules that drive essential processes in cells . To issue such specific sets of instructions , a section of DNA—called a gene—is first copied to make molecules of messenger ribonucleic acid ( or mRNA for short ) in a process called transcription . This process is tightly regulated in all living organisms so that only a subset of genes are actively transcribed at any time in a given cell . A group or ‘complex’ of proteins called TFIID plays an essential role in starting the transcription of genes that encode proteins in humans and other eukaryotic organisms . However , it is tricky to study how TFIID works because mutant cells that are missing individual components of the complex are unable to properly transcribe the required genes and soon die . Consequently , many studies of TFIID have used purified proteins in artificial systems where it is possible to examine particular aspects of TFIID activity in depth . Here , Zhang et al . used a combination of chemistry , biochemistry , and molecular biology techniques to identify a new molecule that can selectively bind to the TFIID complex . In an artificial system containing purified proteins and other molecules , this molecule ‘locks’ TFIID onto DNA and prevents a change in shape that is required for transcription to start . The experiments show that this rearrangement is only required to make the first mRNA copy of a gene because the molecule had no effect on initiating further rounds of transcription on the same DNA . Zhang et al . 's findings reveal that TFIID is very dynamic in controlling transcription , and that subsequent rounds of transcription follow a different path to make mRNAs . The next steps are to use new techniques such as single-molecule imaging to directly visualize the molecules involved in transcription , and to use the new molecule to block the start of transcription in living cells .

2015

Chemical perturbation of an intrinsically disordered region of TFIID distinguishes two modes of transcription initiation

Dynamic post-transcriptional control of RNA expression by RNA-binding proteins ( RBPs ) is critical during immune response . ZFP36 RBPs are prominent inflammatory regulators linked to autoimmunity and cancer , but functions in adaptive immunity are less clear . We used HITS-CLIP to define ZFP36 targets in mouse T cells , revealing unanticipated actions in regulating T-cell activation , proliferation , and effector functions . Transcriptome and ribosome profiling showed that ZFP36 represses mRNA target abundance and translation , notably through novel AU-rich sites in coding sequence . Functional studies revealed that ZFP36 regulates early T-cell activation kinetics cell autonomously , by attenuating activation marker expression , limiting T cell expansion , and promoting apoptosis . Strikingly , loss of ZFP36 in vivo accelerated T cell responses to acute viral infection and enhanced anti-viral immunity . These findings uncover a critical role for ZFP36 RBPs in restraining T cell expansion and effector functions , and suggest ZFP36 inhibition as a strategy to enhance immune-based therapies . Immune responses require precise , dynamic gene regulation that must activate rapidly as threats rise and resolve efficiently as they clear . Post-transcriptional control of mRNA abundance and expression by RNA binding proteins ( RBPs ) is a key layer of this response that can enact rapid , signal-responsive changes ( Kafasla et al . , 2014; Hao and Baltimore , 2009 ) , but knowledge of specific functional roles for dedicated RBPs remains limited . AU-rich elements ( AREs ) in mRNA 3’-untranslated regions ( 3’-UTR ) facilitate post-transcriptional control of many immune functions , including cytokine expression , signal transduction , and immediate-early transcriptional response ( Chen and Shyu , 1994; Shaw and Kamen , 1986; Caput et al . , 1986 ) . Many RBPs bind AREs , with diverse ensuing effects on RNA turnover , translation , and localization ( Stoecklin and Mühlemann , 2013; Tiedje et al . , 2012; von Roretz et al . , 2011 ) . The Zinc finger binding protein 36 ( ZFP36 ) family of proteins are prototypical ARE-binding factors with distinctive , activation-dependent expression in hematopoietic cell lineages ( Raghavan et al . , 2001; Carballo et al . , 1998 ) . The family includes three somatic paralogs: ZFP36 ( a . k . a Tristetraprolin , TTP ) , ZFP36L1 ( a . k . a . butyrate responsive factor 1 , BRF1 ) , and ZFP36L2 ( a . k . a BRF2 ) with highly homologous CCCH zinc-finger RNA binding domains ( Blackshear , 2002 ) . In many contexts , the archetypal paralog ZFP36 de-stabilizes target mRNAs by binding 3’-UTR AREs and recruiting deadenylation and degradation factors ( Brooks and Blackshear , 2013; Lykke-Andersen and Wagner , 2005 ) . More recently , evidence has begun to emerge for roles in translation ( Tao and Gao , 2015; Tiedje et al . , 2012 ) but in vivo function and context have not been established . While many aspects remain unsettled , ZFP36 is clearly critical for immune function , as its loss causes a systemic inflammatory disease in mice ( Taylor et al . , 1996 ) . A key feature of this syndrome is aberrant stabilization and over-expression of Tnf in myeloid cells , particularly macrophages ( Carballo et al . , 1998 ) , where UV cross-linking and immunoprecipitation ( CLIP ) analyses have further supported direct regulation ( Tiedje et al . , 2016; Sedlyarov et al . , 2016 ) . However , this role does not fully account for ZFP36 function in vivo , as underscored by reports that myeloid-specific deletions of Zfp36 do not recapitulate spontaneous autoimmunity ( Qiu et al . , 2012; Kratochvill et al . , 2011 ) . Increasing evidence points to important functions for ZFP36 proteins in adaptive immunity . Dual ablation of paralogs Zfp36l1 and Zfp36l2 in T cells arrests thymopoeisis at the double-negative stage , and causes lethal lymphoma linked to Notch1 dysregulation ( Hodson et al . , 2010 ) . This role in restraining aberrant proliferation was later extended to B-cell development and lymphoma ( Galloway et al . , 2016; Rounbehler et al . , 2012 ) , but the severe phenotype precluded analysis of ZFP36 family function in mature T cells . Consistent with such a function , in vitro studies suggest ZPF36 regulates the expression of T cell-derived cytokines , including IL-2 , IFN-γ , and IL-17 , that mediate lymphocyte homeostasis , microbial response , and inflammation ( Lee et al . , 2012; Ogilvie et al . , 2009; 2005 ) . The landscape of ZFP36 targets beyond these limited cases in T cells is unknown , but will be the key to understanding its emerging roles in inflammation , autoimmunity , and malignant cell growth ( Patial and Blackshear , 2016 ) . To determine ZFP36 functions in T cells , we employed high-throughput sequencing of UV-cross-linking and immunoprecipitation ( HITS-CLIP ) to generate a definitive set of ZFP36 RNA targets . HITS-CLIP utilizes in vivo UV-cross-linking to induce covalent bonds between RBPs and target RNAs , allowing stringent immunopurification and thus rigorous identification of direct binding events ( Licatalosi et al . , 2008; Ule et al . , 2003 ) . These new ZFP36 RNA binding maps pointed to roles in regulating T-cell activation kinetics and proliferation , a function confirmed in extensive functional assays , and in vivo studies demonstrating a critical role in anti-viral immunity . Our results illuminate novel functions for ZFP36 in adaptive immunity , laying groundwork for understanding and modulating its activity in disease . ZFP36 expression is induced upon T-cell activation ( Raghavan et al . , 2001 ) . We examined its precise kinetics following activation of primary mouse CD4 +T cells by Western analysis with custom ZFP36 antisera generated against a C-terminal peptide of mouse ZFP36 . Protein levels peaked ~4 hr post-activation and tapered gradually through 72 hr , and were re-induced by re-stimulation 3 days post-activation ( Figure 1A ) . ZFP36 expression depended on both TCR stimulation , provided by anti-CD3 , and co-stimulation , provided by co-cultured dendritic cells ( DCs ) ( Figure 1B ) . A similar pattern of transient ZFP36 induction occurred in activated CD8 +T cells ( Figure 1—figure supplement 1A ) . Western blot analysis showed multiple bands at ~40–50 kD , indicating several isoforms . Notably , isoforms running above the predicted molecular weight ( MW ) of ZFP36 ( 36 kD ) pre-dominated early after activation , and are consistent with previously reported hyperphosphorylation ( Qiu et al . , 2012 ) . In addition , partial conservation of the immunizing peptide in ZFP36L1 and ZFP36L2 raised the possibility of paralog cross-reactivity . Western analyses with recombinant constructs confirmed ZFP36 , ZFP36L1 , and ZFP36L2 are readily detected with our custom antisera ( henceforth , pan anti-ZFP36; Figure 1—figure supplement 1B ) . Commercial paralog-specific antibodies were identified , and Western analysis showed that both ZFP36 and ZFP36L1 were induced by T-cell activation ( Figure 1—figure supplement 1B–C ) . ZFP36L2 , expected to run at ~62 kD , was not detected under any conditions examined . Analysis of Zfp36 KO T cell lysates with pan ZFP36 antisera showed ~50% reduced signal compared to WT . We conclude that the residual signal is likely due to persistent expression of ZFP36L1 , which is highly homologous and of similar size to ZFP36 . Collectively , these results demonstrate activation-dependent expression of ZFP36 and ZFP36L1 in T cells , and suggest activated Zfp36 KO T cells have partial loss ( ~50% ) of pan ZFP36 expression ( Figure 1—figure supplement 1D ) . The characterization of Zfp36 as an immediate early response gene in various cell types established transcription as a mechanism of its activation-induced expression ( Lai et al . , 1995 ) . Interestingly , Zfp36 ( RPKM = 22 . 5 ) and Zfp36l1 ( RPKM = 8 . 2 ) mRNAs are robustly present in RNAseq data from naïve CD4 +T cells , despite an absence of detectable protein by western blotting . ZFP36L2 was not detected in any conditions examined , but its mRNA was also detected in naïve cells ( RPKM = 30 . 2 ) . These observations indicate that post-transcriptional mechanisms regulate expression of ZFP36 paralogs in T cells . The striking pattern of ZFP36/L1 expression in T cells led us to develop ZFP36 HITS-CLIP as a screen for its biological function ( Figure 1C–F; Figure 1—figure supplement 2 ) . Notably , ZFP36/L1 RNPs isolated by CLIP from activated CD4 +T cells sera exhibited high molecular weight ( MW ) complexes resistant to detergent , heat , and RNAse , consistent with a pattern previously observed in ZFP36 CLIP in macrophages ( Figure 1D , Figure 1—figure supplement 2A; [Sedlyarov et al . , 2016] ) . This RNP signal pattern was dependent on UV irradiation , and was observed with two different anti-pan-ZFP36 but neither pre-immune sera . Given prior evidence that ZFP36 regulates T-helper type-1 ( Th1 ) cytokines ( e . g . TNF-α and IFN-γ ) , we next generated a comprehensive map of ZFP36/L1 binding sites by HITS-CLIP using anti-pan-ZFP36 in WT CD4 +T cells , activated for 4 hr under Th1-polarizing conditions ( Ogilvie et al . , 2009; Carballo et al . , 1998 ) . 5132 robust binding sites were defined , requiring a peak height ( PH ) >5 , and support from at least 3 ( of five total ) biological replicates and two different pan-ZFP36 antisera ( Supplementary file 1A; [Shah et al . , 2017; Moore et al . , 2014] ) . Consistent with identification of bonafide ZFP36/L1 binding events , HITS-CLIP recovered the known AU-rich ZFP36 consensus motif at high significance , along with reported binding sites in Tnf , Ifng and other targets ( Figure 1E; Supplementary file 1A; [Brewer et al . , 2004] ) . Globally , ZFP36/L1 sites confirmed a preponderance of 3’-UTR binding ( >75% ) , and showed substantial binding in coding sequence ( CDS;~6 . 5% ) and introns ( 5 . 4% ) ( Figure 1F ) . Separate analysis of low and high MW RNP complexes showed similar transcript localization and motif enrichment , all consistent with ZFP36 binding ( Figure 1—figure supplement 2B ) . This analysis indicates the presence of large , stable ZFP36 complexes in vivo , consistent with stable multimers ( Cao et al . , 2003; 2004 ) . Subsequently , CLIP reads from different MW regions were pooled to maximize dataset depth . To examine possible paralog specificity , we also mapped ZFP36L1 sites by HITS-CLIP in Zfp36 KO CD4+T cells under identical conditions . As in western analysis ( Figure 1—figure supplement 1C–D ) , Zfp36 KO samples showed reduced but significant CLIP signal compared to WT ( Figure 1D , Figure 1—figure supplement 2A ) , representing ZFP36L1-RNA complexes . Sites in WT and Zfp36 KO cells showed very similar enriched motifs and transcript localizations , indicating that ZFP36 and ZFP36L1 have similar binding profiles in vivo ( Figure 1E–F , Figure 1—figure supplement 2C , Supplementary file 1B ) . Majorities of robust sites ( 53% ) and target mRNAs ( 66% ) identified in WT cells were found independently in Zfp36 KO cells , and site overlap was far greater ( >90% ) for peaks of increasing magnitude ( Figure 1G ) . A subset of potential ZFP36L1-specific sites was identified only in Zfp36 KO cells ( n = 675; Supplementary file 1C ) , although these showed similar features to ZFP36 sites overall ( Figure 1—figure supplement 2C; third panel ) . Thus , these analyses do not exclude paralog specificity at some sites , but indicate broadly overlapping in vivo RNA binding for ZFP36 and ZFP36L1 reflecting their high homology . Secondary enriched motifs revealed additional properties of ZFP36/L1 target sites . The second top motif resembled the known recognition sequence for polyadenylation factor CFI ( m ) 25 ( Venkataraman et al . , 2005 ) . Accordingly , ZFP36 binding in 3’UTRs was most concentrated in the vicinity of expected polyA sites , ~50 nucleotides before transcript ends ( Figure 1—figure supplement 2D ) . Analysis of CDS-specific binding revealed the AU-rich ZFP36 motif , along with strong enrichment of the 5’ splice site ( 5’-ss ) consensus ( Figure 1—figure supplement 2E ) . Cross-link-induced truncations ( CITS ) clarified that CDS peaks are centered within coding exons , but supporting CLIP reads often spanned the exon-intron boundary . Thus , at least a subset of CDS binding by ZFP36 occurs prior to pre-mRNA splicing in the nucleus . We next employed RNA profiling strategies to determine the functional effects of ZFP36/L1 binding . RNAseq analysis in WT and Zfp36 KO CD4 +T cells activated under conditions identical to our HITS-CLIP analyses uncovered two main effects . First , a small number of mRNAs that were silent in WT cells , including numerous immunoglobulin loci , were detected at low to moderate levels in KO cells ( Figure 1—figure supplement 3A ) . These mRNAs lacked evidence of ZFP36/L1 binding , both in HITS-CLIP data and direct motifs searches , suggesting they are not direct targets . Given established chromatin regulation of many of these loci ( i . e . Ig genes ) and the on-off nature of the changes , dysregulated transcriptional silencing is a potential explanation , but is likely to be a secondary effect of ZFP36 loss of function . The second main effect emerged from a global analysis , which showed ZFP36 binding in 3’UTR ( p=4 . 44×10−16; Kolmogorov-Smirnoff [KS] ) and CDS ( p=5 . 33×10−11 ) correlated to subtle but highly significant shifts toward greater mRNA abundance in Zfp36 KO cells , relative to mRNAs with no binding sites ( Figure 1H ) . This correlation was not observed for mRNAs with binding exclusively in introns , and we did not find evidence in these data that ZFP36/L1 binding correlated with altered usage of proximal splice or polyA sites ( not shown ) . The same pattern was also observed when considering a more stringently defined of sites set overlapping statistically robust CITS ( Figure 1—figure supplement 3B ) . The overall trend in transcriptome profiling is consistent with evidence that ZFP36 represses RNA abundance ( Lykke-Andersen and Wagner , 2005 ) . However , stratification of sites by the magnitude of ZFP36 CLIP binding allowed resolution of potentially complex effects . For 3’UTR binding , ZFP36 targets overall showed a significant shift in abundance , but mRNAs containing the top 20% most robust sites ( ranked by peak height [PH] , see Materials and methods ) showed no significant effect . Thus , a higher degree of binding correlated with less effect on RNA abundance in the absence of ZFP36 ( Figure 1—figure supplement 3C ) . This trend was not observed for CDS sites , where the top 20% showed a similar shift to sites overall ( Figure 1—figure supplement 3D ) . Thus , our analyses show a trend of negative regulation of RNA abundance in this context , but with blunted effects for highly robust binding sites in 3’UTR ( see Discussion ) . We next examined in more detail the effects of ZFP36 regulation for HITS-CLIP targets with highly robust 3’UTR binding in T cells . Activation marker CD69 , apoptosis regulator BCL2 , and effector cytokines TNF and IFNG showed significantly increased protein levels in Zfp36 KO versus WT T cells 4 hr after activation ( Figure 2A , Figure 2—figure supplement 1 ) . Of these , only Bcl2 showed increased mRNA abundance . Tnf , Ifng , and Cd69 were all among the top 20% of targets as defined by CLIP binding magnitude ( PH ) , thus supporting the trend in our global analyses that some highly robust binding targets show little regulation at the level of mRNA abundance in this context . The effects on protein level in the absence of changes in mRNA abundance suggested regulation of translation . We tested this principle by constructing GFP fluorescent reporters with an intact 3’UTRs ( WT-UTR ) , or variants with the CLIP-defined ZFP36 binding site deleted ( Δ-UTR; Figure 2B ) , for Cd69 , Tnf , and Ifng . In 293 cells , ZFP36 strongly repressed protein expression for all three WT-UTR reporters , while showing weaker ( Tnf and Ifng ) or no ( Cd69 ) repression of mRNA levels . Of note , the Δ-UTR constructs showed increased protein levels both in the presence and absence of ZFP36 . This indicates that additional , endogenous factors are regulating these AU-rich sites in 293 cells , though Western analyses confirmed that ZFP36 paralogs are undetectable ( Figure 1—figure supplement 1A ) . In addition , ZFP36 over-expression exerted ~2 fold repression of Δ-UTR constructs , which may indicate incomplete ablation of binding or secondary effects of ZFP36 over-expression . However , repression of WT-UTR constructs was consistently greater than for Δ-UTR variants , demonstrating specific ZFP36 repression of the defined binding sites . In summary , these heterologous assays independently confirmed ZFP36 regulation of CLIP-defined targets , and support specific effects of translation , in addition to RNA stability . To directly test a role for ZFP36 in translational regulation in T cells , we next performed ribosome profiling of WT and Zfp36 KO CD4 +T cells activated for 4 hr under Th1 conditions ( Figure 3A and Figure 3—figure supplement 1A; [Ingolia et al . , 2009] ) . We observed robust ribosome association of Zfp36 mRNA in WT cells that was lost completely downstream of the engineered gene disruption in Zfp36 KO cells ( Figure 3—figure supplement 1B ) , consistent with accurate identification of translating mRNAs . Globally , there was a subtle but significant shift toward greater ribosome association in Zfp36 KO cells for mRNAs bound by ZFP36 in 3’UTR ( p=1 . 04×10−11; KS ) or CDS ( p=1 . 58×10−11 ) , relative to mRNAs with no ZFP36 binding ( Figure 3B ) . These shifts mirror those for global RNA abundance , with two notable exceptions . First , mRNAs with ZFP36 binding in CDS showed a significantly larger shift in ribosome association than those with 3’UTR binding ( Figure 3B ) . Second , in contrast to effects on RNA abundance , the top 20% most robust ZFP36 binding sites in 3’UTR showed similar effects on ribosome association to sites overall ( Figure 3—figure supplement 1C ) . Thus , ZFP36 target mRNAs show increased ribosome association in Zfp36 KO cells . Levels of ribosome-associated mRNA are related to total abundance . To evaluate changes in translational efficiency ( ΔTE ) in Zfp36 KO versus WT T cells , ribosome profiling fold-changes were normalized to those from RNAseq . We then used Gene Set Enrichment Analysis ( GSEA ) to examine the distribution of ZFP36 targets among all detected mRNAs ranked by ΔTE ( Subramanian et al . , 2005 ) . Importantly , this analysis compares the observed ΔTE of CLIP-defined ZFP36 target mRNAs to mRNAs with no detected ZFP36 binding sites . ZFP36 3’UTR and , more significantly , CDS binding targets were strongly enriched for increased TE in Zfp36 KO cells ( Figure 3C ) . In addition , ZFP36 targets with highly robust 3’UTR binding showed more significant effects on TE than ZFP36 3’UTR targets overall . This enrichment was not observed for mRNAs with intronic binding sites , indicating specificity for 3’UTR and CDS binding . As a striking confirmation of these results , normalized ribosome coverage on robust 3’UTR targets Tnf and Ifng was significantly higher in Zfp36 KO cells than WT ( Figure 3D ) , despite no detectable difference in overall mRNA abundance ( Figure 2A ) . Crucially , ribosome coverage averaged across all mRNAs was not appreciably different between KO and WT cells , indicating specific effects on ZFP36 targets ( Figure 3E ) . Notably , the pattern of ribosome association along these and other transcripts is remarkably consistent between WT and Zfp36 KO cells , but with altered magnitude . Mechanistically , this observation indicates that ZFP36 prevents association of mRNAs with ribosomes , but does not impact elongation . These results indicate repression of mRNA target translation by ZFP36 during T- cell activation , likely at the level of initiation . ZFP36 target mRNAs pointed to multilayered control of T cell function , including its reported regulation of effector cytokines ( e . g . Il2 , Ifng , Tnf , Il4 , Il10; Figure 4—figure supplement 1 ) . Novel targets spanned direct components of the TCR complex ( e . g . CD3d , CD3e ) , co-stimulatory and co-inhibitory molecules ( e . g . Cd28 , Icos , Ctla4 ) , TCR-proximal signaling ( Fyn , Sos1 , Akt1 ) , and transcriptional response ( e . g . Fos , Nfatc1 , Nfkb1 ) . As an unbiased assessment , we examined the distribution of ZFP36 targets in high-resolution gene expression time courses of CD4 + T-cell activation ( Yosef et al . , 2013 ) . ZFP36 targets were enriched for mRNAs , like its own , that were rapidly induced after T-cell activation , and targets were depleted among mRNAs with stable expression or delayed induction ( Figure 4A ) . Gene Ontology ( GO ) enrichments spanned many basic metabolic and gene regulatory functions , in addition to signal transduction , cellular proliferation , and apoptosis ( Figure 4B; Supplementary file 2 ) . Functional clustering of ZFP36 targets in proliferation and apoptosis prompted us to investigate potential regulation of T cell proliferation . In thymidine incorporation assays , naïve Zfp36 KO CD4 +T cells showed greater proliferation than WT from 16 to 24 hr post-activation ( Figure 4C ) . Similar results were obtained with CD8 +T cells ( Figure 4—figure supplement 2A ) . This increase reflected decreased apoptosis ( Figure 4D ) and increased numbers of proliferating cells ( Figure 4E ) in KO versus WT cultures . We examined whether an action on IL-2 might account for enhanced proliferation , as increased IL-2 production in Zfp36 KO T cells has been reported ( Ogilvie et al . , 2005 ) , and our HITS-CLIP data confirmed direct interaction . Zfp36 KO T cells proliferated more than WT both in the presence of excess recombinant IL-2 or neutralizing anti-IL-2 antibody , as well as in different Th polarizing conditions , indicating the effect is not solely IL-2-dependent ( Figure 4—figure supplement 2B–C ) . In summary , T cells from Zfp36 KO mice show enhanced proliferation shortly after activation under all conditions examined . Anti-CD3 is not a physiologic stimulation , so we next examined proliferative responses to MHC-peptide-mediated TCR binding . First , we bred WT and Zfp36 KO mice with a transgenic , class-II restricted TCR specific for a β-galactosidase-derived antigen ( BG2 ) . BG2 transgenic Zfp36 KO cells showed greater proliferation than WT across a broad titration of cognate peptide , but not irrelevant peptide ( Figure 4F ) . Second , Zfp36 KO T cells also showed greater proliferation than WT in response to allogeneic DCs ( Figure 4G ) . Therefore , Zfp36 KO cells show an exaggerated proliferative response upon MHC-peptide stimulation over a range of signal strengths . Analysis of canonical T-cell activation markers revealed enhanced induction of CD69 and CD25 in Zfp36 KO versus WT cells over the first 24 hr post-activation ( Figure 4H ) . At 40 hr post-activation , a greater proportion of Zfp36 KO versus WT had transitioned from a naïve to effector surface phenotype ( Figure 4I ) . Notably , thymidine incorporation data showed enhanced proliferation of Zfp36 KO cells early after activation , but similar rates in Zfp36 KO and WT cells after 24 hr ( Figure 2C ) . Collectively , these results show accelerated activation kinetics in the absence of ZFP36 . The accelerated activation of Zfp36 KO T cells could in principle reflect the activity of other cell subsets or inflammatory signals in Zfp36 KO mice . To test for a T cell-intrinsic function , we generated mixed bone marrow ( BM ) chimeras , allowing isolation of WT and KO T cells that develop in the same in vivo milieu ( Figure 5A ) . Naïve Zfp36 KO T cells sorted from chimeras showed greater proliferation than WT 24 hr post-activation , indicating cell-intrinsic effects ( Figure 5B ) . To assess the potential impact of secreted factors , chimera-derived WT and Zfp36 KO cells were re-mixed 1:1 ex vivo . Here , differences between Zfp36 KO and WT cells were still significant , but blunted compared to separate cultures . This result indicates cross-regulatory effects between WT and KO cells through secreted or surface factors , but these do not fully account for the observed differences . Interestingly , the reduced proliferation of Zfp36 KO cells in mixed ( Figure 5B , right panel ) versus separate ( left panel ) cultures indicate that WT cells can exert a restraining effect on KO cells . Thus , accelerated activation in Zfp36 KO cells may in part reflect compromised autoregulatory and/or suppressive functions . Three days after activation , mixed cultures remained skewed in favor of Zfp36 KO cells , again confirming accelerated expansion ( Figure 5C ) . These results show that ZFP36 regulation of T-cell activation is cell-intrinsic , and that ZFP36 normally functions to restrain T -cell activation . The efficient in vitro responses of Zfp36 KO T cells suggest they are functional , but respond with altered kinetics . To examine the downstream consequences of this differential regulation , HITS-CLIP and RNAseq analyses were done in Th1-skewed CD4 +T cells 3 days after activation . ZFP36 binding site features in cells activated for 3 days were similar to ones identified at 4 hr ( Figure 6—figure supplement 1A–B; supplementary file 3 ) , but results from transcriptome profiling were strikingly different at these two time points ( Figure 6A ) . First , in contrast to subtle effects observed at the 4 hr time point , many transcripts showed highly divergent expression in Zfp36 KO versus WT T cells 3 days after activation ( Figure 6A ) . However , these differences were not correlated to ZFP36 HITS-CLIP binding at 72 hr ( Figure 6B ) . Thus , the absence of ZFP36 in the early phases of T -cell activation can lead to significant secondary effects downstream . GSEA with RNAseq data from Th1 Zfp36 KO cells 3 days post-activation showed reduced activity of transcription factors driving proliferation ( e . g . Myc and E2F; Figure 6—figure supplement 2A ) and strong overlap with previously described signatures of T cell exhaustion ( Figure 6C; [Crawford et al . , 2014] ) . Consistent with a late activated and exhausted phenotype in Zfp36 KO cells , the enhanced production of IFN-γ at early time points ( Figures 2A and 3D ) gave way to comparable production by day three and reduced production by day 5 ( Figure 6D ) . Moreover , TNF-α production was not significantly different 3 or 5 days post-activation ( Figure 6D ) , despite large differences early on ( Figures 2A and 3D ) . In summary , relieved translational control drives elevated cytokine production in Zfp36 KO cells early post-activation , but enhanced production dissipates downstream due to more rapid expansion and exhaustion . The net result , reflecting both dysregulated cytokine production and more rapid culture expansion ( Figures 4 and 5 ) , is a higher accumulation of IFN-γ and TNF-αin Zfp36 KO culture supernatants 72 hr post-activation ( Figure 6E ) . We examined co-inhibitory and co-stimulatory checkpoint proteins that are linked to T cell exhaustion , and found elevated expression of PD-1 and ICOS at late time points in Zfp36 KO cells , and more rapid peaking of LAG-3 ( Figure 6F ) . Interestingly , these effects were observed in Th1 but not Th0 conditions , suggesting a dependence on Th1 cytokines . To test this dependence , and to examine whether elevated receptor expression was T cell-intrinsic , we analyzed cells derived from mixed BM chimeras . This analysis confirmed differential , T cell-intrinsic expression of these receptors ( Figure 6G ) . However , re-mixing WT and KO cells ex vivo neutralized these differences , indicating they are driven by secreted factors . We tested whether recombinant IFN-γ , supplemented at levels measured in KO cultures , could cause elevated receptor expression in WT Th1 cells , and found it promoted ICOS but not PD-1 upregulation ( Figure 6—figure supplement 2B ) . These results indicate that Th1 cytokines , including but not only IFN-γ , can drive an exhaustion-like phenotype . The absence of ZFP36 promoted this phenotype in vitro , due to more rapid activation and expansion coupled with greater accumulation of Th1 cytokines . The accelerated response of Zfp36 KO T cells , and the potential for accelerated exhaustion , led us to examine the effects of ZFP36 regulation in vivo . We first determined that naïve Zfp36 KO mice had normal T cell levels in peripheral blood and no defects in thymocyte development ( Figure 7—figure supplement 1A–B ) . Total splenocytes , including T cells , were slightly reduced in Zfp36 KO versus WT mice ( Figure 7—figure supplement 1C ) , but proportions of total CD4 + and CD8+T cells were normal ( Figure 7—figure supplement 1D ) . Proportions of naïve CD4 +T cells were also normal in KO mice , and naïve CD8 +T cells only slightly decreased . ( Figure 7—figure supplement 1E ) . Levels of CD25-hi CD4 +cells were not significantly different in spleens of WT and KO mice , consistent with similar levels of natural Tregs ( Figure 7—figure supplement 1F ) . FoxP3 expression was not examined directly ex vivo , but in vitro induction of Tregs from naïve cells CD4 +T cells , enumerated in FoxP3-GFP mice , was not different between WT and KO ( Figure 7—figure supplement 1G; [Haribhai et al . , 2007] ) . To examine the effector T cell populations present in spleen , splenocytes were stimulated directly ex vivo with phorbol myristate acetate ( PMA ) and ionomycin . More CD4 + and CD8+T cells produced IFN-γin KO versus WT splenocytes , but levels of IL-4 and IL17A production were comparable ( Figure 7—figure supplement 1H ) . Therefore , greater numbers of Th1 cells are present at steady state in Zfp36 KO mice . To examine whether loss of ZFP36 causes an intrinsic disposition to the Th1 fate , skewing of sorted , naïve CD4 +T cells was examined . These analyses showed indistinguishable differentiation of Th1 and Th17 subsets ex vivo in WT and Zfp36 KO cells ( Figure 7—figure supplement 1I ) . The greater accumulation of Th1 cells in vivo may therefore reflect a response to factors not replicated in vitro , or the effects of additional cell types . The lymphocytic choriomeningitis virus ( LCMV ) Armstrong strain causes an acute infection leading to massive T cell expansion and viral clearance in 8–10 days ( Dutko and Oldstone , 1983 ) . Using MHC-tetramers , we observed accelerated expansion and recession of virus-specific CD4+ ( Figure 7A ) and CD8+ ( Figure 7B ) T cells in Zfp36 KO versus WT mice in peripheral blood . This result was confirmed in independent experiments focused on early time points post-infection ( p . i . ) , where virus-specific T cells in Zfp36 KO mice showed earlier expansion and more rapid upregulation of CD69 ( Figure 7C–D ) . Enumeration of virus-specific T cells in spleen mirrored dynamics in blood; levels were greater in Zfp36 KO animals 6 days p . i . , but marginally lower by day 10 , consistent with more rapid expansion and resolution ( Figure 7E–F ) . Levels of memory T cells day 40 p . i . were similar in Zfp36 KO and WT mice . In summary , antigen-specific T cell response is clinically functional but accelerated in Zfp36 KO mice during viral infection . Stimulation with LCMV peptides ex vivo revealed higher rates of IFN-γ and TNF-αproduction in Zfp36 KO versus WT CD4+ ( Figure 7G ) and CD8 +T cells 6 days p . i . ( Figure 7—figure supplement 2A ) . Numbers of cytokine-producing cells were proportional to LCMV-specific tetramer +cells ( Figure 7E–F ) . However , levels of IFN-γand TNF-α protein were significantly greater in CD4 +Zfp36 KO cytokine-producing cells versus WT ( Figure 7H ) , and TNF-α levels were also higher for CD8 +cells ( Figure 7—figure supplement 2B ) . In addition , ‘bifunctional’ IFN-γ+TNF-α+T cells were more frequent in Zfp36 KO mice , even when normalized to frequencies of tetramer +cells ( Figure 7I and Figure 7—figure supplement 3C ) . Strikingly , LCMV genomic RNA in spleen was ~10 fold lower day 6 p . i . in Zfp36 KO versus WT animals , consistent with more rapid clearance of LCMV infection ( Figure 7J ) . Viral load correlated inversely with levels of tetramer +CD4+ and CD8+T cells in both Zfp36 KO and WT mice , consistent with the established role of T cell response in LCMV clearance ( Figure 7—figure supplement 2D ) . To examine whether the accelerated LCMV-specific T cell response in Zfp36 KO mice may be T cell-intrinsic , infections were repeated in mixed BM chimeras . Irradiated recipient mice were re-constituted with a 1:1 mix of congenically marked WT or Zfp36 KO BM cells ( Figure 7—figure supplement 3A ) . Ten weeks after re-constitution , pre-infection baseline measurements showed a significantly greater expansion of Zfp36 KO T cells versus WT in blood ( Figure 7—figure supplement 3B ) . However , the kinetics of WT and KO T cell response in these animals upon LCMV infection were indistinguishable ( Figure 7—figure supplement 3C–D ) . Therefore , in a mixed environment in vivo , Zfp36 KO and WT T cells show similar kinetics upon viral challenge . Notably , maximum T cell expansion was observed in mixed chimeras 7–8 days p . i . , which was intermediate to the maxima observed in Zfp36 KO ( 6 days ) and WT ( 10 days ) mice . Collectively , these data demonstrate a remarkable enhancement of anti-viral immunity in the setting of reduced ZFP36 family activity in vivo . This enhancement was marked by an accelerated T cell response and enhanced production of effector cytokines and , based on mixed chimera experiments , may involve immune cell types in addition to T cells . Immune response requires rapid , adaptable gene regulation—features uniquely suited to post-transcriptional control . Our studies illuminate a role for ZFP36 RNA binding proteins in controlling the pace of T cell response , a crucial dimension of adaptive immunity , and tie this to the effectiveness of anti-viral responses in vivo . ZFP36 and ZFP36L1 expression are rapidly induced upon T-cell activation , and gradually recede thereafter . While transcriptional induction has been previously established , our detection of Zfp36 and Zfp36l1 mRNAs in naïve T cells , in the absence of detectable protein , indicates post-transcriptional regulation of their mRNAs in T cells . Moreover , paralog auto- and cross-regulation are likely , as Zfp36 and Zfp36l1 mRNA 3’UTRs possess robust HITS-CLIP binding sites . We did not detect ZFP36L2 in any of our analyses , indicating it is absent or negligibly expressed under conditions examined here . However , the presence of its mRNA further suggests post-transcriptional control of ZFP36 paralog expression , and is consistent with functions in other contexts or stages of T cell function ( [Vogel et al . , 2016; Hodson et al . , 2010] ) . Definitive determination of ZFP36 targets in T cells by HITS-CLIP , coupled with transcriptome and ribosome profiling studies , revealed that ZFP36 attenuates T-cell activation by suppressing the abundance and translation of its mRNA targets . The correlation of ZFP36 binding with reduced mRNA abundance is consistent with reports that ZFP36 can destabilize target mRNAs by recruiting degradation factors ( Fabian et al . , 2013; Lykke-Andersen and Wagner , 2005 ) . However , our ability to stratify the relative magnitude of ZFP36 binding using CLIP resolved a more complex trend , with highly robust 3’UTR binding sites ( top 20% ) showing no detectable correlation with RNA abundance ( Figure 1—figure supplement 3C ) . This non-uniform trend was observed for 3’UTR but not CDS targets , and affected mRNA abundance but not ribosome association . Importantly , effects on protein levels in the absence of changes in mRNA abundance were confirmed independently for robust 3’UTR binding targets Tnf , Ifng , and Cd69 . It is possible that different degrees of ZFP36 association in vivo elicit distinct functional outcomes , through differential RNP localization or downstream effector recruitment . Notably , ZPF36 CLIP showed a broad MW range of ZFP36-RNA complexes with distinct biochemical properties including stability to heat , detergent , and high salt . While the current studies did not uncover distinct mRNA targets across this range , more detailed biochemical studies will be necessary to clarify potentially distinct ZFP36 complexes in vivo , and their potentially distinct roles in different cell types of the immune system . We further present evidence that ZFP36 suppresses translation of its target mRNAs in T cells . Endogenous targets and exogenous reporters showed greater ZFP36-dependent suppression of protein versus RNA levels ( Figure 2 ) , and ribosome profiling in primary T cells confirmed direct effects on translation ( Tao and Gao , 2015; Qi et al . , 2012 ) . The strongest effects were linked to a novel class of AREs in coding sequence , uncovered with ZFP36 binding maps . CDS sites correlated with repressed RNA abundance and translation , but a greater level of repression was evident in ribosome association ( Figure 3B–D ) . Intriguingly , some ZFP36 CLIP reads in CDS sites spanned exon-intron boundaries , indicating these associations can form prior to pre-mRNA splicing in the nucleus ( Figure 1—figure supplement 2E ) . The identification of AREs in the CDS and the possibility of resulting translation control pre-programmed in the nucleus point to novel , unexplored regulatory strategies . Notably , these results differ significantly from iCLIP analyses in macrophages using exogenous GFP-tagged ZFP36 , where only 3’-UTR sites correlated with target repression , and only for a ZFP36 construct with mutated MK2 phosphorylation sites ( Tiedje et al . , 2016 ) . In those studies , the WT ZFP36 construct showed negligible repressive effects , contrasting with our data in 293 and T cells , and data from other contexts ( Tao and Gao , 2015; Ogilvie et al . , 2009 ) . However , iCLIP data for the transduced WT ZFP36 showed low 3’UTR binding ( 23% ) , high intergenic binding ( 38% ) , and a preference for GU-rich motifs , diverging sharply from our analysis of endogenous ZFP36 and prior in vivo and in vitro characterizations ( Brewer et al . , 2004; Worthington et al . , 2002 ) . These differences may reflect distinct ZFP36 phosphorylation , and hence regulatory outcomes , or as yet undefined variables related to the different cellular context . A direct comparison is further confounded by the use of exogenous , transduced constructs in macrophage experiments , in contrast to our analysis of endogenous proteins in T cells . Importantly , the methods and reagents developed here allow for systematic analysis of endogenous ZFP36 proteins in future global and cell-type-specific investigations addressing these issues . The similarity of RNA-binding maps covering both ZFP36 and ZFP36L1 ( WT cells ) or ZFP36L1 alone ( Zfp36 KO cells ) supports redundancy of ZFP36 paralogs , a likely source of robustness in immune regulation . Zfp36 KO cells are thus likely a partial loss-of-function due to robust ZFP36L1 expression , a notion consistent with the relatively subtle regulatory effects on RNA abundance and translation . Phenotypically , loss of Zfp36 led to accelerated activation of mature T cells , but not uncontrolled proliferation or impaired development , which may again may reflect a partial loss of pan ZFP36 activity . Indeed , a prior study reported no effects when Zfp36l1 was deleted in T cells , but drastic dysregulation of thymocyte proliferation upon loss of both Zfp36l1 and Zfp36l2 ( Hodson et al . , 2010 ) . These studies indicate total paralog dosage is critical , but also suggest the importance of the specific balance of ZFP36 paralogs in a defined context . Improved cell profiling methods , such as cell-type-specific tagging of RBPs , may illuminate these complexities in future studies . The mRNA targets defined by ZFP36 HITS-CLIP span from surface molecules engaged in the earliest steps in T-cell activation to downstream signaling and transcriptional effectors . Targets were strongly enriched for regulation of proliferation and apoptosis , extending prior reports that the ZPF36 family regulates proliferation in early T- and B-cell development and cancer . In each case , the reported mechanism was distinct , spanning regulation of Notch , G1/S phase transition , and Myc , respectively ( Galloway et al . , 2016; Hodson et al . , 2010; Rounbehler et al . , 2012 ) . ZFP36 HITS-CLIP identified all of these target pathways in T cells , consistent with a central function in controlling cell proliferation . However , the phenotype of Zfp36 KO T cells is novel and distinct , leading not to uncontrolled proliferation , but to accelerated effector response and resolution . The global effects of ZFP36 repression on RNA abundance and translation were widespread but subtle , and spanned many layers of T cell function . Functional validation of novel ZFP36 targets , including T-cell activation marker Cd69 and apoptosis regulator Bcl2 , suggest factors that likely contribute to this regulation . Analyses with cells sorted from mixed BM chimeras showed that the enhanced activation in Zfp36 KO cells is T cell-intrinsic ( Figure 5 ) . Experiments in which these sorted cells were re-mixed ex vivo exhibited blunted differences , as compared to cells cultured separately ( Figure 5B ) . These results indicate a role for both intracellular and secreted factors in exerting ZFP36 regulatory effects . More broadly , the combined picture of our genomic and functional data are one of many functionally diverse targets contributing to a multifaceted , finely tuned response , which is a hallmark of post-transcriptional regulatory control . Analyses at later time points after activation in Th1 conditions revealed widespread dysregulation of Zfp36 KO cells , with a gene expression signature and surface phenotype resembling T cell exhaustion ( Figure 6; [Crawford et al . , 2014] ) . Notably , and in contrast to analyses early post-activation , changes in RNA abundance 3 days post-activation were not correlated to direct ZFP36 binding . Therefore , secondary effects of ZFP36 loss pre-dominate as T cell expansion and differentiation progress in these settings . Notably , altered expression of co-inhibitory and –stimulatory receptors in Zfp36 KO cells was specific to Th1 cells ( Figure 6F ) , and mixing experiments confirmed dependence on secreted factors ( Figure 6G; Figure 6—figure supplement 2B ) . Thus , a dysregulated Th1 secretion profile including but not limited to elevated IFN-γ can push T cells to a state resembling exhaustion in vitro . More broadly , these results highlight the crucial point that subtle , but direct , regulation of a broad range of targets early after T-cell activation by ZFP36 can have striking downstream consequences in a rapidly expanding cell population . In vivo studies of acute viral infection showed accelerated expansion and recession of virus-specific CD4 + and CD8+T cells in Zfp36 KO animals . Zfp36 KO T cells had higher levels TNF-α and IFN-γ protein expression than WT after peptide stimulation , and more ‘bi-functional’ TNF-α/IFN-γ co-producing cells , thought to be important for anti-viral immunity ( Crawford et al . , 2014 ) . More rapid T cell expansion coincided with lower accumulated viral titers ( or more rapid clearance ) in Zfp36 KO animals , indicating ZFP36 regulates anti-viral immunity . In mixed BM chimeras , the response kinetics of WT and Zfp36 KO T cells to LCMV infection were indistinguishable . This result raises the possibility that other cell types and pathways , such as antigen presenting cells , antibody response , or innate immune regulators , could contribute to the accelerated anti-viral T cell response in Zfp36 KO mice . It is noteworthy that the point of maximum T cell expansion in these chimeras was intermediate to that of Zfp36 KO and WT mice . This result resembles our previous observation that WT and Zfp36 KO cells sorted from BM chimeras show blunted differences in activation kinetics when re-mixed ex vivo , as compared to cells cultured separately ( Figure 5B ) . Collectively , these data suggest that Zfp36 KO and WT cells exhibit cross-regulatory effects in a mixed environment that may obscure or complement cell-intrinsic differences . Given that many prominent ZFP36 targets encode secreted factors , such cross-regulatory effects are a virtual certainty . Collectively , these in vivo studies demonstrate an accelerated T cell response to viral infection in the absence of ZFP36 , which may reflect the heightened activity of multiple cell types in addition to T cells . Regardless of the initiating mechanism , the observation of enhanced LCMV clearance is likely T cell-dependent , given the central role of T cells in LCMV immunity ( Matloubian et al . , 1994 ) . Accordingly , we observed a quantitative , negative correlation between viral load and antigen-specific T cell levels in vivo ( Figure 7—figure supplement 2D ) . Importantly , the enhanced anti-viral response in the chronic absence of ZFP36 in KO mice is accompanied by spontaneous inflammation and autoimmunity that worsen with age . Our results starkly illustrate the delicate balance of protective immunity against destructive inflammation , and reveal post-transcriptional regulation by RBPs as central to this trade-off . Starting from transcriptome-wide RNA binding maps in T cells , we uncovered a crucial function for ZFP36 proteins in regulating adaptive immunity . These data suggest carefully titrated inhibition of ZFP36 might serve as a pharmacologic strategy in contexts where accelerated T cell response to challenge is desirable . Our in vivo LCMV studies demonstrate acute viral infection as one context , but application to other intracellular pathogens warrants investigation . Moreover , the ability to activate T cells to target tumor antigens and the clinical utility of checkpoint inhibitors raise the possibility of exploring ZFP36 inhibition to enhance tumor immunity . ZPF36 HITS-CLIP identified many targets central to these strategies , including Cd274 ( PD-L1 ) , Pdcd1l2 ( PD-L2 ) , Icos , Cd27 , Cd28 , Ctla-4 , Btla , and Lag3 , suggesting a means for concerted regulation . The autoimmune phenotype of the Zfp36 KO mouse highlights an important caveat , common to parallel issues seen with clinical use of checkpoint inhibitors . The tools for cell-type-specific analysis of ZFP36 , its targets , and its inhibition now exist to investigate and refine this balance . All data reported are for independent biological replicates , unless specifically noted in figure legends . In most cases , one mouse was one biological replicate . For CLIP studies 2–4 littermate mice of the same sex and genotype were pooled for each biological replicate . When performed , technical replicates deriving from the same biological replicate were averaged . For ex vivo studies , including genomic analyses , a sample size of 3–5 biological replicates was judged sufficient based on a power analysis using values from pilot studies , requiring p<0 . 05 with 95% power . To account for greater variability , sample sizes were doubled for in vivo studies . Mouse studies were not blinded . Details for statistical analysis appear in figure legends . Comparisons between experimental groups were done with two-sided student’s t-tests , with p<0 . 05 considered significant . Statistics for bioinformatic analyses are detailed below . High throughput sequencing data from this study are available at the NCBI Gene Expression Omnibus ( GEO ) website under accession GSE96076 .

The immune system must quickly respond to anything that may cause disease – from cancerous cells to viruses . For instance , a type of white blood cell called a T cell patrols the body , looking for potential threats . If a T cell identifies such a threat , it “activates” and undergoes various changes so that it can help to eliminate the problem . One way that T cells change is by switching on different genes to make specific proteins . The information in the genes is first used as a template to produce a molecule called a messenger RNA ( mRNA ) , which is then translated to build proteins . So-called RNA-binding proteins help control events before , during and after the translation stage in the process . Previous studies have shown that one particular RNA-binding protein , called ZFP36 , controls the translation of proteins that are important for how the immune system recognizes the body’s own tissue and deals with cancer cells . However , it was less clear if it also helped T cells to activate and defeat viruses . Now , using cutting-edge technology , Moore et al . have identified thousands of new mRNAs controlled by ZFP36 in mice , many of which did indeed make proteins that help T cells activate and spread throughout the body . Further experiments showed that mice that lack ZFP36 in the T cells were much quicker at responding to viruses than other mice . This suggests that ZFP36 actually restrains T cells and slows down the body’s immune system . Knowing more about how T cells work could lead to new treatments for diseases; it may , for example , allow scientists to engineer T cells to better attack cancer cells , However , other studies have shown that mice without ZFP36 often go on to develop autoimmune diseases , which result from the immune system attacking healthy cells by mistake . As such , it seems that there is a fine line between improving the body’s immune system and increasing the risk of autoimmune diseases , and that RNA-binding proteins play an important role in managing this delicate balance .

2018

ZFP36 RNA-binding proteins restrain T cell activation and anti-viral immunity

Pathogens rely on proteins embedded on their surface to perform tasks essential for host infection . These obligatory structures exposed to the host immune system provide important targets for rational vaccine design . Here , we use a systematically designed series of multi-domain constructs in combination with small angle X-ray scattering ( SAXS ) to determine the structure of the main immunoreactive region from a major antigen from Leptospira interrogans , LigB . An anti-LigB monoclonal antibody library exhibits cell binding and bactericidal activity with extensive domain coverage complementing the elongated architecture observed in the SAXS structure . Combining antigenic motifs in a single-domain chimeric immunoglobulin-like fold generated a vaccine that greatly enhances leptospiral protection over vaccination with single parent domains . Our study demonstrates how understanding an antigen’s structure and antibody accessible surfaces can guide the design and engineering of improved recombinant antigen-based vaccines . The molecular details of how surface antigens of a pathogen are exposed to the host’s defenses are highly relevant to rational vaccine development ( Rappuoli et al . , 2016; Dormitzer et al . , 2012; Dormitzer et al . , 2008 ) . The structural context of immunologically accessible epitopes allows for the redesign of recombinant vaccine scaffolds to display the most antigenic surfaces ( Correia et al . , 2014; Malito et al . , 2014 ) . The exploration of recombinant strategies can be particularly fruitful when classical vaccine strategies provide weak protection . Currently available inactivated vaccines against leptospirosis , the most common bacterial zoonosis ( Picardeau , 2017 ) , are inadequate because of the severe side-effects and lack of cross-protection among pathogenic Leptospira species ( Grassmann et al . , 2017; Adler , 2015 ) . Because the estimated worldwide burden of leptospirosis is over 1 million severe cases and ~60 , 000 deaths per year , advances in recombinant leptospiral vaccines are desperately needed and are likely to benefit from a structure-based antigen design strategy ( Picardeau , 2017; Costa et al . , 2015 ) . While several leptospiral antigens have been explored for use in vaccines , the most promising candidates have been derived from the Leptospira immunoglobulin-like ( Lig ) protein family ( Ko et al . , 2009; Grassmann et al . , 2017; Cao et al . , 2011; Yan et al . , 2009; Chang et al . , 2007 ) . Lig proteins are present in only pathogenic species with LigB ( but not LigA or LigC ) being found in all pathogenic Leptospira genomes ( McBride et al . , 2009; Matsunaga et al . , 2003 ) , LigB’s expression during host invasion further suggests an important role in virulence ( Lessa-Aquino et al . , 2017; Choy et al . , 2007 ) . To promote infection , LigB can bind multiple blood factors and extracellular matrix molecules ( ECMs ) to facilitate immune system evasion and tissue colonization ( Choy , 2012; Vieira et al . , 2014; Figueira et al . , 2011 ) . The persistence and exposure of LigB on the leptospiral outer membrane during infection leaves an exploitable vulnerability . Recently , a hamster immunization study has suggested that LigB-derived vaccines have the potential to confer sterile immunity against leptospiral challenge ( Conrad et al . , 2017 ) . The new study ( Conrad et al . , 2017 ) is inconsistent with earlier LigB vaccine studies which only confer partial protection ( Yan et al . , 2009; Silva et al . , 2007 ) and raises the possibility of further improvements in LigB-derived vaccine efficiency through rational design . Our understanding of the LigB structure is limited to the NMR structure of the individual Lig protein Ig-like domain ( Ptak et al . , 2014 ) . LigB is attached to the leptospiral outer surface by a short N-terminal anchor , which is followed by a stretch of twelve consecutive Ig-like domains , and flanked at its C-terminus by an additional non-Ig-like domain ( Figure 1—figure supplement 1A ) . Most host interactions with Lig protein have been identified within the Ig-like domain regions and previously reported LigB-based vaccines have targeted the Ig-like domain region ( Yan et al . , 2009; Conrad et al . , 2017; Breda et al . , 2015; Lin et al . , 2009a ) . A more comprehensive understanding of the domain arrangement would provide a picture of the most accessible antigenic regions and a guide for structure-based vaccine design . In this study , small angle X-ray scattering ( SAXS ) ( Skou et al . , 2014 ) was used to obtain a low-resolution solution structure of the Ig-like domain region’s architecture . The full LigB Ig-like domain region’s extended arrangement with notable bends encouraged the exploration of the highly-exposed surface for immunoreactivity with a library of anti-LigB monoclonal antibodies ( mAbs ) . The capability of these mAbs to bind antigen and to adhere to the surface of pathogenic Leptospira was then correlated with their ability to kill these bacteria in the presence of serum complements . Finally , the identified mAb-reactive domains and previously obtained LigB12 ( LigB 12th Ig-like domain ) NMR structure informed the generation of chimeras on single Ig-like domain scaffolds capable of eliciting an immune response with either side of the β-sandwich domain . To illustrate the potential for rationally engineered antigens , a vaccine containing the chimera LigB10-B7-B7 , which displays identified mAb-interacting surfaces from LigB7 and LigB10 , offered greatly improved protection over LigB7 and LigB10 against leptospiral lethal challenge in hamsters . These findings provide a blueprint for combining immunoreactivity mapping from an mAb library and high-resolution structural information from NMR to engineer epitopes and improve the efficacy of LigB vaccines as well as recombinant vaccines from other pathogens . To determine the architecture of the twelve Ig-like domain stretch in LigB , small angle X-ray scattering ( SAXS ) was used to generate low-resolution solution structures along a sliding multi-domain window . Because the 5-domain length contains four domain-domain linkers , the 5-domain structure proved to be optimal to define the relative angle of two neighboring domain-domain joints with one additional joint on each end ( See Materials and methods for rationale; Figure 1—figure supplement 1 ) . All eight possible 5-domain protein constructs ( LigB1-5 to LigB8-12 ) were analyzed with SAXS ( Figure 1; Figure 1—figure supplement 2 ) . Guinier fits and Porod analysis suggested minimal aggregation for all but LigB5-9 ( Figure 1—figure supplement 2B and Figure 1—source data 1 ) . The simulated fits to the experimental SAXS curves for the eight 5-domain constructs ( Figure 1A; Figure 1—figure supplement 2A ) were used to generate the atomic distance distribution for the molecules ( Figure 1B ) . The longest atomic pair distance deduced from the pair-distance distribution function ( P ( r ) ) is indicative of the length of the 5-domain proteins . The expected length of a fully extended arrangement of five folded LigB Ig-like domains is only slightly longer than the pair distance for most of the eight 5-domain proteins and their corresponding envelopes ( Figure 1B; Figure 1—figure supplement 2C ) . LigB5-9 has an atomic pair distance that exceeds what is possible for a folded 5-domain monomer ( in agreement with aggregation indicated by the difference between experimental and predicted molecular weights , Figure 1—source data 1 ) and was excluded from the final model . Ambiguity assessment using the AMBIMETER program ( Petoukhov and Svergun , 2015 ) shows that the first six structures ( LigB1-5 to LigB6-10 ) have a high degree of uniqueness with ambiguity scores of 0 . 3 or lower . The last two structures ( LigB7-11 and LigB8-12 ) have scores of 1 . 7 and 1 . 4 respectively , which indicates the potential for shape ambiguity . Each of the eight 5-domain SAXS envelopes exhibited distinct structures with a variety of domain-domain angles ( Figure 1—figure supplement 2D ) . A significant degree of bending is present between the first three domains . The stretch of domains between LigB3-6 is particularly straight . Slight bending in the angle between domains in the final five domain stretch produces a gentle spiral shape . By aligning the bends in the four shared domains of neighboring 5-domain regions , a best fit structure was generated for the full twelve Ig-like domains of LigB ( Figure 1C ) . Overall the LigB1-12 structure is only 16% shorter than a fully extended model structure of twelve folded domains . The SAXS-derived structure of LigB1-12 demonstrates that most of the individual Ig-like domains exhibit a high degree of exposed surface area and suggests a high degree of accessibility to host interactions . The extensive exposed surface area is a consequence of the unexpected rigid , rod-like structure , with several well-defined kinks . To explore the exposure of the LigB Ig-like domain region to a host immune response , two purified LigB truncations , LigB1-7 and LigB7-12 ( Figure 2A ) , were used to generate two sets of hybridoma cell lines for mAb production: library C ( 24 mAbs ) and library V ( 36 mAbs ) , respectively . Using ELISA , hybridoma supernatants containing anti-LigB mAbs were qualitatively screened for the ability to bind to their respective LigB truncations ( Figure 2B ) . Based on the distribution of antigen binding efficiencies , a threshold level was set to OD630 = 1 . 0 ( Figure 2—figure supplement 1 ) . Only mAbs with binding above the threshold level were purified for further characterization of binding properties and bactericidal activity . Screening of library V required an additional twelve mAbs in order to equal the nine threshold-level mAbs identified in the library C screen . For each library , nine mAbs were identified to have moderate to high binding efficiency to LigB antigens during the initial mAb screen . The dissociation constants ( KD ) for each of these mAbs was obtained from dose-dependent ELISA curves ( Figure 3; Figure 3—figure supplement 1; Table 1 ) . Based on KD values , library C mAbs were generally able to bind tighter to the LigB antigen than library V mAbs . Several library C mAbs ( C5 , C6 , C7 ) exhibit sub-micromolar KD values while only one library V mAb ( V10 ) was able to bind in the sub-micromolar range . Domain-level specificity of individual LigB mAbs was investigated using a comprehensive set of single Ig-like domain LigB truncates and tandem ( double ) Ig-like domain LigB truncates ( except LigB5-6 ) based on the NMR structure ( Ptak et al . , 2014 ) of the single domain , LigB12 ( Figure 4—figure supplement 1 ) . The LigB-derived antigens were immobilized on microtiter plates and tested for mAb binding specificity using an ELISA binding assay . The ELISA results for individual mAbs were generally able to localize binding to a specific double and/or single Ig-like domain . Figure 4A illustrates the ELISA assays used to identify the binding of mAbs C5 and V10 for specific two-domain regions and for specific single Ig-like domains . A comprehensive survey of domain-level epitope mapping for different mAbs are summarized in Table 1 and Figure 4—figure supplement 1 . For the LigB1-7 antigen-derived library , immunoreactivity of mAbs was weighted towards LigB1-2 and LigB4-5 regions , while for the LigB7-12 antigen-derived library , mAbs were preferentially generated against LigB7-8 and LigB10-11 regions . Only 3 of 10 double domains , LigB3-4 , LigB6-7 , and LigB8-9 , and 2 of 12 single domains , LigB3 and LigB11 , lack immunogenicity for the set of tested mAbs . From an antigenic response to LigB1-7 and LigB7-12 , at least every other individual Ig-like domain is capable of eliciting mAb production and the generated-LigB mAb libraries cover the length of the LigB Ig-like domain region . Several anti-LigB mAbs were tested using fluorescence-based flow cytometry for the ability to recognize native proteins on the surface of Leptospira cells . The spirochetes were incubated with mAbs from library C or V . Anti-LigB mAb-bound Leptospira cells were fluorescently-labelled with anti-mouse IgG antibodies and counted by flow cytometry . The strong fluorescence signal from incubation of the pathogenic L . interrogans serovar Pomona cells with anti-LigB mAb C5 was indicative of a tight cell surface interaction ( Figure 4B ) . Cells incubated with either PBS ( Figure 4B ) or the negative control mAb C22 ( Figure 4—figure supplement 2A ) failed to generate a fluorescence signal after secondary anti-mouse IgG antibody incubation . Additionally , cells from L . biflexa , a non-infectious Leptospira species which lacks the genes for Lig proteins ( Figueira et al . , 2011 ) , also failed to exhibit a fluorescence signal strong enough to indicate binding after anti-LigB mAb C5 incubation ( Figure 4—figure supplement 2A ) . A total of seven library C mAbs and six library V mAbs were measured for Leptospira surface binding character and all produced a fluorescence signal count over three-fold higher than the PBS control and over two-fold higher than the poorly binding mAb controls ( C22 and V33 ) ( Figure 4—figure supplement 2B ) . The cell binding propensities for measured mAbs are listed as mean fluorescence intensity ( MFI ) values in Table 1 . The flow cytometry data supports the presence of mAb-accessible Lig protein Ig-like domains on the surface of Leptospira cells . To test the role of the mAb libraries in promoting complement-mediated killing of Leptospira in vitro , the bactericidal activity of the LigB-binding mAbs was tested by incubating live L . interrogans serovar Pomona cells with complement-containing human serum with or without added antibodies . Using dark-field microscopy , motile bacteria were counted to calculate the survival rate at different time points . The PBS only ( negative control ) condition was unable to effectively kill the live Leptospira cells leaving an 84% survival rate at 120 min post treatment ( Figure 5A ) . The hamster-derived polyclonal antibody ( pAb ) treatment of Leptospira cells led to an enhancement of bactericidal activity with only 63% survival at 120 min post treatment . Leptospira incubated with C5 or V10 mAbs showed only a 2% to 3% survival rate ( a >20 fold improvement over pAb-incubated spirochetes p<0 . 05 ) ( Figure 5A ) . In addition , a high-throughput luciferase-based methodology was utilized to screen the bactericidal activity of the full set of mAbs . The assay measures the luminescence intensity from the expression of the light-generating lux cassette proteins in another pathogenic L . interrogans serovar Manilae to accurately report the viable cell count ( Murray et al . , 2010 ) . Note that the LigB proteins from L . interrogans serovar Manilae and L . interrogans serovar Pomona share 96% identity suggesting that the bactericidal properties provided by the LigB ( Pomona ) mAb library using these two serovars would be similar . Bioluminescent L . interrogans serovar Manilae were again incubated with complement-containing human serum with or without added antibodies . The survival rate was determined by measuring the luminescence intensity of metabolically active Leptospira relative to that from the spirochetes prior to Ab-incubation . Both incubation time and antibody concentration were varied to obtain time-dependent and dose-dependent curves ( Figure 5—figure supplement 1A and C , respectively ) as well as LT50 and LD50 values ( Table 1 ) . Experimental analysis of luminometer measurements for control conditions and representative mAbs are presented in Figure 5B . The two negative controls , PBS only and non-specific mouse IgG1 , as well as the hamster-derived pAbs were unable to kill >50% of cells at the longest time point ( 120 min ) or at the highest dosage ( 100 µg/ml ) . All anti-LigB mAbs were able to eliminate >50% of Leptospira between 40 and 85 min with the exception of mAb C23 ( Figure 5—figure supplement 1B ) . Each of the LigB-specific mAbs had the potential to kill >90% of the active Leptospira cells after a sufficient exposure time . All of the LigB-specific mAbs were able to effectively decrease the number of metabolically active Leptospira cells to less than 50% with a dose value ( LD50 ) between 10 and 30 μg/ml ( Figure 5—figure supplement 1D ) . The relative lack of efficiency of hamster pAbs compared to purified mAbs could be explained by the pAbs being a mixture of high affinity and low affinity antibodies , being a mixture of IgG subclasses with differing abilities to activate the complement system , or being derived from different rodents than the mAbs with potentially different abilities to activate the complement system . The ability of the anti-LigB mAbs to kill Leptospira in the presence of complement is consistent with their capability in binding to LigB protein in pathogenic Leptospira . In the context of a known antigen structure , the results of mAb-antigen interaction and mAb bactericidal activity experiments provide the opportunity to explore potential correlations between these three data sets . The broad mAb accessibility of LigB Ig-like domains supports the extended organization of the LigB1-12 SAXS structure . The LigB structure’s domain-domain arrangement is summarized in 3-domain segments ( Figure 5C ) . The properties for each mAb have been correlated in plots of KD vs . LD50 ( Figure 5D ) or FACS cell binding vs . LD50 ( Figure 5E ) . Additionally , the mAb data points are colored to signify the mAb’s LigB segments binding specificity . A trendline for KD vs . LD50 scatter plots using all data and excluding the two statistically-determined outliers ( Figure 5D; Figure 5—figure supplement 2 ) yields correlations with R²=0 . 508 and R²=0 . 773 , respectively . Antibodies specific to each of the four LigB segments are near the trendline . Thirteen anti-LigB mAbs were also tested for Leptospira surface binding ability . When plotted against LD50 , FACS-derived cell binding values fit to a trendline with an R²=0 . 312 and the single outlier-adjusted trendline of with an R²=0 . 466 ( Figure 5E; Figure 5—figure supplement 2 ) . In both plots , near-trendline mAbs with specificity to the LigB1-3 and LigB4-6 segments have a relatively high binding ability and low LD50 and correlated mAbs with specificity to the LigB7-9 segment have a relatively low binding ability and high LD50 . Because mAbs that do not fit the predicted correlation between binding affinity and bactericidal activity have the potential to inform vaccine studies , the bases for outlier discrepancies were examined further . Only one outlier , C12 , displays enhanced bactericidal activity above what would be expected based from binding correlations . While C12 displays some specificity for the straightest segment ( LigB4-6 ) , it can bind to two single domains ( LigB4 and LigB6 ) as well as a double domain in the segment LigB1-3 . Further , an additional ELISA assay found that C12 is the only library C mAb that can also bind to LigB7-12 ( Figure 5—figure supplement 3 ) . The ability of C12 to bind to multiple Ig-like domains covering a large range of the Ig-like domain regions suggests a reason for the increased bactericidal activity relative to C12’s binding affinity . The only outlier with bactericidal activity below expected is C23 which binds to LigB2-3 . These domain neighbors exhibit the sharpest interdomain bend suggesting that steric limitations can differentially effect LigB binding and bactericidal activity . Indeed , the kinked structure of C23’s target domain may decrease the rate for C23 to initiate bacterial killing relative to its KD ( Figure 5—figure supplement 4 ) . Overall , the link between antigen binding and bactericidal activity complements the highly accessible extended structure of LigB with highly accessible regions having the potential to yield high bactericidal mAbs . The high degree of correlation between antigen-mAb binding and bactericidal activity supports the hypothesis that a mAb’s ability to kill a pathogen can be determined by its antigen-specific binding affinity . To provide proof that a single Ig-like domain can act as a scaffold to display multiple protective epitopes , the binding epitope for anti-LigB mAbs was more specifically mapped to a surface within individual Ig-like domains . Chimeric LigB Ig-like domains were designed to differentiate interactions specific to the surface residues of major folding units . The high degree of homology between LigB Ig-like domains ( Figure 6A and C ) provided an opportunity for tertiary structural elements to be exchanged at two chimeric swapping positions ( Ptak et al . , 2014 ) ( Figure 6A and B ) . The first two segments of the chimeras ( β-strands A-C and β-strands C′-F ) were engineered to separate the top and bottom halves of the Ig-like domain β-sandwich . The third chimera region ( β-strands G-G′ ) was included to determine if an antigenic surface is formed on the untethered edge of the β-sandwich . Four single domain proteins that have been recognized to contain mAb-specific epitopes ( Table 1 ) were paired based on matching length to generate two sets of chimeras ( LigB5/LigB12 and LigB7/LigB10 ) . Including the wild-type domains , eight possible chimeric combinations were generated on single Ig-like domains ( Figure 6D and E ) . All mAbs showed binding to a subset of the eight proteins with a clear region-specific pattern ( Figure 6D and E ) . Two LigB7-specific mAbs ( V12 and V34 ) , two LigB10-specific mAbs ( V1 and V10 ) , and one LigB12-specific mAb ( V2 ) could bind only to the N-terminal β-strands A-C from their respective interacting domains . The LigB5-specific mAb C5 and LigB7-specfic mAb V3 interacted with proteins that contained β-strands C′-F from LigB5 and LigB7 , respectively . Unique to the LigB5/LigB12 mAb set , mAb V11 and mAb C6 required both the β-strands C′-F and the terminal β-strands G-G′ from LigB12 and LigB5 , respectively . The mAb binding patterns demonstrate that the surface contribution of both β-strands A-C and β-strands C′-F ( or C′-G′ ) can provide distinct epitopes for direct mAb targeting . The ultimate goal of combining structural and immunoreactivity studies of antigens is to explore a strategy for the rational engineering of improved vaccines . A chimeric LigB Ig-like domain was evaluated for the potential to elicit an immune response to multiple regions of LigB and to thereby enhance vaccine protection against leptospiral infection . The chimera LigB10-B7-B7 ( β-strands A-C: B10 and β-strands C′-F and G-G′: B7 ) was chosen as the best candidate for animal studies because the chimeric domain possesses the ability to bind highly bactericidal mAbs specific to each parent domain ( Figure 6—figure supplement 1 ) . Additionally , LigB10-B7-B7 was similar to wild-type LigB7 and LigB10 in protein expression levels and in overall secondary structure ( circular dichroism analysis , Figure 7—figure supplement 1 ) . To generate a protective response , hamsters were immunized with 50 µg of LigB7 , LigB10 , LigB10-B7-B7 , or PBS ( as a negative control ) for two times at 3 week intervals , and then challenged with 2 . 5 × 102 of triple passages of L . interrogans Pomona ( Conrad et al . , 2017; Kunjantarachot , 2014 ) ( Figure 7A ) . On day 8 , the leptospiral challenge was lethal for five of six hamsters inoculated with PBS group , three of five hamsters inoculated with wild-type LigB7 , and four of five hamsters inoculated with wild-type LigB10 . By day 10 , all of the wild-type LigB7 and LigB10 immunized hamsters had either died or were euthanized due to severe clinical signs . In contrast , all five of the hamsters that had been inoculated with LigB10-B7-B7 and subsequently challenged with Leptospira survived until the end of the experiment ( day 21 post infection ) . The survival rate of the LigB10-B7-B7 group is significantly higher ( 100% ) than either control group ( 17% ) or individual wild-type domain groups ( 0% ) ( p<0 . 05 ) . Serological fluids collected during the hamster studies were tested for the ability of the immunizations to generate an effective humoral response using a direct ELISA binding assay . The sera from LigB domain immunized hamsters provided a strong antibody response against the corresponding immobilized recombinant LigB domain ( Figure 7—figure supplement 2 ) . LigB7 , LigB10 , and LigB10-B7-B7 boosters were able to further enhance the antibody response . The Anti-LigB10-B7-B7 sera was also reactive with wild-type LigB7 and LigB10 but failed to react with the negative control , LigB12 . LigB10-B7-B7 immunization led to an increase in post-booster secondary response for IgG but not IgM antibodies implying the generation of a typical immune memory response ( Figure 7—figure supplement 3 ) . The serological tests indicate that targeted antibodies can be effectively generated by vaccines based on individual LigB Ig-like domains and chimeric domains . Several complementary methods were employed to test for the vaccine’s ability to reduce bacterial burdens of host tissue . Liver , kidney , and urinary bladder tissue from all immunization groups were examined by real-time quantitative reverse transcription polymerase chain reaction ( RT-qPCR ) to identify the Leptospira specific gene , LipL32 ( Figure 7B ) ( Levett et al . , 2005; Haake et al . , 2000 ) . The hamster group that received immunizations containing recombinant chimeric protein , LigB10-B7-B7 , exhibited a significant reduction in leptospiral burden when compared to bacterial loads from wild-type LigB7 or LigB10 immunized hamsters . The immunized hamsters were further examined for histopathologic changes of livers , kidneys and lungs ( Figure 7C–H and Figure 7—figure supplement 4 ) . Except for one apparently normal hamster from PBS group , all other hamsters from PBS , LigB7 and LigB10 immunization groups presented severe clinical signs and prominent macroscopic lesions on spirochete-targeted organs ( e . g . multifocal pulmonary ecchymoses , icteric liver and enlarged kidney ) . Lung lesions included thickening of alveolar septa due to edema , interstitial leukocytes infiltration , endothelial cell swelling , and extensive hemorrhage ( Figure 7C ) . Leptospira-infected livers were infiltrated by various inflammatory cells , indicating moderate to severe hepatitis . Focal necrosis was also found in parenchymal hepatocytes , leading to loss of normal tissue integrity ( Figure 7D ) . Severe tubulointerstitial nephritis with locally extensive hemorrhage in uriniferous spaces and tubules was present in Leptospira-infected kidneys ( Figure 7E ) . Furthermore , greater than 50% of renal tubules were lost and replaced by lymphoplasmacytic cells infiltration and severe fibrosis . In contrast , all hamsters immunized with chimeric LigB10-B7-B7 presented no visible macroscopic/microscopic lesions in livers , kidneys or lungs ( Figure 7F–H ) . In agreement with high survival rate of LigB10-B7-B7 immunized hamsters , vaccines containing the chimeric antigen provide superior protection and potentially sterilizing immunity for the animals from leptospiral infection . Leptospira , like other pathogenic bacteria , have evolved a wide variety of disease-related surface proteins to initiate colonization , to combat host defense systems and to reach target organs based on tissue-specific tropism ( Ko et al . , 2009; Vieira et al . , 2014 ) . Lig proteins contribute to pathogenesis by acting as both surface adhesins to mediate attachment to host extracellular matrix ( ECM ) molecules ( Lin et al . , 2009a; Vieira et al . , 2014; Figueira et al . , 2011 ) and recruiters of complement regulators to evade attack by the host’s innate immune system ( Breda et al . , 2015; Choy et al . , 2007; Hsieh et al . , 2016; Castiblanco-Valencia et al . , 2012 ) . In addition , Ig-like domains from LigB are currently some of the strongest recombinant protein-based vaccine candidates in protecting hamsters from lethal challenges although discrepancies in effectiveness suggest that vaccine redesign efforts could help to create an improved vaccine ( Cao et al . , 2011; Conrad et al . , 2017; Yan et al . , 2009; Adler , 2015 ) . This study provides new information regarding the architecture and accessibility of the main host-interacting region of LigB . The elongated structure presents insight into the mechanism of Lig protein adhesion , but , more importantly , an analysis of the mAb exposed regions and lack of long range inter-domain interactions has provided guidance for the optimization of recombinant leptospirosis vaccines using homologous Lig protein domains . The domain arrangement of the LigB1-12 Ig-like domain region was determined using low resolution SAXS structures . The SAXS-derived LigB1-12 structure provides a visual depiction that complements the more therapeutically applicable information from mAb accessibility . The domain accessibility predicted by the LigB architecture also suggested that including the full set of Ig-like domains as antigens for mAb library generation was likely to be worthwhile . While previous SAXS studies have generated composite multi-domain structures ( Jeffries et al . , 2011; Morgan et al . , 2012 ) , the full LigB structure is the first example of using a sliding window of medium length , 5-domain constructs to generate a much longer 12-domain structure . The advantage of using a sliding window is that the high level of redundancy within the 5-domain structures allows for an increased accuracy in orienting and positioning the segments relative to each other . The sequence of domains LigB1-6 is identical to LigA domains 1–6 so the SAXS structure provides a partial structure of the LigA Ig-like domain region . While LigA is not present in all pathogenic Leptospira species , LigA-based vaccines , like LigB-based vaccines , enhance immunity to leptospirosis ( Palaniappan et al . , 2006; Silva et al . , 2007; Faisal et al . , 2009 ) . Recombinant antigen-based vaccine technologies are the most promising avenue for the development of optimized protective strategies against leptospirosis . In recombinant antigen design , the basic antigen scaffold must present a native and well-folded epitope to the immune system while a minimized antigen can eliminate nonessential , disruptive features ( Kulp and Schief , 2013 ) . Examples of vaccines against different variants of Neisseria meningitides and Lyme disease Borreliae have been generated from chimeric recombinant antigens and have been capable of inducing broad spectrum bactericidal mAbs ( Earnhart and Marconi , 2007; Scarselli et al . , 2011 ) . Ig-like domains are found in a variety of prokayotic and eukaryotic extracellular proteins and readily fold to their native structure ( Bodelón et al . , 2013 ) . By incorporating various epitopes onto a homologous scaffold , the overall structural integrity of the antigenic surface can be maintained on the chimeric LigB Ig-like domain construct . Based on the high resolution LigB12 structure , each of the three chimeric segments encompassed a 900 to 2000 Å2 surface and could potentially make a full conformational epitope ( Ptak et al . , 2014 ) . While only two of three LigB chimeric segments were able to present as a full epitope experimentally , the third segment was required for binding by several of the mAbs . Indeed , initial experiments suggest that a redesigned chimera can harbor three separate epitopes on a single LigB Ig-like domain scaffold . Interestingly , the single domain LigB10-B7-B7 provides better protection to hamsters than two longer constructs , LigB7-12 ( which provided poor protection ) and LigB1-7 ( which provided good protection ) ( Yan et al . , 2009; Conrad et al . , 2017 ) . Exposure of epitopes in single domain antigens which would otherwise be blocked by host factors ( Beernink et al . , 2011 ) ( i . e . , ECM or serum proteins [Lin et al . , 2009a; Breda et al . , 2015; Hsieh et al . , 2016] ) in multi-domain constructs may enhance the efficiency . Two future goals of LigB recombinant vaccine studies will be to optimize further the chimeric antigens , consciously limiting the functionality of host-interacting sites , and to offer cross-species protection against different serovars . This study and other recent studies ( Ptak et al . , 2014; Conrad et al . , 2017 ) have significantly advanced our understanding of Lig protein structure and the potential of recombinant vaccines . The field of leptospirosis is poised to take advantage of the new insights and could make significant improvements in vaccines and other treatments to reduce the agricultural and human impact of the pathogen , Leptospira . A series of single ( LigB1 to LigB12 ) , double ( LigB1-2 to LigB11-12 , excluding LigB5-6 ) , five-domain ( LigB1-5 to LigB8-12 ) , and multiple ( LigB1-7 , LigB7-12 , and LigB5-12 ) Ig-like domains of LigB from Leptospira interrogans serovar Pomona ( GenBank , FJ030916 ) ( illustrated in Figure 1—figure supplement 1A , Figure 2—figure supplement 1A , and Figure 4—figure supplement 1 ) were constructed on the vector pET28-His-Sumo between BamHI and HindIII ( or XhoI ) restriction enzyme sites as previously described ( Manford et al . , 2010; Lin et al . , 2009b; Lin et al . , 2009a ) . The constructed plasmids containing LigB genes were transformed to E . coli Rosetta strain and the protein expression was induced with 1 mM isopropyl β-D-thiogalactopyranoside ( IPTG ) at 20 ˚C for 16 hr . After the cells were lysed using a high-pressure cell disruptor , the cell lysates were spun down and the supernatants were purified by Ni-NTA resin ( Qiagen ) . The His-Sumo tagged LigB proteins were eluted with phosphate buffered saline ( PBS; 137 mM sodium chloride , 10 mM sodium phosphate , 2 . 7 mM potassium chloride , 1 . 8 mM potassium phosphate ) containing 300 mM imidazole and then digested with Sumo protease Ulp-1 while dialyzing against PBS buffer at 4 ˚C overnight ( Manford et al . , 2010 ) . Afterwards , the digested proteins were applied to a second Ni-NTA resin to remove the His-Sumo tag , and the untagged proteins were collected in the flow-through fraction . To gain higher protein purity , the untagged Lig proteins were further separated from other contaminates by size exclusion chromatography ( SEC ) Superdex75 ( GE Healthcare ) , resulting in only one major species migrating on SDS-PAGE . For secondary structure analysis , circular dichroism ( CD ) spectra of LigB proteins were measured on an Aviv 202–01 spectropolarimeter ( Aviv Biomedical , Lakewood , NJ ) and predicted secondary structure composition was obtained using BeStSel ( Micsonai et al . , 2015 ) . Chimeric LigB5/LigB12 and LigB7/LigB10 constructs were created by overlapping extension PCR to produce all six possible swapped genes from three protein segments ( Figure 6 ) . The three segments , strand A-C , strand C′-F and strand G-G′ , and specific residue boundaries were identified based on the high-resolution NMR structure of LigB12 ( PDB ID 2MOG ) ( Ptak et al . , 2014 ) and the sequence alignment of parent domains . The percent identity matrix for the set of Ig-like domains was obtained using the EMBL-EBI web service Clustal Omega ( McWilliam et al . , 2013 ) . After ligation into pET28-Sumo vectors , the selected positive constructs were isolated from kanamycin resistant colonies and confirmed by DNA sequencing . Constructs corresponding to two , five and eight domains were generated to identify the optimal domain number required to obtain useful SAXS-derived envelopes ( Figure 1—figure supplement 1 ) . The 8-domain long construct ( LigB5-12 ) failed to generate envelopes with distinct domain regions representative of possible 8-domain structures . High average normalized spatial discrepancy ( NSD ) values for a reconstruction indicate increased variability among the trial models used to build the final envelope . Increasing average NSD with longer constructs may reflect a wider range of conformations expected with longer chains but additional factors could influence NSD statistics for longer constructs . The maximum diameter of a model is limited by the minimum q value measured in the scattering profile , as dictated by the Shannon Limit ( Putnam et al . , 2007 ) . Thus , for these experiments , q_min = 0 . 1 implies that models can be no longer than pi/0 . 1 = 314 Å , under ideal conditions . In reality , minor aggregation and noise at low q may also place additional limits on model length . Mylonas and Svergun have also shown that reconstructions of long rods are not as reliable as more compact structures ( Mylonas and Svergun , 2007 ) . For these reasons , we have chosen to assemble a full-length model from shorter overlapping constructs . SAXS envelopes derived from two or five sequential domains were generally comprised of distinct domain regions and accurately reflected the number and size of expressed domains . Of the 2- and 5-domain structures , only the envelopes of five sequential domains provided the relative orientation of multiple neighboring domains . LigB-derived proteins were exchanged into PBS buffer ( pH 7 . 4 ) and concentrated to 12–20 mg/ml as assessed by absorption at UV280 . SAXS data for individual samples was measured for 15 scans at 1× , 2/3× , and 1/3 × protein concentrations after dilution with PBS buffer and centrifugation at 14 , 000 rpm for 10 min . Capillary cells were robotically loaded with 30 μL samples from a 96 well-plate maintained at 4°C ( Nielsen et al . , 2012 ) . Between each sample , the capillary cell was thoroughly washed with detergent and water and then dried with air . All SAXS experiments were collected at the Cornell High-Energy Synchrotron Source ( CHESS ) ’s F2 or G1 beamline using a dual Pilatus 100 K-S SAXS/WAXS detector ( Acerbo et al . , 2015; Skou et al . , 2014 ) . Background subtraction of SAXS buffer and further data and statistical analysis were performed using the free open-source software , RAW ( Nielsen et al . , 2012; Hopkins et al . , 2017 ) . The GNOM program from the ATSAS suite was used to determine P ( r ) plots . Optimal Rmax was determined by screening at 5 Å intervals and the P ( r ) plots were normalized to the first peak . Profiles best representing the dilute ( ideal ) solution limit were used to generate ab initio models with DAMMAVER and DAMMIF programs from the ATSAS suite ( Petoukhov et al . , 2012; Volkov and Svergun , 2003 ) . 10 initial models were used to calculate the final model . Models with a normalized spatial distribution > mean + 2*standard deviation were treated as outliers and not included in determining the final model . No more than one model was excluded for each structure . Model uniqueness was evaluated using the AMBIMETER program ( Petoukhov and Svergun , 2015 ) . AMBIMETER score ranges < 1 . 5 indicate a unique ab initio shape determination , 1 . 5–2 . 5 indicate some potential for alternate solutions , and >2 . 5 indicate that multiple shape solutions will fit the data . A 5 mg total of LigB1-7 , and separately 5 mg of LigB7-12 , was used to immunize and boost the antibody production from five BALB/c mice . Hybridoma clone generation was conducted in the laboratory of Dr . William Davis ( fee-for-service ) as previously described ( Park et al . , 2015 ) . Standard ELISA assays were used for in-house screening of the hybridoma clones for positive supernatants . A total of 24 clones of IgG-type mAbs against LigB1-7 ( library C; Lig protein Conserved region ) and 36 clones of IgG-type mAbs against LigB7-12 ( library V; Lig protein Variable region ) were generated . Each library produced 9 mAbs which were found to be of moderate to high binding efficiency and utilized for further characterization . Purification of mAbs was conducted by protein A/G chromatography ( Pierce ) using the manufacturer-recommended procedure with minor modification ( Eliasson et al . , 1989 ) . Briefly , the hybridoma supernatant was dialyzed against the binding buffer ( 100 mM potassium phosphate , 150 mM sodium chloride , pH 8 . 0 ) overnight at 4 ˚C . The dialyzed sample was applied to a protein A/G column ( pre-equilibrated with binding buffer ) . After washing with 10–15 ml of binding buffer to remove the unbound fraction , the bound mAb was eluted with 100 mM glycine at pH 3 . 0 and neutralized immediately with 1 M Tris at pH 9 . 0 . The eluate containing mAb was dialyzed against PBS buffer at pH 7 . 0 , concentrated to 3–4 mg/ml , and stored at −80 ˚C . LigB-binding mAbs were selected using an ELISA assay ( Lin et al . , 2009a; Lin et al . , 2011 ) . Microtiter plates ( Nunc MaxiSorp , ThermoFisher Scientific ) were coated with 1 μg of LigB1-7 or LigB7-12 in coating buffer ( 0 . 2 M NaHCO3 , pH 9 . 4 ) at 4˚C overnight . After blocking with 3% BSA in PBS buffer for 1 hr , hybridoma supernatants were prepared at 1/500 dilutions and individually applied to LigB-coated wells for initial screening . To obtain dissociate constants , selected mAbs with moderate to high binding affinity to LigB were serially diluted in PBS ( 0 . 00686 , 0 . 0137 , 0 . 0412 , 0 . 123 , 0 . 370 , 1 . 11 , 3 . 33 and 10 µM ) and then individually applied to LigB-coated wells . Between each step , PBS containing 0 . 05% Tween-20 ( PBS-T ) was used to wash the plates for three times . Subsequently , anti-mouse IgG antibody conjugated with HRP ( 1:5000 , Invitrogen ) was added to detect the binding of mAbs presented in hybridoma supernatants . Finally , 100 μl of TMB peroxidase substrate ( KPL ) was applied to each well , the optical density of which was recorded at 630 nm by ELx808 Absorbance Microplate Reader ( BioTek ) . OD630 values represent the mean of three independent trials ± the standard deviation . For each trial , samples were assayed in two replicates . For examining the domain specificities of mAbs , single domain LigB fragments ( LigB1 to LigB12 ) , double domain LigB fragments ( LigB1-2 to LigB11-12 , excluding LigB5-6 ) , positive control ( LigB1-7 , LigB7-12 ) , or negative control ( BSA ) were coated with a fixed concentration ( 1 μg/well ) on the ELISA plates . Each mAb against LigB1-7 or LigB7-12 was then added to corresponding single-domain or double-domain LigB coated wells . For binding analysis of each chimera , the set of six chimeric LigB proteins plus the two parent LigB proteins were coated on the ELISA plates ( 1 μg/well ) . Each mAb with activity against either of the two parent LigB proteins was screened for binding activity on the LigB coated wells . For measuring the antibody responses triggered by LigB-based recombinant vaccines , sera from hamsters immunized with LigB7 , LigB10 and LigB10-7-7 were collected at different time points ( pre-immunization , post-immunization and post-booster ) . Then , these serum samples ( 1:500 ) were added to corresponding LigB Ig-like domain coated wells ( 1 μg/well ) , and also applied to LigB12 coated wells ( negative control ) . Finally , anti-hamster IgG or IgM antibody conjugated with HRP ( 1:1000 , KPL or SouthernBiotech ) was used to detect the LigB-bound antibodies from hamster sera . All single and double domain ELISA experiments including single chimeric domain experiments were conducted in three trials , the mean ±S . D . of which were shown in bar charts . Leptospira interrogans freshly harvested from hamsters challenged with serovar Pomona or overnight NaCl-treated low passage Leptospira were prepared in PBS buffer containing 5 mM MgCl2 at 108 cells/ml . Various anti-LigB mAbs ( 100 µg/ml ) were individually applied to the bacterial suspension at 1:500 dilution . After a 1 hr incubation at room temperature , the bacteria-mAbs mixtures were spun down at 2 , 000 g for 7 min and then washed with PBS containing 1% BSA . Subsequently , goat anti-mouse antibodies conjugated with fluorescein ( FITC ) ( 1:1000 ) were used as secondary antibodies for probing the bacteria-bound mAbs . After another PBS-BSA wash , the bacteria-mAbs mixtures were fixed by 0 . 5% formaldehyde in PBS . The non-pathogenic Leptospira biflex , which does not express LigB , was subjected to the same procedure by incubating with selected mAbs for the control experiments ( Figueira et al . , 2011 ) . Another negative control was conducted by treating the pathogenic Leptospira with unrelated mouse IgG ( isotypic mouse IgG purchased from ThermoFisher Scientific ) . Flow cytometry was performed at the Cornell University Flow Cytometry Core facility using a BD LSR-II ( BD Biosciences ) instrument with the excitation laser at 488 nm and the emission wavelength at 525/575 nm . Unstained Leptospira was identified by forward scatter ( FSC ) and side scatter ( SSC ) . Selected mAb ( C5 ) treated L . interrogans without secondary antibodies ( FITC-conjugated goat anti-mouse IgG ) was used as a negative control to set the gating region . The mean FITC-positive ( MFI ) count was obtained for the gated region of the 525 nm vs . 575 nm scatter plots using BD FACSDiva software . For each sample , at least 20 , 000 cells were analyzed in two independent trial of two replicates . To examine if the mAbs were effective at killing bacteria in vitro , the serum bactericidal activity ( SBA ) assay was conducted . 108 cells/ml of low passage , high virulent L . interrogans seorvar Pomona and Manilae strain M1307 were prepared in PBS buffer containing 2 mM MgCl2 and 1 mM CaCl2 , and then mixed with respective mAbs , plus 25% of normal human serum ( ImmunoReagents ) as a complement source . The mixtures were incubated at 37°C for 1 hr , and the viability of the bacteria were examined using dark-field microscopy . The survival rate of leptospira was calculated as the number of motile ( alive ) cells in every 100 counts separately by two researchers ( blind ) . The mean value was obtained from the two independent ( blind ) measurements as a single technical replicate . Three independent trials of two replicates were measured . The viability of the leptospira was also accessed from the luminescence emitted by metabolically active L . interrogans Manilae strain M1307 ( Murray et al . , 2010 ) . The same preparations were examined at different time points ( 0 , 30 , 60 and 90 min ) , and the luminescence intensity of each sample was measured by GloMax 96 Microplate Luminometer ( Promega ) . The survival rate of leptospira was calculated by the intensity at a specific time point divided by the intensity at 0 min . In addition , a series of dilutions of mAbs ( 100 μg/ml to 3 . 125 μg/ml ) were evaluated for dose dependent bactericidal efficacy of antibodies . The time ( LT50 ) and dose ( LD50 ) required to reach 50% lethality were obtained from fitted logistic and dose inhibition curves , respectively ( Origin 7 . 0 ) . Hamsters ( Harlan Sprague Dawley Laboratory ) were housed in isolation units approved by the Cornell University Institutional Animal Care and Use Committee ( Protocol number: 2015–0133 ) . The Golden Syrian Hamsters used in vaccine trials were allowed to run free in the cage , were fed a commercial ration , and were provided water ad libitum as previously described ( Kunjantarachot , 2014 ) . Six 5-week-old hamsters each were vaccinated subcutaneously for each LigB-based recombinant vaccine containing adjuvant 2% Alhydrogel ( InvivoGen ) at 3 week intervals for a total of two injections . The control group was injected with adjuvant only . Three weeks after the final vaccination , all animals were challenged with 2 . 5 × 102 of triple passaged L . interrogans seorvar Pomona through the intraperitoneal route as previously described ( Kunjantarachot , 2014 ) . Kidneys , livers , lungs and urinary bladders were biopsied from hamsters within 1 hr after euthanasia ( Palaniappan et al . , 2006 ) . Leptospiral loads found in livers , kidneys and urinary bladders from all immunization groups were examined by real-time RT-qPCR ( Levett et al . , 2005 ) . The total RNA was extracted from target tissues and then reverse transcribed to cDNA . Subsequently , the Leptospira specific gene , LipL32 ( Haake et al . , 2000 ) , was amplified and detected with fluorescence by 7500 Fast Real-Time PCR system . Histopathological tissue slices were fixed with 10% neutral buffered formalin and stained with hematoxylin and eosin . Tissue samples were imaged and scored by light microscopy .

Vaccines encourage the immune system to develop a protection against disease-causing bacteria and viruses . Some types of immune cells release antibodies , which recognize particular proteins on the surface of the invading microbe . A vaccine that contains these surface proteins allows immune cells to develop the antibodies that can help to fight off an infection at a later date . Studying the shape and structure of the surface proteins can reveal how they are detected by our immune systems and can further be used to design more effective vaccines . Leptospirosis is the most common bacterial disease to affect both humans and animals . The symptoms of the disease include fever , muscle pain and bleeding from the lungs . New vaccines against leptospirosis are desperately needed because current ones have severe side effects and do not fully protect against the disease . The most promising new vaccine candidates are the Lig proteins , which are found on the surface of leptospirosis causing bacteria cells , but little was known about their molecular structure . The region of the Lig protein that is recognized by the immune system consists of a series of twelve connected ‘immunoglobulin-like’ domains . Hsieh , Ptak et al . used X-ray scattering to determine the structure of this region and found that the protein is highly elongated . Additional experiments showed that the individual domains provoke immune responses to different extents . Antibodies that can interact more strongly with the Lig protein were also better able to kill the bacteria . Based on this information , Hsieh , Ptak et al . combined parts of the individual domains that bind strongly to antibodies to design a new protein that , when used as a vaccine , protected hamsters against leptospirosis much better than other Lig protein-based vaccines . Further engineering and testing are required to develop an optimized , commercial leptospirosis vaccine , but the work of Hsieh , Ptak et al . shows the effectiveness of structure-based vaccine design methods . In the future , similar methods could be used to develop better vaccines and treatments for other infectious diseases .

2017

Extended low-resolution structure of a Leptospira antigen offers high bactericidal antibody accessibility amenable to vaccine design

RNA granules are non-membrane bound cellular compartments that contain RNA and RNA binding proteins . The molecular mechanisms that regulate the spatial distribution of RNA granules in cells are poorly understood . During polarization of the C . elegans zygote , germline RNA granules , called P granules , assemble preferentially in the posterior cytoplasm . We present evidence that P granule asymmetry depends on RNA-induced phase separation of the granule scaffold MEG-3 . MEG-3 is an intrinsically disordered protein that binds and phase separates with RNA in vitro . In vivo , MEG-3 forms a posterior-rich concentration gradient that is anti-correlated with a gradient in the RNA-binding protein MEX-5 . MEX-5 is necessary and sufficient to suppress MEG-3 granule formation in vivo , and suppresses RNA-induced MEG-3 phase separation in vitro . Our findings suggest that MEX-5 interferes with MEG-3’s access to RNA , thus locally suppressing MEG-3 phase separation to drive P granule asymmetry . Regulated access to RNA , combined with RNA-induced phase separation of key scaffolding proteins , may be a general mechanism for controlling the formation of RNA granules in space and time . RNA granules are concentrated assemblies of RNA and RNA-binding proteins that form without a limiting membrane in the cytoplasm or nucleoplasm of cells ( Courchaine , 2016 ) . RNA granules are ubiquitous cellular structures and several classes of cytoplasmic RNA granules have been described , including stress granules , P bodies , neuronal granules and germ granules ( Anderson and Kedersha , 2006 ) . Cytoplasmic RNA granule components typically exchange rapidly between a highly concentrated pool in the granule and a more diffuse , less concentrated pool in the cytoplasm ( Weber and Brangwynne , 2012 ) . In addition to RNA-binding domains , proteins in RNA granules often contain prion-like , low complexity , or intrinsically-disordered regions ( IDRs ) ( Courchaine , 2016 ) . In concentrated solutions , IDRs spontaneously de-mix from the aqueous solvent to form liquid droplets ( liquid-liquid phase separation or LLPS ) or hydrogels ( Li et al . , 2012; Weber and Brangwynne , 2012; Elbaum-Garfinkle et al . , 2015; Guo and Shorter , 2015; Lin et al . , 2015; Nott et al . , 2015 ) . Like RNA granules in vivo , proteins in LLPS droplets and hydrogels exchange with the solvent ( Kato et al . , 2012; Li et al . , 2012; Elbaum-Garfinkle et al . , 2015; Lin et al . , 2015 ) . These findings have suggested that LLPS or reversible gelation drives the assembly of RNA granules in vivo ( Guo and Shorter , 2015 ) . In cells , RNA granule assembly is regulated in space and time . For example , stress granules assemble within seconds of exposure to toxic stimulants that require the temporary removal of mRNAs from the translational pool ( Anderson and Kedersha , 2006 ) . In eggs , germ granules assemble in the germ plasm , a specialized area of the cytoplasm that is partitioned to the nascent germline during the first embryonic cleavages ( Voronina et al . , 2011 ) . How phase separation , a spontaneous process in vitro , is regulated in vivo to ensure that RNA granules form at the correct place and time is not well understood . The germ ( P ) granules of C . elegans are an excellent model to study the mechanisms that regulate granule assembly ( Updike and Strome , 2010 ) . For most of development , P granules are stable perinuclear structures , but in the transition from oocyte-to-embryo , P granules detach from the nucleus and become highly dynamic ( Pitt et al . , 2000; Wang et al . , 2014 ) . As the oocyte is ovulated in the spermatheca , P granules disassemble and release their components in the cytoplasm . After fertilization , P granule proteins reassemble into dynamic granules that undergo repeated cycles of assembly and disassembly in synchrony with cell division . Live imaging in the 1-cell zygote has revealed that these cycles are spatially patterned along the anterior-posterior axis of the embryo: granule assembly is favored in the posterior and granule disassembly is favored in the anterior ( Brangwynne et al . , 2009; Gallo et al . , 2010 ) . By the first mitosis , P granules are found exclusively in the posterior cytoplasm together with other germ plasm components . As a result , P granules are segregated to the posterior germline blastomere P1 and excluded from the anterior somatic blastomere AB . P granule asymmetry is under the control of the PAR polarity network which divides the zygote into distinct anterior and posterior domains ( Motegi and Seydoux , 2013 ) . The PAR-1 kinase is enriched in the posterior cytoplasm and restricts the RNA-binding protein MEX-5 ( and its redundant homolog MEX-6 ) to the anterior cytoplasm ( Griffin et al . , 2011 ) . MEX-5 and MEX-6 in turn restrict P granules to the posterior ( Schubert et al . , 2000; Gallo et al . , 2010 ) . In mex-5 mex-6 double mutants , P granule still undergo cycles of assembly and disassembly but these are no longer patterned along the anterior-posterior axis , and small granules remain throughout the cytoplasm ( Gallo et al . , 2010 ) . An attractive hypothesis is that MEX-5 blocks phase separation of P granule components in the anterior cytoplasm ( Brangwynne et al . , 2009; Lee et al . , 2013 ) . The mechanism of MEX-5 action and the critical P granule component ( s ) regulated by MEX-5 , however , are not known . P granule assembly in zygotes requires several P granule proteins , including the RNA-binding protein PGL-1 ( and its redundant paralog PGL-3 ) and the intrinsically-disordered protein MEG-3 ( and its redundant paralog MEG-4 ) ( Hanazawa et al . , 2011; Wang et al . , 2014 ) . PGL-1 and PGL-3 are RGG domain proteins that self-associate and recruit other RNA-binding proteins to the granules , including the GLH family of RNA helicases ( Updike and Strome , 2010; Hanazawa et al . , 2011 ) . MEG-3 and MEG-4 are redundant , serine-rich proteins that bind to PGL-1 in vitro and are essential for P granule assembly in embryos . In zygotes lacking meg-3 and meg-4 , PGL-1 and GLH-2 form transient assemblies that do not segregate asymmetrically and are not maintained in later stages ( Wang et al . , 2014 ) . Phosphorylation of MEG-3 by the DYRK kinase MBK-2 promotes granule disassembly in zygotes , but the mechanism that favors disassembly specifically in the anterior cytoplasm is not known ( Wang et al . , 2014 ) . In this study , we use a combination of in vivo and in vitro experiments to examine the contribution of the MEX , MEG , and PGL proteins to P granule assembly and asymmetry . We show that MEG-3/4 , but not PGL-1/3 , are essential for granule assembly and asymmetry , and that MEX-5 localizes MEG-3 in a posterior-rich gradient . We demonstrate that MEG-3 is an RNA-binding protein that is stimulated by RNA to undergo phase separation . MEX-5 is sufficient to block MEG-3 phase separation in vitro and to block MEG-3 granule formation in vivo . Our findings are consistent with a model whereby MEX-5 antagonizes MEG-3’s access to RNA to inhibit MEG-3 condensation in the anterior cytoplasm . To determine the genetic hierarchy that controls granule asymmetry , we first compared the distributions of MEX-5 , MEG-3 , PGL-1 and GLH-1 during the earliest stages of zygote polarization . We monitored MEX-5 , MEG-3 and GLH-1 localization using tagged alleles generated by genome editing ( Materials and methods , Supplementary file 1 ) and PGL-1 localization using a polyclonal antibody that recognizes PGL-1 ( Strome and Wood , 1983 ) . Before polarization , MEX-5 , MEG-3 , PGL-1 and GLH-1 were all distributed evenly throughout the cytoplasm . MEX-5 appeared mostly diffuse in the cytoplasm with a few foci , whereas MEG-3 , PGL-1 , and GLH-1 appeared both diffuse and enriched in many small ( <1 micron diameter ) foci ( Figure 1A ) . During polarization ( pronuclear formation and migration ) , MEX-5 and MEG-3 began to redistribute into opposing cytoplasmic gradients along the long axis of the zygote ( anterior-posterior axis ) with MEG-3 beginning to form large ( ~1 micron ) granules in the posterior ( Figure 1A ) . Total levels of MEG-3 do not change during this period , consistent with redistribution of existing MEG-3 protein from anterior to posterior ( Figure 1B , Figure 1—figure supplement 1 ) . In contrast to MEX-5 and MEG-3 , PGL-1 and GLH-1 remained uniformly distributed during polarization . After polarization ( mitosis ) , all proteins were localized , with MEX-5 in the anterior cytoplasm and MEG-3 , PGL-1 and GLH-1 in large granules in the posterior cytoplasm ( Figure 1A ) . 10 . 7554/eLife . 21337 . 003Figure 1 . Localization of P granule proteins during zygote polarization . ( A ) Photomicrographs of live wild-type ( mCherry::MEX-5 and GLH-1::eGFP ) or fixed meg-4 zygotes ( MEG::3 OLLAS and PGL-1 ) at three different stages: before polarization ( pronuclear formation ) , during polarization ( pronuclear migration ) and after polarization ( mitosis ) . meg-4 ( ax3052 ) zygotes were co-immunostained for MEG-3::OLLAS ( anti-OLLAS , Novus Biological ) , and PGL-1 ( K76 , DSHB ) . meg-4 is required redundantly with meg-3 for P granule assembly , and each is sufficient to support localized granule assembly ( Wang et al . , 2014 ) . In this and subsequent figures , dashed lines outline each embryo , embryos are oriented with anterior to the left and posterior to the right and are ~50 μM long . At least three embryos were examined per genotype shown . ( B ) MEG-3::meGFP levels in the anterior and posterior halves of the zygote during polarization . Values represent average fluorescence intensity over time ( relative to initial levels ) in the anterior ( red ) and posterior ( blue ) . Averages come from values measured from three different embryos . Error bars represent standard deviation of the mean . ( C ) Photomicrographs of fixed zygotes after polarization immunostained for OLLAS , PGL-1 , or FLAG . mex-5/6 zygotes were derived from wild-type hermaphrodites treated with mex-5 and mex-6 RNAi . meg-3/4 zygotes were derived from meg-3 ( ax3055 ) ; meg-4 ( ax3052 ) hermaphrodites . pgl-1/3 zygotes were derived from pgl-3 ( bn104 ) hermaphrodites treated with pgl-1 RNAi ( see Figure 1—figure supplement 1 for additional examples of pgl-1 ( RNAi ) ;pgl-3 ( bn104 ) zygotes also stained for PGL-1 to verify loss of PGL-1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21337 . 00310 . 7554/eLife . 21337 . 004Figure 1—figure supplement 1 . MEG-3 localizes before PGL-1 and does not require PGL-1 or PGL-3 to assemble granules . ( A ) Photomicrographs of live MEG-3::meGFP in zygotes at three different stages: before polarization ( pronuclear formation ) , during polarization ( pronuclear migration ) and after polarization ( mitosis ) . These and similar images taken from 3 zygotes at 14 time points were used to quantify MEG-3::meGFP fluorescence levels over time as shown in Figure 1B . ( B ) Photomicrographs of fixed wild-type and meg-3/4 zygotes after polarization immunostained for PGL-3 . meg-3/4 zygotes were derived from meg-3 ( ax3055 ) ; meg-4 ( ax3052 ) hermaphrodites . PGL-3 staining was done using the KT3 antibody ( DSHB ) . ( C ) Photomicrographs of fixed zygotes stained with anti-FLAG ( top ) and K76 anti-PGL-1 ( bottom ) to show MEG-4 localization at the 1-cell ( left ) and 4-cell ( right ) stages . pgl-1/3 zygotes were derived from pgl-3 ( bn104 ) ; meg-4 ( ax2080FLAG tag ) hermaphrodites treated with pgl-1 RNAi . DOI: http://dx . doi . org/10 . 7554/eLife . 21337 . 004 To determine the interdependence of these localizations , we examined the effect of removing MEX-5/6 , MEG-3/4 or PGL-1/3 using RNA-mediated interference ( RNAi ) or genetic mutants ( GLH-1 has already been shown to depend on PGL-1/3 for asymmetry [Hanazawa et al . , 2011] ) . In zygotes derived from mothers treated with double-stranded RNA against mex-5 and mex-6 ( mex-5/6 ( RNAi ) zygotes ) , the MEG-3 gradient did not form and MEG-3 and PGL-1 granules remained uniformly distributed throughout the cytoplasm ( Figure 1C , ( Gallo et al . , 2010 ) . In meg-3; meg-4 double mutant embryos , the MEX-5 gradient was unaffected but neither PGL-1 nor PGL-3 granules segregated properly ( Wang et al . , 2014 ) , Figure 1C , Figure 1—figure supplement 1 ) . In pgl-1 ( RNAi ) ; pgl-3 ( bn104 ) zygotes , the MEG-3 and MEG-4 gradients were unaffected and MEG-3/MEG-4 granules formed in the posterior as in wild-type , except that the granules appeared smaller especially in zygotes ( Wang et al . , 2014 ) , Figure 1C , Figure 1—figure supplement 1 ) . These analyses suggest that MEX-5 and MEX-6 regulate granule asymmetry by localizing MEG-3 and MEG-4 to the posterior , which in turn are required to localize PGL-1 and PGL-3 . PGL-1 and PGL-3 are not required to localize MEG-3 or MEG-4 or to assemble MEG-3/4 granules , but contribute to the size of MEG-3/4 granules as reported previously ( Wang et al . , 2014 ) . Using mex-5 transgenes , Griffin et al . ( 2011 ) showed that the formation of the MEX-5 gradient requires phosphorylation of serine 404 in the C-terminus of MEX-5 by the kinase PAR-1 . To determine whether the MEX-5 gradient is required to pattern MEG-3 , we mutated serine 404 to alanine ( S404A ) at the mex-5 locus using CRISPR/Cas9-mediated genome editing ( Paix et al . , 2015 ) . We introduced the S404A mutation in two strains: one where the mex-5 locus had been previously tagged with mCherry to monitor MEX-5 localization , and one where MEG-3 had been previously tagged with GFP to monitor MEG-3 localization ( Supplementary file 1 ) . As expected , we found that MEX-5 ( S404A ) failed to form a gradient during zygote polarization and remained uniformly distributed ( Figure 2A ) . Using the MEG-3::GFP strain , we found that zygotes derived from mothers homozygous for mex-5 ( s404a ) ( mex-5 ( S404A ) zygotes ) , MEG-3 did not form a gradient or granules . Instead , MEG-3 remained uniformly distributed in the cytoplasm throughout the 1-cell stage ( Figure 2B ) . We conclude that MEX-5 is sufficient to suppress the formation of MEG-3 granules throughout the cytoplasm . 10 . 7554/eLife . 21337 . 005Figure 2 . MEX-5 is necessary and sufficient to disassemble MEG-3 granules in vivo . ( A ) Photomicrographs of mCherry tagged live zygotes [wild-type and mex-5 ( S404A ) ] or fixed zygotes expressing MEX-5 tagged with an OLLAS epitope [mex-5 ( ZF- ) ;mex-6 ( RNAi ) ] at pronuclear meeting to show MEX-5 localization . Wild-type MEX-5 is in an anterior-rich gradient , whereas MEX-5 ( S404A ) and MEX-5 ( ZF- ) are uniformly distributed . Numbers indicate number of zygotes exhibiting phenotype shown / total number of zygotes examined . ( B ) Photomicrographs of live embryos expressing MEG-3::meGFP . Genotypes at the mex-5 locus are as indicated . Numbers indicate the numbers of zygotes examined as in A . In 1/10 mex-5 ( ZF- , S404A ) ; mex-6 ( RNAi ) zygotes , MEG-3 granules were asymmetric possibly due to incomplete depletion of MEX-6 by RNAi . DOI: http://dx . doi . org/10 . 7554/eLife . 21337 . 00510 . 7554/eLife . 21337 . 006Figure 2—figure supplement 1 . Expression of MEX-5 , and MEX-5 ( ZF- ) , MEG-3 , and MEG-3IDR . ( A ) MEX-5 functions redundantly with MEX-6 to localize MEG-3: Photomicrographs of live wild-type and mex-5 ( RNAi ) zygotes co-expressing mCherry:MEX-5 and MEG-3::meGFP . Loss of mCherry fluorescence in the mex-4 ( RNAi ) zygotes was used to confirm efficient loss of mCherry::MEX-5 . Numbers indicate number of zygotes exhibiting phenotype shown / total number of zygotes examined . ( B ) Western blot of embryo lysates co-blotted with α OLLAS and α tubulin ( loading control ) . Expected sizes for protein-epitope fusions are indicated on the right by a < symbol . N2 lysate is a control lysate expressing no tagged proteins , all others are lysates from embryos expressing tagged proteins as indicated . ( C ) Bar graph showing relative expression of MEG-3 and MEX-5 . Western blot in Figure 2—figure supplement 1 was quantified using Image J . MEX-5 is 10X more abundant than MEG-3 . DOI: http://dx . doi . org/10 . 7554/eLife . 21337 . 006 The MEX-5 RNA-binding domain is comprised of two zinc fingers that bind with high affinity to poly-U stretches . Pagano et al . ( 2007 ) have shown that mutation of a single amino acid in each finger ( R247E and K318E ) reduces MEX-5 binding affinity for poly-U by 35-fold . Using mex-5 transgenes , Griffin et al . ( 2011 ) showed the finger mutations R247E and K318E also disrupt formation of the MEX-5 gradient in vivo . To determine the effect of these mutations on MEG granule assembly , we introduced R247E and K318E ( hereafter referred to as ZF- ) at the mex-5 locus by CRISPR/Cas9 genome editing into an OLLAS::MEX-5 tagged line and the MEG-3::GFP line . Like MEX-5 ( S404A ) , MEX-5 ( ZF- ) did not form a gradient and was uniformly distributed in zygotes ( Figure 2A ) . In contrast to mex-5 ( S404A ) zygotes , however , mex-5 ( ZF- ) zygotes assembled posterior MEG-3 granules as in wild-type ( data not shown ) . mex-5 is partially redundant with its paralog , mex-6 ( Schubert et al . , 2000 ) , which is sufficient to polarize MEG granules in the absence of MEX-5 ( Figure 2—figure supplement 1 ) . Consistent with this redundancy , we found that depletion of mex-6 by RNAi in mex-5 ( ZF- ) zygotes caused MEG-3 granules to assemble throughout the cytoplasm , as in mex-5/6 ( RNAi ) zygotes ( Figure 2B ) . These observations suggest that mex-5 ( ZF- ) is a loss-of-function allele . The loss of mex-5 activity was not due to reduced expression as MEX-5 ( ZF- ) was expressed at the same level as wild-type MEX-5 ( Figure 2—figure supplement 1 ) . To determine whether high-affinity RNA binding is also required for MEX-5 ( S404A ) ability to suppress MEG-3::GFP granule assembly throughout the cytoplasm , we introduced the S404A mutation by genome editing into mex-5 ( ZF- ) hermaphrodites . We found that mex-5 ( ZF- , S404A ) zygotes assembled posterior MEG-3::GFP granules , as is observed in mex-5 ( ZF- ) and wild-type zygotes . Depletion of mex-6 by RNAi in this background yielded zygotes with uniform MEG-3::GFP granules , as expected for a mex-5/6 loss-of-function phenotype ( Figure 2B ) . We conclude that suppression of MEG-3 granule assembly by MEX-5 depends on MEX-5’s ability to bind RNA with high affinity . Unlike MEX-5 , MEG-3 does not have a recognizable RNA-binding domain ( Wang et al . , 2014 ) . MEG-3 contains a long predicted intrinsically-disordered region ( IDR ) at its N-terminus ( aa1-550 ) followed by a region with lower predicted disorder ( aa550-862 ) [IUPRED predictions , ( Dosztányi et al . , 2005 ) . To determine whether MEG-3 binds RNA , we expressed and purified as His-tagged fusions full length MEG-3 , MEG-3 ( aa1-544 ) ( hereafter referred to MEG-3IDR ) and MEG-3 ( aa545-862 ) ( hereafter referred to MEG-3Cterm ) ( Figure 3—figure supplement 1 ) . We tested each for binding to poly-U30 RNA using electrophoretic mobility shift assays ( EMSA ) and fluorescent polarization ( FP ) assays ( Pagano et al . , 2007 ) . EMSA experiments revealed that MEG-3 and MEG-3IDR interact with poly-U30 RNA to form complexes that migrate as a discrete band during electrophoresis ( Figure 3A ) . Using FP , we calculated the apparent dissociation constant ( Kd , app ) of MEG-3 for poly-U30 to be ~32 nM , similar to that of MEX-5 ( Kd , app= ~ 29 nM ) ( Pagano et al . , 2007 ) . MEG-3IDR also bound RNA but with ~15 fold lower affinity ( Kd , app= ~ 460 nM ) . MEG-3Cterm did not bind RNA significantly by EMSA or in the FP assay ( Kd , app > 3000 nM ) ( Figure 3A , Figure 3—figure supplement 1 ) . We conclude that MEG-3 binds RNA with high affinity and that this activity resides primarily within the MEG-3 IDR , although high affinity binding also requires the MEG-3 C-terminus . 10 . 7554/eLife . 21337 . 007Figure 3 . MEG-3 Binds RNA in vitro . ( A ) Binding of MEG-3 to poly-uridine 30 ( poly-U30 ) is shown by electrophoretic mobility shift assay ( EMSA ) using fluorescein-labeled poly-U . EMSAs are shown for ( top to bottom ) full length MEG-3 , MEG-3IDR , and MEG-3Cterm . Unbound poly-U30 is denoted by an asterisk ( * ) . For each image shown , n ≥ 3 technical replicates . ( B ) Fluorescence Polarization of poly-U30 by MEG-3 . Fluorescence polarization values normalized relative to saturation are shown for full length MEG-3 ( green ) , MEG-3IDR ( black ) , and MEG-3Cterm ( red ) . Values represent averages of ≥3 technical replicates . A fit of the polarization as a function of protein concentration is plotted and used to calculate the given Kd , app . Error bars report S . E . M . Expanded graphs are shown in Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 21337 . 00710 . 7554/eLife . 21337 . 008Figure 3—figure supplement 1 . MEG-3 RNA binding . ( A ) Coomassie stained gel showing His-tagged MEG-3 proteins . Expected sizes are indicated on the right . Each lane is from a separate gel . ( B ) RNA-RNA competition EMSA . 300 nM MEG-3 ( 300 nM ) was incubated with 50 nM poly-U30 and increasing amounts of unlabeled 30-mer RNAs as indicated to the left of each panel . The first lane of each gel contains only labeled poly-U RNA . The last lane of each gel contains poly-U RNA and the highest concentration of unlabeled competitor RNA ( 5000 nM ) . The unbound poly-U RNA is denoted by an asterisk ( * ) . Double asterisks ( ** ) in the poly-G assay denote a band that arises due to an interaction between poly-G and poly-U ( independent of MEG-3 ) . For each image shown , n = 2 technical replicates . ( C ) Fluorescence Polarization of polyuridine by MEG-3 ( expanded from Figure 3 ) . Fluorescence polarization values normalized relative to saturation are shown for full length MEG-3 ( top , green ) , MEG-3IDR ( middle , black ) , and MEG-3Cterm ( bottom , red ) . Values represent averages of ≥3 technical replicates . A fit of the polarization as a function of protein concentration is plotted and used to calculate the given Kd , app . Error bars report S . E . M . DOI: http://dx . doi . org/10 . 7554/eLife . 21337 . 008 To begin to explore the specificity of MEG-3 RNA binding , we challenged MEG-3/poly-U30 complexes with increasing concentrations of competitor RNAs and examined their behavior by EMSA . We found that poly-U , and to a lesser extent poly-A , were effective competitors , but not poly-C or poly-G ( Figure 3—figure supplement 1 ) . These observations suggest that MEG-3’s affinity for RNA is affected by nucleotide composition . Concentrated ( <1 μM ) solutions of RNA-binding proteins containing IDRs spontaneously phase separate when switched from high to low salt ( 150 mM NaCl ) ( Elbaum-Garfinkle et al . , 2015; Lin et al . , 2015; Nott et al . , 2015 ) . We were not able to maintain high concentrations of MEG-3 or MEG-3IDR in solution in the presence of high salt ( Materials and methods ) . Therefore to examine the phase separation properties of MEG-3 , we used concentrated ( 100–320 μM ) preparations maintained in 6M urea , diluted these into aqueous buffer ( 150 mM NaCl ) and used light microscopy to immediately observe the mixture ( Method ) . We found that MEG-3 and MEG-3IDR readily formed phase-separated condensates within 10 min at room temperature . Control proteins ( BSA and MBP ) subjected to the same treatment did not phase separate ( data not shown ) . MEG-3 condensates were observed across a range of protein concentrations ( 0 . 5 μM to 5 μM ) and became larger and more abundant with increasing protein concentration ( Figure 4—figure supplement 1 ) . MEG-3 and MEG-3IDR behaved similarly to each other , except that in the low concentration range ( <5 μM ) , MEG-3 formed more condensates than MEG-3IDR , and in the high concentration range ( >5 μM ) MEG-3IDR tended to form larger condensates ( Figure 4—figure supplement 1 ) . RNA can stimulate the phase transition of IDR proteins that bind RNA ( Guo and Shorter , 2015 ) . To determine the effect of RNA on MEG-3 phase separation , we added poly-U30 RNA to the phase separation buffer before diluting in MEG-3 . We found that 0 . 1 μM poly-U30 was sufficient to increase the number of visible condensates especially at low MEG-3 protein concentrations ( <1 μM ) ( Figure 4A , Figure 4—figure supplement 1 ) . Higher concentrations of RNA increased the number of MEG-3 condensates even further . For a given concentration of poly-U30 , MEG-3 formed more condensates than MEG-3IDR ( Figure 4A , Figure 4—figure supplement 1 ) . Addition of sub-stochiometric amounts of fluorescently tagged poly-U30 confirmed that the RNA phase separates with MEG-3 ( Figure 4—figure supplement 2 ) . We conclude that MEG-3 and MEG-3IDR have an intrinsic propensity to phase separate that can be stimulated by RNA . 10 . 7554/eLife . 21337 . 009Figure 4 . Stimulation of MEG-3 phase separation by RNA . ( A ) Bar graph showing the number of condensates formed by 0 . 5 μM MEG-3 or MEG-3IDR in the presence of increasing poly-U30 . Values represent averages from three technical replicates . Error bars indicate S . E . M . ( B ) Photomicrographs of phase separation assay showing condensate formation of 5 μM full length MEG-3 incubated with poly-U RNA and/or MEX-5 as indicated . ( C ) Violin plots showing condensate size and number for each experiment represented in B . The height of the plot shows the area of condensates in μm2 . The width of the plot correlates to the proportion of condensates of that size . Numbers inside each violin plot are the total number of condensates pooled from three technical replicates for each condition . An additional control is shown in Figure 4—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 21337 . 00910 . 7554/eLife . 21337 . 010Figure 4—figure supplement 1 . Phase Separation assay titration . DIC images of MEG-3 solutions in the presence of increasing concentrations of protein and RNA . MEG-3 was diluted into phase separation buffer ( 25 mM HEPES , pH7 . 4 , NaCl adjusted to a final concentration of 150 mM ) at room temperature . The solution was transferred to a glass bottom dish and the dish surface was photographed using an inverted DIC microscope 10 min after the initial dilution ( room temperature ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21337 . 01010 . 7554/eLife . 21337 . 011Figure 4—figure supplement 2 . Phase Separation assay controls . ( A ) Same as above except 3’ fluorescein labeled poly-U RNA was included in the buffer . DIC images ( left ) show the MEG-3 condensates and fluorescence images ( 488 channel ) show the 3’ fluorescein labeled poly-U RNA concentrated in the condensates . ( B ) Violin plots showing condensate size and number comparing MEG-3 alone and MEG-3 + MEX-5 ( Additional data ) . As in Figure 4C , the height of the plot shows the area of condensates in μm2 . The width of the plot correlates to the proportion of condensates of that size . Numbers inside each violin plot are the total number of condensates pooled from three technical replicates for each condition . Data for the MEG-3 alone sample are the same as in Figure 4C and shown here only for reference . DOI: http://dx . doi . org/10 . 7554/eLife . 21337 . 011 To examine whether MEX-5 can affect MEG-3 phase separation , we purified the MEX-5 RNA-binding domain and C-terminus ( aa236-468 ) as a His fusion ( we were not able to obtain soluble full length MEX-5 , Materials and methods ) . We pre-incubated recombinant MEX-5 with poly-U in buffer for 30 min before adding MEG-3 . We found that MEX-5 strongly inhibited MEG-3 phase separation induced by RNA . Addition of excess RNA ( 3-fold increase ) restored robust phase separation in the presence of MEX-5 ( Figure 4B , Figure 4C ) . Additionally , MEX-5 had no effect on the phase separation of MEG-3 in the absence of RNA ( Figure 4—figure supplement 2 ) . These observations suggest that MEX-5 does not interfere with MEG-3 phase separation directly , but interferes with the ability of poly-U30 to induce phase separation . Our in vitro experiments indicate that MEG-3IDR is sufficient to promote RNA binding and phase separation , but does so less efficiently than full length MEG-3 at low protein concentrations . To examine the behavior of MEG-3IDRin vivo , we deleted the C-terminus of MEG-3 by CRISPR/Cas9 genome editing to generate a meg-3 allele that only expresses MEG-3IDR ( Materials and methods , Supplementary file 1 ) . We found that , like full-length MEG-3 , MEG-3IDR is a cytoplasmic protein that redistributes into a posterior-rich gradient during polarization of the zygote ( Figure 5A ) . Unlike MEG-3 , however , MEG-3IDR did not coalesce into prominent , micron-sized granules in zygotes ( Figure 5A ) . Distinct MEG-3IDR granules were observed starting in the 2-cell stage as MEG-3IDR segregates into the progressively smaller P blastomeres ( Figure 5A ) . In mex-5/6 ( RNAi ) zygotes , MEG-3IDR did not form a gradient and did not form granules ( Figure 5B ) . These observations indicate that MEG-3IDR is partially defective in granule formation , while remaining sensitive to MEX-5/6 . The loss of meg-3 activity was not a result of reduced expression as MEG-3IDR was expressed at greater levels than wild-type MEG-3 ( Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 21337 . 012Figure 5 . Coalescence of MEG-3IDR can be stimulated by blocking mRNA turnover in vivo . ( A ) Photomicrographs of fixed embryos expressing MEG-3IDR tagged with OLLAS epitope . First row are zygotes ( one-cell stage ) and second row are later stage embryos as indicated . ( B ) Photomicrographs of fixed zygotes expressing MEG-3IDR tagged with OLLAS epitope . Genotypes are indicated above each embryo ( left to right: wild-type , mex-5/6 ( RNAi ) , and let-711 ( RNAi ) ) . Numbers indicate number of zygotes exhibiting phenotype shown / total number of zygotes examined . In 1/8 let-711 ( RNAi ) zygotes , MEG-3IDR formed granules but these were smaller and confined to the posterior half of the zygote , possibly due to incomplete depletion of let-711 . DOI: http://dx . doi . org/10 . 7554/eLife . 21337 . 01210 . 7554/eLife . 21337 . 013Figure 5—figure supplement 1 . MEG-3IDR in vivo . Western blot of mixed-stage embryo lysates co-blotted with α OLLAS and α tubulin . Expected sizes for protein-epitope fusions are indicated on the right by a < symbol . Dashed line indicates a break in the original gel . Genotype is indicated on the bottom . The MEG-3IDR is present at higher level than MEG-3 , likely due to the fact that unlike MEG-3 , MEG-3IDR persists longer in somatic cells ( see Figure 5 and data not shown ) . The significance of this difference is not known . n = 3 technical replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 21337 . 013 In vitro , the weaker phase separation properties of MEG-3IDR at low protein concentrations can be stimulated by RNA . To determine whether excess RNA could also rescue granule formation by MEG-3IDR in zygotes , we blocked maternal mRNA turnover by depleting LET-711 by RNAi . LET-711 is the scaffolding component of the CCF/NOT1 deadenylase , the main deadenylase that promotes mRNA turnover in oocytes and early embryos ( DeBella et al . , 2006; Nousch et al . , 2013 ) . Depletion of LET-711 has been shown to increases poly-adenylation and to block the turnover of maternal nos-2 RNA in early embryos ( Gallo et al . , 2008; Nousch et al . , 2013 ) . We found that MEG-3IDR formed numerous micron-sized granules in let-711 ( RNAi ) zygotes . The MEG-3 IDR granules and cytoplasmic gradient extended further towards the anterior compared to wild-type ( Figure 5B ) . These observations suggest that , as we observed in vitro , excess RNA can overcome the inhibitory effects of MEX-5 and boost MEG-3 coalescence in vivo . MEG-3 contains a long N-terminal IDR but no recognizable RNA-binding domain . We have found that MEG-3 binds RNA ( poly-U30 ) with nanomolar affinity in vitro ( Kd , app = 32 nm ) . The MEG-3IDR is essential for binding , but on its own binds with lower affinity ( Kd , app = 460 nm ) . One possibility is that the MEG-3 IDR extends beyond the region predicted by IUPRED ( Dosztányi et al . , 2005 ) . The region immediately C-terminal scores close to the IUPRED cut-off ( Wang et al . , 2014 ) , and may contribute to RNA binding . IDRs are over-represented among RNA-binding domains ( Varadi et al . , 2015; Castello et al . , 2016 ) . Electrostatic interactions between positively-charged amino acids and the negatively-charged RNA backbone are often invoked as a possible mechanism for RNA binding by IDRs ( Guo and Shorter , 2015; Basu and Bahadur , 2016 ) . MEG-3 is rich in basic residues , but shows a strong preference for poly-U over poly-C and poly-G , suggesting that non-charged interactions are also involved . Several recent studies have demonstrated that RNA-binding proteins containing IDRs phase separate in aqueous solutions ( Guo and Shorter , 2015 ) . MEG-3 follows this paradigm: MEG-3 readily formed condensates within minutes of dilution from urea to an aqueous buffer ( 150 mM NaCl ) . MEG-3 phase separation could be stimulated by RNA: addition of poly-U30 to the phase separation buffer increased the number of MEG-3 condensates especially at 1 μM and lower protein concentrations . MEG-3IDR behaved similarly to full-length MEG-3 , except that MEG-3IDR required higher concentrations of RNA to phase separate at low protein concentrations . Consistent with this in vitro behavior , MEG-3IDR did not form large granules in wild-type zygotes , but could be induced to do so in zygotes defective in mRNA deadenylation and turnover ( RNAi depletion of the LET-711/NOT1 ) . These observations suggest that the IDR confers on MEG-3 an intrinsic tendency for phase separation that is tunable by RNA . RNA-induced phase separation has also been observed for Whi3 , a fungal RNA-binding protein , and for hnRNPA1 , a stress granule protein ( Lin et al . , 2015; Zhang et al . , 2015 ) . In these proteins , the IDR and RNA-binding domain are distinct and RNA-induced phase separation requires both domains . It will be interesting to determine whether the MEG-3IDR in fact contains separable domains for phase separation and RNA binding . MEX-5 has been hypothesized to regulate P granule asymmetry by creating a supersaturation gradient of critical granule component ( s ) along the anterior-posterior axis of the zygote ( Lee et al . , 2013 ) . Our observations suggest that the critical component regulated by MEX-5 is RNA . MEX-5 binds RNA with nanomolar affinity ( Kd , app= ~ 29 nM , Pagano et al . , 2007 ) and is 10-fold more abundant than MEG-3 in embryos ( Figure 2—figure supplement 1 ) . In our in vitro phase separation assay , the MEX-5 RNA-binding domain was sufficient to inhibit RNA-induced phase separation of MEG-3 . Similarly , in vivo , uniform MEX-5 was sufficient to inhibit MEG-3 granule assembly throughout the cytoplasm and this activity was disrupted by mutations that lower MEX-5’s affinity for RNA . Together , these results support a model where MEX-5 suppresses MEG-3 phase separation by limiting MEG-3’s access to RNA . How MEX-5 regulates MEG-3’s access to RNA in vivo remains to be determined . As suggested by the in vitro observations , MEX-5 could compete directly with MEG-3 for RNA binding . Alternatively , MEX-5 could function indirectly by recruiting factors to MEG-3/RNA complexes that reduce MEG-3’s affinity for RNA . The MEX-5 gradient arises as a consequence of phosphorylation by the posteriorly-enriched kinase PAR-1 , which decreases the size of MEX-5 complexes and increases MEX-5’s diffusion rate ( Griffin et al . , 2011 ) . One possibility is that phosphorylation by PAR-1 prevents MEX-5 from binding RNA creating a pool of ‘MEX-5-free’ RNA available to phase separate with MEG-3 in the posterior cytoplasm . While MEX-5 was sufficient to inhibit the RNA-induced phase separation of MEG-3 in vitro , it was not able to reverse it . Addition of MEX-5 to preformed MEG-3 condensates was not sufficient to dissolve them ( data not shown ) . In contrast , in vivo , MEG-3 granules dissolve in the anterior cytoplasm as MEX-5 concentration increases ( Figure 1A ) . We showed previously that , in vivo , P granule dynamics are also regulated by phosphorylation . The MBK-2 kinase and the PP2A phosphatase promote granule disassembly and assembly , respectively ( Wang et al . , 2014 ) . One possibility is that cycles of MEG phosphorylation and de-phosphorylation promote cycles of granule disassembly and re-assembly throughout the cytoplasm . By limiting RNA access , MEX-5 inhibits the re-assembly of MEG-3 condensates in the anterior cytoplasm freeing more MEG-3 for granule assembly in the posterior cytoplasm . MEX-5’s high affinity for poly-U stretches , which are present in 91% of C . elegans 3’ UTRs ( Pagano et al . , 2007 ) , suggests that MEX-5 interacts with most mRNAs in zygotes and thus could function as a general ‘mRNA sink’ . Consistent with this view , the MEX-5 gradient patterns the distribution of three other RNA-binding proteins that , like MEG-3 , form posterior-rich gradients ( Wu et al . , 2015 ) . The observation that blocking mRNA turnover stimulates MEG-3 coalescence into macroscopic granules is consistent with the idea that the mRNA pool accessible to MEG-3 is limiting in zygotes . A limiting mRNA pool has also been suggested to regulate the balance of P bodies and stress granules in cells ( Buchan and Parker , 2009 ) . We propose that regulated access to mRNAs , combined with RNA-induced phase separation of key scaffolding proteins , may be a general mechanism for controlling the formation of RNA granules in space and time . A recent theoretical study has suggested that MEX-5 patterns P granules by regulating the phase separation of a different P granule protein: PGL-3 ( Saha et al . , 2016 ) . This model is based on the assumption that PGL-3 is essential to assemble P granules and on three in vitro observations: ( 1 ) PGL-3 binds RNA using a C-terminal RGG domain , ( 2 ) RNA-binding via the RGG domain stimulates phase separation of PGL-3 and ( 3 ) MEX-5 can interfere with the formation of PGL-3/RNA condensates . Using estimates for the in vivo concentrations and diffusion rates of PGL-3 , MEX-5 and RNA , Saha et al . built a mathematical model that simulates a competition for RNA between MEX-5 and PGL-3 . In the simulation , introduction of the MEX-5 gradient is sufficient to dissolve PGL-3 granules in the anterior cytoplasm . This model is inconsistent with several in vivo observations . First , Hanazawa et al . ( 2011 ) have reported that PGL-3 lacking the RGG domain still forms asymmetric granules in zygotes . Second , the formation of the MEX-5 gradient does not correlate temporally with , and is not sufficient for , PGL-3 asymmetry in vivo ( Figure 1—figure supplement 1 ) . Third , PGL-3 and its paralog PGL-1 are not required to assemble or localize MEG granules in vivo ( Figure 1C , Figure 1—figure supplement 1 ) . The in vivo data demonstrate that , contrary to the model proposed by Saha et al . ( 2016 ) , PGL-3 does not require RNA binding to localize to the granules , does not sense the MEX-5 gradient directly , and does not drive granule assembly or asymmetry . The in vivo data indicate instead that MEG-3/4 are responsible for P granule asymmetry and recruitment of PGL-3 to P granules . In vitro , MEG-3 binds to PGL-1 and PGL-1 binds to PGL-3 ( Hanazawa et al . , 2011; Wang et al . , 2014 ) , raising the possibility that the MEGs recruit the PGLs by direct protein-protein interactions . We showed previously that MEG-3 overlaps but does not co-localize perfectly with PGL-3 in P granules ( Wang et al . , 2014 ) . Formation of stable PGL assemblies in embryos also requires LAF-1 , a DEAD-box RNA helicase that phase separates in vitro independently of RNA ( Elbaum-Garfinkle et al . , 2015 ) . P granules therefore appear to comprise multiple phases , each with distinct properties and components that have an affinity for one another but do not fully mix . We suggest that MEG-3 and MEG-4 form the main RNA-dependent phase of P granules . The MEG phase functions as a scaffold that recruits other P granule components to build multi-phasic assemblies . An important question will be to determine the physical and biochemical properties of the MEG phase . The conclusion that P granules are liquid droplets was based mostly on observations of the PGL phase ( Brangwynne et al . , 2009 ) . Whether the MEG phase is also liquid or is a more ordered gel-like or solid condensate remains to be determined . How the MEG scaffold specifies the unique RNA/protein composition of P granules will be an exciting area for future inquiry . C . elegans was cultured according to standard methods at 20°C ( Brenner , 1974 ) . Genome editing was performed using CRISPR/Cas9 as described in Paix et al . ( 2015 ) . Alleles used in this study are listed in Supplementary file 1 . RNAi knock-down experiments were performed by feeding on HT115 bacteria ( Timmons and Fire , 1998 ) . Feeding constructs were obtained from the Ahringer or Openbiosystems libraries and transformed into HT115 bacteria . pL4440 was used as a negative control empty feeding vector . Bacteria were grown at 37°C in LB + ampicillin ( 100 µg/mL ) for 5 hr , induced with 5 mM IPTG for 45 min , plated on NNGM ( nematode nutritional growth media ) + ampicillin ( 100 µg/mL ) + IPTG ( 1 mM ) , and grown overnight at room temperature . Embryos isolated by bleaching from gravid hermaphrodites were added to the RNAi plates and transferred to fresh plates as L4 larvae before examination of their progeny . All RNAi experiments were performed at 20°C . All purifications were performed using an AKTA pure FPLC protein purification system ( GE Healthcare Silver Spring , MD ) . Adult worms were placed into M9 salt solution on epoxy autoclavable slides ( thermo-fisher Waltham , MA ) and squashed with a coverslip to extrude embryos . Slides were frozen by laying on pre-chilled aluminum blocks for 20 min ( chilled using dry ice ) . Embryos were permeabilized by freeze-cracking ( removal of coverslips from slides ) followed by incubation in methanol at −20°C for >15 min , and in acetone ( pre-chilled at −20°C ) at room temperature for 10 min . Slides were blocked in PBS-Tween ( 0 . 1% ) BSA ( 0 . 5% ) for 15 min x 2 , and incubated with 50 ul primary antibody overnight at 4°C in a humid chamber . Antibody dilutions ( in PBST/BSA ) : K76 ( 1:10 , DSHB ) , Rat α OLLAS-L2 ( 1:200 , Novus Biological Littleton , CO ) , mouse α FLAG ( 1:500 , Sigma St . Louis , MO ) . Secondary antibodies were applied for 2 hr at room temperature . Fluorescence microscopy was performed using a Zeiss Axio Imager with a Yokogawa spinning-disc confocal scanner . Images were taken and stored using Slidebook v 6 . 0 software ( Intelligent Imaging Innovations ) using a 63x objective . For live imaging , embryos were dissected from adult hermaphrodites in M9 salt solution and mounted onto 3% agarose pads . All embryo images are z stack maximum projections using a z step size of 1 μm , spanning the entire width of the embryo . Equally normalized time-lapse images were quantified using Slidebook v 6 . 0 . Average fluorescence intensity relative to area of anterior ( 60% ) and posterior ( 40% ) of zygote were quantified and average fluorescence intensity relative to area of background ( outside of zygote ) was subtracted from each of these values . For each time-point , anterior and posterior fluorescence were expressed as fractions of total fluorescence and then normalized to T0 ( 14 min prior to mitosis ) . Final values represent average of three embryos . Error bars show standard deviation of the mean . EMSA were carried out as described in Pagano et al . ( 2007 ) . Reactions consisted of 50 nM 3’ Fluorescein-labeled RNA oligonucleotides ( Dharmacon -GE Lifesciences Silver Springs , MD ) incubated with protein for 2 hr or more at room temperature or for 30 min followed by 2 hr incubation with unlabeled competitor . Samples were run on 1% agarose gel in 1x TAE . Gels were scanned immediately using typhoon FLA-9500 with blue laser at 473 nm . Equilibration reactions were performed using same protocol as for EMSAs . Reactions were transferred to 384 well microplates ( Greiner Bio-One Monroe , NC ) . The apparent fluorescence polarization was determined using a Clariostar monochromator microplate reader with fluorescein-sensitive filters and polarizers . Polarization values were normalized relative to saturation polarization value . For each experiment , values of three reads were averaged . Average values and standard errors from at least three technical replicates were calculated and plotted against each protein concentration . These data were fit to a quadratic equation ( Equation 1 , where b is the base polarization , m is the maximum polarization , R is the labeled nucleic acid concentration , and P is the total protein concentration ) as in ( Pagano et al . , 2011 ) , to calculate the apparent dissociation constant . The reported values are the dissociation constants calculated using the polarization values averaged from all technical replicates . The reported errors are standard error values calculated from the dissociation constant of each individual technical replicate . ( 1 ) ϕ=b+ ( m−b ) ×[R+P+Kd , app− ( R+P+Kd , app ) 2− ( 4RP ) 2R] His::tagged MEG-3 fusions were quickly diluted out of urea into condensation Buffer ( 25 mM HEPES , pH7 . 4 , NaCl adjusted to a final concentration of 150 mM ) in the presence and absence of poly-U30 RNA . Dilutions were performed by adding buffer to protein in low-binding siliconized Eppendorf tubes and mixing briefly by pipetting . The reaction was either spun at 3000 rpm for 2 min or transferred directly into 35 mm glass bottom dish ( Cat . No . P35G-1 . 5–14 C MatTek Corp Ashland , MA ) for imaging . Phase separation assays with MEX-5 were performed by pre-incubating 3 . 5 μM 6xHis::MBP::MEX-5 protein ( dialyzed into condensation buffer ) with condensation buffer and poly-U30 RNA for 30 min before diluting in MEG-3 . Differential interference contrast ( DIC ) images were obtained on an Olympus inverted microscope , using a 100X objective . Images were taken and stored using Slidebook v 4 . 0 and 5 . 0 . All images are a single focal plane focused on the slide surface . For each phase separation experiment , we took three separate images of an 80×80 micron field and counted the condensates using Image J64 . To recognize condensates , background was subtracted from each image using a rolling ball radius of 10 pixels , a pixel brightness threshold was set to 15–255 . Remaining pixels were smoothed three times and size and number of objects greater than 0 . 032 μm2 were quantified . Quantifications were manually verified for each image used . At least three technical replicates were quantified for each condition . Average number of condensates and S . E . M were calculated using all technical replicates . Western blots were performed by running worm lysates on 7% Tris Acetate SDS PAGE precast gels ( Bio-Rad Hercules , CA ) . Protein was transferred to nitrocellulose membrane which was pre-blocked in 5% Milk diluted in PBS-Tween ( 0 . 1% ) for 5 min ( three times ) . The membrane was then incubated with Primary antibody for at least 18 hr at 4° C or 2 hr at room temperature . Membranes were washed and blocked in 5% milk for 5 min ( three times ) and incubated with secondary HRP conjugated antibody for 45 min at room temperature . Membranes were washed in 5% milk for 5 min ( two times ) and PBST for 5 min ( one time ) . The membranes were then exposed to ECL substrate for 1 min and then exposed to film . Primary antibody dilutions ( in 5% Milk PBST ) : Rat α OLLAS-L2 ( 1:1000 , Novus Biological Littleton , CO ) , Mouse α Tubulin ( 1:1000 , Sigma St . Louis , MO ) . All in vivo biological replicates refer to experiments performed on independently isolated hermaphrodites ( in the case of mutants , this refers to separate strains isolated from independent editing events ) or independently treated hermaphrodites ( in the case of RNAi , this refers to wild-type worms exposed to independent RNAi treatments ) . In vitro biological replicates refer to experiments performed with independently purified protein preps . All in vivo technical replicates refer to observations in the same strain from separate zygotes . In vitro technical replicates refer to separate experiments performed using the same purified protein preps .

Animal cells contain many smaller compartments known as organelles that perform particular roles . For example , a compartment called the nucleus stores most of the cell’s genetic information . The nucleus and many other organelles form inside layers of membrane that physically separate them from the rest of the cell . However , some organelles , such as the germ granule , do not have a membrane . It is thought that these organelles may form in the same way that oil droplets tend to come together when mixed with water . However , oil droplets form in water spontaneously and do not fall apart , so it is not clear how cells could control the assembly and destruction of such organelles . The germ granules inside the cells of a worm called C . elegans are destroyed and reassembled in cycles . Smith et al . investigated how the worm cells control these cycles . The experiments show that a protein called MEG-3 is required to allow the components of granules to transition from behaving like individual molecules dissolved in water ( similar to being dissolved in cell fluid ) to assembling into droplets . When MEG-3 is mixed with molecules of ribonucleic acid ( RNA ) it can bind very tightly to the RNA and then separate out from the rest of the fluid to form distinct droplets . Smith et al . also show that another protein called MEX-5 can destroy these droplets by attaching itself to RNA in place of MEG-3 , which causes MEG-3 to dissolve back into the rest of the fluid . The physical properties of the MEG-3 droplets are still not known and so the next step following on from this work will be to find out whether germ granules behave like liquids , gels or hard solids .

2016

Spatial patterning of P granules by RNA-induced phase separation of the intrinsically-disordered protein MEG-3

Healthy pregnancy depends on proper placentation—including proliferation , differentiation , and invasion of trophoblast cells—which , if impaired , causes placental ischemia resulting in intrauterine growth restriction and preeclampsia . Mechanisms regulating trophoblast invasion , however , are unknown . We report that reduction of Inverted formin 2 ( INF2 ) alters intracellular trafficking and significantly impairs invasion in a model of human extravillous trophoblasts . Furthermore , global loss of Inf2 in mice recapitulates maternal and fetal phenotypes of placental insufficiency . Inf2−/− dams have reduced spiral artery numbers and late gestational hypertension with resolution following delivery . Inf2−/− fetuses are growth restricted and demonstrate changes in umbilical artery Doppler consistent with poor placental perfusion and fetal distress . Loss of Inf2 increases fetal vascular density in the placenta and dysregulates trophoblast expression of angiogenic factors . Our data support a critical regulatory role for INF2 in trophoblast invasion—a necessary process for placentation—representing a possible future target for improving placentation and fetal outcomes . Implantation and placentation involve complex synchronization between the developing embryo and decidualization of the uterus . Extravillous trophoblasts ( EVTs ) differentiate from column cytotrophoblasts ( CTBs ) , invade through the endometrium to the myometrium , and remodel decidual spiral arteries to form high-capacity , low-resistance vessels , supplying maternal blood to the lacunae surrounding the developing placental villi ( Damsky et al . , 1992; Red-Horse et al . , 2004 ) . Shallow invasion by EVTs and failed spiral artery remodeling yield peripheral vasoconstriction and high-resistance vessels thought to comprise the first stage of the development of preeclampsia ( PE ) . Together with reduced arterial compliance , these vascular changes result in the hypertensive phenotype characteristic of this disease ( Bosio et al . , 1999; Wolf et al . , 2001 ) . The cause of shallow EVT invasion is unknown and under-investigated due to a lack of relevant animal models ( McCarthy et al . , 2011 ) . For example , the reduced uterine perfusion pressure ( RUPP ) rat model of PE—which recapitulates systemic changes in maternal renal , immune , and circulatory functions—necessitates physical occlusion of the abdominal aorta and uterine arteries ( Alexander et al . , 2001; Li et al . , 2012 ) . The resulting placental ischemia begins at midgestation , well after artery remodeling . Several mouse models have been developed to understand the pathophysiology of this disease ( McCarthy et al . , 2011 ) such as the nitric oxide synthase knockout mouse ( Huang et al . , 1993; Huang et al . , 1995; Shesely et al . , 2001 ) , the catechol-O-methyltransferase deficient mouse ( Kanasaki et al . , 2008 ) , and the glial cells missing hypomorphic mouse ( Bainbridge et al . , 2012 ) . These models recapitulate aspects of PE but the etiology of shallow EVT invasion , the early cause of placental ischemia , is still unknown . Successful cellular invasion depends on formation of invasive structures such as invadopodia and podosomes ( Parast et al . , 2001; Patel and Dash , 2012 ) , suggesting cytoskeletal integrity and appropriate remodeling is critical for EVT migration and invasion . Formins , a multi-domain family of proteins highly expressed in reproductive tissues ( Figure 1 ) , have been identified as critical in the regulation of cytoskeletal assembly and organization through actin polymerization and microtubule bundling , mediating processes such as cellular migration , division , and intracellular transport ( Antón et al . , 2008; Antón et al . , 2011; Chhabra and Higgs , 2007; Gaillard et al . , 2011; Higgs and Peterson , 2005; Ness et al . , 2013; Pollard , 2007; Schönichen and Geyer , 2010 ) . Phylogenetic analyses indicate that the structure of these proteins is highly conserved ( Figure 1—figure supplement 1 ) and examination of evolutionary rates of mammalian formins showed no evidence of positive selection acting on either the INF clade or the INF2 clade ( Table 1 ) . Several formin family genes have previously been associated with pregnancy and reproductive phenotypes , including preterm birth ( Cruickshank et al . , 2013; Elovitz et al . , 2014; Montenegro et al . , 2009 ) . Furthermore , there is evidence of increased expression of a formin activator , RhoA-GTP , during pregnancy ( Hudson and Bernal , 2012 ) . Inverted formin 2 is unique among the formins due to its ability to sever and depolymerize actin filaments in addition to traditional formin functions such as microtubule bundling and actin polymerization ( Chhabra and Higgs , 2006; Ramabhadran et al . , 2012 ) . Severing and depolymerization of actin filaments allows INF2 to generate highly transient filaments ( Chhabra et al . , 2009 ) . Transient activation of cofilin—one known regulator of actin filament assembly and disassembly—has been shown to be important in stimulated cell motility ( Yamaguchi and Condeelis , 2007 ) , suggesting tight regulation of actin dynamics may be vital to extravillous trophoblast invasion . Furthermore , polymerization of actin filaments by INF2 is important for mitochondrial fission , a process that may be important in regulating trophoblast metabolism ( Burton et al . , 2017; Korobova et al . , 2013 ) . Importantly , INF2 is necessary for intracellular transport—responsible for mobilizing cargo such as SRC kinases , which are responsible for EVT degradation of extracellular matrix ( Patel and Dash , 2012 ) and invasiveness . One SRC-like tyrosine kinase trafficked by INF2 , lymphocyte-specific protein tyrosine kinase ( LCK ) , has previously been shown to play a role in tumor metastasis ( Andrés-Delgado et al . , 2010; Mahabeleshwar and Kundu , 2003 ) —perhaps one of the many reasons EVT invasion is frequently compared to the metastatic invasion in cancer . Given the central role of formin proteins in reproduction , actin cytoskeleton dynamics , and the intracellular transport of LCK ( Andrés-Delgado et al . , 2010 ) , we hypothesized that INF2 is necessary for successful trophoblast invasion and vascular remodeling . Inf2-deficient mice demonstrated impaired spiral artery remodeling , hypertension , fetal growth restriction , and altered placental development , identifying the Inf2 null mouse as a novel model of placental insufficiency . INF2 targeted siRNA successfully reduced expression in an in vitro model of human EVTs ( HTR-8/SVneo; Figure 2A and B; p=0 . 0046 ) . Reduction of INF2 in these cells did not impact phalloidin content ( Figure 2—figure supplement 1A; p=0 . 58 ) , however , mitochondrial volume was significantly increased ( Figure 2—figure supplement 1B and C; p=0 . 0048 ) . INF2 knockdown impaired invasion of HTR-8/SVneo cells by 73% compared to nonsense siRNA- and vehicle-treated cells ( Figure 2C; p=0 . 0005 ) . To determine if INF2 is necessary for transcytosis in EVTs as in other cells ( Andrés-Delgado et al . , 2010; Madrid et al . , 2010 ) , we visualized intracellular localization of MAL2 and LCK in vehicle- or knockdown siRNA-treated HTR-8/SVneo cells . MAL2 is dispersed throughout the cytoplasm in vehicle-treated HTR-8/SVneo cells with no change in localization after INF2 knockdown ( Figure 2—figure supplement 2 ) . Reduction of INF2 restricted LCK to the perinuclear region of cultured HTR-8/SVneo cells while LCK was distributed throughout the cytoplasm in controls ( Figure 2D ) . There was no change in LCK protein expression ( Figure 2B ) . Treatment of HTR-8/SVneo cells with the LCK/FYN-specific inhibitor PP1 reduced invasion by 69% ( Figure 2E; p=0 . 011 ) while the SRC inhibitor TX1123 reduced invasion by 98% compared to controls ( Figure 2E; p=0 . 0007 ) . Inf2 expression in C57Bl/6J placentas was significantly increased at gestational day 15 . 5 ( E15 . 5; Figure 3A; p=0 . 026 ) compared to E13 . 5 and E18 . 5 placentas . By E18 . 5 , Inf2 mRNA returned to earlier pregnancy levels . We demonstrate dense , specific staining of trophoblast cells throughout the labyrinth , junctional zone , and decidua in control mice and none in the knockout mice ( Figure 3B ) . Lck is restricted to the perinuclear region in Inf2−/− trophoblasts while it is distributed throughout the cytoplasm in Inf2+/+ trophoblasts ( Figure 3C ) . Immunofluorescence staining revealed co-localization of Inf2 with the pan-trophoblast marker cytokeratin-7 ( Ck7 ) and the trophoblast giant cell ( TGC ) marker proliferin ( Figure 3D ) . We visualized lectin-labeled maternal spiral arteries in cleared , depth-coated placentas . Fully extended spiral artery numbers were counted in placentas rendered in 3D ( Videos 1 and 2 ) . At E19 . 0 , the number of spiral arteries in Inf2−/− placentas was significantly reduced compared to wildtype placentas ( Figure 4A and B; p=0 . 023 ) . Using the volumetric pressure cuff system to monitor blood pressure changes throughout pregnancy , systolic blood pressure dropped from pre-pregnancy levels at E15 . 5 in all females . In contrast , at E17 . 5 blood pressure was significantly elevated in Inf2−/− females compared to Inf2+/+ ( Figure 4C; p=0 . 012 ) . By postnatal day 2 ( P2 ) , the systolic blood pressure of all females was comparable to pre-pregnancy levels . No significant differences in total urinary protein were measured in non-gravid females ( n = 6; 26984 ± 2936 vs 29428 ± 3441 ng/μL ) or females at E17 . 5 ( n = 3 , 4; 33834 ± 4644 vs 34727 ± 7028 ng/μL; data not shown ) . As maternal hypertension in pregnancy may result from abnormal placental production of angiogenic factors , we measured these levels in serum . Despite a trend of higher placental growth factor-2 ( Plgf-2 ) in the maternal circulation of Inf2−/− females at E15 . 5 ( Figure 4—figure supplement 1; p=0 . 16 ) , no significant differences were detected in either Plgf-2 ( p=0 . 16 , 0 . 97 ) or FMS-like tyrosine kinase 1 ( Flt1; p=0 . 90 , 0 . 87 ) levels at E15 . 5 or E18 . 5 . To evaluate the significance of Inf2 in gestation , we compared pregnancy outcomes in Inf2+/+ and Inf2−/− mice . Gestation length was increased by 9 . 8 hr in Inf2−/− mice ( Figure 5A; p=0 . 009 ) with no impact on pup weight at birth ( p=0 . 96 ) , litter size ( p=0 . 83 ) , or total litter weight ( Figure 5B and Figure 5—figure supplement 1A and B; p=0 . 51 at E18 . 5 , 0 . 31 at p0 ) . Despite extended gestational length , there were no detectable differences in serum progesterone ( p=0 . 64 ) , uterine prostaglandins F2α ( p=0 . 64 ) and E2 ( p=0 . 99 ) , or oxytocin receptor mRNA expression at E18 . 5 ( Figure 5—figure supplement 2A–D; p=0 . 33 ) ( Bezold et al . , 2013 ) . While normal weight at birth , fetal weight at E18 . 5 was significantly reduced in Inf2−/− pups compared to Inf2+/+ pups ( Figure 5B; p=0 . 020 ) . There were no differences in placental weight ( Figure 5C; p=0 . 82 ) ; however , the ratio of fetal to placental weight was significantly reduced in Inf2−/− mice ( Figure 5D; p=0 . 019 ) . Previous studies showed that altered fetal growth in late pregnancy is preceded by changes in placental nutrient transport ( Jansson et al . , 2006 ) , however , there were no detectable differences in mRNA expression of the amino acid or glucose transporters studied here at E18 . 5 ( Figure 5—figure supplement 3A–D; p=0 . 19–0 . 65 ) . Altered end-diastolic flow and pulsatility index may indicate the presence of intrauterine growth restriction ( IUGR ) and/or PE ( Bond et al . , 2015; Krebs et al . , 1996; Turan et al . , 2008 ) . To assess vascular capacity and placental function , we performed umbilical artery and vein Doppler in pregnant Inf2+/+ and Inf2−/− dams at E18 . 5 ( Figure 6A ) . End-diastolic velocity ( EDV ) and pulsatility index ( PI ) were significantly elevated in Inf2−/− fetuses ( Figure 6B and C; p=0 . 045 and 0 . 022 ) , with no significant differences in resistance index ( p=0 . 33 ) , peak systolic velocity ( p=0 . 12 ) , or fetal heart rate ( Figure 6—figure supplement 1A–C; p=0 . 06 ) . Moreover , some umbilical vein waveforms appeared pulsatile ( Figure 6A ) . Fetal vascular density in the labyrinth of placentas ( Figure 6D ) at E18 . 5 was significantly higher in Inf2−/− placentas ( Figure 6E; p=0 . 018 ) and the proportion of placental depth consisting of labyrinth but not junctional zone was significantly reduced in Inf2−/− compared to Inf2+/+ placentas ( Figure 6F and G; p=0 . 030 and 0 . 105 ) . To determine if INF2 regulates angiogenic factor expression , we utilized an in vitro model of the crosstalk between CTBs ( BeWo choriocarcinoma ) with reduced INF2 mRNA expression ( Figure 7A; p<0 . 0001 ) and human placental microvascular endothelial cells ( HPMVECs ) . Knockdown of INF2 significantly increased PGF mRNA in the BeWo cell line ( Figure 7B; p=0 . 0007 ) . HPMVEC exposure to cultured media significantly increased soluble vascular endothelial growth factor receptor type 1 ( sVEGFR-1; sFLT1 ) mRNA ( Figure 7D; p=0 . 0083 ) in response to INF2 deficiency; therefore , we hypothesized that PGF protein secreted by INF2-knockdown BeWo cells would also be increased and underlie the sFLT1 response in the HPMVECs . Knockdown of INF2 significantly upregulated secretion of PGF compared to vehicle-treated cells ( Figure 7C; p=0 . 037 ) . PGF secretion by nonsense siRNA-treated cells , however , did not differ significantly from vehicle-treated cells ( Figure 7C; p=0 . 39 ) . Global loss of Inf2 in mice , however , did not change placental Pgf or sFlt1 mRNA levels in vivo ( data not shown ) . Establishment of a healthy pregnancy is dependent on proper embryo implantation and the differentiation , invasion , and communication of trophoblast cells with the uterine milieu—processes that continue throughout gestation as the placenta develops and grows ( Kokkinos et al . , 2010 ) . Alteration in CTB differentiation disrupts placental architecture ( Cross , 2005 ) and changes in placental vascularization results in placental insufficiency . At E18 . 5 , Inf2-deficient fetuses were significantly growth restricted and , interestingly , the ratio of fetal weight to placental weight was significantly reduced in these mice—consistent with IUGR in human pregnancies as a result of inefficient placentas ( Hayward et al . , 2016 ) . Unlike the human situation , birth weights of growth-restricted Inf2−/− pups did not differ from wildtype pups , which we hypothesize is due to the increased gestation length in Inf2−/− animals allowing these growth restricted fetuses to catch up in weight prior to birth . We did not detect change in placental nutrient transporter mRNA expression at E18 . 5 , however previous studies have demonstrated that alterations in placental nutrient transporters may occur prior to changes in birth weight and not concurrently ( Jansson et al . , 2006 ) . Therefore , it is possible that Inf2 may play a role in nutrient transporter localization and should be interrogated in future studies . Umbilical ultrasound at E18 . 5 revealed significant increases in end-diastolic velocity and pulsatility index in Inf2−/− fetuses , consistent with IUGR and poor placental function . A reduction in labyrinth depth in Inf2−/− placentas with a significant increase in fetal vessel density confirms aberrant placental architecture and placental vasculopathy . Placental vascularization and development depends on both autocrine and paracrine signaling between trophoblasts and endothelial cells ( Charnock-Jones and Burton , 2000; Ong et al . , 2000; Troja et al . , 2014 ) . Reduction of INF2 in BeWo cells is sufficient both increase endogenous PGF expression and secretion , indirectly increasing endothelial sFLT1 mRNA expression . These data suggest INF2 is necessary for angiogenic balance in trophoblasts , however , further studies to determine the precise mechanisms by which INF2 modulates expression and secretion of these factors in trophoblasts are required . While global loss of Inf2 in vivo did not reflect this , we posit that cell-specific differences that are easily ascertained in cell culture are masked by the in vivo milieu . Differentiation and invasion of EVTs is critical for normal spiral artery remodeling and placentation ( Moffett-King , 2002 ) . Shallow invasion by EVTs is thought to underlie failed remodeling , the primary placental insult leading to ischemia and subsequent development of PE and maternal hypertension . We report that loss of Inf2 significantly reduced the number of fully extended maternal spiral arteries , suggesting shallow invasion and remodeling by TGCs . Like other models of PE , Inf2−/− mice display new onset of maternal hypertension in late pregnancy that resolves after delivery—consistent with the peripheral vasoconstriction and reduced arterial compliance seen in patients with PE . However , we did not detect any differences in total urinary protein or serum pro- and anti-angiogenic factors in mid- or late gestation in Inf2−/− mice as reported in patients clinically ( Tsatsaris et al . , 2003 ) . Proteinuria may be masked in our transgenic line , however , as C57BL/6 mice are resistant to developing proteinuria ( Ishola et al . , 2006 ) . Women with PE and severe IUGR commonly deliver preterm due to clinical intervention to avoid maternal and fetal demise; therefore , it is unknown whether the length of gestation would naturally be increased in these women as it is in our mouse model to allow for continued growth , resulting in an average weight at birth . These data suggest a limitation—that despite shallow invasion , limited spiral artery remodeling , and hypertension in late gestation , our model may not fully recapitulate human PE but may represent a model of placental insufficiency with maternal hypertension . Successful tumor cell invasion is dependent on the generation of invasive actin-rich structures ( Parast et al . , 2001; Patel and Dash , 2012 ) , such as invadopodia . Formation of invadopodia involves cytoskeletal remodeling , suggesting that this process may also be essential for EVT invasion . INF2 is a cytoskeletal modulator that is important for regulating cell polarity , mitochondrial fission , intracellular trafficking , as well as cell and tissue morphogenesis ( Chhabra and Higgs , 2006; Chhabra et al . , 2009; Goode and Eck , 2007; Madrid et al . , 2010 ) . Despite its major function in regulating cellular actin dynamics , reduction of INF2 did not alter overall cytoskeletal F-actin as determined by comparing cytoplasmic phalloidin content in EVTs . In functional assays , this reduction significantly altered invasion , suggesting EVT invasion is dependent on INF2 expression but independent of cytoskeletal F-actin maintenance . Pharmacologic inhibition of the SRC-like tyrosine kinase LCK—which is dependent on INF2 for proper intracellular trafficking ( Andrés-Delgado et al . , 2010 ) —similarly reduced invasion . Consistent with published results in human T lymphocytes ( Andrés-Delgado et al . , 2010 ) , reduction of INF2 in HTR-8/SVneo cells restricted LCK to the perinuclear region , which was reflected in Inf2−/− placentas . Together , these data suggest a novel model by which EVT invasion is mediated by INF2-dependent targeting of LCK to the plasma membrane ( Figure 8 ) . Loss of Inf2 and the subsequent restriction of Lck in the cytoplasm would inhibit TGC invasion and spiral artery remodeling in the mouse , leading to maternal hypertension late in gestation . Consistent with previous studies , loss of INF2 significantly increased mitochondrial size in vitro ( Korobova et al . , 2013 ) . Our data , therefore , supports a role for INF2 in mitochondrial fission . Studies suppressing dynamin-related protein 1 ( Drp1 ) —a fission protein whose recruitment to mitochondria is facilitated by INF2 ( Jin et al . , 2017; Korobova et al . , 2013 ) —in breast cancer cells inhibited formation of lamellipodia and suppressed their migration and invasion capabilities ( Zhao et al . , 2013 ) . However , as knockdown of Drp1 results in a significantly more severe mitochondrial phenotype , we support the suggestion put forth by Korobova et al . that there are INF2-independent pathways in mitochondrial fission . Therefore , we surmise that INF2-dependent trophoblast invasion seen in our study is not only mitochondrial-dependent . This presents an alternative mechanism in the regulation of invasion that requires meticulous dissection between mitochondrial dynamics and metabolism , the precise role of INF2 in mitochondrial fission , and mitochondrial recruitment to lamellipodia in EVTs . Overall , our data represent one mechanism of EVT invasion and proffer new avenues for future study in the molecular biology of trophoblast cells of the placenta . We present a novel genetic model of placental insufficiency encompassing critical aspects of IUGR and late-onset PE , despite several differences between the mouse and human placenta . Expression data on the 15 human formins in placenta , fallopian tube , breast , ovary , endometrium , and uterus were collected from the Human Protein Atlas ( Uhlén et al . , 2015 ) where RNA-sequence expression is measured in TPM ( Transcripts Per Kilobase Million ) . To identify genes that belong to the formin gene family we used the defining FH2 domain as a query against the Pfam database ( Finn et al . , 2005 ) . Using UniProtKB identifiers for 11 mammalian species ( Homo sapiens , Pan troglodytes , Gorilla gorilla , Pongo abelii , Nomascus leucogenys , Macaca mulatta , Calithrix jacchus , Otolemur garnettii , Mus musculus , Rattus norvegicus , and Canis lupus familiaris ) , coding and protein sequences of the 15 genes of the formin family were collected from ENSEMBL , release 89 ( Cunningham et al . , 2015 ) . Exceptions to this gene family were found in P . troglodytes and P . abelii ( P . troglodytes lacks DIAPH1 and FMN2 and P . abelii lacks FMNL1 ) . Most proteins were annotated as members of the family ( 155 ) ; however , others were annotated as novel predictions ( 7 ) . Only the longest coding sequence for each ortholog was kept . Protein sequences for all mammalian formins were aligned with the alignment program MAFFT , v7 . 310 ( Katoh and Standley , 2013 ) and edited in SeaView ( Gouy et al . , 2010 ) to build a domain phylogeny . Creation of a FH2 phylogeny was based upon previous formin family phylogenetic analyses ( Chalkia et al . , 2008 ) . A maximum likelihood ( ML ) phylogenetic tree was built using the RAxML software , v8 . 2 . 9 ( Stamatakis , 2014 ) with the PROTGAMMAAUTO model of substitution and 100 bootstrap replicates; the best fit model was the JTT model ( Jones et al . , 1992 ) . The tree was midpoint-rooted using phylogenetic visualization software FigTree , v1 . 4 . 3 ( Rambaut , 2007 ) . Unaligned coding sequences of mammalian formins were derived using Pal2Nal , v14 ( Suyama et al . , 2006 ) . Coding sequence alignment was reformatted using the program trimAl , v1 . 3 ( Capella-Gutiérrez et al . , 2009 ) . The FH2 mammalian tree was edited to distinguish INF clade branches as the foreground from the remaining background branches . Another FH2 phylogeny was edited distinguishing only INF2 clade branches from the rest of the tree . To examine ancient positive selection on the mammalian FH2 domain in the INF and INF2 clades , analyses of evolutionary rates were conducted using the codeml package in PAML , v4 . 8 ( Yang , 2007 ) . Branch models were used to predict the rates of codon substitution ( dn/ds ) based on the following hypotheses: all branches reflect equivalent rates of amino acid substitution ( H0 ) , positive selection on the INF clade ( H1 ) , and positive selection on the INF2 clade ( H2 ) . First trimester human EVT ( HTR-8/SVneo ) cells were cultured in RPMI-1640 medium supplemented with 5% FBS at 37°C under 5% CO2 ( Graham et al . , 1993 ) . BeWo choriocarcinoma cells , which are male in origin , were cultured in nutrient Mixture F-12 Hams medium supplemented with penicillin ( 100 U/mL ) , streptomycin ( 100 μg/mL ) , and 10% FBS at 37°C under 5% CO2 . HPMVECs , the sex of which are undetermined , were maintained at 37°C under 5% CO2 in EGM-2 plated on Attachment Factor-coated culture flasks ( Troja et al . , 2014 ) . BeWo choriocarcinoma cells were authenticated by STR profiling and tested negative for mycoplasma contamination by ATCC ( ATCC , Manassas , VA ) . HTR-8/SVneo cells , a generous gift from Dr . Charles Graham at Queen's University ( Graham et al . , 1993 ) , have also been authenticated by STR profiling and tested negative for mycoplasma contamination by ATCC . HPMVECs were isolated from normal term pregnancies under IRB approval and were tested for expression of endothelial cell markers CD31 and von Willebrand factor ( Troja et al . , 2014 ) . Contamination by smooth muscle cells was assessed by immunocytochemistry . Mycoplasma contamination was not assessed in these cells . HTR-8/SVneo and BeWo cells were seeded at a density of 2 . 5 × 105 cells per well of a 6-well plate and allowed to reach 70% confluency at the time of transfection . Transfections were performed in duplicate ( triplicate for invasion assays ) with 50 nM of MISSION siRNA Universal Negative Control #1 ( Millipore Sigma , St . Louis , MO ) , 50 nM of INF2 siRNA ( Thermo Fisher Scientific , Waltham , MA ) , or vehicle following the Lipofectamine 3000 Reagent standard protocol ( Thermo Fisher Scientific ) . 24 hr post-transfection , the media was replaced on cells and replaced with either growth media or serum-free media . 48 hr post-transfection , cells were harvested for mRNA or protein analysis . Additional HTR-8/SVneo cells were cultured for 7 days in DMSO , 1 μM PP1 ( Cayman Chemical , Ann Arbor , MI ) , or 1 μM TX1123 ( Millipore Sigma ) prior to invasion assays . RNA was isolated from whole mouse placentas or uterus using TRIzol reagent ( Thermo Fisher Scientific ) per manufacturer’s protocols . RNA was isolated from cells using the Qiagen RNeasy Mini Kit per the manufacturer’s protocols . cDNA was synthesized from 1000 ng of RNA ( Qiagen , Hilden , Germany ) per manufacturer’s protocols . EXPRESS SYBR GreenER ( Thermo Fisher Scientific ) was used for qPCR analysis . Primer sequences were generated using the NCBI primer BLAST tool for human INF2 and mouse Inf2 mRNAs . All mouse genes analyzed were normalized to Ribosomal Protein S20 ( Rps20 ) expression . All human genes analyzed were normalized to Beta-Actin ( ACTB ) expression . Gene expression data were generated and calculated using the ΔΔCt method on the StepOnePlus real time PCR system ( Applied Biosystems , Foster City , CA ) . Cell pellets were homogenized in RIPA Buffer ( Millipore Sigma ) supplemented with Protease Inhibitor Cocktail ( Millipore Sigma ) and Phosphatase Inhibitor Cocktails I and II ( Millipore Sigma ) and protein concentration determined by BCA Assay ( Thermo Fisher Scientific ) . 30 μg was loaded per lane of a 4–12% gradient Bis-Tris polyacrylamide gel ( Thermo Fisher Scientific ) and transferred to a Hybond enhanced nitrocellulose membrane using a semidry transfer system ( BioRad , Hercules , CA ) . Membranes were blocked with 5% milk in TBS with 0 . 1% Tween-20 ( Millipore Sigma ) . Blots were probed with overnight at 4°C with anti-INF2 ( Millipore Sigma , ABT61 , 1:500 ) or anti-LCK ( Abcam , Cambridge , UK; ab208787 , 1:2000 ) . Binding of the secondary goat anti-rabbit secondary antibody ( Santa Cruz , Dallas , TX; sc-2004 , 1:10 , 000 ) was determined using SuperSignal West Dura Extended Duration Substrate ( Thermo Fisher Scientific ) . Blots were stripped with Restore Western Blot Stripping Buffer ( Thermo Fisher Scientific ) and reprobed with the anti-actin antibody ( Millipore Sigma , 1:30 , 000 ) as an internal control . Serum starved HTR-8/SVneo cells were plated on BioCoat Matrigel Invasion Chambers ( Corning , Corning , NY ) at a density of 2 . 0 × 105 cells in 200 μL of serum-free media . Each chamber was placed in a well of a 24-well plate containing 600 μL of RPMI medium with 10% FBS for 24 hr , after which the Matrigel and non-invading cells were removed from the membrane . Inserts were fixed in 4% PFA and washed in PBS . Nuclei of invaded cells were stained with DAPI and membranes were mounted on slides . Invaded cells were counted in five random fields at 10x magnification in three inserts per treatment and in three independent experiments . Drug-treated HTR-8/SVneo cells were similarly tested for invasiveness after culture for 7 days in DMSO , 1 μM PP1 ( Cayman Chemical ) , or 1 μM TX1123 ( Millipore Sigma ) in three independent experiments . Data are represented as percent invaded cells normalized to vehicle-treated HTR-8/SVneo cells . 24 hr post-transfection , cells were transferred to a chamber slide ( ibidi ) . 48 hr post-transfection , one subset of slides were treated for 20 min with 100 mM MitoTracker Red CMXRos ( Thermo Fisher Scientific , M7512 ) at 37°C . Media was removed from remaining slides and cells were washed with PBS and fixed in 4% PFA . After blocking in 10% normal horse serum in 0 . 1% Tween-PBS , cells were incubated overnight at 4°C with MAL2 ( Abcam , ab75347 , 1:100 ) or LCK ( Abcam , ab208787 , 1:100 ) . Positive staining was detected using a fluorescent donkey anti-rabbit IgG secondary antibody ( Thermo Fisher Scientific , 1:200 ) . Cells were counterstained with Alexa Fluor 594 phalloidin ( Thermo Fisher Scientific , 1:500 ) and DAPI ( Thermo Fisher Scientific , 1:10 , 000 ) . Cells were washed and post-fixed with 4% PFA and were imaged on a Nikon Ti-E inverted microscope with a Nikon A1R and a 100x oil immersion objective in the Confocal Imaging Core at Cincinnati Children’s . Images were processed in NIS-Elements and depicted as maximum intensity projections and are representative from three independent experiments . Imaris , v9 . 0 . 1 ( Bitplane , Zurich , Switzerland ) , was used to determine mitochondrial volume and phalloidin content of HTR-8/SVneo cells . Cytoplasmic phalloidin content was determined by taking the ratio of cytoplasmic phalloidin volume to the total cellular phalloidin volume , which we multiplied by 100 . Inf2-deficient mice ( Inf2tm1 . 1 ( KOMP ) Vlcg , abbreviated Inf2−/− ) used for this research project were generated by the trans-NIH Knock-Out Mouse Project ( KOMP ) on a C57BL/6NTac background and obtained from the KOMP Repository ( Dickinson et al . , 2016 ) . Using the ZEN-Ub1 cassette in VGB6 ES cells , 12 , 623 bp of Inf2 ( Chr12:112 , 600 , 006–612 , 628 ) were deleted , with an insertion of a LacZ reporter between exons 1 and 23 , removing 1270 amino acids . Mice were housed on a 14-/10 hr light-dark cycle with access to chow and water ad libitum . Colonies were maintained as Inf2+/− x Inf2+/− matings , allowing for Inf2−/− animals to be compared to littermate controls . Females between 10 weeks and 6 months of age were used for studies . All animal procedures were approved by the Cincinnati Children’s Medical Center Animal Care and Use Committee and were in accordance with the National Institutes of Health guidelines . Nulliparous females were set up with males ( so pups were homozygous ) for timed matings at 1700 hr and separated the following morning at 0800 hr . A copulatory plug marked 0 . 5 days post-coitum ( dpc ) . Placentas at gestational days 15 . 5 and 18 . 5 were collected and fixed overnight in 4% PFA at 4°C . Tissues were washed in PBS , halved , and immersed in 70% ethanol prior to processing and paraffin embedding . Placentas were sectioned at 5 μm . Slides were baked at 60°C overnight , deparaffinized , and rehydrated . Antigen retrieval was performed using 10 mM citrate buffer ( pH 6 . 0 ) followed by PBS washes . Endogenous peroxidase activity was removed by treatment with 3% H2O2 for DAB IHC . Non-specific binding was blocked by incubating slides for 1 hr in 4% normal horse serum in 0 . 1% Tween-PBS . Inf2 was detected using a rabbit anti-INF2 primary antibody ( Millipore Sigma , HPA000724 , 1:25 ) and using a biotinylated horse anti-rabbit IgG secondary antibody at ( Vector Laboratories , Burlingame , CA; BA-1100 , 1:200 ) followed by treatment with ABC peroxidase complex ( Vector Laboratories ) . Slides were developed with either DAB ( Vector Laboratories ) , counterstained with nuclear fast red and mounted with PROTOCOL Mounting Medium ( Thermo Fisher Scientific ) , or developed with Cyanine 5 tyramide ( Perkin Elmer , Waltham , MA ) , counterstained with DAPI ( Thermo Fisher Scientific ) and mounted with ProLong Gold Antifade Mountant ( Thermo Fisher Scientific ) . Immunofluorescent co-localization of INF2 with: endothelial cells using a goat anti-Endomucin-2 antibody ( R&D Systems , Minneapolis , MN; AF4666 , 1:200 ) , canal and spiral artery-associated TGCs using a goat anti-Proliferin antibody ( R&D Systems , AF1623 , 1:200 ) , and trophoblasts using Alexa Fluor 594 rabbit anti-Cytokeratin 7 ( Abcam , Cambridge , UK; ab209600 , 1:100 ) . Lck localization was determined using a rabbit anti-Lck antibody ( Abcam , ab3885 , 1:200 ) . Nuclei were counterstained with DAPI ( Thermo Fisher Scientific , 1:10 , 000 ) . Images are representative of placentas from three dams per genotype . At E19 . 0 , 300 μL of 400 μg/mL DyLight 649 labeled Lycopersicon Esculentum ( Tomato ) Lectin ( Vector Laboratories ) in 100 U/mL heparin sulfate and PBS was injected into the tail vein . 20 min were allowed for lectin circulation . Placentas were fixed in 4% PFA overnight at 4°C . The following day , placentas were washed in PBS . Tissue was cleared via an active CLARITY technique ( Lee et al . , 2016 ) . Placentas were incubated overnight at 4°C in a 4% acrylamide hydrogel monomer containing 0 . 25% Wako VA-044 photoinitiator . Tissues were incubated for 3–4 hr in a 37°C water bath to polymerize the gel . Tissues were lightly washed , then placed in a Logos Biosystems X-CLARITY machine for 6–8 hr at 1 . 5 Amperes and 37°C . Post clearing , tissues were washed in 37°C water overnight , then multiple washes with PBS to remove any residual SDS . Placentas were then dehydrated in a methanol series and placed in benzyl alcohol/benzyl benzoate for final clearing and mounted in a custom-made aluminum chamber slide for upright microscopy . Images were acquired on a Nikon FN1 upright microscope with a Nikon A1R-MP in single-photon confocal mode and a 16X/0 . 8 NA water immersion objective in the Confocal Imaging Core at Cincinnati Children’s . Images were stitched , analyzed and processed in NIS-Elements . A researcher blinded to mouse genotype counted and recorded number of spiral arteries . Images and videos are representative . Systolic blood pressure was measured in conscious animals using a non-invasive Volume Pressure Recording method ( Kent Scientific , Torrington , CT ) , a previously validated method ( Feng et al . , 2008 ) . Females were trained in the restraints on the warming platform for two weeks prior to the study . Baseline measurements were an average of measurements on virgin females for three consecutive days . Plugged females resumed training at E7 . 5 to avoid interfering with implantation while non-plugged females resumed training until plugged . Measurements were taken longitudinally from E14 . 5 through postnatal day 7 . Three cycles used to acclimate females to tail cuff inflation and discarded from analysis , using acceptable reads from the following 12 cycles . Investigator observed all measurement cycles and manually discarded reads with signal artifacts . Data collected by Kent Scientific software were analyzed off-line . All blood pressure measurements were obtained between 0700 and 1000 hr . Maternal blood was collected in serum separator tubes at E15 . 5 and E18 . 5 by submandibular phlebotomy . Blood was allowed to clot at room temperature for 30 min and serum was removed and stored per the manufacturer’s conditions until assayed by the Research Flow Cytometry Core at CCHMC on a MILLIPLEX MAP Mouse Angiogenesis/Growth Factor Magnetic Bead Panel and MILLIPLEX MAP Mouse Soluble Cytokine Receptor Magnetic Bead Panel ( Millipore Sigma ) . Nulliparous Inf2−/− or Inf2+/+ females were set up with homozygous males of corresponding genotype for timed matings at 1700 hr and separated the following morning at 0800 hr . Visualization of a copulatory plug marked 0 . 5 days post-coitum ( dpc ) . Cages were checked four times daily ( 0600 , 1000 , 1400 , 1800 hr ) to calculate gestation length , determined by the timing of birth of the first pup . Number of pups in a litter and pup weights were recorded at birth ( P0 ) or at E18 . 5 after euthanization . Serum progesterone concentration at E18 . 5 ( by submandibular phlebotomy as described above ) was measured by ELISA according to manufacturer’s protocols ( BioVendor , Brno , Czech Republic; RTC008R ) . PGF2α and PGE2 were measured on snap frozen uterus at E18 . 5 . Prostaglandins were extracted after weighing frozen tissue by homogenization in 100% ethanol . Centrifugation removed debris and the supernatant dried down under inert gas , resuspended and assayed per the manufacturer’s instruction ( Oxford Biomedical Research , Rochester Hills , MI ) . A Vevo 2100 ultrasound machine ( Fugifilm VisualSonics , Toronto , Canada ) equipped with a 40 MHz transducer was used to perform fetal ultrasounds on E18 . 5 . Dams were anesthetized with 1 . 0% inhaled isoflurane , abdominal hair removed with a depilatory agent , and positioned on a warmed platform to maintain euthermia . After fetal number and placement were determined , each fetus was examined consecutively around the uterine horn . Seven fetuses per litter were scanned in each dam . Umbilical vessels were identified using 2-dimensional and color Doppler imaging with the vessels traced from the fetus to the site of insertion into the placenta . A freely mobile loop of umbilical cord was interrogated . Color Doppler images of the umbilical artery and vein were recorded , typically capturing flow in both vessels simultaneously . The pulsed wave Doppler sample volume was adjusted and subtle positional changes of the transducer made to obtain umbilical vessel interrogation as close to parallel flow as possible , adjusting the beam angle from 0 to 60 degrees as needed to provide the best alignment . Pulsed wave Doppler is recorded at a sweep speed of 5 . 1 m/second with peak velocity scaled to a maximum of 150 mm/second to optimize tracings for off line analysis . Images were analyzed using the vascular package included in the Vevo 2100 software by investigators blinded to genotype . Off line measurements include fetal heart rate , umbilical artery peak systolic velocity ( PSV ) , end diastolic ( EDV ) and the umbilical artery velocity time integral ( VTI ) , the latter of which provides the umbilical artery mean velocity . Umbilical vein flow was qualitatively assessed for abnormalities such as pulsatile diastolic flow or flow reversal . Images highlight the largest differences . Vascular density was measured in placentas at E15 . 5 and E18 . 5 by counting the number of blood vessels , identified using IF for Endomucin , per high-powered field ( 40X magnification ) , blinded to genotype . Vessel numbers in each of 10 random fields were averaged in a single section per placenta across 2–3 placentas per dam . An average of three measurements made across 20X H&E scans of each the labyrinth , junctional zone , and whole placenta were taken using the Nikon Elements Software . Results were reported from 2 to 3 placentas per dam at E18 . 5 . Images are representative . 24 hr post-transfection , media on transfected BeWo cells was replaced with EGM-2 ( Lonza , Basel , Switzerland ) . 48 hr post-transfection , BeWo cells were harvested for RNA analysis while the media was removed and placed on HPMVECs at 70% confluence grown in 6-well plates coated with attachment factor ( Thermo Fisher Scientific ) . In a subset of BeWo cells , media and cells were collected for PGF ELISA analysis ( R&D Systems , DPG00 ) . ELISA data were normalized to total protein . HPMVECs were cultured in BeWo cell conditioned medium for 48 hr prior to harvesting for RNA analysis . Data were analyzed by Student’s t test , one- , or two-way ANOVA test ( Prism 7 . 0c software; GraphPad Software , Inc . , San Diego , CA ) as indicated in Figure Legends and a p≤0 . 05 was considered significant . The n represents either dams or fetuses as indicated in Figure Legends . Results are reported as ±SEM .

The placenta is an organ that develops with the baby during pregnancy and links the baby with his or her mother . This connection allows mom and baby to communicate throughout the pregnancy to share nutrition and growth signals , and to coordinate their immune systems . Abnormal placental growth can have lasting , harmful effects on the health of the mother and baby . Specialized cells in the placenta called trophoblasts help the embryo implant into the mother’s womb and direct the flow of nutrient- and oxygen-rich blood from the mother to the baby . If trophoblasts do not penetrate deeply enough into the womb , the mother will be at risk for developing a life-threating condition called preeclampsia , which occurs when her blood pressure becomes dangerously high . The only treatment is to deliver the baby . Her baby will also be at risk of poor growth and premature delivery . Scientists still do not know exactly how the trophoblasts invade the womb and what goes wrong that causes placental abnormalities . Now , Lamm et al . show that losing a gene called Inverted Formin 2 , or Inf2 for short , which helps cells to form structures , causes placental abnormalities and preeclampsia symptoms in mice . In the experiments , trophoblasts in mice without Inf2 were unable to invade the womb properly . The Inf2-lacking mice had fewer blood vessels feeding the placenta . These mice developed high blood pressure late in pregnancy , which returned to normal after their babies were born and the placentas expelled . During pregnancy , the placentas of Inf2-lacking mice were less efficient in transporting nutrients and gases , and their fetuses grew slowly and showed signs of distress . This suggests that the INF2 gene is necessary for the placenta to develop properly . Learning more about what can go wrong as the placenta forms might help physicians predict or prevent preeclampsia , fetal growth problems , and other placental abnormalities . More studies could determine if treatments targeting INF2 would improve the development of the placenta , protect mothers from preeclampsia , and prevent conditions that slow down the babies’ growth .

2018

Inverted formin 2 regulates intracellular trafficking, placentation, and pregnancy outcome

The yeast Target of Rapamycin Complex 1 ( TORC1 ) plays a central role in controlling growth . How amino acids and other nutrients stimulate its activity via the Rag/Gtr GTPases remains poorly understood . We here report that the signal triggering Rag/Gtr-dependent TORC1 activation upon amino-acid uptake is the coupled H+ influx catalyzed by amino-acid/H+ symporters . H+-dependent uptake of other nutrients , ionophore-mediated H+ diffusion , and inhibition of the vacuolar V-ATPase also activate TORC1 . As the increase in cytosolic H+ elicited by these processes stimulates the compensating H+-export activity of the plasma membrane H+-ATPase ( Pma1 ) , we have examined whether this major ATP-consuming enzyme might be involved in TORC1 control . We find that when the endogenous Pma1 is replaced with a plant H+-ATPase , H+ influx or increase fails to activate TORC1 . Our results show that H+ influx coupled to nutrient uptake stimulates TORC1 activity and that Pma1 is a key actor in this mechanism . The Target of Rapamycin Complex 1 ( TORC1 ) plays a pivotal role in controlling cell growth in probably all eukaryotic organisms . It operates by integrating upstream signals such as growth factors ( GFs ) and nutrients to modulate , by phosphorylation , multiple downstream effectors , mostly proteins involved in anabolic processes ( e . g . protein synthesis , ribosome biogenesis ) or catabolic processes ( e . g . autophagy , bulk endocytosis of plasma membrane transporters ) ( González and Hall , 2017; Powis and De Virgilio , 2016; Saxton and Sabatini , 2017 ) . The central role of TORC1 in regulating cell growth is illustrated by the many reported cases of mTORC1 dysfunction associated with diseases including cancers ( Eltschinger and Loewith , 2016; Saxton and Sabatini , 2017 ) . In human cells , GFs and amino acids are key signals of mTORC1 regulation . GFs act through activation of Rheb , a GTPase present at the lysosomal membrane that stimulates mTORC1 activity ( Durán and Hall , 2012; Zheng et al . , 2014 ) . Amino acids such as leucine and arginine act via a complex of two GTPases , namely RagA or B and RagC or D , to promote recruitment of mTORC1 to the lysosome , where it is activated by Rheb . The heterodimeric Rag GTPase complex recruits mTORC1 to the lysosome when RagA/B is bound to GTP and RagC/D to GDP ( Kim et al . , 2008; Sancak et al . , 2008 ) . The activity of GTPases is typically modulated by GTPase Activating Proteins ( GAPs ) and Guanine nucleotide Exchange Factors ( GEFs ) . Recent studies have aimed to identify these GAPs and GEFs and their upstream regulators and to better understand how these factors are controlled in response to variations in the cytosolic concentrations of amino acids and/or to their transport across the plasma or lysosomal membrane ( González and Hall , 2017; Powis and De Virgilio , 2016; Saxton and Sabatini , 2017 ) . Importantly , Sestrin and Castor proteins have recently been found to act , respectively , as cytosolic leucine and arginine sensors . When bound to their specific amino acids , these sensor proteins lose the ability to inhibit GATOR2 , a negative modulator of the GATOR1 GAP complex which inhibits RagA/B , and this results in TORC1 activation ( Wolfson and Sabatini , 2017 ) . Specific lysosomal amino acid transporters and the V-ATPase complex also contribute importantly to mTORC1 control ( Goberdhan et al . , 2016; Zoncu et al . , 2011 ) . All this illustrates the complexity of the mechanisms through which amino acids regulate mTORC1 activity . The TOR kinase that is part of TORC1 was originally identified in yeast , after isolation of dominant TOR mutations conferring resistance to rapamycin ( Rap ) ( Heitman et al . , 1991; Loewith and Hall , 2011 ) . The protein components of TORC1 , the RagA/B and C/D proteins , and their upstream GATOR-type regulatory complexes also exist in yeast ( Hatakeyama and De Virgilio , 2016; Loewith and Hall , 2011 ) . For instance , RagA/B and RagC/D correspond , respectively , to the yeast Gtr1 and Gtr2 proteins , which are part of a vacuole-associated complex ( EGO ) ( Dubouloz et al . , 2005 ) similar to the Rag-binding Ragulator of human cells ( Sancak et al . , 2010 ) . When cells are grown in nutrient-rich medium , yeast TORC1 is active and stimulates by phosphorylation a wide variety of proteins . It notably stimulates the Sch9 kinase ( Urban et al . , 2007 ) under conditions promoting anabolic functions and cell growth . Active TORC1 also inhibits the Tap42-PP2A phosphatase , which stimulates autophagy , stress resistance , and nitrogen ( N ) transport and utilization ( Loewith and Hall , 2011 ) . In contrast , TORC1 is inhibited in N-starved and Rap-treated cells , so that anabolic processes , including protein synthesis , are inhibited and cell responses such as autophagy , bulk endocytosis of transporters , utilization of secondary N sources , and stress resistance are stimulated ( Hatakeyama and De Virgilio , 2016; Loewith and Hall , 2011 ) . One Tap42-PP2A target protein is the protein kinase Npr1 ( Nitrogen permease reactivator 1 ) , which is phospho-inhibited when TORC1 is active ( Schmidt et al . , 1998 ) . Once Npr1 is inhibited , various permeases of nitrogenous compounds undergo intrinsic inactivation ( Boeckstaens et al . , 2014; Boeckstaens et al . , 2015 ) or downregulation via ubiquitylation , endocytosis , and degradation ( MacGurn et al . , 2011; Merhi and AndreAndré , 2012 ) . Stimulation of TORC1 activity in yeast is usually monitored by visualizing the degree of Sch9 and/or Npr1 kinase phosphorylation . Sch9 and Npr1 are moderately phosphorylated in cells grown on a poor N source such as proline , but hyperphosphorylated upon addition of a preferential N source such as glutamine ( Gln ) or NH4+ ( Schmidt et al . , 1998; Stracka et al . , 2014; Urban et al . , 2007 ) . In a study using Sch9 phosphorylation as readout , addition of any amino acid to proline-grown cells was found to result in rapid but transient Rag/Gtr-dependent TORC1 activation , whereas longer term TORC1 activation was observed only upon addition of an N source supporting optimal growth , for example Gln or NH4+ , and it appeared not to depend on the Rag GTPases ( Stracka et al . , 2014 ) . Furthermore , sustained activation of TORC1 in response to NH4+ is impaired in mutant cells lacking the glutamate dehydrogenases involved in assimilation of NH4+ into amino acids ( Fayyad-Kazan et al . , 2016; Merhi and AndreAndré , 2012 ) . The upstream signals and molecular mechanisms involved in activation of yeast TORC1 in response to amino acid uptake and/or assimilation remain poorly known . For instance , although Gln behaves as a key signal for sustained TORC1 stimulation ( Crespo et al . , 2002; Stracka et al . , 2014 ) , no Gln sensor has been identified to date , and yeast seems to lack Sestrin and Castor proteins . Furthermore , no study has evidenced any particular role of vacuolar amino acid transporters in TORC1 regulation . The yeast leucyl-tRNA synthetase is reported to play a role in sensing balanced levels of isoleucine , leucine , and valine and to act as a GEF for Gtr1 ( Bonfils et al . , 2012 ) , whereas the equivalent mammalian enzyme is proposed to control mTORC1 as a GAP for RagD ( Han et al . , 2012 ) . On the basis of current knowledge , it would thus seem that the upstream signals and mechanisms controlling TORC1 according to the N or amino acid supply conditions might differ significantly between yeast and human cells . The present study began with an unexpected observation regarding the uptake of β-alanine into yeast cells: this amino acid , which cannot be used as an N source ( i . e . it is not a source of amino acids ) , stimulates TORC1 activity . Analysis of this effect has revealed that the general signal triggering Rag/Gtr-dependent activation of TORC1 in response to amino acid uptake is the influx of H+ coupled to transport via H+/amino-acid symporters . We further show that the Pma1 H+-ATPase establishing the H+ gradient at the plasma membrane is essential to this TORC1 activation , and suggest that Pma1 modulates TORC1 via signaling . In cells growing under poor N supply conditions ( e . g . in a medium containing proline as sole N source ) , the yeast general amino acid permease Gap1 is active and stable at the plasma membrane . Under these conditions , TORC1 is only moderately active ( Schmidt et al . , 1998 ) . Activation of TORC1 upon NH4+ uptake and assimilation into amino acids triggers Gap1 ubiquitylation , followed by its endocytosis and degradation in the vacuole ( Merhi and AndreAndré , 2012 ) . Thanks to isolation of a Gap1 mutant insensitive to this TORC1-dependent ubiquitylation , we have shown that substrate transport by Gap1 can also trigger Ub-dependent endocytosis and degradation of this transporter ( Ghaddar et al . , 2014b ) . This type of control , shared with other transporters of fungal and non-fungal species , probably enables cells to avoid excess uptake of external compounds ( Gournas et al . , 2016 ) . To further investigate , without interference from the TORC1-dependent pathway , the mechanism of transport-elicited Gap1 ubiquitylation we sought to identify a Gap1 amino acid substrate unable to activate TORC1 . We focused on beta-alanine ( β-ala ) and confirmed its reported status as a Gap1 substrate ( Stolz and Sauer , 1999 ) by comparing the uptake of [14C]-β-ala ( 0 . 5 mM ) in proline-grown wild-type and gap1∆ mutant cells ( Figure 1A ) . β-Ala , however , cannot sustain growth when used as sole N source , that is it cannot serve as a source of amino acids ( Figure 1A ) . This contrasts with 4-aminobutyrate ( GABA ) , an amino acid differing from β-ala by a single additional CH2 group ( Figure 1A ) and whose catabolism depends on a specific GABA transaminase ( Andersen et al . , 2007 ) . We thus tested whether β-ala transport by Gap1 triggers ubiquitylation and downregulation of the transporter . This proved to be the case , as addition of β-ala ( 0 . 5 mM ) caused the appearance , above the immunodetected Gap1 signal , of two slowly migrating bands that were not observed with the non-ubiquitylable Gap1 ( K9R , K16R ) mutant ( Figure 1B ) . Upon β-ala addition , furthermore , Gap1 initially present at the cell surface underwent endocytosis and targeting to the vacuole , whereas Gap1 ( K9R , K16R ) remained stable at the plasma membrane ( Figure 1C ) . An inactive Gap1 mutant ( Gap1-126 ) ( Ghaddar et al . , 2014b ) failed to be ubiquitylated and downregulated upon β-ala addition ( Figure 1B and C ) . These results are those expected if β-ala elicits Gap1 ubiquitylation specifically via the transport-elicited pathway . Yet we sought to make sure that β-ala does not activate TORC1 . To our surprise , addition of β-ala to proline-grown cells caused a typical manifestation of TORC1 activation: a Rap-sensitive reduction of the electrophoretic mobility of HA-tagged Npr1 , indicative of increased phosphorylation via TORC1 ( Merhi and AndreAndré , 2012; Schmidt et al . , 1998 ) ( Figure 1D ) . β-Ala similarly caused a Rap-sensitive increase in the phosphorylation of Sch9 kinase residue Thr737 ( Figure 1E ) , a known TORC1 target ( Urban et al . , 2007 ) . It thus seemed that activation of TORC1 , largely impaired in the gap1∆ mutant ( Figure 1D and E ) , could contribute to the observed β-ala-induced downregulation of Gap1 . This assumption was confirmed in additional experiments ( Figure 1—figure supplement 1 ) . Gap1-mediated uptake of β-ala thus results in TORC1 activation . We therefore hypothesized that , although β-ala cannot be used as an N source , it might be converted to certain amino acids capable of stimulating TORC1 . β-Ala uptake , however , was found not to increase the intracellular concentrations of individual amino acids , measured in cell extracts , apart from that of β-ala itself ( Figure 1F ) . We next tested whether this β-ala-induced activation of TORC1 involves the Rag A/B and C/D GTPases , encoded by the GTR1 and GTR2 genes , respectively . Increased phosphorylation of Npr1 upon β-ala addition was indeed impaired in gtr1∆ gtr2∆ mutant cells ( Figure 1G ) , and this effect was not due to reduced uptake of β-ala ( Figure 1H ) . We conclude that Gap1-mediated uptake of β-ala elicits TORC1 activation via the Rag GTPases , and that this effect is not due to conversion of intracellular β-ala to other amino acids . Gap1 has been reported to be a ‘transceptor’ , that is a protein combining the properties of transporters and receptors , capable of activating protein kinase A ( PKA ) in a cAMP-independent manner ( Donaton et al . , 2003 ) . According to this model , conformational changes of Gap1 , triggered by binding and/or transport of amino acids , would stimulate a PKA-targeting signaling pathway ( Schothorst et al . , 2013 ) . We thus hypothesized that this transceptor function of Gap1 might also promote TORC1 activation in a Rag/Gtr-dependent manner . This hypothesis is potentially supported by a previous report that Gap1 interacts with Gtr2 ( Gao and Kaiser , 2006 ) . An alternative view is that β-ala entering cells through Gap1 might be detected by a cytosolic amino acid sensor capable of promoting TORC1 activation . To explore these possibilities , we tested whether β-ala uptake via another permease might also activate TORC1 . Confirming a previous prediction ( Gournas et al . , 2015 ) , we found the high-affinity proline permease Put4 also to catalyze β-ala transport . This contribution of Put4 was visible at least in proline-free media , such as a medium where the sole N source was urea ( another poor N source ) . Under these conditions , the gap1∆ mutant displayed residual uptake of β-ala ( 2 mM ) , and this uptake was abolished in the gap1∆ put4∆ mutant ( Figure 2A ) . Importantly , this Put4-dependent β-ala uptake was associated with a Rap-sensitive hyperphosphorylation of Npr1 ( Figure 2B ) . We next expressed in the gap1∆ mutant a heterologous amino acid transporter known to be active in S . cerevisiae , namely the Gap1 permease of the fungus Hebeloma cylindrosporum ( Wipf et al . , 2002 ) . HcGap1 shares ~ 30% sequence identity with Gap1 and Put4 . In proline-grown gap1∆ cells , HcGap1 restored high β-ala uptake activity , roughly similar to that conferred by Put4 to urea-grown cells ( Figure 2C ) . Remarkably , this uptake of β-ala was also associated with Rap-sensitive hyperphosphorylation of Npr1 ( Figure 2D ) . In conclusion , transport of β-ala via endogenous Gap1 and/or Put4 or via the heterologous HcGap1 permease elicits TORC1 activation . Although these observations do not rule out the possibility that all three tested permeases might function as transceptors , they seem to favor the view that intracellular β-ala itself , or the process of its transport across the plasma membrane , stimulates TORC1 activity . Uptake of arginine ( Arg ) by proline-grown cells is known to be mediated by Gap1 and the arginine-specific permease Can1 ( Wiame et al . , 1985 ) . Consistently , we measured high Arg uptake activity in the wild type and in gap1∆ and can1∆ single mutants , but none in the gap1∆ can1∆ double mutant ( Figure 3A ) . According to a previous report , Can1 , in contrast to Gap1 , does not stimulate PKA upon substrate transport ( Donaton et al . , 2003 ) . This suggests that Can1 does not function as a transceptor . We thus sought to determine whether Arg transport via Gap1 or Can1 alone supports TORC1 activation . Arg uptake into wild-type cells was indeed found to induce Rap-sensitive Npr1 hyperphosphorylation ( Figure 3A and B ) . This response was impaired in the gtr1∆ gtr2∆ mutant , an effect not due to decreased Arg uptake ( Figure 3B ) . Arg-elicited TORC1 activation also resulted in Sch9 phosphorylation ( Figure 3C ) , as previously reported ( Stracka et al . , 2014 ) . Increased phosphorylation of Npr1 upon Arg addition was also detected in gap1∆ and can1∆ single mutants , but not in the gap1∆ can1∆ strain ( Figure 3A ) . This shows that both permeases can promote Arg-induced TORC1 activation . We next expressed HcGap1 in the gap1∆ can1∆ strain and found it to restore high Arg uptake ( Figure 3D ) associated with Rap-sensitive Npr1 hyperphosphorylation ( Figure 3E ) . Arginine catabolism requires arginase ( Car1 ) ( Wiame et al . , 1985 ) , so a car1 mutant fails to grow on Arg as sole N source ( Figure 3F ) . Arg addition to the car1 mutant also resulted in Rap-sensitive Npr1 hyperphosphorylation ( Figure 3G ) . In conclusion , TORC1 is activated upon Arg uptake via the endogenous Gap1 and/or Can1 or the heterologous HcGap1 permease . This activation of TORC1 involves the Rag GTPases and occurs even if Arg is not catabolized . The simplest way to explain the above observations is that intracellular β-ala and Arg are detected by one or several internal amino acid sensors promoting Rag/Gtr-dependent TORC1 activation . These sensors could , for instance , act like the human Castor and Sestrin proteins , recently shown to function as arginine and leucine sensors , respectively , and to modulate upstream regulators of mTORC1 ( Wolfson and Sabatini , 2017 ) . These sensor proteins , however , do not seem to exist in yeast . Furthermore , one would not expect a cytosolic sensor capable of activating TORC1 in response to β-ala . Alternatively , a TORC1-activating signal might arise from a common feature of the permease-mediated Arg and β-ala transport reactions . In yeast , transport by secondary active plasma membrane transporters is coupled to H+ influx . This transport is thus driven by the plasma membrane H+ gradient established by the Pma1 H+-ATPase . We therefore hypothesized that an influx of H+ coupled to amino acid uptake might initiate a signal stimulating TORC1 activity . To evaluate this hypothesis , we first checked that the β-ala and Arg transporters tested above are H+-symporters . In support of this view , incubation of cells with the FCCP protonophore caused a strong reduction of β-ala uptake via Gap1 , Put4 , or HcGap1 and of Arg uptake via Gap1 , Can1 , or HcGap1 ( Figure 4A and B ) . Furthermore , each permease was rapidly inhibited when the cells were shifted to glucose-free medium ( Figure 4A and B ) , a condition known to cause rapid inhibition of the Pma1 H+-ATPase and thus collapse of the plasma membrane H+ gradient ( Kane , 2016 ) . Hence , as expected , all four permeases analyzed in our study , including HcGap1 , behave as H+-symporters . We next examined whether active uptake of another metabolite , not present in the growth medium , also elicits Rag/Gtr-dependent TORC1 activation . We chose cytosine , whose uptake via the Fcy2 permease is known to be coupled to H+ influx ( Pinson et al . , 1997 ) . Interestingly , addition of cytosine did cause rapid activation of TORC1 , as judged by increased Npr1 and Sch9 phosphorylation ( Figure 4C and D ) . Furthermore , this cytosine-elicited TORC1 activation was largely impaired in the gtr1∆ gtr2∆ mutant ( Figure 4E ) . Cytosine can be used as sole N source , and thus as a source of amino acids . Yet in an fcy1 mutant lacking cytosine deaminase and thus unable to use cytosine as an N source , Npr1 was still phosphorylated upon cytosine addition , unless Rap was also present ( Figure 4F ) . H+-coupled uptake of cytosine thus stimulates TORC1 in a Rag/Gtr-dependent manner , even when the nucleobase is not assimilated into amino acids . This observation is compatible with the proposed view that H+ influx is the signal initiating TORC1 stimulation . To further assess this model , we sought to analyze the activity of TORC1 upon equivalent uptake of the same external compound by either a facilitator or an H+-coupled symporter . Hexoses including fructose are known to enter cells via several Hxt transporters that function as facilitators ( Wieczorke et al . , 1999 ) . Yet particular S . cerevisiae strains are reported to also express an H+-coupled specific fructose transporter called Fsy1 ( Galeote et al . , 2010; Rodrigues de Sousa et al . , 2004 ) . We thus used the hxt null strain , lacking the HXT1 to −17 and GAL2 genes and therefore unable to assimilate hexoses ( Wieczorke et al . , 1999 ) , in which we expressed the FSY1 gene behind its own promoter , or none hexose transporter gene , and we analyzed in parallel the wild-type ( from which the hxt null mutant derives ) expressing the endogenous Hxt facilitators . The strains were initially grown on maltose as hxt null cells can utilize this disaccharide . They were then shifted for a few hours to ethanol because the FSY1 gene is more highly expressed on this carbon source ( Rodrigues de Sousa et al . , 2004 ) . As in previous experiments , the N source was proline . Using these growth conditions , we measured equivalent 14C-fructose uptake in Hxt- and Fsy1-expressing cells ( Figure 4G ) . None significant fructose uptake was detected in the hxt null mutant , as expected ( Figure 4G ) . Furthermore , fructose uptake via Fsy1 was inhibited in the presence of FCCP , but uptake mediated by the Hxt facilitators was not ( Figure 4H ) . We finally assayed TORC1 activity under these growth and fructose uptake conditions . We observed a Rap-sensitive increase of Sch9 phosphorylation upon fructose uptake by Fsy1-expressing cells . Such a TORC1 activation was observed neither in the wild-type incorporating fructose via the Hxt facilitators nor in the hxt null mutant expressing none fructose transporter ( Figure 4I ) . TORC1 activation in response to fructose uptake thus occurred only when this transport was coupled to H+ influx . This result fully supports the view that the H+ influx is what generates the TORC1 activation signal . In all the above experiments , TORC1 thus seems activated in response to the H+-influx coupled to a nutrient transport reaction . We next determined whether the sole diffusion of H+ via a protonophore such as FCCP might also elicit TORC1 activation . Using a strain stably expressing pHluorin , we first observed that the ionophore caused the cytosolic pH to drop to about 6 . 1 , the pH of the buffered growth medium ( Figure 5A ) . Remarkably , this rapid acidification of the cytosol coincided with hyperphosphorylation of Npr1 , and this response was inhibited by Rap ( Figure 5B ) . Hence , H+ influx mediated by a protonophore also results in TORC1 stimulation . Yet this TORC1 activation , intriguingly , did not lead to increased phosphorylation of Sch9 ( Figure 5C ) . A likely explanation is that an additional control elicited when the cytosol becomes too acidic ( a stressful condition ) impedes phosphorylation of Sch9 by activated TORC1 . This is in keeping with a previous report that when the cytosolic pH drops to values around 6 ( as occurs under glucose starvation or when expression of the Pma1 H+-ATPase is repressed ) , the action of TORC1 on Sch9 is inhibited , whereas nitrogen control of the Gat1 and Gln3 transcription factors via TORC1 remains unaltered ( Dechant et al . , 2014 ) . TORC1 activation in response to an FCCP-mediated H+ influx , thus visible on immunoblots for HA-Npr1 , was largely impaired in the gtr1∆ gtr2∆ strain ( Figure 5D ) , indicating that it is Rag/Gtr-dependent . The multisubunit SEACIT complex antagonizes TORC1 by acting as a GAP on the Gtr1 GTPase , and its function is itself negatively controlled by the multisubunit SEACAT complex ( Panchaud et al . , 2013a , 2013b ) . To determine whether these GATOR-like upstream regulators of Gtr1 are involved in H+ influx-elicited TORC1 activation , FCCP was added to cells lacking Seh1 , a component of the SEACAT complex , or Iml1 , a component of the SEACIT complexe ( Panchaud et al . , 2013a , 2013b ) . TORC1 activation was largely impaired in the seh1∆ mutant ( Figure 5E ) . In the iml1∆ mutant , a high basal phosphorylation of HA-Npr1 was detected , as expected , and FCCP did not significantly further increase this phosphorylation , at least during the first minutes after its addition ( Figure 5F ) . These results indicate that the SEACIT/SEACAT upstream regulators of Gtr1 are involved in H+-influx-elicited stimulation of TORC1 activity . We also found the amount of HA-Npr1 to be much reduced in seh1∆ mutant cells , and a similar effect though less pronounced was observed in the gtr1∆ gtr2∆ strain ( Figures 1G , 4E and 5D ) . This suggests that the abundance of HA-Npr1 is influenced by the activation state of TORC1 . Furthermore , phosphorylation of HA-Npr1 in the above-analyzed mutants was found to increase after prolonged incubation with FCCP ( Figure 5D , E and F ) . It thus seems that FCCP stimulates another mechanism of TORC1 activation that does not depend on the Rag/Gtr GTPases . For instance , prolonged incubation with FCCP might promote release of amino acids from the vacuole or mitochondria , and this could promote TORC1 activation independently of Gtr1/2 . It has in fact been reported that the vacuole-associated Pib2 protein containing a FYVE domain acts in parallel with Gtr1 to promote TORC1 activation ( Kim and Cunningham , 2015; Varlakhanova et al . , 2017 ) . Furthermore , in an in vitro TORC1 kinase assay using isolated vacuoles , the addition of glutamine was found to stimulate TORC1 activity in a manner dependent on Pib2 but not Gtr1 ( Tanigawa and Maeda , 2017 ) . We thus also analyzed the role of Pib2 and found that HA-Npr1 is normally hyperphosphorylated after FCCP addition to pib2∆ mutant cells ( Figure 5G ) . In conclusion , the above experiments indicate that H+ influx mediated even by a protonophore elicits a cellular response resulting in Rag/Gtr-dependent , Pib2-independent , TORC1 activation . They further suggest that if the cytosol becomes too acidic , an additional control likely impedes Sch9 phosphorylation by activated TORC1 . We observed that addition of amino acids , cytosine , or NH4+ to growing cells does not detectably change their cytosolic pH ( data not shown ) . This was expected , given the high buffering capacity of the cytosol and the compensating H+ efflux activity of the Pma1 H+-ATPase , which is stimulated under acidic conditions as long as glucose is present ( Eraso and Gancedo , 1987; Ullah et al . , 2012 ) . We then examined whether an increase of cytosolic H+ , imposed without changing the composition of the external medium , might also lead to TORC1 activation . An H+ increase can in principle be caused by inhibition of the vacuolar V-ATPase , as this enzymatic complex catalyzes ATP-dependent uptake of H+ into the vacuole in order to acidify the organelle , to compensate for the constant H+ efflux mediated by H+-coupled vacuolar transporters , and to control the cytosolic pH ( Kane , 2016 ) . We thus tested the effects of two inhibitors of the V-ATPase , concanamycin A ( CMA ) and bafilomycin A ( BAF ) ( Figure 6 ) . Addition of CMA to proline-grown cells did not significantly change the cytosolic pH of the cells ( Figure 6A ) . This suggests that Pma1-dependent efflux and the buffering capacity of the cytosol prevented the expected increase in cytosolic H+ . BAF addition did cause a slight but significant drop in the cytosolic pH , suggesting that this treatment caused stronger inhibition of the V-ATPase ( Figure 6E ) . Remarkably , both CMA and BAF treatment resulted in stimulation of TORC1 activity , as judged by Rap-sensitive hyperphosphorylation of Npr1 ( Figure 6B and Figure 6F , respectively ) and by a transient increase in Sch9 phosphorylation ( Figure 6C and Figure 6G , respectively ) . No increase in Npr1 phosphorylation was detected in either CMA- or BAF-treated gtr1∆ gtr2∆ mutant cells ( Figure 6D and H , respectively ) . We conclude that inhibition of the V-ATPase is associated with efficient Rag/Gtr-dependent stimulation of TORC1 activity , even when this inhibition is not sufficient to cause a detectable lowering of the cytosolic pH . Activation of TORC1 in the above-described situations ( H+ influx , increased in cytosolic H+ ) might involve an uncharacterized sensor of intracellular H+ , capable of transmitting this signal to TORC1 . Alternatively , the Pma1 H+-ATPase might control TORC1 activity upon sensing H+ influx or increase in cytosol . For instance , Pma1 activity increases under acidic conditions , and this coincides with a reduction of its Km for ATP , possibly via allosteric control ( Eraso and Gancedo , 1987; Ullah et al . , 2012 ) . We hypothesized that the particular state adopted by the H+-ATPase in response to increased H+ might stimulate certain factors controlling TORC1 activity . To test this possibility , we thought of expressing in yeast , instead of the endogenous Pma1 , a heterologous H+-ATPase known to be catalytically active in yeast . We reasoned that if TORC1 activation depends on a signaling capability of Pma1 , an H+-ATPase from a distant species should fail to activate TORC1 in response to an increase in cytosolic H+ . According to previous reports , several plant H+-ATPases are active when expressed in yeast strains where the essential PMA1 gene and its non-essential paralog PMA2 ( expressed to a much lower level ) are deleted or repressed ( Morsomme et al . , 2000; Palmgren and Christensen , 1994 ) . The wild-type forms of these plant H+-ATPases typically compensate only partially for the lack of Pma1 . It is possible , however , to isolate mutant derivatives sustaining faster growth , particularly on low-pH media where cells need high H+-ATPase activity to maintain a neutral cytosolic pH ( Morsomme et al . , 2000 ) . For instance , the H+-ATPase Pma4 of tobacco ( Nicotiana plumbaginifolia ) restores limited growth to a pma1∆ pma2∆ double-null mutant . A truncated Pma4 protein called Pma4882ochre , lacking the last 71 C-terminal amino acids as a result of an ochre nonsense mutation in codon 882 of the PMA4 gene , is able to support faster growth of yeast pma1∆ pma2∆ cells ( Luo et al . , 1999 ) . We thus studied TORC1 activation in cells expressing either the endogenous PMA1 gene or the tobacco plant PMA4822ochre gene . As most the above experiments were carried out with strains having the ∑1278b background , we first isolated a GAL1-PMA1 pma2∆ derivative of this strain , where the PMA1 gene is placed under the control of the galactose-inducible , glucose-repressible GAL1 promoter . This strain can grow on galactose but not glucose , unless it contains a plasmid expressing the endogenous PMA1 gene or the tobacco PMA4822ochre gene under the control of the PMA1 promoter ( Figure 7A ) . We cultured GAL-PMA1 pma2∆ cells expressing PMA1 or PMA4822ochre on glucose proline medium , as in the above-described experiments . We found cells expressing PMA4822ochre to grow more slowly ( Figure 7B ) . This shows that the mutant plant H+-ATPase does not fully compensate for the lack of Pma1 . Accordingly , compared to the cytosolic pH of PMA1-expressing cells , that of PMA4822ochre-expressing cells was slightly lower ( Figure 7C ) . H+-coupled uptake of β-ala ( 1 mM ) was also significantly lower in the latter cells ( data not shown ) . We therefore lowered the concentration of β-ala provided to PMA1-expressing cells in order to reach an uptake rate equivalent to that measured in PMA4822ochre-expressing cells ( Figure 7D and F ) . As expected , and whichever gene was expressed , [14C]-β-ala uptake was inhibited after a brief treatment of the cells with FCCP . This shows that in both cases , β-ala uptake is coupled to H+ influx ( Figure 7D and F ) . Upon transfer of the cells to a glucose-free medium , uptake of β-ala into PMA1-expressing cells was also strongly reduced ( Figure 7D and F ) . This was expected , since Pma1 is inhibited under these conditions . This reduction was much less pronounced in PMA4822ochre-expressing cells ( Figure 7D and F ) . This result can be readily explained by the fact that inactivation of H+-ATPases upon glucose starvation , a regulation conserved between yeast and plant H+-ATPases , requires a C-terminal auto-inhibitory region which the truncated protein Pma4822ochre lacks ( Morsomme et al . , 2000; Portillo , 2000 ) . Next , under conditions of equal H+-coupled β-ala uptake into PMA1- and PMA4822ochre-expressing cells , we analyzed TORC1 activity . Phosphorylation of Npr1 was found to increase moderately upon addition of β-ala at low concentration ( 0 . 1 mM ) to PMA1-expressing cells , but this variation was significant , as judged by its sensitivity to Rap ( Figure 7E ) . Remarkably , no increase in Npr1 phosphorylation was detected in PMA4822ochre-expressing cells ( Figure 7E ) . Furthermore , basal phosphorylation of Npr1 in these cells did not increase , despite the lower pH of their cytosol ( Figure 7E and C ) . Non-activation of TORC1 in PMA4822ochre-expressing cells was even clearer when Sch9 phosphorylation was used as readout ( Figure 7G ) . Basal phosphorylation of Sch9 before β-ala addition was also reduced in PMA4822ochre-expressing cells ( Figure 7G ) . Non-activation of TORC1 in PMA4822ochre-expressing cells upon β-ala uptake was also observed in the background of another strain deleted of both PMA1 and PMA2 ( Figure 7—figure supplement 1 ) . We next analyzed TORC1 activation after addition of glutamine , leucine , or arginine , each activation being deficient in the gtr1Δ gtr2Δ strain ( Figure 7—figure supplement 2 ) . As with β-ala , the external concentration of each amino acid was first adjusted to reach equivalent uptake in PMA1- and PMA4822ochre-expressing cells . Using these conditions , we observed a Rap-sensitive activation of TORC1 in the cells expressing PMA1 but not in those expressing PMA4822ochre ( Figure 7—figure supplement 2 ) . We also analyzed TORC1 activation upon FCCP addition . In both PMA1- and PMA4822ochre-expressing cells , as expected , the ionophore caused a rapid drop in the cytosolic pH , to a value close to the pH of the buffered external medium ( Figure 7C ) . Also as expected , this induced strong hyperphosphorylation of Npr1 in control PMA1-expressing cells ( Figure 7H ) . In contrast , FCCP addition did not increase Npr1 phosphorylation in PMA4822ochre-expressing cells ( Figure 7H ) . Similar results were obtained after treatment with BAF: Npr1 phosphorylation was found to increase in PMA1-expressing but not in PMA4822ochre-expressing cells ( Figure 7I ) . These results show that the endogenous Pma1 H+-ATPase plays an essential role in Rag/Gtr-dependent TORC1 activation in response to increased cytosolic H+ . The importance of Pma1 in stimulating TORC1 activity was also illustrated by the ability of PMA1-expressing cells to resume growth following exposure to Rap , whereas PMA4822ochre-expressing cells failed to do so ( Figure 7J ) . It has been reported that the Rag GTPases are not required for sustained activation of TORC1 in the presence of NH4+ ( Stracka et al . , 2014 ) . In support of this view , NH4+ addition to proline-grown cells caused Rap-sensitive hyperphosphorylation of Npr1 in both wild-type and gtr1∆ gtr2∆ mutant cells ( Figure 7K ) . Furthermore , sustained TORC1 activation after NH4+ addition is reported to depend on the enzymes converting NH4+ to glutamate ( Fayyad-Kazan et al . , 2016; Merhi and AndreAndré , 2012 ) , the main N donor in amino acid biogenesis reactions . Interestingly , upon NH4+ addition to proline-grown cells , we found TORC1 to be properly activated by NH4+ regardless of the H+-ATPase produced ( Pma1 or the plant Pma4822ochre ) ( Figure 7K ) . This result shows that TORC1 can be properly activated in PMA4822ochre-expressing cells . It also suggests that Pma1 is required for TORC1 activation in response to H+ influx but not to an increase in internal amino acids . Uptake of amino acids into N-deprived yeast cells causes Rag/Gtr-dependent activation of TORC1 . The actual signal and the underlying mechanism of this cellular response remain unknown . Our study shows that what triggers this activation is the H+ influx coupled to transport by H+/amino-acid symporters . A similar response is observed upon H+-dependent uptake of cytosine , or even fructose , but not when an equivalent amount of fructose enters the cells via passive transport systems . The activation of TORC1 can even be elicited by diffusion of extracellular H+ via a protonophore or by inactivation of the vacuolar V-ATPase ( which also causes an increase in cytosolic H+ ) . We further show that TORC1 activation in response to an H+ influx and/or an increase in cytosolic H+ requires the Rag GTPases encoded by the GTR genes . It does not , however , require the Pib2 protein recently shown to act in parallel with the Rag/Gtr proteins to activate TORC1 . Finally , we show that the Pma1 H+-ATPase plays a central role in this TORC1 activation pathway . Yeast cells typically adapt to starvation for any nutrient by reducing TORC1 activity . This is probably because mechanisms capable of sensing the nutritional status of the cell impede the activation of TORC1 , and not just because of a possible drop in H+-coupled uptake of this nutrient . This reduction of TORC1 activity typically coincides with increased synthesis of large amounts of high-affinity H+-symporters able to assimilate replenishing compounds , for example , Gap1 for amino acids , Pho84 for phosphate , Sul1 and Sul2 for sulfate , Zrt1 for zinc , Fcy2 for cytosine , and Fur4 for uracil . Furthermore , many permeases for the non-limiting nutrients likely undergo parallel increased endocytosis and degradation , as such responses have been observed in rapamycin-treated cells ( Crapeau et al . , 2014 ) . This permease reconfiguration at the plasma membrane probably causes an overall reduction of the H+ influx . Our results suggest that the H+-symporters that are derepressed under starvation conditions are potentially able to reactivate TORC1 once their substrate becomes available again in the medium . In other words , H+ influx via these transporters could provide a general signal for reactivating TORC1 upon relief from diverse starvation conditions . On the other hand , sustained activation of TORC1 probably requires efficient assimilation of the internalized nutrient . Accordingly , longer term TORC1 activation after addition of a preferential N source such as NH4+ is reported to require glutamine accumulation and/or synthesis ( Stracka et al . , 2014 ) . The influx of H+ coupled to uptake of any growth-limiting nutrient might thus provide a general signal for rapid , transient TORC1 reactivation in order to prepare cells for subsequent growth acceleration or restart once the nutrient has been properly assimilated . It could also contribute to the reactivation of TORC1 observed upon addition of glucose to glucose-starved cells , as although glucose enters cells via Hxt facilitators it also reactivates Pma1 , which in turn drives the H+-coupled uptake of amino acids and other nutrients . Uptake of amino acids by N-starved cells is also reported to elicit transient activation of PKA , resulting in stimulation of trehalase by phosphorylation ( Donaton et al . , 2003 ) . According to current models , PKA could be activated via a signaling pathway stimulated by Gap1 acting as a transceptor , and a similar function has been described for other H+-coupled nutrient transporters found to be derepressed under particular starvation conditions ( Schothorst et al . , 2013 ) . Yet , the mechanisms underlying this signaling remain unknown . It is tempting to envisage an alternative model according to which the signal eliciting PKA activation is the H+ influx coupled to the nutrient uptake reaction , as in the case of TORC1 . This model is worth considering , since addition of FCCP to N-starved cells results in stimulation of trehalase activity ( Figure 5—figure supplement 1 ) . In a previous study , addition of excess Gln to proline-grown cells was found to cause a rapid , transient activation of TORC1 , not observed in the gtr1∆ mutant , followed by a more sustained TORC1 activity , still observed in gtr1∆ cells ( Stracka et al . , 2014 ) . This Gtr1-independent TORC1 activation is reminiscent of the situation described in mouse cells , where Gln activates mTORC1 in a manner independent of RagA and RagB ( Jewell et al . , 2015 ) . Similarly , direct addition of Gln to isolated vacuoles elicits TORC1 activation in vitro in a Gtr1-independent manner ( Tanigawa and Maeda , 2017 ) . In contrast , this response requires the Pib2 protein proposed to act in parallel with Gtr1 to activate TORC1 ( Kim and Cunningham , 2015; Varlakhanova et al . , 2017 ) . These observations suggest that Gln uptake first elicits a transient , Gtr1-dependent activation of TORC1 , and we propose that the signal of this early activation is the H+ influx coupled to Gln transport . The subsequent sustained activation of TORC1 , in contrast , is suggested to be promoted by the intracellular accumulation of Gln ( Stracka et al . , 2014 ) . The actual function of Pib2 in this process needs further investigation . The same applies to Gtr1/2 because , while sustained TORC1 activation occurs normally after Gln addition to gtr1∆ cells ( Stracka et al . , 2014 ) , we failed to observe it in the double gtr1∆ gtr2∆ mutant , in keeping with another study ( Varlakhanova et al . , 2017 ) . We have found the Sch9 kinase to be stimulated by TORC1 in response to an increase in cytosolic H+ . This observation is interesting , in the light of the recent finding that Sch9 contributes to pH homeostasis by controlling the assembly and activity of the vacuolar V-ATPase ( Wilms et al . , 2017 ) . This is relevant because the latter , together with the plasma membrane H+-ATPase , contributes importantly to controlling the cytosolic pH ( Kane , 2016 ) . Yet according to other reports , Sch9 phosphorylation and cell growth are reduced when the cytosol becomes acidic ( e . g . after a drop to a pH near 6 ) , for instance under glucose starvation or when Pma1 synthesis is reduced ( Dechant et al . , 2014; Orij et al . , 2012; Ullah et al . , 2012 ) . It thus seems that even though Sch9 is activated by an H+ influx and/or by an increase in cytosolic H+ , it is inhibited when the cytosol becomes too acidic , and this causes growth inhibition . In FCCP-treated cells , accordingly , TORC1 appears to be efficiently activated ( as judged by hyperphosphorylation of the Npr1 kinase ) , but this is not accompanied by Sch9 phosphorylation . This suggests that a particular mechanism sensitive to acidic conditions hampers Sch9 phosphorylation by activated TORC1 . Such a control seems physiologically relevant , as acidification of the cytosol is stressful for the cell and stimulation of growth under these conditions would be inappropriate . Accordingly , other stresses are reported to promote dephosphorylation of Sch9 without affecting the TORC1-regulated Tap42-PP2A branch controlling Npr1 phosphorylation ( Hughes Hallett et al . , 2014 ) . A key question raised by our work is: what is the molecular mechanism responsible for stimulation of TORC1 activity in response to H+ influx and/or an increase in cytosolic H+ ? Our data indicate that the plasma membrane Pma1 H+-ATPase plays a central role in this cellular response . Specifically , TORC1 activity fails to be stimulated in response to H+-coupled uptake of amino acids , ionophore-mediated H+ diffusion , or inhibition of the V-ATPase in cells producing the tobacco plant Pma4822ochre instead of Pma1 . Furthermore , these cells display a strongly reduced ability to restart growth after exposure to rapamycin . At least two models can be proposed to account for these observations . On the one hand , non-activation of TORC1 might result indirectly from the inability of Pma4822ochre to fully compensate for the lack of Pma1 activity . We did find PMA4822ochre -expressing cells to grow more slowly and their cytosol to be slightly acidic . This might trigger adaptive feedback mechanisms impeding TORC1 reactivation . Yet in PMA4822ochre-expressing cells we found TORC1 to be properly activated after NH4+ addition , and this shows that TORC1 activity can be efficiently stimulated at least via the Rag/Gtr-independent pathway seemingly responding to internal amino acids ( Stracka et al . , 2014 ) . Alternatively , the essential role of Pma1 in TORC1 activation in response to H+ influx might reflect the ability of Pma1 to stimulate a signaling pathway controlling TORC1 activity . For instance , the activity of Pma1 is known to increase when the concentration of H+ in the cytosol rises , and this control involves a decreased Km for ATP ( Eraso and Gancedo , 1987; Ullah et al . , 2012 ) . This activity increase also likely occurs when protons are co-transported with nutrients via plasma membrane H+-symporters . This stimulation of Pma1 activity , possibly involving a conformational change of the H+-ATPase , might be transmitted to cytosolic factors that would in turn modulate TORC1 activity . A role of Pma1 in signaling to TORC1 is attractive , because Pma1 is the main ATP-consuming enzyme of yeast and is thus ideally positioned for sensing cellular ATP levels . Furthermore , as mentioned above , Pma1 stimulation by H+ influx could also give cells a general mechanism for sensing relief from starvation for any nutrient and for reactivating TORC1 in response to this relief . Other observations support a role of Pma1 in signaling to TORC1 . For instance , TORC1 inhibition has been observed when Pma1 synthesis is reduced ( Dechant et al . , 2014 ) . Furthermore , yeast TORC1 is rapidly inhibited under glucose starvation ( Urban et al . , 2007 ) , this coinciding with polymerization of the kinase complex into a single , vacuole-associated cylindrical structure ( Prouteau et al . , 2017 ) . Although a specific mechanism involving phosphorylation of the Kog1 subunit is reported to contribute to this TORC1 inhibition ( Hughes Hallett et al . , 2015 ) , a role of Pma1 might also be considered , as this H+-ATPase is subject to rapid and reversible auto-inhibition under these conditions ( Portillo et al . , 1989; Serrano , 1983 ) . Interestingly , TORC1 has been reported recently to be required for full activity of Pma1 ( Mahmoud et al . , 2017 ) , suggesting the existence of some crosstalk between Pma1 and TORC1 . The model according to which Pma1 is capable of controlling TORC1 via signaling also seems reasonable in the light of previous works showing that the Na+/K+-ATPase of animals cells , a P-type ATPase structurally similar to Pma1 and other H+-ATPases , is engaged in dynamic interactions with other proteins , including the Src tyrosine kinase . The interaction with Src is modulated by the conformation of the ion pump and initiates signal transduction processes ( Cui and Xie , 2017 ) . As the cytosolic region of the Na+/K+-ATPase interacting directly with Src ( Lai et al . , 2013 ) is relatively well conserved in the yeast H+-ATPase , we introduced several substitutions into this region of Pma1 with a view of disrupting possible interactions with other factors . The generated Pma1 variants , however , behaved normally in TORC1 activation assays . In conclusion , our results show that cytosolic H+ and Pma1 are major actors in TORC1 activation in response to active nutrient uptake . They also raise the interesting possibility that Pma1 might control TORC1 via signaling . Further work is needed to evaluate this model , which would open important prospects for work on nutritional signaling in yeast and other organisms . The yeast strains used in this study ( Table 1 ) derive from the Σ1278b wild type , the only exceptions being YPS14-4 ( W303 ) , JW00035 ( W303 ) , CEN . PK2-1c ( VW1A ) , EBY . VW 4000 , and I3 . Cells were grown at 29°C on a minimal medium buffered at pH 6 . 1 ( Jacobs et al . , 1980 ) , with glucose ( Gluc ) ( 3% w/v ) , maltose ( 3% w/v ) , galactose ( Gal ) ( 3% w/v ) , or ethanol ( EtOH ) ( 1% v/v ) as a carbon source . For cultures in Gal medium , a low concentration of Gluc ( 0 . 3% w/v ) was also added to boost initiation of growth . The nitrogen ( N ) sources added to liquid growth media were NH4+ as ( NH4 ) 2SO4 ( 20 mM ) , proline ( Pro ) ( 10 mM ) , or urea ( 10 mM ) . For strain YPS14-4 and its derivative expressing PMA4882-ochre , cells were grown on the same buffered minimal medium adjusted to pH 6 . 5 to improve growth . In all experiments , cells were examined or collected during exponential growth , a significant and regular number of generations after seeding . Our experience is that these precautions and the use of a minimal medium that is buffered considerably improve the reproducibility of data between biological replicates ( Wiame et al . , 1985 ) . When indicated , rapamycin ( Rap ) at 200 ng/ml concentration was added for 30 min . The ura3 mutation present in all strains was complemented by transformation with a plasmid , for example , pFL38 . Comparative analyses of growth were performed by growing cells in a Greiner 24-well microplate incubator coupled to a SYNERGY multi-mode reader ( BioTek Instruments ) . The plasmids used in this study are listed in Table 2 . Growing cells were laid on a thin layer of 1% agarose and viewed at room temperature with a fluorescence microscope ( Eclipse E600; Nikon ) equipped with a 100 differential interference contrast , numerical aperture ( NA ) 1 . 40 Plan-Apochromat objective ( Nikon ) and appropriate fluorescence light filter sets . Images were captured with a digital camera ( DXM1200; Nikon ) and ACT-1 acquisition software ( Nikon ) and processed with Photoshop CS ( Adobe Systems ) . In each figure , we typically show only a few cells , representative of the whole population . Labeling of the vacuolar membrane with CMAC fluorescent dye was performed by adding 1 µl of the dye to 5 ml of culture at least 30 min prior to visualization . For western blot analysis , crude cell extracts were prepared as previously described ( Hein et al . , 1995 ) . Proteins were transferred to a nitrocellulose membrane ( Schleicher and Schuell; catalog number NBA085B ) and probed with mouse anti-GFP ( Roche; catalog number 11 814 460 001 ) , anti-hemagglutinin ( anti-HA ) ( 12CA5; Roche ) , or anti-yeast 3-phosphoglycerate kinase ( anti-PGK ) ( Invitrogen ) or with rabbit anti-Pma1 ( De Craene et al . , 2001 ) , anti-phospho-Thr737-Sch9 , or anti-Sch9Total ( see below ) . Primary antibodies were detected with horseradish-peroxidase-conjugated anti-mouse or anti-rabbit immunoglobulin G secondary antibodies ( GE Healthcare ) , followed by enhanced chemiluminescence ( Roche; catalog number 12 015 196 001 ) . Each Western blot was carried two to four times , a representative experiment is presented . The antibody was produced by and purchased from the GeneCust company . The CKFAGF ( pT ) FVDESAID peptide containing phosphorylated Thr737 was injected into rabbit . The affinity of the antibody preparation was tested in an ELISA for the phosphorylated peptide . Antibody specificity was tested by western blot analysis of cell extracts isolated from proline-grown wild-type ( w-t ) and sch9∆ strains , before and after addition of NH4+ , well known to stimulate Sch9 phosphorylation ( Figure 1—figure supplement 2 ) . The anti-Sch9Total antibody was a kind gift of Robbie Loewith . Yeast strains expressing a single pHluorin gene integrated into the genome or containing a multicopy plasmid expressing the pHluorin gene were grown at 29°C on Gluc proline buffered medium , pH 6 . 1 , to OD660 ~0 . 2 . Fluorescence intensities were recorded with a SYNERGY multi-mode microplate reader ( BioTek Instruments ) with emission filter 512/9 nm and excitation filters 395/9 and 475/9 nm , as previously reported ( Orij et al . , 2009; Zimmermannova et al . , 2015 ) . To eliminate the background fluorescence , pHluorin-nonproducing wild-type cells were grown in parallel , and the corresponding values at each excitation wavelength were subtracted from those of pHluorin-producing cells . The I395 nm to I475 nm emission intensity ratio was used to calculate the cytosolic pH . The fluorescence intensities of each strain were typically recorded in four separate cultures ( 1 ml culture per well ) within one experiment ( technical replicates ) , and the presented data are means ±SD of at least two independent experiments ( biological replicates ) . The calibration curve was generated as described previously ( Orij et al . , 2009; Zimmermannova et al . , 2015 ) , with minor changes . The cell culture ( 100 ml , OD660 = 0 . 2 ) was filtered , washed , resuspended in 8 ml phosphate-buffered saline ( Sigma ) containing digitonin ( 175 μg/ml ) , and incubated for 15 min at RT . Digitonin was washed out and the cells were resuspended in 8 ml PBS ( the OD of the cell suspension was about 2 . 5 ) and placed on ice . Then 40 µl aliquots were transferred to CELLSTAR black polystyrene clear-bottom 96-well microtiter plates ( Greiner Bio-One ) containing , per well , 160 μl citric acid/Na2HPO4 buffer at a pH ranging from 5 . 6 to 7 . 6 ( in this volume , the OD was 0 . 5 ) . Recording of pHluorin fluorescence emission and background subtraction were performed as described above . The I395 nm to I475 nm intensity ratio was calculated , plotted against the corresponding buffer pH , and fitted to a third-degree polynomial regression curve . The accumulation of [14C]-labeled amino acids or [14C]-labeled-fructose was measured at the time points indicated as previously described ( Ghaddar et al . , 2014a; Grenson et al . , 1966 ) . The radiolabeled compounds were purchased either from Perkin-Elmer or from Hartmann analytic . Data points represent averages of two biological replicates; error bars represent standard deviations ( SD ) . Yeast cultures ( 50 ml ) were collected by centrifugation ( 7000 g for 3 min ) and washed twice with 10 ml Milli-Q water . The final pellet was resuspended in 2 ml Milli-Q water and boiled for 15 min . To remove cell debris , suspensions were centrifuged at 13 , 000 g for 1 min and filtered ( Millipore 0 . 45 μm ) . The resulting soluble fractions were subjected to amino acid analysis after AccQ Tag pre-column derivatization ( Waters ) . For this an AccQ Tag Ultra UPLC column ( Waters ) with UV detection at 260 nm was used according to the manufacturer’s recommendations ( Fayyad-Kazan et al . , 2016 ) . Cells growing on Gluc NH4+ medium were collected by filtration , and after washing and resuspension in Gluc medium without any N source , they were incubated overnight at 29°C with shaking . Cells were filtered , washed , and transferred again to fresh N-free Gluc medium for 30 min before addition of FCCP ( 20 μM ) . Culture samples were collected at various times and trehalase activity was measured in permeabilized cells as previously described ( De Virgilio et al . , 1991 ) . Glucose levels were measured using the ‘Glucose assay kit’ ( Sigma-Aldrich , Belgium ) .

Cells adapt their growth rate depending on the amount of nutrients available . The protein complex called TORC1 plays a central role in this . When nutrients are abundant , TORC1 is very active and stimulates the production of proteins and other molecules needed for the cell to grow . However , when nutrients such as amino acids become scarce , TORC1 reduces its activity and allows the cells to adapt to starvation . This TORC1-mediated control of the metabolism is crucial for the cell to survive , and faulty TORC1 proteins have been associated with several diseases including cancers . TORC1 was originally discovered in yeast , which provides a powerful model to study this control system . However , until now , it was not known how TORC1 is reactivated when amino acids are added to cells that have been starved of these molecules . Knowing the answer to this question would allow us to better understand how the availability of nutrients controls the activity of TORC1 . Now , Saliba et al . have discovered that TORC1 is not reactivated by the amino acids themselves , but by protons , which are positively charged hydrogen ions that travel into the cell together with the amino acids . This influx of protons is the driving force behind the active transport of amino acids and other nutrients into the cell , and potentially serves as a general signal to activate TORC1 in response to the uptake of nutrients , especially when cells have been starved . Furthermore , the results showed that a specific enzyme in the cell membrane plays an essential role in activating TORC1 . This enzyme pumps the protons out of the cell to compensate for their influx and to maintain the proton gradient in the membrane that drives the absorption of nutrients . When this enzyme was replaced with an equivalent plant enzyme , the proton-coupled nutrient uptake did not activate TORC1 in the yeast cells . These findings may help scientists who are interested in how TORC1 is regulated in organisms other than mammals , such as plants or fungi . A next step will be to find out how exactly the proton pump in the cell membrane helps to activate TORC1 .

2018

The yeast H+-ATPase Pma1 promotes Rag/Gtr-dependent TORC1 activation in response to H+-coupled nutrient uptake

Decision-making behavior is often characterized by substantial variability , but its source remains unclear . We developed a visual accumulation of evidence task designed to quantify sources of noise and to be performed during voluntary head restraint , enabling cellular resolution imaging in future studies . Rats accumulated discrete numbers of flashes presented to the left and right visual hemifields and indicated the side that had the greater number of flashes . Using a signal-detection theory-based model , we found that the standard deviation in their internal estimate of flash number scaled linearly with the number of flashes . This indicates a major source of noise that , surprisingly , is not consistent with the widely used 'drift-diffusion modeling' ( DDM ) approach but is instead closely related to proposed models of numerical cognition and counting . We speculate that this form of noise could be important in accumulation of evidence tasks generally . Subjects performing perceptual decision-making tasks display a large amount of trial-to-trial behavioral variability . Determining the sources of this variability could provide insight into the neural mechanisms of decision-making and produce more accurate predictions of behavior . It has been proposed that behavioral variability is caused in part by noise accruing during the process of evidence accumulation . This noise may have a variety of origins depending on the behavioral task . It can be inherent in the natural world , produced by the signal detection limits of sensory organs themselves ( Barlow and Levick , 1969 ) , or it may reflect the variability of neural responses in the brain at different stages of processing . Previous studies have attempted to trace the sources of noise using a combination of behavioral and neurophysiological approaches . At the behavioral level , many models of the evidence accumulation process are known as ‘drift-diffusion’ models ( DDMs ) , because a major component of the noise is modeled as diffusion noise in the accumulator , i . e . , noise that is independent for each time point ( Bogacz et al . , 2006; Smith and Ratcliff , 2004 , but see Zariwala et al . , 2013 ) . A recent study by Brunton and colleagues ( Brunton et al . , 2013 ) developed an accumulation of evidence task , which they modeled with a DDM-like process , to isolate different sources of noise . In their task , evidence was delivered in randomly timed but precisely known pulses: two randomly generated streams of discrete auditory clicks were presented from left and right speakers , and subjects ( rats and humans ) were trained to indicate the side with the greater number of clicks . The precise timing of the stimuli combined with large numbers of behavioral trials enabled fitting of a detailed and statistically powerful behavioral model . The results suggested that in most subjects the diffusion noise in the accumulator was essentially zero ( noiseless ) , and the behavioral variability was best explained by noise that was added with each pulse of evidence . However , it was assumed that each pulse introduced independent noise . This implied that although time-locked to the stimulus , the noise associated with the stimulus was diffusion-like , and the standard deviation of the total stimulus-induced noise scaled as the square root of the number of pulses . Overall , the model was qualitatively consistent with the widely-assumed diffusion-like noise in the evidence accumulation process . However , similar to other reports , diffusion-like noise alone was unable to fully predict behavioral variability , and specifically failed to account for errors on the easiest trials . Following common practice , Brunton et al . ( 2013 ) included an additional noise parameter , termed lapse rate , which described variability that did not depend on the stimulus or trial duration . The need for the lapse rate term suggests the existence of additional sources of noise that are not diffusion-like . To better characterize sources of noise during perceptual decision-making , here we developed a visual analogue of the Brunton et al . accumulation of evidence task with two key features ( Erlich et al . , 2013 , SFN , abstract ) . First , rats could perform the task during voluntary head-restraint ( Girman , 1980; 1985; Kampff et al . , 2010 , SFN , abstract ) , allowing for the potential of cellular resolution imaging ( Scott et al . , 2013 ) and perturbation in future studies ( Rickgauer et al . , 2014 ) . Second , the sensory stimuli were designed to attempt to isolate internally generated noise by minimizing noise inherited from the stimulus . In our task , rats were presented with two series of brief ( 10 ms ) , pseudorandomly timed LED flashes to the left and right visual hemifields . Following the form of the Brunton et al . ( 2013 ) task , the side with the greater total number of flashes indicated the location of a water reward . However , whereas the acoustics of the behavior chamber could distort localization of auditory stimuli , here our visual stimuli were presented from well-separated LEDs positioned to the left and right visual hemifields , suggesting that the rats would have no difficulty distinguishing the left versus right origin of each flash . Additionally , low numbers of flashes were presented in well-separated time bins . Behavioral analyses indicated that rats solved the task by accumulating all available evidence , and that flash-associated noise was the predominant source of behavioral variability . We implemented a signal detection theory-based model to evaluate the assumption that noise in the accumulator value was diffusion-like , i . e . that noise was independent for each flash and the noise ( standard deviation ) of the accumulator value scaled as the square root of the number of flashes . The signal detection theory-based model was designed to find the scaling relationship between noise in the accumulator value and number of flashes that best described the data . Surprisingly , model fits revealed that the noise in the subjects’ numerical estimates scaled linearly with the number of flashes presented , not as the square root . This relationship , called scalar variability , has previously been observed in tasks that require subjects to estimate the duration of a stimulus ( Gibbon , 1977 ) or count the number of stimuli ( Gallistel and Gelman , 2000 ) . Moreover , a behavioral model implementing scalar variability predicted imperfect behavioral performance on the easiest trials , without the need for a non-zero ‘lapse rate’ parameter . Linear scaling of the standard deviation of the noise with the number of flashes reveals a source of noise that does not treat individual pulses independently . Therefore , the noise is not diffusion-like; unlike a lapse rate it depends on total number of evidence pulses; and it is introduced after single-flash sensory processing . We suggest that taking into account this form of noise will be a critical factor for understanding variability in decision-making behaviors . An orthogonal approach for measuring noise in the accumulation process is to relate the variability of neuronal responses to behavior ( Shadlen et al . , 1996; Cohen and Newsome , 2009; Mazurek et al . , 2003 ) . Ideally , this approach would include recordings from multiple neurons across brain regions and precise neural circuit manipulations . To this end , we developed a version of our task that rats could perform during voluntary head restraint . Voluntary head restraint was developed as an alternative to forced head restraint in which the initiation and termination of the restraint period are under the control of the animal ( Girman , 1980; 1985; A . R . Kampff et al . , 2010 , SFN , abstract ) . Recently we reported that rats can be trained to perform voluntary head restraint in a high-throughput semi-automated behavioral facility and described a voluntary head restraint system that provides the stability needed for in vivo cellular resolution imaging ( Scott et al . , 2013 ) . Here we show that performance of head restrained rats was essentially identical to the performance of rats trained on the unrestrained version of the task . These results demonstrate that rats can perform complex cognitive behaviors during voluntary head restraint and provide a platform for characterizing noise during decision-making across multiple brain regions . Accumulation of evidence involves two processes: maintaining a memory of the evidence and adding new evidence to that memory . To assess whether noise ( and thus behavioral variability ) was more closely associated with the memory of the accumulator or with incoming sensory evidence , we initially fit the Brunton et al . model to our visual task data . This produced results consistent with those found for Brunton et al . ’s auditory task , including near-zero estimates of accumulator memory noise ( i . e . , a predominant role for incoming sensory evidence noise ) , and long accumulation time constants ( Figure 2—figure supplement 1 ) . However , further analysis described below led us to question Brunton et al . ’s assumption of independent noise across pulses of sensory evidence . We therefore took a model-free approach to estimate whether noise was more closely associated with the memory of the accumulator or with incoming sensory evidence . When trial duration and flash difference were held constant , errors increased with total number of flashes presented , suggesting that noise increased with each flash ( Figure 2A ) . Next , we sought to directly compare the effects of flashes and time on behavioral performance ( % correct ) . First , looking across trials with identical differences in flash number ( fixed |#R-#L| = ΔF ) but with varying total flash number ( #R+#L=ΣF ) , we calculated the fraction of correct responses as a function of ΣF , relative to the average performance ( Δ Performance; Figure 2C ) . With the difference in flashes ΔF thus controlled for , we found that trials with greater numbers of flashes showed a substantial decrease in performance . Linear regression suggested that each additional flash decreased performance by 1 . 14% ( +/-0 . 1% ) . With an average of 4 flashes presented per second , increasing total flashes at a fixed flash difference thus produced an average decrement in performance of 4 . 56% per second . Then , to estimate the effect of time on performance , we calculated Δ performance across trials with both identical flash differences ( ΔF ) and identical total flash number ΣF ) , but with different overall trial duration . This analysis , which controls for the number of flashes , revealed that the purely time-dependent decrement in performance was only 0 . 037% ( +/- . 77% ) for each second of trial duration . These results suggest that error rates depend far more strongly on the number of flashes than on trial duration . We found this both for the group data of Figure 2 and for individual rats as well ( Figure 2—figure supplement 2; Figure 2—figure supplement 3 ) . 10 . 7554/eLife . 11308 . 006Figure 2 . Error rate increases with number of flashes , not trial duration . ( A ) For trials of fixed duration and fixed difference in number of flashes ( Δ-flashes , colored lines ) , behavioral performance decreased with more flashes . ( B ) Reverse correlation analysis indicating the relative contribution of flashes occurring at different times in the trial to the subject’s behavioral choice . Each point on the upper line represents the probability that on trials in which the subject poked to the right , there was an extra right flash in each time bin . The lower lines represent the same analysis for trials in which the subject poked left . The flatness of the lines suggests that rats use early , middle and late flashes equally to guide their decision . Lines and error bars represent mean and standard error across rats . This result suggests a long time constant of accumulation . ( C ) Changes in behavioral performance ( % correct ) as a function of the number of total flashes presented . Data points indicate behavioral performance relative to the average performance ( Δ Performance ) across trials with identical differences in flash number ( |#R-#L| = ΔF ) but with varied total flash number ( #R+#L=F ) . Red lines are regression lines to those points , weighted by the number of behavioral trials contributing to each point . ( D ) Changes in behavioral performance as a function of the trial duration . Performance across all delay durations was computed for each unique combination of total number of flashes and difference in the number of flashes . For each unique combination of flash number and difference , performance relative to that average performance ( Δ Performance ) was computed for different trial durations binned in 50 ms bins ( black circles ) . Red lines are regression lines to those points , weighted by the number of behavioral trials contributing to each point . DOI: http://dx . doi . org/10 . 7554/eLife . 11308 . 00610 . 7554/eLife . 11308 . 007Figure 2—figure supplement 1 . Fits of drift diffusion-like model to individual rats . ( A ) Schematic of the accumulation model of Brunton et al . ( 2013 ) used here to compare the contribution of flash- and time-associated noise to behavioral variability . At each moment in time the model represents the accumulated evidence as a decision variable , a ( t ) ( black line ) . Colored arrows indicate the timing of left ( blue ) and right ( red ) flashes . σs2 parameterizes the noise added with each flash , σa2 parameterizes the noise added at each time point , τ ( or 1/λ ) parameterizes the memory time constant of a ( t ) . λ<0 suggests that the memory decays to a = 0 with time , λ>0 suggests that the magnitude of a ( t ) increases over time . B parameterizes sticky bounds: if a ( t ) ever reaches +/-B , integration stops and the animal is committed to that decision ( go right/go left , respectively ) . There are a few other terms in the model that are not represented in the schematic . A bias term represents an offset of a ( t ) at the beginning of each trial . A lapse rate parameterizes the percent of trials on which the animal behaves randomly . φ and τφ parameterize sensory adaptation dynamics . After a flash , φ is a constant that scales the effect of the flash; it recovers to an unadapted/facilitated magnitude with time constant τφ . φ >1 indicates that successive flashes facilitate; φ <1 indicates that they depress . Fits to those parameters are not shown here , because we found that the inter-flash-intervals presented here were sufficiently long to minimize any adaptation/facilitation effects of subsequent flashes . Light gray lines indicate alternative runs of the model on the same trial . ( B ) Model fits of flash-associated noise ( σs2; blue circles ) and noise associated with time ( σa2; red circles ) . To evaluate these parameters in comparable units , σs2 was divided by the average number of flashes per second for each rat . For each subject , noise associated with time ( σa2 ) is close to zero , whereas noise correlated with flashes ( σs2 ) is predominant , consistent with previous studies ( Brunton et al . , 2013; Hanks et al . , 2015 ) . ( C ) The drift in the accumulator’s memory is parameterized by λ . A leaky integrator would have negative values of λ , an impulsive integrator would have positive values . The time constant of the integrator , τ , is 1/λ . For many , but not all , of the rats , the time constant is close to or greater than 2 s ( λ <= 0 . 5 ) . ( D ) The bound value for all except two rats is larger than the maximum number of flashes ( on one side ) that each rat experienced . This suggests that rats accumulated/used all of the flashes to inform their decision . ( E ) The bias term for the each rat . ( F ) The lapse rate term for each rat , represented as the percent of trials in which animals behaved randomly . ( G ) The number of behavioral trials that were used to fit the model parameters for each rat , represented on a logarithmic scale . DOI: http://dx . doi . org/10 . 7554/eLife . 11308 . 00710 . 7554/eLife . 11308 . 008Figure 2—figure supplement 2 . Effect of flash number on performance of individual rats . Changes in behavioral performance ( % correct ) for each rat as a function of the number of total flashes presented . Data points indicate behavioral performance relative to the average performance ( Δ Performance ) across trials with identical differences in flash number but with varied total flash number . Red lines are regression lines to those points , weighted by the number of behavioral trials contributing to each point . DOI: http://dx . doi . org/10 . 7554/eLife . 11308 . 00810 . 7554/eLife . 11308 . 009Figure 2—figure supplement 3 . Effect of trial duration on performance of individual rats . Changes in behavioral performance ( % correct ) for each rat as a function of trial duration . Δ Performance was computed by estimating the average performance ( % correct ) across all trials with a unique combination of #R-#L and #R+#L , regardless of trial duration . Then , for each unique combination of flash number and difference , performance relative to the average performance in that condition ( Δ Performance ) was computed for different delay durations , binned in 50 ms bins ( black circles ) . Red lines are regression lines to those points , weighted by the number of behavioral trials contributing to each point . DOI: http://dx . doi . org/10 . 7554/eLife . 11308 . 00910 . 7554/eLife . 11308 . 010Figure 2—figure supplement 4 . Psychophysical reverse correlation for each individual rat . Reverse correlation analysis indicating the relative contribution of flashes occurring at different times in the trial to the subject’s behavioral choice . Each point on the upper ( red ) lines represents the probability that on trials in which the subject poked to the right , there was an extra right flash in each time bin . The lower ( blue ) lines represent the same analysis for trials in which the subject poked left . The flatness of the lines for all rats suggests that they use evidence from early , middle and late flashes equally to guide their decision . S142 did not have enough trials with 8 time bins to be included in this analysis . These results suggest time constants of accumulation that are longer than the duration of the trial . DOI: http://dx . doi . org/10 . 7554/eLife . 11308 . 010 To assess whether the rats exhibited long accumulation time constants we next computed the ‘psychophysical reverse correlation’ ( Brunton et al . , 2013; Kiani et al . , 2008; Nienborg and Cumming , 2009 ; see Materials and methods ) . For each time bin , we computed the probability that there was an excess flash in that bin on the side to which the subjects subsequently oriented . This analysis indicated that flashes across all time bins contribute equally to subjects’ decisions , both when data were pooled across rats ( Figure 2B ) , and for individual rats ( Figure 2—figure supplement 4 ) . This result suggests accumulation time constants that are longer than the trial duration , since subjects’ choices were equally influenced by early , middle , and late flashes . Taken together , behavioral analyses and behavioral model fits suggested that rats based their choices on evidence accumulated over the entire trial duration and that the noise associated with each flash was the predominant source of noise in the accumulation process . As described in the introduction , in the drift diffusion framework assumed in previous behavioral models of evidence accumulation , the standard deviation of the total flash-associated noise scales as the square root of the total number of flashes ( Brunton et al . , 2013 ) . However , two popular alternative models , called scalar variability and subitizing , have been proposed in which noise scales differently with evidence . In the scalar variability model , which has been used to describe both time estimation and counting , the standard deviation of the estimate shows a linear relationship with the quantity represented . The subitizing model predicts that the numerical representation of the first few numbers , up to three or four , is essentially noiseless , and after five its standard deviation grows linearly with the number of flashes . Subitizing has been proposed in tasks in which stimuli are presented simultaneously ( Trick and Pylyshyn , 1994 ) and also sequentially ( Camos and Tillmann , 2008; von Glasersfeld , 1982 ) . To quantify the amount of noise associated with different numbers of flashes , we used a signal detection theory-based framework that abstracted away the sequential nature of the stimulus presentation , and focused instead on the total number of flashes on each side , and noise associated with those totals , as the key determinants of the animal’s decisions ( Figure 3A ) . In this approach , the subjects’ estimate of the number of flashes on one side is modeled as a random variable drawn from a Gaussian distribution . For n flashes presented on a side , the mean of this distribution is equal to n , and the variance is a free parameter , σn2 . If there can be up to 15 flashes presented on a side , then , there will be 16 free parameters in the model , σ02 through σ152 . On each trial the subject selects two such random variables , one for the total number of flashes on the left and one for total flashes on the right , compares the two variables and orients to the side for which the random sample was greater . Correct responses occur when the random sample associated with the side that had the greater number of flashes is greater than the random sample from the other side . 10 . 7554/eLife . 11308 . 011Figure 3 . Signal detection theory-based model reveals linear scaling of standard deviation of numerical estimates . ( A ) Schematic of the model used to determine the standard deviation ( σ ) of the subjects’ estimate of flash number . Left panel indicates the stimuli from an example trial in which four flashes were presented to the left side ( green ) and six flashes were presented to the right ( orange ) . The model assumes that on any given trial , a subjects’ estimate of flash number on each side is a continuous random variable drawn from a Gaussian distribution whose mean is the number of flashes on that side ( n ) , and whose variance is a free parameter ( σn2 ) ( middle panel ) . The choice on each trial is determined by comparing these two random variables . Errors occur when the difference of the two random variables ( greater magnitude – lesser magnitude ) is less than zero . The variance for each Gaussian representing a given flash number ( σ02 … σ152 ) was fit to the behavioral data ( right panel ) using maximum likelihood estimation . ( B ) Model fits of the standard deviations ( σ0 …σ15 ) in the rats’ estimate for different numbers of flashes . Error bars indicate the 95% confidence intervals for the mean based on one thousand-fold resampled data . Note the deviation from pure linear dependence of the σn parameters on n for n<2 . ( C ) Comparison of the behavioral data ( left panel ) with the predictions of the model ( right panel ) based on the σn values calculated as shown in Figure 3B . Color indicates the percentage of trials on which the subject responded correctly . ( D ) Comparison of psychometric performance of the rats ( data , green triangles ) and model prediction of performance ( model , blue squares ) . ( E ) Three models that predict how the standard deviation ( σ ) of the numerical estimate scales with the number of flashes . Scalar variability predicts that σ scales linearly with the number of flashes ( SV , yellow ) . Subitizing predicts that σ is zero until a limit ( 3 or 4 ) , and then follows scalar variability prediction ( black dashed ) . The drift diffusion models predict that the variance of the estimate scales linearly , and σ scales with the square root of the number of flashes ( LV , blue ) . Purple triangles are model estimates of σ , replotted from Figure 3B . Each model was fit using linear regression to the model estimate of σ , weighted by the number of data points contributing to each triangle . Additionally , all models were constrained to intersect with the origin . ( F ) Goodness of fit of the σs shown in Figure 3B to subitizing ( SUB+SV , black ) , the drift diffusion model ( LV , blue ) and scalar variability ( SV , yellow ) using least squares regression . Analysis indicates that the data is best fit by a scalar variability model . Error bars represent the 95% confidence intervals based on fits derived from a thousand-fold resampling of the data . DOI: http://dx . doi . org/10 . 7554/eLife . 11308 . 01110 . 7554/eLife . 11308 . 012Figure 3—figure supplement 1 . Signal detection theory-based model fit to individual rats . Results of fitting the model described in Figure 3 for each rat: model fits of the standard deviations ( σ0 …σ15 ) in each rat’s estimate for different numbers of flashes . Error bars indicate the 95% confidence intervals for the mean based on one thousand-fold resampled data . Red lines are regression lines weighted by the number of behavioral trials contributing to each data point . DOI: http://dx . doi . org/10 . 7554/eLife . 11308 . 01210 . 7554/eLife . 11308 . 013Figure 3—figure supplement 2 . SDT model prediction vs . data for each rat . Comparison of psychometric performance of each rat ( data , blue circles ) and the signal detection theory-based model prediction of performance ( model , green lines ) . For each rat , the model recapitulates the behavioral data , including the lapse rate ( i . e . imperfect performance on the easiest trials ) , which was not explicitly parameterized in the model . DOI: http://dx . doi . org/10 . 7554/eLife . 11308 . 01310 . 7554/eLife . 11308 . 014Figure 3—figure supplement 3 . Behavior and SDT model approximate scalar variability . Scalar variability ( and Weber’s law ) predicts that scaling the number of flashes by the same factor would lead to identical discriminability/performance . Here we sought to test that directly . Circles represent behavioral performance ( pooled across rats ) on trials with different ratios of flashes on the unrewarded ( i . e . “low” ) and rewarded ( i . e . “high” ) sides . Error bars are 95% confidence intervals for a binomial distribution . Each color represents a fixed low:high ratio; each point within a color group represents a fixed ratio of low:high flash numbers multiplied by a constant factor . The solid lines represent the signal detection theory model prediction on trials with different low:high ratios . If performance were constant for a fixed ratio of low:high flashes , points within a color group should lie on a horizontal line . The data and model prediction approximate scalar variability , but interestingly , deviate from it for trials with flashes on only one side ( purple circles , line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11308 . 01410 . 7554/eLife . 11308 . 015Figure 3—figure supplement 4 . Signal detection theory-based model fit to the auditory ( clicks ) data . Each panel shows model fits of the standard deviations ( σ ) in the rat’s estimate of the effective number of clicks for an individual rat performing the auditory accumulation of evidence task . Error bars indicate the standard deviations of the mean based on repeated calculations of one thousand-fold resampled data . Effective numbers of clicks were estimated based on sensory adaptation parameters ( see Materials and methods ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11308 . 01510 . 7554/eLife . 11308 . 016Figure 3—figure supplement 5 . Permutation test comparing goodness-of-fit of scalar variability and linear variance to the auditory ( clicks ) data . A nonparametric permutation procedure ( see Materials and methods: Model Comparison ) was used to evaluate whether the MLE fits of the standard deviations in the rat’s estimate of number of effective clicks ( shown in Figure 3—figure supplement 4 ) were better fit by scalar variability or linear variance . The black line is a histogram of the difference of r2 values for scalar variability ( SV ) and linear variance ( LV ) ( rSV2−rLV2 ) for a thousand bootstrapped samples . The gray line is a histogram of the null hypothesis distribution for a thousand bootstrapped samples . The red line indicates the p-value for the permutation test: the area under the gray line at the value corresponding to the observed difference in the distribution of r2 values ( i . e . the mean of the black distribution ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11308 . 01610 . 7554/eLife . 11308 . 017Figure 3—figure supplement 6 . Chronometric plots for the auditory ( clicks ) data and predictions of scalar variability . Each panel shows the performance of an individual rat on the auditory accumulation task as a function of the duration of the cue period . Colored dots indicate the data and colored lines indicate the prediction of pure scalar variability: σ = kN , where N is the number of clicks after the simultaneous click . Color indicates trial difficulty , parameterized by γ , the log of the ratio of the rates of the clicks on each side . Across all animals , performance increases as a function of cue duration for a fixed γ ( colored dots ) , however , pure scalar variability predicts flat psychometric function ( colored lines ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11308 . 01710 . 7554/eLife . 11308 . 018Figure 3—figure supplement 7 . Chronometric plots for the auditory ( clicks ) data and predictions of scalar variability with an offset . Each panel shows the performance of an individual rat on the auditory accumulation task as a function of the duration of the cue period . Colored dots indicate the data and colored lines indicate the prediction of scalar variability with an offset: σ = k0+kN , where N is the number of clicks after the simultaneous click . Color indicates trial difficulty , parameterized by γ , the log of the ratio of the rates of the clicks on each side . Across all animals , performance increases as a function of cue duration for a fixed γ ( colored dots ) . Whereas pure scalar variability predicts flat psychometric function ( see Figure 3—figure supplement 6 ) , inclusion of an offset term captures increased performance with cue duration . DOI: http://dx . doi . org/10 . 7554/eLife . 11308 . 018 We used maximum likelihood estimation to calculate the standard deviations of each distribution ( σ0 . . . σ15 ) given the number of flashes and behavioral choices across all trials . We found that the best-fit standard deviation values scaled approximately linearly with the number of flashes ( Figure 3B; Figure 3—figure supplement 1 ) . We next compared the predictions of the model with the behavioral data ( Figure 3C and D ) . The model was able to capture a large number of features of the data , including the subjects’ imperfect performance on the easiest trials , even though no lapse rate parameter was included in the model ( Figure 3D ) . This was true both for the group data in Figure 3 and for individual rats ( Figure 3—figure supplement 2 ) . Note , although our model represents numerical estimates as a scalar random variable on each trial , our results are also consistent with probabilistic numerical representations in the brain ( Kanitscheider et al . , 2015 ) . We used least squares regression combined with bootstrapping to estimate the best-fit scaling of σn as a function of n according to each of the three models of noise: scalar variability ( SV , linear scaling ) , subitizing ( SUB +SV , constant at zero up to some value of n , then linear scaling ) and linear variance ( LV , square root scaling ) ( Figure 3E ) . Comparison of goodness-of-fit value ( r2 ) using a nonparametric permutation test confirmed that the MLE standard deviation estimates ( σ0 . . . σ15 ) were better fit by the scalar variability model than the subitizing or linear variance models ( Figure 3F; see Materials and methods: Model Comparison ) . Scalar variability suggests that scaling the number of flashes on each side by the same factor should lead to identical discriminability: if σn= k×n , then the probability of choosing right ( equation in Figure 3A ) is a function of the ratio r = NR/NL , where NR and NL are the number of flashes on right and left , respectively . For the easiest trials ( e . g . , NL >> 1 and NR = 0 ) , performance asymptotes at 12π∫1k∞e−x2/2dx producing a non-zero lapse rate even without an explicit lapse rate parameter ( see mathematical appendix , Appendix 1 ) . To test the prediction that performance should be constant for trials with a fixed ratio r , we compared behavioral performance on trials with constant ratios of flashes on the rewarded and unrewarded sides . Supporting scalar variability , both the data and best-fit signal detection theory-based model predictions exhibited roughly constant performance for trials with fixed ratios of flashes ( Figure 3—figure supplement 3 ) . Interestingly , however , behavior deviated from scalar variability , especially for trials with very low number of flashes . For these trials , there was an improvement in performance with additional flashes , even though the ratio of flashes was fixed ( Figure 3—figure supplement 3 ) , which is not predicted by pure scalar variability . Nevertheless , this deviation from scalar variability was well predicted by the model parameters of Figure 3B . It is thus likely driven by the deviation from pure linear dependence of the σn parameters on n for n<2 ( Figure 3B ) . We repeated the signal detection theory-based analysis on behavioral data collected from rats performing a previously developed auditory accumulation of evidence task ( Brunton et al . , 2013; Hanks et al . , 2015 ) . We accounted for adaptation effects , which are prominent in the auditory task due to high click rates , and computed the ‘effective’ number of clicks presented after implementing the adaptation dynamics described in Equation ( 6 ) ( see Materials and methods ) . On trials in which this approach yielded fractional number of clicks , we rounded the effective click number ( after adaptation ) to the nearest integer . This kept the number of parameters in the auditory task model roughly equal to the number in the model for the visual task . We then used maximum likelihood estimation to calculate the standard deviations of each distribution ( σ0 , … , σn ) given the effective number of clicks and the subjects’ behavioral choices . Consistent with our results with the visual flashes task , the goodness-of-fit value ( r2 ) revealed that the MLE standard deviation estimates in the auditory task were better fit by a model of scalar variability plus a constant offset ( SV , σn= k0+ k×n ) compared to the linear variance previously assumed by sequential sampling models ( LV , σn2 =k0+ k×n ) ( r2 = 0 . 97 for SV vs r2 = 0 . 89 for LV , p << 0 . 001 based on nonparametric permutation test , Figure 3—figure supplement 4 , 5 ) . Our findings are thus not specific to the visual modality . Notably , perfect scalar variability ( σn =k×n without the constant k0 ) would predict no improvement as a function of time for a fixed click ratio , yet such an improvement is clearly seen in the clicks data ( Figure 3—figure supplement 6 ) . The SV model accounted for this by having a significantly non-zero offset , k0 , and was thus able to fit the clicks data quite well ( Figure 3—figure supplement 7 ) . A non-zero fixed offset , k0 , thus represents a source of noise that is added to the accumulators , independently of the stimulus that was presented . Notice that this source of noise does not account for the lapse rate , since k0 does not appear in the expression for the lapse rate ( see mathematical appendix , Appendix 1 ) . A larger offset ( k0 ) was the main qualitative difference we found between fitting the auditory clicks data ( for which k0 = 2 . 10 +/- 0 . 20 , mean +/ s . e . m . across rats ) and the visual flashes data ( for which k0 = 0 . 49 +/- 0 . 142 ) . The signal detection theory-based model described above makes two assumptions about the nature of the accumulation process . First it assumes that subjects maintain separate estimates of left and right flashes during a behavioral trial . Second it assumes that noise depends only on the final value of the left and right integrator . However , many quantitative models of accumulation of evidence propose that subjects maintain a single decision variable ( e . g . , a running estimate of the difference in number of right minus left pulses; Bogacz et al . , 2006 ) . Other binary decision models have proposed two separate accumulators in which noise is added throughout the accumulation process ( Ratcliff et al . , 2007; Usher and McClelland , 2001 ) . Therefore , we decided to evaluate the assumption of a dual accumulator and the assumption that the noise depends only on the final value of the integrator ( s ) , by comparing the performance of multiple variants of our signal detection theory-based behavioral model ( Figure 4 ) . We probed three independent questions , the combined answers to which produced eight different models: ( 1 ) Is performance best fit by dual accumulators or by a single accumulator ? ( 2 ) Is performance best fit by gradual accrual of noise throughout stimulus presentation , or by noise that depends only on the final value of the integrator ? And finally , as in the previous section , ( 3 ) Is variability best fit as standard deviation being linear in flash number , or by variance being linear in flash number ? Each model ( a-h ) had two free parameters β1 and β2 that determined the relationship between the number of flashes , n , and the standard deviation of the flash number , σn ( Figure 4B ) . Four models ( a , b , e , f ) assumed scalar variability ( σ = β1*n+β2 ) and four ( c , d , g , h ) assumed linear variance ( σ2 = β1*n+β2 ) . Four models assumed a single accumulator ( a , c , e , g ) and four assumed independent left and right accumulators ( b , d , f , h ) . Four models ( a-d ) assumed that noise was based only on the final estimate of the accumulator ( static sampling ) , and four assumed ( e-h ) that noise was added to the estimate of the accumulator at the time of each flash ( sequential sampling ) . 10 . 7554/eLife . 11308 . 019Figure 4 . Comparison of behavioral models suggests the presence of at least two accumulators . ( A ) General forms of the signal detection theory-based models that were compared assumed either single or dual accumulators . Each model determines which choice to make on each trial , by randomly selecting a value ( a ) from a Gaussian with mean equal to the difference in total right minus left flashes , and standard deviation σn . If a>0 , the model decides “Right” , and if a<0 the model decides “Left” . Noise , parameterized as the standard deviation of the distributions of flash number or flash difference ( σn ) , enters the accumulation process differently in each model . Single accumulator models assume that noise depends only on the difference in the number of flashes ( R-L ) while dual accumulator models assume that noise depends on the total number of flashes ( R+L ) . ( B ) For each class of models ( single vs . dual accumulators ) , we implemented versions that assumed scalar variability or linear variance and that assumed a static or a sequential sampling process . For the static model , equivalent to our original signal detection theory-based model , noise scales with the number of flashes seen at the end of the trial . For the sequential sampling model , noise is added to the subject’s estimate with each flash , and the magnitude of the noise depends on the value of the accumulator at the time of the flash ( a ( t ) ) . ( C ) Cladogram indicating the details of each of the eight models ( a-h ) . ( D ) Comparison of the likelihood of the model given the data for each of the eight model versions . A bootstrapping procedure was used , in which behavioral trials were resampled ( with replacement ) and each model was fit to each resample . Error bars indicate the 95th percentiles of the model likelihoods using the best-fit parameters for all resamples . The number of parameters is equal across all models , allowing direct comparison of likelihoods . ( E ) The permutation test used to assess statistical significance . Black line indicates the distribution of the differences of likelihoods between the two models with the largest likelihood , scalar variability , dual accumulator static version ( b ) and linear variance , dual accumulator , static version ( d ) , across all resamples . Gray line indicates the distribution under the null hypothesis ( see Materials and methods: Model comparison ) . Model b has a significantly larger likelihood than model d and all other models ( p<< . 001; see Materials and methods: Model comparison ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11308 . 01910 . 7554/eLife . 11308 . 020Figure 4—figure supplement 1 . Model comparisons for each rat . Bar plots indicate the difference in the likelihoods between the scalar variability , dual accumulator static sampling model ( b ) and the seven other model versions , described in Figure 4 , for each rat . Error bars indicate the 95th percentiles of the model likelihoods using the best-fit parameters for all resamples . The number of parameters were equal across all models , allowing direct comparison of likelihoods . A bootstrapping procedure was used , in which behavioral trials were resampled ( with replacement ) and each model was fit to each resample . Permutation test revealed that in 6/7 rats the scalar variability , dual accumulator static sampling model ( model b ) had a significantly greater likelihood than all other models ( a , c-h ) ( p<0 . 001 ) . For rat S142 model b was not significantly greater than other models ( p>0 . 001 ) except model ( f ) ( p<0 . 001 ) . Note that rat S142 may not have performed enough trials ( n<2000 ) to accurately fit the model . DOI: http://dx . doi . org/10 . 7554/eLife . 11308 . 02010 . 7554/eLife . 11308 . 021Figure 4—figure supplement 2 . Model predictions versus data . Performance of each model described in Figure 4 . Red circles are data pooled from all rats . Green lines are model predictions , using parameters fit to the pooled data . The 'Scalar variability dual accumulator static' model exhibited the highest likelihood of all the models , for the pooled data and for individual rats ( see Figure 4; Figure 4—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11308 . 021 We used maximum likelihood estimation to find , for each model , the values of β1 and β2 that best fit the behavioral data . Despite the closely related structure of the eight models , and having precisely the same number of free parameters , the likelihood of the data at the best-fit parameter values for nearly all of the two-accumulator versions ( b , d , h ) of the model was significantly greater than the corresponding one-accumulator versions ( a , c , g ) ( p< . 001; nonparametric test using bootstrapping; see Materials and methods ) , indicating better performance for the two-accumulator model ( Figure 4D , E ) . Overall we found that the static , dual accumulator scalar variability model had the highest likelihood across all models ( p< . 001; nonparametric test using bootstrapping see Materials and methods: Model Comparison; Figure 4—figure supplements 1 , 2 ) . This suggests that the main noise source , in the limited regime of models considered here , operates on the final value of the accumulators , rather than being incrementally and sequentially added to the accumulators as the process unfolds . We emphasize , however , that we have not made an effort to systematically explore the full space of possible models in which noise is added gradually during the stimulus presentation , so a conclusion that noise is added only to the final value of the accumulator remains tentative . Scalar variability has been proposed both for numerical and interval timing estimation , raising the possibility that rats could be accumulating the summed duration of the flash ‘ON’ times . Alternatively , it is possible that rats could be treating each flash as a distinct event , independent of its duration , and integrating the number of flashes . To test which strategy the rats used , we performed an additional experiment in which we trained a second cohort of rats ( n=7 ) on a new version of the task . As before , rats were rewarded for orienting to the side with the greater number of flashes , but in this version the durations of each flash and inter-flash interval were randomly jittered ( see Materials and methods ) . Flash duration was drawn from a Gaussian distribution with a mean of 10 ms and standard deviation of 1 . 5 ms ( Figure 5A , B ) . Because LED duration is correlated with number of flashes on most trials , it is difficult to evaluate whether animals use flash duration or flash number on most trials ( Figure 5C , D ) . However , on trials in which there was no difference in the number of flashes ( Δflashes = 0 ) , the jittering of flash duration led to the generation of some trials with a difference in the duration presented from the right or left LEDs ( Figure 5B ) . Since , for example , 20 ms of light is the equivalent of a difference of two flashes , we reasoned that on these trials , if the subjects were integrating light duration , they would be more likely to orient to the side with longer LED durations , and a regression line fit to the data in Figure 5E should have a positive slope . Conversely , if they were integrating discrete flash number , they should not exhibit a preference to the side with longer flash duration , and a regression line fit to the data in Figure 5E should have a slope of zero . Consistent with integrating flash number , we found that on trials with 0 Δflashes , rats did not exhibit a preference for the side with longer flash durations ( Figure 5E , F ) . These results indicate that the rats were accumulating the number of flashes , not their total duration . 10 . 7554/eLife . 11308 . 022Figure 5 . Rats accumulate flash number , not duration . ( A ) Distribution of flash durations . In a subset of experiments , flash durations were drawn from a Gaussian distribution with a mean of 10 ms . As a consequence , on some trials in which the same number of flashes were presented to both sides ( #R=#L ) , there was a greater overall flash duration on one side ( up to 50 ms ) ( B ) Schematic of the timing and duration of left ( blue ) and right ( red ) flashes on an example trial in which #R=#L , but there was greater overall flash duration from the right LED . ( C ) Percent of trials on which the animal went right as a function of difference in right-left LED duration ( specifically ( durR-durL ) / ( durR+durL ) ) across all combinations of left and right flash number . Because LED duration is correlated with number of flashes on most trials , this looks very similar to performance as a function of flash number . Yellow line is the regression line to the data . ( D ) Percent of trials on which the animal went right as a function of difference in flash number ( ( #R-#L ) / ( #R+#L ) ) . Green line is the regression line to the data . ( E ) Behavioral choice as a function of the difference in the overall flash duration ( ( durR-durL ) / ( durR+durL ) ) on trials with equal numbers of flashes on both sides ( #R=#L ) . Black circles indicate the percent of trials in which animals went right on trials of different overall flash durations . Error bars indicate 95% confidence intervals for the mean assuming a binomial distribution . Solid gray line is the regression line to the data , whose slope is displayed in panel F . ( F ) Bar plots indicate the slope of the regression lines in panels C and D for trials with equal numbers of flashes . Error bars are standard error of the coefficient ( slope ) estimates . These represent different models of what the slope of the regression line to the data would be if the animal were integrating flash duration ( ‘Duration model’ , yellow ) or flash number ( ‘Flash model’ , green ) . The slope of the regression line fit to the data ( ‘Data’ ) is significantly different from the line relating the LED duration to choice , but not significantly different from zero . This suggests that rats integrate flash number . DOI: http://dx . doi . org/10 . 7554/eLife . 11308 . 022 One potential additional source of variability in the rats’ behavior is the location of reward on previous trials . In some behavioral tasks , subjects display a win-stay-lose-switch approach to decision-making , in which choices that lead to reward tend to be repeated , while alternative unrewarded choices tend to be abandoned . To assess whether the memory of reward location on previous trials biased the subjects’ decisions on future trials , we identified trials following a reward and compared behavioral performance depending on whether the reward had been delivered to the left or right ( Figure 6A ) . On these trials subjects continued to perform the accumulation task , but were biased toward the side where they had previously obtained reward . The effect of this bias was modest but significant , and was constant across a range of numerical flash differences , producing a vertical shift in the psychometric function ( Figure 6A ) . Similarly , if the subjects made an error on one side , they exhibited a bias toward the other side on the subsequent trial ( Figure 6B ) . 10 . 7554/eLife . 11308 . 023Figure 6 . Trial history contributes to behavioral variability . ( A ) Reward biases decision on subsequent trials . Plot indicates psychometric performance on trials following correct right ( red ) and correct left ( blue ) trials . Black line indicates the mean . Data is pooled across all sessions for all rats . Error bars are 95% confidence intervals for a binomial distribution . ( B ) Psychometric performance on trials following errors . Making an error on one side modestly decreased the probability that subjects would orient to that side on the subsequent trial . ( C ) Following a rewarded trial , subjects exhibited increased probability of returning to the same side poke up to three trials in the future ( black line ) . Following an error trial , they exhibited a decreased probability of orienting to the same side up to three trials in the future ( gray ) . ( D ) Effects of reward history for correct trials are additive . Black line indicates the probability of returning to the same side given 1-6 consecutive rewards on the side . Brown line indicates the bias predicted from the linear sum of the biases observed for non-consecutive rewards as shown in Figure 6C . DOI: http://dx . doi . org/10 . 7554/eLife . 11308 . 023 Next , we extended this analysis to determine how long this bias persists . We computed the probability that subjects chose to return to the same side where they had previously obtained a reward and repeated this computation for progressively later choices ( Figure 6C , black line ) . The behavioral side bias caused by a reward decreased steadily over the next three trials and after three trials no significant bias was observed . We repeated this analysis for error trials and found that the bias was smaller in magnitude , but displayed the same steady decrease for three trials into the future ( Figure 6C , gray line ) . After three trials no significant bias was observed . Finally , when rewards were consecutive ( i . e . more than two correct responses to the right in a row ) the bias observed was additive . The side bias observed from two consecutive rewards on the same side is not significantly different from the sum of the reward biases from one and two trials back ( Figure 6D ) . Together these results suggest that ( 1 ) both rewards and errors bias the subject’s choices up to three trials in the future and ( 2 ) reward history effects combine linearly to influence the behavior of the subject . Similar results have been observed in the auditory , Brunton et al . , version of the accumulation of evidence task ( Chan , Brunton , and Brody , unpublished data ) . Future studies into the biological mechanisms of decision-making will be facilitated by the ability to perform precise neuronal circuit perturbations and cellular resolution imaging . Behavioral tasks that can be performed during head restraint allow the experimenter to perform techniques such as cellular resolution optical stimulation and large-scale two-photon calcium imaging that are difficult to employ in unrestrained preparations . Therefore we developed a version of the visual accumulation of evidence task that could be performed during voluntary head-restraint . In the head-restrained version of the task , subjects initiate a behavioral trial by guiding a surgically implanted titanium headplate into a custom headport mounted on one wall of an operant training chamber ( Figure 7A , left and 7B ) . The trial begins when the leading edge of the headplate contacts two miniature snap action switches ( contact sensors ) mounted on the headport . Voltage-controlled pneumatic pistons then deploy to immobilize the headplate and , using the principles of kinematic mounts , register the head to within a few microns . During the restraint period , which lasts 2–3 s , head-restrained animals are presented with up to six flashes on each side ( Figure 7A , middle ) . After a brief ( 500 ms ) delay , subjects are released from restraint , and can obtain water reward by orienting to one of two side ports mounted to the wall flanking the headport ( Figure 7A , right ) . Head-restrained subjects can also choose to terminate the restraint period early by operating a release switch located on the floor of the chamber . Trials that are terminated early result in a brief timeout ( ~2 s ) in which no reward can be obtained and no trial can be initiated . 10 . 7554/eLife . 11308 . 024Figure 7 . Head-restrained and unrestrained rats exhibit comparable task performance . ( A ) Schematic of the accumulation of evidence task during head restraint . A rat initiates a behavioral trial by inserting his headplate into a custom headport along one wall of an operant training box ( left panel ) . While voluntarily head restrained , the rat is presented with a series of pseudo-randomly timed flashes from blue LEDs to the left and right in front of him ( middle panels ) . Following flash presentation , the subject is released from restraint and is free to withdraw his head from the headport and orient to one of the side ports to obtain reward ( right panel ) . ( B ) Image of a rat training on voluntary head restraint . The left and right stimulus LEDs , which are mounted in light diffusing boxes , are pseudo-colored in blue . ( C–F ) Direct comparison of behavioral performance between seven head restrained rats and seven unrestrained rats . ( C ) Comparison of psychophysical performance indicates similar sensitivity to the difference in the number of flashes between restrained ( black ) and unrestrained rats ( red ) . Error bars indicate 95% binomial confidence intervals on the data . ( D ) Comparison of behavioral performance between restrained ( upper panel ) and unrestrained ( lower panel ) rats across all different stimulus types in the restrained version of the task . Color indicates the percentage of trials on which the subjects oriented to the right . Gray indicates no data . ( E ) Comparison of the noise ( standard deviation ) in the estimates of the number of flashes for restrained rats ( black ) and unrestrained rats ( red ) using the procedure described in Figure 3 . Lines indicate the best linear fit for the restrained ( black ) and unrestrained ( red ) cohorts . Error bars indicate the 95% estimates of the means of σn . ( F ) Reverse correlation analysis indicating the relative contribution of flashes occurring at different times in the trial to the subjects’ behavioral choice for restrained ( transparent ) and unrestrained ( solid ) rats . Data from unrestrained rats are a subset of the data plotted in Figure 2F ( color convention is also adopted from Figure 2F ) . Lines and error bars are mean and standard error across rats . DOI: http://dx . doi . org/10 . 7554/eLife . 11308 . 024 To compare behavioral performance to unrestrained rats , we trained seven rats on the head-restrained version of this task . The initial stages of training were performed in a high-throughput facility , while acclimation to the pistons and data collection from fully trained animals were performed in a separate facility with a single dedicated behavioral chamber . Rats completed the initial stages of training in similar time to the unrestrained rats . However , a portion of rats ( ~40% ) were slower to acclimate to the pistons than the remainder of their cohort and were excluded from further study . Behavioral analysis revealed comparable performance between voluntarily head-restrained and unrestrained rats ( Figure 7C and D ) . Fitting the accumulator model produced model parameters in the head-restrained subjects that was similar to the best fit parameters for the unrestrained rats: negligible accumulator noise , predominant flash-associated noise , accumulation time constants longer than the trial duration and bounds higher than the number of possible flashes . Fitting the signal detection theory-based model revealed similar scaling of noise between restrained and unrestrained versions of the task ( Figure 7E ) . We computed the ‘psychophysical reverse correlation’ to further assess whether the rats exhibited long accumulation time constants ( see Materials and methods ) . For each time bin , we computed the probability that there was an excess flash in that bin on the side to which the subjects subsequently oriented . This analysis indicated that flashes in each time bin contribute equally to subjects’ decisions ( Figure 7F ) . This was true for both voluntarily head-restrained and unrestrained rats . Head-restraint is an experimental tool widely used in psychology and neuroscience to immobilize the head of a subject . The technique is typically employed in behavioral experiments to facilitate reliable presentation of sensory stimuli to the head or face and precisely control the subjects’ movements , and in neurophysiological experiments to minimize motion between the brain and recording apparatus . However , it can take some species , notably rats and primates , weeks to acclimate to forced head restraint . Moreover forced head restraint requires skilled experimenter intervention and is difficult to incorporate into automated training procedures . Voluntary head restraint has been proposed to reduce the stress normally associated with forced head restraint and facilitate training on sophisticated behavioral tasks ( Scott et al . , 2013 ) . Here , we show that rats are able to perform a complex cognitive task , involving both numerical representation and integration of evidence , while voluntarily head-restrained . This provides a foundation for identifying and characterizing the neural circuit mechanisms of accumulation of evidence and numerical representation in future studies using in vivo cellular resolution imaging and perturbation ( Rickgauer et al . , 2014 ) . The behavioral task described here exhibited a number of features that enabled detailed characterization of noise during accumulation of evidence . The visual stimuli presented in this study were: ( 1 ) low in number; ( 2 ) presented reliably to left and right visual hemifields; ( 3 ) presented from well-separated light sources in front of the rat’s face; ( 4 ) presented at sufficient intervals to avoid adaptation or facilitation between successive stimuli . All of these features attempted to minimize sources of noise other than noise in the subjects’ internal estimates . Consistent with previous reports ( Brunton et al . , 2013; Kiani et al . , 2013 ) , we found that behavioral variability was not affected by the trial duration but was strongly correlated with the number of flashes presented . However , we identified a number of additional sources of noise . First , rats exhibit a modest side bias due to reward history occurring up to three trials in the past ( Figure 6 ) . In dynamic foraging tasks , strategies that incorporate trial history ( such as win-stay , lose-switch strategies ) often maximize rewards ( Corrado et al . , 2005; Herrnstein , 1961; Sugrue et al . , 2004 ) . However , in the visual accumulation of evidence task , the rewarded side is assigned independently on each trial , so trial history adds noise to the decision process . The influence of trial history on subsequent decisions has been observed in a number of other studies , suggesting that it is a general source of behavioral variability that should be incorporated into accumulation of evidence models ( Busse et al . , 2011; Gold et al . , 2008 ) . We also found that the standard deviation in the accumulated estimate scaled linearly with the number of flashes . This type of noise scaling , called scalar variability , was consistent across the range of stimuli presented ( 1-15 flashes , with minor deviations at the lowest number of flashes ) suggesting that the subjects did not use a subitizing approach , i . e . , they did not use rapid accurate counting for numbers less than five . We applied our analysis to a closely related previous auditory task that , as in our current visual task , also delivered sensory evidence in discrete pulses ( Brunton et al . , 2013 ) . We found that our conclusions also applied to the auditory data , with the standard deviation of the noise after N flashes well-described by σn= k0+k×n . Our observation that performance in the auditory task improves with increased integration times is qualitatively consistent with data from tasks in which evidence is delivered in a continuous stream rather than in discrete pulses . This leads us to suggest that scalar variability should be an important consideration in the analysis of data from continuous tasks . But whether or not scalar variability is the dominant form of noise in continuous stream tasks , as it was in the pulsatile tasks analyzed here , remains to be determined . The primary focus of this work has been to provide a quantitative description of behavior during accumulation of evidence and to evaluate the sources of behavioral variability . However , an important distinction should be made between quantitative models of behavior and mechanistic models of the underlying neural processes ( Marr , 1982 ) . Our results do not suggest an underlying neural mechanism , however they do place important constraints on such mechanistic models . One such constraint is scalar variability , which has been observed in a wide variety of tasks requiring animals to estimate magnitudes ( such as time or number Gallistel and Gelman , 2000; Gibbon , 1977 ) . Two possible neural mechanisms have been proposed to explain this phenomenon . The first is that the brain represents magnitude on a logarithmic scale with constant noise ( Fechner , 1860; Dehaene and Changeux , 1993; Meck and Church , 1983; Nieder and Dehaene , 2009 ) . Neurophysiological recordings , particularly in sensory systems , have provided support for this mechanism . For example , neurons in the rodent somatosensory cortex exhibit firing rates that scale logarithmically with frequency of tactile stimulation ( Kleinfeld et al . , 2006 ) . In addition , neurons in primate parietal cortex appear to encode numerosity on a nonlinear scale ( power law or logarithmic; Nieder and Miller , 2003 ) . The second possibility is that the brain represents magnitude on a linear scale with noise that is proportional to the magnitude . The results described here did not allow us to differentiate between these two models , and indeed , both models make identical psychophysical predictions ( Dehaene , 2001 ) . However , how they would be instantiated in the brain is quite different . For the logarithmic case , the signal diminishes with the quantity being represented , but noise added at each time point is independent . For the linear case , the signal is constant ( implying near perfect summation of signals ) , but noise scales linearly with the magnitude being estimated . Sequential sampling models of accumulation of evidence represent the decision process as the movement of one or more latent decision variables towards a criterion value . Such models have been useful to study a wide range of decision processes . One key advantage is that they provide a moment-by-moment estimate of the decision variable on each trial . This time-varying estimate can be compared to the observed neuronal dynamics , allowing analysis of the correlation between cellular firing rates and the subject’s internal representation of evidence ( Hanks et al . , 2015 ) . Our results suggest three modifications that could improve the accuracy of sequential sampling models of accumulation of evidence . First we found that reward and error trials introduce a bias that persists up to three trials into the future . Second we found that standard deviation in the decision variable increased linearly with the total number of flashes . Note that this observation is inconsistent with the common assumption made in drift diffusion models , that noise is added independent of time and the value of the decision variable . Third , the data was better fit by models that accumulate left and right flashes separately than by models that accumulate a single decision variable . The existence of two accumulators ( Ratcliff et al . , 2007; Usher and McClelland , 2001 ) has been previously observed in the vertebrate brain , for example the oculomotor system in the goldfish contains two lateralized accumulators that integrate transient motor commands to maintain a memory of gaze position ( Aksay et al . , 2007 ) , and is consistent with electrophysiological recordings in primates ( Bollimunta and Ditterich , 2012 ) and rodents ( Hanks et al . , 2015 ) . In the future , it will be interesting to develop a reaction-time version of the task , in which subjects are allowed to determine the time during which they observe the stimulus . This would allow testing predictions from recent pulse-accumulation models ( Simen et al . , 2015 ) . Variability in behavior and neuronal networks has been proposed to result , in part , from biophysical constraints of the nervous system . In some systems , however , noise is thought to serve an important functional role . For example it has been proposed that random fluctuations in neuronal membrane potential contributes to contrast invariance in the visual system ( Anderson et al . , 2000 ) . Moreover , during motor learning , noise is actively introduced into highly reliable systems to create the behavioral variability required for trial and error learning ( Olveczky et al . , 2005 ) . An interesting question is whether noise also facilitates learning in perceptual tasks . One potential function of noise in the estimate of flash number is that it could allow animals , which have been trained on a reduced stimulus set , to generalize , and perform correctly on trials with numbers of flashes they have never encountered . For example , if rats were only trained on trials with 5 and 6 flashes , because of the variability of their estimate of 5 and 6 flashes , on a portion of those trials they will have effectively perceived 7 and 8 flashes , and could learn the correct response to those trials by trial and error . Thus noise may allow the observer to experience a larger range of stimuli than they actually encountered , enabling generalization . High-throughput operant conditioning combined with detailed , quantitative analysis of behavioral variability allowed us to characterize the process of evidence accumulation in rats . This analysis revealed that animals’ choices were likely derived from a circuit with multiple accumulators and that decisions were influenced by noise from multiple sources . The major source of noise displayed ‘scalar variability’ in the sense that its standard deviation scaled linearly with the number of pulses of evidence . This observation is inconsistent with the assumption made by most previous models of accumulation for decision-making , namely that the primary noise source is diffusion-like , added to the accumulator independent of time or accumulator value . Our observation is instead consistent with previous models of numerical cognition , and suggests that evidence accumulation and numerical cognition may be subserved by similar or related circuits and neural mechanisms . We speculate that noise that obeys scalar variability could be an important component in models of accumulation of decision-making evidence generally . Finally , the observation that voluntarily head-restrained rats exhibit similar behavioral performance to unrestrained rats , and can be trained to perform complex decision-making tasks , suggests the possibility of using cellular resolution optical imaging and perturbation technologies to characterize the neural substrates of cognition . Animal use procedures were approved by the Princeton University Institutional Animal Care and Use Committee ( IACUC; Protocols #1837 and #1853 ) and carried out in accordance with National Institutes of Health standards . All subjects were male Long-Evans or Sprague Dawley rats weighing between 200 and 500 g ( Taconic , NY ) . Rats were placed on a water schedule in which fluids are provided during behavioral training . If rats consumed less than 3% of their body weight in water , they received ad lib water for 1 hr . Rats progressed through several stages of an automated training protocol before performing the task as described in the results . All data described in this study were collected from fully trained rats . Sessions with fewer than 100 completed trials were excluded from analyses . These sessions were rare and usually caused by hardware malfunctions . From the sessions included in this paper , on average , rats completed 385 trials per session ( median: 362 trials per day; range: 128–1005 trials per day ) , and performed at 70% correct ( median: 71%; range: 56–88% correct; see Figure 1—figure supplement 1 ) . The cue period consisted of fixed time bins , each 250 ms in duration . Unrestrained rats experienced up to 15 bins , whereas head-fixed rats experienced up to 6 bins . For a given trial , each of the two LEDs had a fixed generative probability of producing a flash in each bin . In the final training stage for the unrestrained rats , the generative probabilities for the high-probability LED and low-probability LED ranged from 70-80% and 20-30% , respectively . For restrained rats , the generative probabilities were 70% and 30% . However , the rats were rewarded based on the number of flashes that were actually presented , not on the underlying generative probabilities . In other words , in the rare event that the LED designated ‘high-probability’ happened to produce fewer flashes than the LED designated ‘low-probability’ , the animal was rewarded for orienting to the side that had more flashes . The flashes were 10 ms in duration ( except for an experimental group in which the flash durations were jittered; see Figure 5 ) , and they occurred in the first 10 ms of each cue bin . For the unrestrained rats , the inter-flash intervals ranged from 240 ms ( two flashes in subsequent bins ) to 3 . 45 s ( two flashes separated by 13 bins; see Figure 1—figure supplement 1 ) . For the head-fixed rats , the inter-flash intervals ranged from 240 ms to 1 . 24 s . In a subset of experiments , inter-flash intervals and flash durations were jittered . In these experiments , flash durations were drawn from a Gaussian distribution with a mean of 10 ms , and standard deviation of 1 . 5 ms . Inter-flash intervals were drawn from a Gaussian distribution with a mean of 250 ms and standard deviation of either 25 or 50 ms . In the unrestrained version of the task , there was a 500 ms pre-cue period before the beginning of the cue period . For the majority of trials , the duration of the post-cue ( or memory ) period was randomly drawn from a uniform distribution from 5 to 500 ms . In a small subset of trials , the post-cue period was drawn from a broader distribution ranging up to 6 . 5 s , but those trials represent a minority of the dataset ( see Figure 1—figure supplement 1 ) . In the head-fixed version of the task , the pre-cue period was 1 s in duration , to allow time for the pistons to actuate and fix the headplate in place . The post-cue period was 500 ms . Operant training chambers were controlled by the freely available , open-source software platform Bcontrol ( Erlich et al , 2011; Scott et al . 2013 ) . Bcontrol consists of an enhanced finite state machine , instantiated on a computer running a real-time operating system ( RTLinux ) , and capable of state transitions at a rate of 6 kHz , plus a second computer , running custom software written in MATLAB . Each behavioral trial consisted of a sequence of states in which different actuators—for example , opening of a solenoid valve for water reward—could be triggered . Transitions between the states were either governed by elapsed times ( e . g . , 40 ms for water reward ) or by the animal’s actions , which caused changes to the voltage output of a sensor in the chamber . Surgical procedures for headplate implantation have been previously described ( Scott et al . , 2013 ) . Briefly , we anesthetized animals with isoflurane in oxygen and gave Buprenorphine as an analgesic . Once anesthetized , the scalp and periosteum were retracted , exposing the skull . Dental cement ( Metabond ) was used to bond the headplate to the skull . After a 2-week recovery period , implanted animals began training in voluntary head restraint . The choice-conditioned reverse correlation analysis reveals the degree to which flashes in each time bin contribute to the decision . We selected trials in the final training stage that had the same generative probabilities ( γ ) for generating flashes ( p ( flash on high-probability side ) = 0 . 7; p ( flash on low-probability side ) = 0 . 3 ) . We computed the average number of left ( f¯L ( t ) ) and right ( f¯R ( t ) ) flashes in each time bin ( t ) conditioned on whether the animal went right or left , to obtain the average difference in flash number: ( 1 ) fwent_right ( t ) =fR ( t ) -fL ( t ) ( 2 ) fwent_left ( t ) =fL ( t ) -fR ( t ) We next computed the expected difference in flash number for each trial ( 0 . 7– . 3 = 0 . 4 for correct trials , -0 . 4 for error trials ) , and averaged across trials conditioned on choice . We subtracted this number ( the average expected difference in flash number in each bin ) from the observed average difference in flash number: ( 3 ) Fwent_right ( t ) =fwent_right ( t ) -f ( t ) |γ ( 4 ) Fwent_left ( t ) =fwent_left ( t ) -f ( t ) |γ We multiplied F¯went_left by -1 so that the two vectors ( F¯ ) were both in units of excess right flashes . F¯went_right and F¯went_left are the red and blue lines plotted in Figure 2B . Positive/negative values in F¯ are time bins in which more right/left clicks occurred than expected by chance , respectively . To obtain a moment-by-moment description of the decision process , we implemented a behavioral model fit to the trial-by-trial data that has been described previously ( Brunton et al . , 2013; Hanks et al . , 2015 ) . On each trial the model converts the incoming stream of discrete left and right flashes into a scalar quantity a ( t ) that represents the gradually accumulating difference between flashes presented to the two sides . At the end of the trial , the model predicts whether the animal would go right or left if a is positive or negative , respectively . The rat’s behavior is used to fit parameters that govern how a ( t ) evolves . These parameters quantify sensory noise and noise associated with time , leakiness/instability of the accumulation process , sensory depression/facilitation , side bias , and a lapse rate that corresponds to a fraction of trials on which a random choice is made . The dynamics of a ( t ) are implemented by the following equation: ( 5 ) da=σadW+ ( δt , tR*ηR*C−δt , tL*ηL*C ) dt+λadt where δt , tR , L are delta functions at the flash times , η are Gaussian variables drawn from N ( 1 , σs ) , dW is a white-noise Wiener process , and C parametrizes adaptation/facilitation of subsequent flashes . Adaptation/facilitation dynamics of C are implemented by the following equation: ( 6 ) dCdt=1−Cτϕ+ ( ϕ−1 ) C ( δt , tR+δt , tL ) In addition , a lapse rate parameter represents the fraction of trials on which the rat responds randomly . To describe how noise in the decision process scales with number of flashes , we implemented a model to estimate the width of the distribution of animals’ internal estimates of flash number . This signal detection theory-based model assumes that on a given trial , the animal’s estimate of flash number presented to each side is a random variable drawn from a Gaussian distribution whose mean is the number of flashes on that side , and whose variance is a free parameter in the model . The difference of those Gaussians predicts the subjects’ performance: correct trials occur when the difference of Gaussians is positive ( i . e . the random variable drawn from the distribution representing the larger number is in fact larger than the random variable drawn from the distribution representing the smaller number ) . This was implemented by the following equation: ( 7 ) p ( correct ) =∫0∞NL−S , σL2+σS2d ( L−S ) Where L and S are the means of the distributions for the larger and smaller numbers , respectively . σL is the standard deviation of noise when L stimuli has been shown for the ‘larger/correct’ side , and σS is the standard deviation of noise when S stimuli has been shown for the ‘smaller/incorrect’ side . In other words , σL2 and σS2 are the variances of the distributions of the numerical estimates of L and S . The values of σL2 and σS2 that maximized the likelihood of the animals’ behavioral choices were fit using the Matlab function fmincon . The values of σN2 reported in this paper were derived using a bootstrapping approach . One thousand surrogate data sets were created by selecting behavioral trials at random with replacement from the original data set . The signal detection theory-based model was then fit to each surrogate data set , producing 1000-fold estimates for σ02 … σ152 . The σn2 values reported in the paper represent the mean of over these values and the confidence intervals were derived from the standard deviation of these values . To evaluate the goodness-of-fits of the different models , we implemented a nonparametric permutation test . First , for each model , we computed a distribution of bootstrapped r2 values by resampling the behavioral trials with replacement , performing the fitting procedure , and computing the r2 value , on 1000 iterations . For pairwise comparisons , ( for example , between the scalar variability ( SV ) and linear variance ( LV ) models in Figure 3F ) , the null hypothesis was that the r2 values for each distribution derive from a common distribution . To test this hypothesis , we combined the bootstrapped r2 values from the SV and LV models into a single distribution , and from that combined distribution , created two arbitrary distributions of fake SV and LV r2 values , and computed the average of those arbitrary distributions . We repeated this procedure for many ( 100 , 1000 , and 10000 ) iterations to compute a distribution of arbitrary SV and LV r2 values and consequently , the difference of those arbitrary distributions , which represents the null hypothesis . We treated the area under the null distribution corresponding to the difference between the true SV and LV r2 distributions as the p-value . This procedure yielded identical results with 100 , 1000 , and 10000 permutations . It was performed for the model comparisons in Figure 3F , as well as for evaluation of the SV vs . LV models for the auditory ( clicks ) data . To compare the different versions of the signal detection theory-based model ( Figure 4 ) , for example single accumulator version vs two accumulator version or two accumulator version vs . the two accumulator time-varying version , we used a bootstrapping approach . One hundred surrogate datasets were created by selecting behavioral trials at random with replacement from the original dataset . The signal detection theory-based model was then fit to each surrogate dataset , and the log likelihood for the best-fit parameters was recorded . This gave us a distribution of log likelihoods for each version of the model . We then performed the nonparametric permutation test described above on the distributions of log likelihoods to make pairwise comparisons between the best-fit model and all other models . The authors would like to thank Klaus Osorio and Jovanna Teran for assistance with behavioral training and animal husbandry . Bingni Brunton contributed data that were collected for a previous paper ( Brunton et al . , 2013 ) . Tim Hanks provided valuable comments on the manuscript and was very helpful in early stages of the analysis . Sam Lewallen and Mikio Aoi suggested useful analytic approaches . In addition the authors would like to thank all members of the Tank and Brody labs for useful discussions .

Perceptual decision-making , i . e . making choices based on observed evidence , is rarely perfect . Humans and other animals tend to respond correctly on some trials and incorrectly on others . For over a century , this variability has been used to study the basis of decision-making . Most behavioral models assume that random fluctuations or 'noise' in the decision-making process is the primary source of variability and errors . However , the nature of this noise is unclear and the subject of intense scrutiny . To investigate the sources of the behavioral variability during decision-making , Scott , Constantinople et al . trained rats to perform a visual 'accumulation of evidence' task . The animals counted flashes of light that appeared on either their left or their right . Up to 15 flashes occurred on each side , in a random order , and the rats then received a reward if they selected the side that the greatest number of flashes had occurred on . The rats chose correctly on many occasions but not on every single one . Using a computer-controlled rat training facility or 'rat academy' , Scott , Constantinople et al . collected hundreds of thousands of behavioral trials from over a dozen rats . This large dataset provided the statistical power necessary to test the assumptions of leading models of behavioral variability during decision-making , and revealed that noise grew more rapidly with the number of flashes than previously predicted . This finding explained patterns of behavior that previous models struggled with , most notably the fact that individuals make errors even on the easiest trials . The analysis also revealed that animals maintain two separate running totals – one of stimuli on the left and another of stimuli on the right – rather than a single tally of the difference between the two . Scott , Constantinople et al . further demonstrated that rats could be trained to perform this task using a new system that enables functional brain imaging . The next step is to repeat these experiments while simultaneously recording brain activity to study the neural circuits that underlie decision-making and its variability .

2015

Sources of noise during accumulation of evidence in unrestrained and voluntarily head-restrained rats

TANGO1 binds and exports Procollagen VII from the endoplasmic reticulum ( ER ) . In this study , we report a connection between the cytoplasmic domain of TANGO1 and SLY1 , a protein that is required for membrane fusion . Knockdown of SLY1 by siRNA arrested Procollagen VII in the ER without affecting the recruitment of COPII components , general protein secretion , and retrograde transport of the KDEL-containing protein BIP , and ERGIC53 . SLY1 is known to interact with the ER-specific SNARE proteins Syntaxin 17 and 18 , however only Syntaxin 18 was required for Procollagen VII export . Neither SLY1 nor Syntaxin 18 was required for the export of the equally bulky Procollagen I from the ER . Altogether , these findings reveal the sorting of bulky collagen family members by TANGO1 at the ER and highlight the existence of different export pathways for secretory cargoes one of which is mediated by the specific SNARE complex containing SLY1 and Syntaxin 18 . Collagens are the most abundant secretory proteins , comprising 25–30% of the human body dry weight ( Pataridis et al . , 2008 ) . They are required for cell attachment , tissue organization and remodeling , and for the differentiation of chondrocytes to produce mineralized bones ( Gelse et al . , 2003; Wilson et al . , 2011 ) . There are at least 28 different kinds of collagens , composed of homo or hetero trimers of polypeptide chains coiled around each other to form a triple helix ( Shoulders and Raines , 2009 ) . These unbendable triple helices , which can be up to 450 nm long , as in the case of Collagen VII , are too big to fit into the conventional transport carriers of the secretory pathway that have been identified thus far ( Malhotra and Erlmann , 2011 ) . How are these bulky proteins exported from the ER ? While the debate on the trafficking of collagen-like molecules across the Golgi stack goes unabated , new data are beginning to unravel the mechanism by which collagens are exported from the ER . A protein called TANGO1 has been identified for its requirement in the export of Procollagen VII ( PC VII ) from the ER in tissue culture cells ( Saito et al . , 2009 ) . The knockout of TANGO1 in mice results in the production of a pup that dies at birth due to defective bone mineralization . The cause is a block in the secretion of multiple collagens needed for the differentiation of chondrocytes ( Wilson et al . , 2011 ) . TANGO1 is also required for collagen secretion in Drosophila melanogaster ( Pastor-Pareja and Xu , 2011; Lerner et al . , 2013 ) . TANGO1 binds PC VII via its SH3 domain in the lumen of the ER ( Saito et al . , 2009 ) . On the cytoplasmic side , TANGO1 binds cTAGE5 and both these proteins contain a proline rich domain that interacts with the COPII components SEC23/24 ( Saito et al . , 2009 , 2011 ) . We have proposed that binding of PC VII to TANGO1 in the lumen promotes the binding of TANGO1's proline rich domain to SEC23/24 . This retards the recruitment of SEC13/31 to SEC23/24 and thus delays the events leading to the biogenesis of the COPII vesicle ( Malhotra and Erlmann , 2011 ) . Upon growth to a size that is large enough to encapsulate PC VII , TANGO1 dissociates from both PC VII and SEC23/24 . The binding of SEC13/31 to SEC23/24 completes the assembly of COPII components on a patch of the ER . These events then lead to the export of PC VII , presumably in a mega carrier from the ER ( Saito et al . , 2009 , 2011 ) . Ubiquitination of SEC31 by the CUL3-KLHL12 ligase complex has been reported to control the exit of Procollagen I ( PC I ) from the ER ( Jin et al . , 2012; Malhotra , 2012 ) . Sedlin is reported to help in the export of PC I and II from the ER by regulating the cycling of SAR1 activation state that is essential for COPII assembly at the ER ( Venditti et al . , 2012 ) . TANGO1 is not required for PC I export from the ER , and it is not known whether PC II export is TANGO1 dependent . Together these data indicate that COPII components are required for the export of procollagens from the ER , however , they also suggest the possibility that not all procollagens exit the ER by the same mechanism . We now show the involvement of SLY1 ( or SCFD1 ) in specific ER export events . SLY1 is a member of the STXBP/unc-18/SEC1 family of proteins that regulate the assembly or the activity of SNAREs in membrane fusion events ( Carr and Rizo , 2010 ) . The yeast ortholog SLY1 , an essential gene , has been described as a single copy suppressor of the YPT1 deletion ( Dascher et al . , 1991 ) and implicated in forward and retrograde trafficking ( Ossig et al . , 1991; Li et al . , 2005 ) . In contrast with its essential roles in yeast , a temperature sensitive mutant of Sly1 in zebra fish is not lethal on the cellular level but rather creates developmental defects in embryonic stages ( Nechiporuk et al . , 2003 ) . In mammals , SLY1 has been reported to function in conjunction with Syntaxin 5 ( STX5 ) in the ER to Golgi transport and might also function in the assembly of pre-Golgi intermediates ( Rowe et al . , 1998 ) together with Syntaxin 18 ( STX18 ) ( Yamaguchi et al . , 2002 ) and Syntaxin 17 ( STX17 ) ( Steegmaier et al . , 2000 ) . SLY1 has been shown to interact with the COG4 complex and suggested to play a role in intra Golgi and retrograde transport ( Laufman et al . , 2009 ) . It is important to note that in mammalian cells , these proposed roles of SLY1 in traffic between ER and Golgi membranes are based entirely on the use of artificial temperature sensitive mutant protein Vesicular Stomatitis Virus ( VSV ) -Glycoprotein ( G ) protein and the artificial cargo signal sequence ( ss ) -Green Fluorescent Protein ( GFP ) . The role of SLY1 in the trafficking of endogenous cargoes and its potential mechanism of action is therefore a matter of debate . We describe in this study , our data that reveal the existence of different export routes for secretory cargoes from the ER: of specific interest is the finding that SLY1 and the ER specific t-SNARE STX18 are necessary for the export of PC VII but not of the equally bulky PC I from the ER . To search for proteins that interact with cytoplasmically oriented portions of TANGO1 , we expressed a Myc-His tagged version of a truncated form of TANGO1 ( TANGO1ct ) that lacks the luminal domain in HeLa cells . After crosslinking with membrane permeable DSP and lysis , proteins were recovered on a Nickel agarose column and analyzed by mass spectrometry . Of interest was the finding of SLY1 in the pool of proteins cross-linked to TANGO1 . To further ascertain the mass spectrometry data , we immunoprecipitated Myc-His-tagged TANGO1ct from transfected and crosslinked HeLa cells as described above and western blotted the bound material with anti-Myc and SLY1 antibodies , respectively . Our data show the presence of SLY1 in the TANGO1 immunoprecipitate ( Figure 1A ) . SLY1 is a cytoplasmic protein but our findings suggest that it interacts with the ER exit sites anchored TANGO1 , so is there a pool of SLY1 associated with ER exit sites where TANGO1 resides ? We have previously documented the use of a human cell line called RDEB/FB/C7 that stably expresses and secretes Collagen VII ( Chen et al . , 2002; Saito et al . , 2009 ) . We used lentivirus to express SLY1-GFP in RDEB/FB/C7 cells . In fixed cells , the GFP signal is localized throughout the cytoplasm with a slight accumulation in the perinuclear region . After prepermeabilization of the cells with saponin and extensive washes to remove the cytoplasmic pool of proteins before fixation , a pool of SLY1-GFP was found to colocalize with the ER exit sites specific COPII component SEC31 ( Figure 1B ) . Arrowheads mark single exit sites that label for SEC31 and SLY1 . In addition , a perinuclear SLY1-containing and SEC31-lacking pool was observed . This perinuclear pool of SLY1 partially overlaps with GM130 , a marker of the early Golgi cisternae ( Figure 1B ) . Our data show that although most of SLY1 was Golgi localized , we detected a small pool that co-localized with Sec31 . At present , we do not know whether recruitment of SLY1 to the ER exit site is by direct binding to TANGO1 or indirectly through other proteins on the cytoplasmic side . 10 . 7554/eLife . 02784 . 003Figure 1 . SLY1 immunoprecipitates with TANGO1 and localizes to ER exit sites . ( A ) Lysates from Myc-His-TANGO1 ( cytoplasmic tail ) transfected or control HeLa cell lysates were incubated with Nickel beads , bound proteins ( 75% ) and total lysates ( 10% ) were separated by SDS-PAGE , endogenous SLY1 and Myc-His-TANGO1 ( cytoplasmic tail ) was detected by Western Blot using SLY1 and Myc antibodies . ( B ) RDEB/FB/C7 cells stably expressing SLY1-GFP ( green ) were fixed with 4% PFA and permeabilized with Triton-X100 or pre-extracted with saponin , washed and then fixed with 4% PFA prior to permeabilization with Triton-X100 . Cells were labeled with an anti-SEC31 or an anti-GM130 antibody ( red ) and DAPI ( blue , Scale bar: 20 µm ) ; arrowheads indicate colocalization of SLY1-GFP and SEC31 . DOI: http://dx . doi . org/10 . 7554/eLife . 02784 . 003 The identification of SLY1 in a pool of proteins that are cross linked to TANGO1 prompted us to test its involvement in PC VII export from the ER . In addition to the RDEB/FB/C7 cell line in which Collagen VII was exogenously introduced , we tested the immortalized human esophageal cell line Het1a to monitor the involvement of SLY1 in the export of endogenous PC VII . RDEB/FB/C7 and Het1a cells were transfected with siRNA oligos specific for SLY1 , TANGO1 , or a scrambled oligo as a control . RDEB/FB/C7 cells were transfected twice to increase knockdown efficiency . The levels of SLY1 and TANGO1 after knockdown were monitored by western blotting and revealed an approximately 90% reduction compared with control cells . Knockdown of TANGO1 did not significantly alter SLY1 protein levels and vice versa ( Figure 2A ) . 48 hr after the last siRNA transfection , the cells were washed and then incubated with fresh medium containing ascorbic acid ( 2 µg/ml ) for 20 hr . The medium and the cell lysates were western blotted with an anti-Collagen VII antibody . Knockdown of SLY1 inhibited PC VII secretion ( Figure 2B ) . Quantitation of the ratio of the secreted to intracellular PC VII revealed a 75% reduction upon SLY1 knockdown compared with control cells . PC VII secretion was 90% reduced upon TANGO1 knockdown , compared with control cells ( Figure 2C ) . 10 . 7554/eLife . 02784 . 004Figure 2 . SLY1 knockdown by siRNA inhibits Procollagen VII secretion . RDEB/FB/C7 cells were transfected with siRNAs directed against SLY1 , TANGO1 , or a scrambled siRNA . Het1a cells were transfected with the same set of siRNAs . ( A ) Knockdown efficiency was tested after 72 hr by western blotting cell lysates with anti-TANGO1 or anti-SLY1 antibodies . ( B ) PC VII secretion was measured by western blotting RDEB/FB/C7 cell lysates and supernatants collected for 20 hr in the presence of ascorbic acid with an anti-Collagen VII antibody . Equal protein loading and cell lysis were controlled by blotting with an anti-Tubulin antibody . ( C ) In three independent experiments , intensities of the PC VII signal in the lysate and the supernatant was recorded by densitometry . The ratio of external vs internal Collagen VII was normalized to quantify secretion in control cells as 100%; Error bars: standard error of the mean ( SEM ) . siRNA-treated RDEB/FB/C7 ( D ) or Het1a ( E ) cells were seeded on coverslips and 20 hr after addition of ascorbic acid , cells were fixed and visualized with the indicated antibodies and DAPI ( blue ) by fluorescence microscopy ( scale bars: 20 µm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02784 . 004 We then transfected RDEB/FB/C7 or Het1a cells with SLY1 siRNA , TANGO1 siRNA , or a control siRNA , as described above , treated the cells with ascorbic acid and after 72 hr the cells were fixed and incubated with antibodies against Collagen VII , the ER-resident chaperone Hsp47 and GM130 . In control cells , we did not detect intracellular PC VII because it is rapidly exported from the ER and secreted by the cells . In SLY1 knockdown cells , on the other hand , PC VII accumulated in large structures that also contained the collagen-specific protein chaperone Hsp47 ( Saga et al . , 1987 ) . Similar large patches of PC VII are also evident in RDEB/FB/C7 cells depleted of TANGO1 , which is well known for its role in the export of PC VII from the ER ( Saito et al . , 2009 , 2011 ) ( Figure 2D , upper panel ) . We did not observe accumulation of PC VII in the Golgi membranes upon knockdown of either SLY1 or TANGO1 ( Figure 2D , lower panel ) . The very low levels of Collagen VII expressed in Het1a cells are difficult to detect by the standard western blotting approaches , however , the accumulation of PC VII in the ER and colocalization with Hsp47 by fluorescence microscopy is also evident in SLY1- and TANGO1-depleted Het1a cells ( Figure 2E ) . Based on these data , we suggest that SLY1 knockdown arrests PC VII in the ER where it accumulates in large patches . To further ascertain the site of PC VII localization in SLY1 knockdown cells , we visualized PC VII by immunoelectron microscopy . In control cells , gold-conjugated secondary antibodies used to visualize anti-Collagen VII antibody were found distributed throughout the ER . In SLY1 knockdown cells , however , PC VII was observed in large patches within dilated ER ( Figure 3 ) but smaller vesicles that could correspond to intermediates en route were not visible . These patches , we suggest , correspond to the patches of PC VII observed in the fluorescence micrographs of SLY1 knockdown cells ( Figure 2D ) . Similar patches of PC VII were also evident upon knockdown of TANGO1 ( Figure 3 ) . 10 . 7554/eLife . 02784 . 005Figure 3 . Immunoelectron microscopy localizes intracellular Procollagen VII in SLY1 knockdown cells to the ER . Control cells and cells depleted for SLY1 or TANGO1 were fixed for immunoelectron microscopy . Ultrathin cryosections were labeled with a polyclonal rabbit antibody against Collagen VII followed by 10 nm protein A-gold complex . Arrowheads mark ER exit sites . Bars , 250 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 02784 . 005 We then tested whether SLY1 knockdown affected the stability of proteins involved in the transport of cargoes between the ER and the Golgi . For this , we monitored the levels of COPI and COPII components in the lysates of SLY1 , TANGO1 knockdown or control siRNA-treated RDEB/FB/C7 cells . The level of SAR1 , the initiator of the COPII vesicle budding process , did not change upon SLY1 or TANGO1 knockdown . Also unchanged were the levels of the other COPII components that are required for the export of secretory cargo , including components of the extracellular matrix , from the ER , specifically SEC23A ( Boyadjiev et al . , 2006 ) and SEC31 ( Jin et al . , 2012 ) . Because of the possible involvement of SLY1 in the retrograde trafficking pathway , we also tested components of the COPI coat: β-COP and ε-COP , and their levels were also unchanged ( Figure 4 ) . 10 . 7554/eLife . 02784 . 006Figure 4 . SLY1 knockdown does not affect the levels of COPI and COPII components . Cell lysates from RDEB/FB/C7 cells transfected with siRNAs oligos against SLY1 , TANGO1 , or a scrambled siRNA were western blotted with antibodies to the indicated COPI and COPII components . DOI: http://dx . doi . org/10 . 7554/eLife . 02784 . 006 We tested the involvement of SLY1 in retrograde trafficking by monitoring the localization of a protein called GRP78/BIP that contains a KDEL sequence and cycles between the ER and the early Golgi cisternae . A block in retrograde transport leads to an accumulation of GRP78/BIP in the Golgi apparatus and its subsequent secretion from the cells ( Yamamoto et al . , 2003 ) . GRP78/BIP was localized to the ER in SLY1 knockdown cells and control cells ( Figure 5A ) . Moreover , GRP78/BIP was not detected in the medium collected from control , SLY1- or TANGO1-depleted cells ( Figure 5B ) . To further quantitate the involvement of SLY1 in transport between the Golgi apparatus and the ER , we used a procedure to trap cycling proteins in the ER as described previously ( Pecot and Malhotra , 2006 ) . HeLa cells treated with control siRNA or cells depleted of TANGO1 or SLY1 were transfected with the following constructs: one encoding ERGIC53 , a protein that cycles between the early Golgi/ERGIC compartment and the ER , fused to FKPB and GFP , and another encoding the invariant chain of the major histocompatibility complex class II receptor ( Ii ) fused to FRAP-HA . In HeLa cells , the Ii chain lacking its MHC counterpart is retained in the ER ( Lotteau et al . , 1990 ) . Rapamycin is known to bind FRAP and this dimer tightly binds FKBP . This procedure can therefore be used to test if FRAP- and FKBP-containing proteins reside in close proximity and thus are accessible to each other . We observed the cells by fluorescence microscopy at steady state and 15 , 30 , and 60 min after the addition of Rapamycin or in control cells without the addition of Rapamycin . Colocalization of the two constructs was assessed by calculating the Pearson's colocalization coefficient . Knockdown of TANGO1 or SLY1 did not affect the kinetics of ERGIC53 redistribution to the ER ( Figure 5C ) . We therefore conclude that under our experimental conditions , knockdown of SLY1 or TANGO1 does not affect retrograde trafficking . 10 . 7554/eLife . 02784 . 007Figure 5 . SLY1 knockdown does not affect retrograde trafficking . ( A ) RDEB/FB/C7 cells depleted of SLY1 or control cells were visualized using an anti-GRP78/BIP , an anti-GM130 antibody and DAPI by fluorescence microscopy ( scale bars: 20 µm ) . ( B ) Lysates and 20 hr medium of SLY1 , TANGO1 or control knockdown RDEB/FB/C7 cells were analyzed by SDS-PAGE and western blotted with an anti-GRP78/BIP antibody . ( C ) In HeLa cells transfected with the indicated siRNAs , the colocalization of FKBP-ERGIC53-GFP and the ER localized Ii-FRAP-HA was determined at each indicated time points after addition of Rapamycin . Trapping , and thus colocalization , was measured by calculating the Pearsons colocalization coefficient between the GFP signal and the Alexa594-stained Ii-FRAP-HA . The average values of at least 30 cells analyzed per experiment for each condition are shown; Error bars: SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 02784 . 007 Retention of PC VII in the ER , upon knockdown of SLY1 or TANGO1 , is therefore not because of a defect in cycling of a component that is specifically required for the constitutive trafficking between the ER and the early Golgi cisternae . Is bulk protein secretion affected by SLY1 knockdown ? RDEB/FB/C7 cells were transfected with SLY1 , TANGO1 or a scrambled siRNA oligo , and after 72 hr , the cells were washed , cultured in methionine and cysteine free medium for 1 hr , pulsed with 35S-methionine for 20 min in methionine free medium , washed and then incubated in complete medium . Another control was included in which scrambled siRNA oligo transfected cells were incubated with Brefeldin A ( BFA ) during the chase period . The medium was collected after 2 hr and then analyzed by SDS-PAGE/autoradiography to identify the secreted polypeptides . Treatment with BFA , as expected , inhibited the secretion of newly synthesized proteins . Surprisingly , SLY1 knockdown did not have any appreciable effect on the secretion of newly synthesized proteins ( Figure 6A ) . This approach does not reveal the specific polypeptides secreted in a SLY1-dependent manner , but shows that the overall pattern of polypeptides detected by SDS-PAGE is not affected by SLY1 knockdown . 10 . 7554/eLife . 02784 . 008Figure 6 . SLY1 knockdown does not block the export of endogenous secretory cargo . ( A ) RDEB/FB/C7 cells depleted for SLY1 or TANGO1 or control cells were pulsed with 35S-methionine for 20 min and chased for 2 hr in complete medium that included 5 mg/ml BFA where indicated . The medium from the cells was collected and analyzed by SDS-PAGE and autoradiography . ( B ) HeLa ss-HRP cells were transfected with control siRNA , SLY1 or TANGO1 specific siRNA oligos or treated with BFA as a positive control for total block in secretion . At the indicated time points , the medium and cell lysates were harvested to measure HRP activity . The graph shows the ratio of secreted to intracellular ss-HRP activity . Average values of three independent experiments are shown; Error: SEM . ( C ) RDEB/FB/C7 cells were transfected with control siRNA , SLY1 or TANGO1 specific siRNA oligos . Collagen I secretion was measured by western blotting of RDEB/FB/C7 cell lysates and supernatants collected for 20 hr in the presence of ascorbic acid with an anti-Collagen I antibody . The samples were western blotted with an anti-Tubulin antibody to monitor loading control and cell lysis . ( D ) Saos2 cells were transfected with control , SLY1 or TANGO1 siRNAs . Knockdown efficiency was tested after 72 hr by western blotting cell lysates with anti-TANGO1 or anti-SLY1 antibodies . Actin was used as a loading control . ( E ) Collagen I secretion was measured by western blotting Saos2 cell lysates and supernatants collected for 20 hr in the presence of ascorbic acid with an anti-Collagen I antibody . The samples were also western blotted with an anti-Tubulin antibody to monitor loading control and cell lysis . DOI: http://dx . doi . org/10 . 7554/eLife . 02784 . 008 It has been shown that SLY1 depletion inhibits the trafficking of widely used model cargoes such as the temperature sensitive VSVG ( Dascher and Balch , 1996 ) and signal sequence GFP ( Gordon et al . , 2010 ) . We tested the effect of SLY1 and TANGO1 depletion on the secretion of signal sequence—Horseradish peroxidase ( ss-HRP ) . HeLa cells stably expressing ss-HRP were transfected with siRNAs targeting either SLY1 or TANGO1 or a scrambled siRNA as a control . After 72 hr , cells were washed and ssHRP was accumulated in the ER for 2 hr by incubating cells at 15°C . New protein synthesis was then blocked by Cycloheximide ( CHX ) treatment and the cells were incubated for the indicated time at 37°C to restart export and trafficking of secretory cargo from the ER . The HRP activity in the lysate and the supernatant was measured as described previously ( Bard et al . , 2006 ) . The ratio of secreted to internal HRP shows that depletion of SLY1 efficiently blocked secretion ss-HRP and was comparable to known secretion inhibitory drug BFA . TANGO1 depletion reduced the secretion of ss-HRP to 50% of that from control cells ( Figure 6B ) . The data presented thus far reveal the requirement of SLY1 in PC VII export , the export of exogenously expressed secretory cargo ss-HRP , but not the bulk of endogenous secretory cargoes . This is similar to the known effects of TANGO1 knockdown on the selective export of secretory cargo from the ER . Interestingly , TANGO1 knockdown by siRNA does not affect the secretion of the bulky PC I ( Saito et al . , 2009 , 2011 ) . Is SLY1 required for the trafficking of PC I ? RDEB/FB/C7 cells were transfected with siRNA oligos specific for SLY1 , TANGO1 or scrambled siRNA as a control and treated with ascorbic acid , as described above . 20 hr after the last change of medium , the cell lysates and the medium were western blotted with anti-Collagen I antibody . Our results reveal that the secretion of collagen I is not blocked by knockdown of SLY1 or TANGO1 ( Figure 6C ) . To further ascertain that SLY1 has no role in PC I export , we transfected Saos 2 cells with siRNA oligos specific for SLY1 , TANGO1 or a scrambled siRNA and after 48 hr Saos 2 cells were washed , and then incubated with fresh medium containing ascorbic acid for 20 hr . The knockdown efficiency was determined by western blotting the cell lysates and revealed a reduction greater than 75% in the levels of SLY1 and TANGO1 ( Figure 6D ) . The intracellular and secreted PC I was measured by western blotting the cell lysate and the medium , respectively . As in RDEB/FB/C7 cells , knockdown of SLY1 or TANGO1 in Saos 2 cells did not affect the secretion of PC I ( Figure 6E ) . In fact , in both cells lines , we detected a marginal increase in the secretion of Collagen I that is accompanied by an increase in the internal levels of Collagen I in the knocked down cells . This suggests a potential effect on the transcriptional regulation of Collagen I expression rather than a defect in its trafficking . The assembly of the COPII coat is initiated by the activation of the small GTPase SAR1 by the ER-resident protein SEC12 . Active SAR1 recruits the inner coat proteins SEC23/24 . The assembly of the outer coat proteins SEC13/31 completes the biogenesis of COPII vesicles by inactivation of SAR1 ( Zanetti et al . , 2012 ) . TANGO1 links PC VII to the inner COPII coat by interacting with the SEC23/24 complex ( Saito et al . , 2009 ) . To further ascertain the involvement of COPII coat assembly at the ER for PC VII export , we depleted either of the SAR1 isoforms A and B or both in RDEB/FB/C7 cells using an siRNA mixture described earlier ( Cutrona et al . , 2013 ) . Probing of cell lysates with anti SAR1 antibody revealed a doublet of bands at the predicted size of SAR1 . As one can observe in the single knockdowns these bands correspond to the two SAR1 isoforms , the upper , weaker band being SAR1 B and the lower , stronger SAR1 A . Single knockdowns result in an almost complete depletion of the individual isoforms , with double knockdown the efficiency is approximately 75% for SAR1 A while SAR1 B is still fully depleted ( Figure 7A ) . The same procedure of SAR1 knockdown was repeated and 48 hr after transfection with the specific siRNA oligos , the cells were washed , and then incubated with fresh medium containing ascorbic acid for 20 hr . The medium and the cell lysates were western blotted with an anti-Collagen VII antibody . Knockdown of SAR1 A alone had marginal effect on the secretion of PC VII , whereas the knockdown of SAR1 B was largely ineffective . However , knockdown of both SAR1 A and B led to an almost total block in PC VII secretion , which accumulated in the cells ( Figure 7B ) . To visualize the intracellular site of PC VII accumulation , we transfected RDEB/FB/C7 cells to deplete both SAR1 A and B as described above , and 24 hr later ascorbic acid was added to the culture medium . 48 hr later , the cells were fixed and stained with antibodies against Collagen VII and SEC31 , respectively . In cells depleted of SAR1 A and B , PC VII accumulated in large structures ( Figure 7C ) similar to the patches of PC VII seen in RDEB/FB/C7 cells depleted of TANGO1 or SLY1 ( Figure 2D ) . As expected the COPII component SEC31 localized to punctate structures resembling ER exit sites in control cells , but not in cells depleted of SAR1 A and B . Here , SEC31 localized mostly in the cytosol and almost no SEC31 positive punctae were visible ( Figure 7C ) . 10 . 7554/eLife . 02784 . 009Figure 7 . SAR1 A and B knockdown blocks Collagen VII secretion . RDEB/FB/C7 cells were transfected with siRNAs directed against SAR1A , SAR1B , both or a scrambled control siRNA . ( A ) Knockdown efficiency was determined after 72 hr by western blotting cell lysates with an anti-SAR1 antibody . Tubulin was used as a loading control . *-unspecific band . ( B ) Collagen VII secretion was measured by western blotting RDEB/FB/C7 cell lysates and supernatants collected for 20 hr in the presence of ascorbic acid using an anti-Collagen VII antibody . ( C ) RDEB/FB/C7 cells treated with ascorbic acid were seeded on coverslips and 72 hr after transfection with control or SAR1 A and B siRNA , the cells were fixed and visualized by fluorescence microscopy with the indicated antibodies and DAPI ( blue , scale bars: 20 µm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02784 . 009 Recruitment of SEC13/31 to SEC23/24 is the ultimate known event in the assembly of COPII coats that leads to membrane fission and the separation of COPII vesicles from the ER . Is SLY1 activity required before or after the COPII recruitment to the ER exit sites ? TANGO1 , as shown previously , connects PC VII to the COPII subunits through interaction with SEC23/24 . The question we asked was whether the location of COPII components and PC VII in cells that are depleted of SLY1 is different compared with control cells that are exporting Collagen VII . Since PC VII is rapidly exported from the ER after its synthesis , our positive control , to test this hypothesis , was cells transfected with scrambled siRNA oligos for 72 hr and then kept for 3 hr at 15°C to accumulate secretory cargo in the ER and then shifted to 37°C for 30 min to restart cargo export from the ER . These cells were visualized by fluorescence microscopy to monitor the location of PC VII and either the COPII component SEC31 or TANGO1 . In the control cells , PC VII accumulated in the ER in large patches similar to that observed in SLY1 or TANGO1 knockdown cells . The majority of SEC31 and TANGO1 were localized to specific sites on the ER that contained large patches of PC VII ( Figure 8A ) . 10 . 7554/eLife . 02784 . 010Figure 8 . SLY1 is required for post coat assembly events in PC VII export . RDEB/FB/C7 cells were transfected with control siRNA ( A ) or siRNAs directed against SAR1 A+B ( B ) , TANGO1 ( C ) or SLY1 ( D ) . Cells were incubated in medium containing ascorbic acid for 72 hr . In control cells ( A ) protein export from the ER was arrested for 3 hr at 15°C and cells were fixed after a 30 min release to 37°C . All other samples were fixed without a 15°C temperature block ( B–D ) . The ER exit sites were visualized using either anti-SEC31 or anti-TANGO1 antibodies ( green ) in STED mode . PC VII patches were visualized using either anti-Collagen VII ( in combination with anti-SEC31 ) or the colocalizing HSP47 ( in combination with anti-TANGO1 ) antibodies ( red ) in normal confocal mode . Edge length of the zoom boxes is approximately 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 02784 . 010 We then compared the location of SEC31 and TANGO1 in RDEB/FB/C7 cells that had been transfected to knockdown specifically SAR1A and B , TANGO1 , or SLY1 by fluorescence microscopy with anti-Collagen VII and SEC31 or TANGO1 antibodies ( Figure 8B–D ) . In SAR1 ( A and B ) knockdown cells , as expected , SEC31 was distributed mostly in the cytoplasm and not associated with the ER exit sites . Interestingly TANGO1 was still present at the collagen-enriched patches , suggesting that the TANGO1–Collagen VII binding is independent of TANGO1–SEC23/24 interaction ( Figure 8B ) . In TANGO1 knockdown cells , the SEC31 positive ER exit sites appeared fewer in number on the collagen patches and more random in localization ( Figure 8C ) . Interestingly , in SLY1-depleted cells , the number and localization of SEC31 sites resembled that of control cells ( Figure 8D ) . SLY1 knockdown did not affect the location of TANGO1 at the ER exit sites . These findings suggest that knockdown of SLY1 does not interfere with the capture of PC VII by TANGO1 nor with the assembly of a complete COPII coat . A large number of SNAREs have been localized and reported to traffic cargoes between the ER and the Golgi ( Table 1 ) . SLY1 interacts with several of these ER to Golgi SNARE proteins ( Peng and Gallwitz , 2004 ) . To further elucidate the function of SNAREs in the export of PC VII , we tested SLY1 interacting SNAREs , specifically the ER localized STX18 and STX17 as well as the ER and Golgi membrane associated STX5 ( Rowe et al . , 1998; Steegmaier et al . , 2000; Yamaguchi et al . , 2002 ) . RDEB/FB/C7 or Saos2 cells were transfected with siRNA oligos for the respective t-SNAREs . The knockdown efficiency of the cognate mRNAs was assessed by rtPCR ( Figure 9A ) . The medium from RDEB/FB/C7 cells treated with siRNAs for STX5 , 17 and 18 was collected as previously described and western blotted for the presence of PC VII . We observe that both knockdown of STX5 and 18 reduced the amount of PC VII being secreted . STX17 had only marginal effect on the secretion of PC VII ( Figure 9B , C ) . The secretion of Collagen I to the media was assessed in Saos2 cells . We observed a reduction of secreted protein , when cells were treated with siRNAs for STX5 . Neither depletion of STX17 nor depletion of STX18 affected the levels of secreted Collagen I ( Figure 9F ) . Alongside , we quantified by immunofluorescence the number of cells that accumulate PC VII and I intracellularly . For this , we counted at least 30 cells in each of five random fields . In STX18 and STX5 knockdown samples more than 80% of the cells accumulated PC VII in the ER compared to 25% of the cells in the control ( Figure 9D , E ) . These numbers are comparable to SLY1- or TANGO1-depleted cells . Accumulation of PC I was only detectable in STX5-depleted cells ( 80% ) . STX18 , like TANGO1 or SLY1 knockdown did not arrest PC I in the ER ( Figure 9G , H ) . 10 . 7554/eLife . 02784 . 011Table 1 . SNAREs involved in the early secretory pathwayDOI: http://dx . doi . org/10 . 7554/eLife . 02784 . 011Namet- or v-SNARESub-cellular localizationSecretion pathwayCargoes associatedSTX 18tER ( Nakajima et al . , 2004; Itakura et al . , 2012 ) Golgi to ER , ER to GolgiCollagen VII ( this study ) , ssGFP ( Gordon et al . , 2010 ) STX 17tER ( Itakura et al . , 2012 ) AutophagyGFP-LC3 ( Itakura et al . , 2012 ) USE1/P31tER ( Nakajima et al . , 2004; Okumura et al . , 2006 ) Golgi to ERERGIC-53 , KDEL-R ( Aoki et al . , 2009 ) SEC20tER ( Nakajima et al . , 2004 ) Golgi to ER , ER to GolgissGFP ( Gordon et al . , 2010 ) STX 5tGolgi ( Rowe et al . , 1998; Aoki et al . , 2009 ) ER to Golgi , Golgi to ERCollagen VII ( this study ) , Collagen I ( this study ) , ssGFP ( Gordon et al . , 2010 ) BET1tER-ERGIC-Golgi ( Hay et al . , 1996; Uemura et al . , 2009 ) ER to GolgiChylomicrons ( Siddiqi et al . , 2006 ) BET1LtGolgi ( Tai et al . , 2004 ) unknownSEC22BvER-ERGIC-Golgi ( Hay et al . , 1996; Tai et al . , 2004 ) ER to Golgi , Golgi to ERssGFP ( Gordon et al . , 2010 ) YKT6vER-ERGIC-Golgi ( Zhang and Hong , 2001; Volchuk et al . , 2004 ) ER to Golgi , Golgi to ERssGFP ( Gordon et al . , 2010 ) VTI1avER-ERGIC-Golgi ( Flowerdew and Burgoyne , 2009 ) ER to GolgiChylomicrons ( Siddiqi et al . , 2006 ) t , targeting; v , vesicular; ER , endoplasmic reticulum; ERGIC , ER to Golgi intermediate compartment; ssGFP , signal sequence GFP . 10 . 7554/eLife . 02784 . 012Figure 9 . The t-SNARE Syntaxin 18 is necessary for Procollagen VII but not Procollagen I export . To evaluate PC VII ( B–E ) and PC I export ( F–H ) , RDEB/FB/C7 and Saos2 cells were transfected with siRNAs directed against STX5 , STX17 , STX18 or a scrambled siRNA . ( A ) Knockdown efficiency was assessed 48 hr after transfection by RT-PCR . PC VII secretion ( B ) and PC I secretion ( F ) was measured by western blotting cell lysates and supernatants collected for 20 hr in the presence of ascorbic acid from RDEB/FB/C7 and Saos2 cells , respectively . Equal protein loading and cell lysis was controlled by blotting with an anti-Tubulin antibody . ( C ) Intensities of Collagen VII in the lysate and the supernatant was recorded by densitometry in four independent experiments . The ratio of external vs internal Collagen VII was normalized to quantify secretion in control cells as 100%; Error bars: standard error of the mean ( SEM ) . ( D and G ) siRNA treated RDEB/FB/C7 cells were seeded on coverslips and 20 hr after addition of ascorbic acid , cells were fixed and visualized with Collagen VII ( D ) or Collagen I ( G ) antibodies ( green ) , and DAPI ( blue ) by fluorescence microscopy . The percentage of cells that accumulate PC VII ( E ) or PC I ( H ) intracellularly was determined by counting at least 30 cells in five random fields . The number of cells accumulating PC VII in the case of STX5 and STX18 siRNA was significantly different from control cells ( p<0 . 05 ) . Accumulation of PC VII in STX17 siRNA cells was not significantly different from the control situation ( p>0 . 1 ) . Error bars: standard error of the mean ( SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02784 . 012 A large number of secretory cargoes bind receptors of the ERGIC53 , ERV , and p24 families; the receptors in turn bind the inner COPII coat components and are then packed into a standard COPII vesicle ( Zanetti et al . , 2012 ) . Our previous findings indicate that TANGO1 , which localizes to ER exit sites , binds PC VII in the lumen and SEC23/24 of the COPII coats on the cytoplasmic site ( Saito et al . , 2009 ) . The function of TANGO1 is facilitated by cTAGE5 ( Saito et al . , 2011 ) . cTAGE5 and TANGO1 interact through their second ( proximal to the proline rich domain ) coiled–coiled domains at the ER exit sites . cTAGE5 does not have a luminal domain and cannot therefore bind PC VII directly ( Saito et al . , 2011 ) . The TANGO1–cTAGE5 complex is proposed to stall the COPII vesicles at the ER exit sites as they are being packed with PC VII and this procedure is proposed to result in the production of a transport intermediate for the export of PC VII from the ER ( Malhotra and Erlmann , 2011 ) . The process by which PC I in the lumen of the ER is connected to COPII coat components remains unclear . In our experimental conditions , SLY-1 and the ER-associated t-SNARE STX18 are specifically required for the export of PC VII but not of the general cargoes and PC I . This strongly indicates the existence of different export pathways from the ER ( Figure 10A ) . These findings also highlight that the sorting of cargoes into different export routes is not driven solely by the size of the secretory cargoes . While the mechanism regulating this sorting event remains unclear it is worth highlighting that TANGO1 in Drosophila is localized to the basal endoplasmic reticulum specific exit sites ( Lerner et al . , 2013 ) . Moreover , TANGO1 in Drosophila is required for the export of Collagen IV , which is secreted into the basal face of the cells ( Lerner et al . , 2013 ) . It is therefore tempting to propose that secretory cargoes are sorted in the lumen of the ER based on their final destination . 10 . 7554/eLife . 02784 . 013Figure 10 . A working model for procollagen export from the ER . ( A ) Multiple exit routes from the ER . Cargo receptors of the ERGIC53 , ERV , and p24 families bind secretory cargoes in the lumen and the inner COPII coat on the cytoplasmic side of the ER . These receptors bound to cargoes are collected into COPII vesicles for export from the ER . TANGO1 connects PC VII in the lumen of the ER with COPII coats on the cytoplasmic side of the ER . The mechanism by which PC I is connected with the COPII components is not known . Sedlin and ubiquitination of SEC31 are required for PC I secretion but their role in PC VII export is not known . ( B ) Building a Procollagen VII containing mega carrier by fusion of recycling ERGIC membranes . Post concentration of PC VII by TANGO1 at the ER exit sites , many of the ER exit sites are concentrated to generate a patch enriched in PC VII . This patch containing STX18 then promotes SLY1 dependent fusion of membranes from the ERGIC . Accretion of membranes by this process grows PC VII-enriched ER patch . The COPII coats and the TANGO1 remain at the neck of the growing patch . These components at the neck promote fission and the resulting mega container enriched in PC VII and ERGIC membrane components is in fact the first post ER compartment that matures to move PC VII forward . DOI: http://dx . doi . org/10 . 7554/eLife . 02784 . 013 Human cells contain a second SLY1 like protein ( SCFD2 , Gene ID: 152579 ) . A TANGO1 like protein called MIA2 that is expressed only in the liver and small intestine has also been identified ( Pitman et al . , 2011 ) . The cargoes exported by these proteins are not known but they could , in combination with SAR1 , facilitate the export of other collagens , including Collagen I and perhaps many other bulky molecules , from the ER . At this juncture , it is also important to comment on our original finding that TANGO1 inhibited the secretion of an artificial cargo ss-HRP in Drosophila S2 cells ( Bard et al . , 2006 ) . There is no evidence that TANGO1 binds HRP in the lumen of the ER and yet both knockdown of TANGO1 and SLY1 , as shown here , affect ss-HRP secretion in mammalian cells . We suggest that the exogenously expressed ss-HRP or ss-GFP exits the ER by any route available and blocking any one , for example by TANGO1 or SLY1 knockdown , affects its export because of a block in one of the possible exit routes . Our surprising finding is that SLY1 , which is known to interact with STX18 , a t-SNARE for membrane fusion ( Hatsuzawa et al . , 2000 ) , is required for the export of PC VII from the ER . The function of SLY1 , which is a member of the STXBP/unc-18/SEC1 protein family , remains unclear . For example , it is not known whether SLY1 regulates the activity of the SNARE complex , or helps with the fusion or the disassembly of the SNAREs ( Carr and Rizo , 2010 ) . SLY1 at the ER -Golgi interface has been shown to interact with a large number of SNAREs and could thus regulate a number of different SNARE complexes ( Dascher and Balch , 1996; Steegmaier et al . , 2000; Yamaguchi et al . , 2002 ) . The t- and v-SNAREs of the ER Golgi anterograde and retrograde trafficking routes are summarized in Table 1 . So far the SNARE complexes reported for the forward trafficking from ER to Golgi are complexes of Golgi t-SNAREs and a specific v-SNARE . However , there is no evidence that the ER attached t-SNAREs such as STX17 , STX18 , Sec20 and USE1/P31 are required for the forward transport to the Golgi complex . It could be argued that SLY1 and STX18 function to recycle a specific component , a v-SNARE; for example , from ERGIC/early Golgi cisternae and without which , the cells terminate the export of PC VII out of the ER . However , cycling of KDEL containing protein BIP and ERGIC53 as well as the export of other secretory cargoes including the bulky PC I is unaffected under these experimental conditions . Additionally , we do not see any obvious accumulation of PC VII containing large vesicles in between the ER and the Golgi . We therefore favor the possibility that SLY1 and STX18 are required for the fusion of membranes to the ER to generate an export carrier for PC VII . The working hypothesis for the export of PC VII from the ER follows ( Figure 10B ) . A SAR1 dependent reaction initiates the assembly of COPII subunits SEC23/24 at the ER exit sites . Recruitment of TANGO1-cTAGE5 complex into this reaction with its binding of PC VII in the lumen and SEC23/24 on the cytoplasmic site . The binding of TANGO1-cTAGE5 to SEC23/24 retards the recruitment of SEC13/31 and prevents premature pinching of membranes ( Saito et al . , 2009; Malhotra and Erlmann , 2011; Saito et al . , 2011 ) . These ER exit sites containing TANGO1 bound PC VII are then concentrated to generate a specific patch , which for the sake of simplicity we call PCP ( PC VII concentrated patch ) . A special pool of recycling membranes; for example , the ERGIC , then fuse with PCP by a reaction that requires SLY1 and STX18 . As the PCP grows , TANGO1 dissociates from the collagen , which then promotes the recruitment of SEC13/31 to the neck . This reaction also prevents the packing of TANGO1-cTAGE 5 into the PC VII enriched bud ( PEB ) . The neck of this large PEB , concentrated in COPII coats , undergoes fission by the same mechanism as that of a standard COPII vesicle . The PC VII enriched large membrane compartment separated from the ER also contains recycled ERGIC membranes . This membrane bound compartment in principle is the first post ER compartment that matures to move PC VII in the anterograde direction for secretion . Interestingly , Luini et al . have reported that export of PC I from the ER requires its concentration at a domain close to the ER exit sites , which they suggest protrudes and then separates from the ER to generate a PC I enriched carrier ( Mironov et al . , 2003 ) . Their model is reminiscent of our scheme in which we place molecular components such as TANGO1 for the concentration of PC VII followed by the fusion of membranes to grow the PC VII enriched ER domain . However , the TANGO1 mediated concentration of PC VII requires the involvement of COPII components , SEC23/24 . Additionally , the protrusion visualized by the microscopy based analysis might correspond to growth of ER patch by SLY1 and STX18 dependent fusion of the recycling ( ERGIC ) membranes . The challenge is to find proteins that connect PC I with the COPII machinery , the SNAREs required for the trafficking of PC I , and the visualization of membranes that we propose are added to grow the patch of PC VII containing ER membrane into a vectorial carrier en route to the cell surface for secretion . TANGO1 constructs were as described previously ( Saito et al . , 2009 ) . Full-length SLY1 was amplified by RT-PCR from the total mRNA of HeLa cells . SLY1 was cloned into the retroviral vector pLJM1-GFP ( Addgene , Cambridge MA ) using Gibson cloning strategy ( Gibson et al . , 2009 ) . SnapGene software ( from GSL Biotech , Chicago , IL; available at www . snapgene . com ) was used for molecular cloning procedures . The Ii-FRAP-HA plasmid has been previously described ( Pecot and Malhotra , 2006 ) . The FKBP-ERGIC53-GFP plasmid was a gift from Dr Hauri ( Ben-Tekaya et al . , 2005 ) . RDEB/FB/C7 cells , HeLa , Het1a , and Saos2 cells were grown at 37°C with 5% CO2 in complete DMEM with 10% fetal bovine serum . Plasmids were transfected in HeLa , with TransIT-HeLaMONSTER ( Mirus Bio LLC , Madison , WI ) or X-tremeGENE9 ( Roche , Indianapolis , IN ) according to the manufacture's protocols . siRNAs were transfected in HeLa , RDEB/FB/C7 , Het1a and Saos2 with Hiperfect ( Qiagen , Venlo , Netherlands ) according to the manufacture's protocols . In the case of RDEB/FB/C7 , siRNAs were transfected twice , on day 0 and on day 1 . For lentiviral infection of SLY1-GFP into RDEB/FB/C7 cells , lentiviral particles were produced by co-tranfecting HEK293 cells with a third generation packaging vector pool and pJLM1-SLY1-GFP using TransIT-293 ( Mirus Bio LLC ) . At 72 hr post transfection the viral supernatant was harvested , filtered , and directly added to RDEB/FB/C7 cells . siRNAs for SLY1 , TANGO1 , STX5 , STX17 , STX18 , and SAR1 A and B were purchased from Eurofins MWG Operon ( Huntsville , AL ) : SLY1 siRNA- AGACUUAUUGAUCUCCAUA; TANGO1 siRNA- GAUAAGGUCUUCCGUGCUU; STX5 siRNA- GGACAUCAAUAGCCUCAAC; STX17 siRNA- GACUGUUGGUGGAGCAUUU; STX18 siRNA- CAGGACCGCUGUUUUGGAUUU; for knockdown of SAR1A+B a pool of 4 siRNAs were used as previously described ( Cutrona et al . , 2013 ) . Control siRNAs consisted of a pool of ON-TARGETplus Non-Targeting siRNAs ( D-001810-10-05 , Thermo Scientific , Waltham , MA ) . Antibodies used in conventional western blotting , immunofluorescence , and immunoelectron microscopy are the following: SLY1 , Collagen VII , Collagen I , GRP78/BIP , SEC23A ( Abcam , Cambridge , UK ) ; TANGO1 ( LifeSpan Biosciences , Seattle , WA ) ; SAR1 ( Milipore , Billerica , MA ) ; HSP47 ( Enzo Life Sciences , Farmingdale , NY ) ; SEC31 , GM130 ( BD Biosciences , San Jose , CA ) ; β-COP , myc , Tubulin , Actin ( Sigma-Aldrich , St . Louis , MO ) ; HA ( Covance , Princeton , NJ ) ; ε-COP ( a gift from Dr Rothman ) . HeLa cells were transfected with a plasmid encoding MycHis-tagged cytoplasmatic tail of TANGO1 . 24 hr post transfection proteins were cross linked by incubation with the cell permeable cross linker DSP ( Pierce , Rockford , IL USA ) for 30 min at room temperature . Residual cross linker was quenched by adding Tris-HCl to a final concentration of 150 mM . The cells were scraped into lysis buffer ( 50 mM Tris , pH 7 . 5 , 150 mM NaCl , 1% TX100 , 1 mM sodium orthovanadate , 10 mM sodium fluoride , and 20 mM β-glycerophosphate , 10 mM Imidazol ) and lysates were clarified by centrifugation at 16 , 000×g for 10 min at 4°C . The lysates were then incubated with Ni+-Agarose ( Qiagen ) beads for 2 hr at 4°C on a rotating platform . The beads were washed three times with lysis buffer and the specifically bound proteins eluted using PBS containing 500 mM Imidazol . RDEB/FB/C7 or Saos2 cells were transfected with siRNA as described above . 24 hr after the last siRNA transfection , the medium was replaced with a fresh medium containing 2 μg/ml ascorbic acid . Fresh medium containing ascorbic acid was added again to the cells 20 hr before being collected . Upon collection , the media were centrifuged at low speed to remove any cell in suspension and the supernatant boiled for 5 min with Laemmli SDS-sample buffer . For cell lysis , the cells were washed with PBS , lysed and centrifuged at 14 , 000 rpm for 15 min at 4°C . The supernatants were boiled for 5 min with Laemmli SDS-sample buffer . Both media and cell lysate were subjected to SDS-PAGE ( 6% acrylamide ) and western blotting with Collagen VII , Collagen I , and Tubulin antibodies . ImageJ ( NIH , Bethesda , Maryland ) was used for quantitation . Cells grown on coverslips were fixed either with cold methanol for 10 min at −20°C or with 4% paraformaldehyde in PBS for 10 min followed by permeabilization with 0 . 2% Triton X100 at room temperature , and then incubated with blocking reagent ( Roche ) for 30 min at room temperature . Primary antibodies were diluted in blocking reagent and incubated overnight at 4°C . Secondary antibodies conjugated with Alexa 488 , Alexa 594 or Alexa 647 were diluted in blocking reagent and incubated for 1 hr at room temperature . For SLY1 localization studies , RDEB/FB/C7 expressing SLY1-GFP were transfected with the indicated siRNAs; after 72 hr after transfection cells were processed for immunofluorescence . Briefly , cells were washed twice with room temperature KHM buffer ( 125 mM potassium acetate , 25 mM HEPES [pH 7 . 2] , and 2 . 5 mM magnesium acetate ) . Cells were then permeabilized by incubation in KHM containing 0 . 1% Saponin for 5 min on ice followed by wash for 7 min at room temperature with KHM buffer . Cells were subsequently fixed in 4% paraformaldehyde and processed for immunofluorescence microscopy . Images were taken with a Leica TCS SPE confocal with a 63x objective . SEC31 and TANGO1 localization images on Figure 8 were taken with a Leica TCS SP5II CW-STED system in STED mode with a 100 × 1 . 4NA objective at a pixel-size of 20 nm using a 592 nm depletion laser and HyD detectors . Cryoimmunoelectron microscopy was performed as described previously ( Martinez-Alonso et al . , 2005 ) . Control and SLY1 or TANGO1 depleted RDEB/FB/C7 were fixed with 2% paraformaldehyde and 0 . 2% glutaraldehyde in 0 . 1 M sodium phosphate buffer , pH 7 . 4 . After washing in buffer , the cells were pelleted by centrifugation , embedded in 10% gelatin , cooled on ice and cut into 1 mm3 blocks . The blocks were infused with 2 . 3 M sucrose at 4°C overnight , frozen in liquid nitrogen and stored until cryo-ultramicrotomy . Sections ( ∼50 nm-thick ) were cut at −120°C with a diamond knife in a Leica Ultracut T/FCS . Ultrathin sections were picked up in a mix of 1 . 8% methylcellulose and 2 . 3 M sucrose ( 1:1 ) . tc\l 1 ‘Preparation of immunogold labeled cryosections’ . Cryosections were collected on carbon and formvar-coated copper grids and incubated with rabbit polyclonal antibodies against Collagen VII , followed by protein A-gold . After labeling , the sections were treated with 1% glutaraldehyde , counterstained with uranyl acetate pH 7 and embedded in methyl cellulose–uranyl acetate pH 4 ( 9:1 ) . HeLa cells were transfected with the respective siRNAs as mentioned above . Past 48 hr , the cells were transfected with FKBP-ERGIC53-GFP and Ii-FRAP-HA plasmids as described previously ( Pecot and Malhotra , 2006 ) . 24 hr after plasmid transfection , cells were incubated with media only ( control ) , or media + CHX ( 100 µg/ml ) + Rapamycin ( 200 nM ) . At the indicated time points , the cells were fixed and analyzed by fluorescence microscopy . ERGIC53 was visualized by GFP fluorescence , and Ii with a HA-antibody and an Alexa594-labeled secondary antibody . Colocalization of the two proteins was assessed in at least 30 cells from five different focus fields of an representative experiment by calculating the Pearson's colocalization coefficient using ImageJ software and the Colocalization Analysis plug-in . Pearson's colocalization coefficient of 1 means total colocalization whereas 0 indicates no colocalization . Control and siRNA-treated cells were cultured in DMEM without L-methionine and L-cysteine for 60 min and pulsed with 100 µCi 35Smethionine ( Hartmann Analytics ) for 20 min . The cells were washed three times with PBS and chased with DMEM containing 10 mM unlabeled methionine for 2 hr . For BFA treatment , 10 µg/ml BFA was added to the medium for the last 10 min of incubation and kept for the whole experiment . Cells were lysed with PBS containing 1% Triton X-100 . 20 µl of cell extracts was mixed with scintillation cocktail , and the radioactivity was determined for normalization . The collected medium was precipitated with TCA and analyzed by SDS-PAGE/autoradiography . HeLa cells stabily expressing ss-HRP were cultured in a 12-well plate and transfected with control siRNA , SLY1 or TANGO1-specific siRNA oligos . 72 hr later , cells were washed with medium and incubated with 1 ml of complete medium at 15°C for 2 hr . Then , cells were incubated with 500 μl of complete medium at 37°C in presence of 100 μM CHX . At the time points indicated , the medium and cells were harvested for HRP secretion assay . 50 μl of the medium was collected and cells were lysed in 100 μl of lysis buffer ( 50 mM Tris , pH 7 . 4 , 150 mM NaCl , 0 . 1% sodium dodecyl sulfate ( SDS ) , 1% Nonidet P-40 ( NP- 40 ) and 0 . 5% sodium deoxycholate ) supplemented with protease inhibitors , 1 mM Na3VO4 , and 25 mM sodium fluoride and centrifuged at 16 , 000×g for 15 min . For the chemiluminescence assay , 50 μl of medium and cell lysate were mixed with an ECL reagent ( Thermo Scientific ) and luminescence measured with a Victor 3 plate reader ( PerkinElmer , Waltham , MA ) .

Collagens are long proteins that join individual cells together to build tissues and organs . They also provide strength and elasticity to bones , tendons , and blood vessels . Like many other proteins , collagens are produced inside cells: they are folded in a compartment called the endoplasmic reticulum , and then packaged and transported to another compartment called the Golgi . Collagens are then directed from the Golgi to their final destination , which is typically the outside of the cell . Small proteins travel from the endoplasmic reticulum to the Golgi inside packages called vesicles . However it is not clear how large proteins like collagens are transported between these two compartments . It is known that a protein called TANGO1 is needed to direct a collagen called Procollagen VII to the outside of the cells . TANGO1 binds to Procollagen VII , and it is thought that TANGO1 delays the release of Procollagen VII from the endoplasmic reticulum , so that the vesicle can grow to a size that is able to accommodate such a bulky cargo . Nogueira , Erlmann et al . have now discovered that TANGO1 binds to another protein called SLY1 , and that this protein must also be present if Procollagen VII is to be exported from the endoplasmic reticulum . In contrast , the transport of a different type of collagen—Collagen I—does not require TANGO1 or SLY1 . SLY1 helps to fuse the membranes that enclose the structures involved in protein trafficking—such as the endoplasmic reticulum , the Golgi , and the vesicles—and this allows the cargoes of vesicles to pass from one compartment to another . Nogueira , Erlmann et al . also found that a second protein ( called Syntaxin 18 ) is also required for the export of Procollagen VII . Nogueira , Erlmann et al . propose that collagen VII export involves TANGO1 delaying the release of collagen from the endoplasmic reticulum so that SLY1 and Syntaxin 18 can fuse other cellular membranes to the growing transport vesicle . Following this work , the next challenge is to uncover how different types of collagens are separated from each other , and identify which specific vesicles are involved in their export .

2014

SLY1 and Syntaxin 18 specify a distinct pathway for procollagen VII export from the endoplasmic reticulum

Though neurotransmitters are essential elements in neuronal signal transduction , techniques for in vivo analysis are still limited . Here , we describe an organic electrochemical transistor array ( OECT-array ) technique for monitoring catecholamine neurotransmitters ( CA-NTs ) in rat brains . The OECT-array is an active sensor with intrinsic amplification capability , allowing real-time and direct readout of transient CA-NT release with a sensitivity of nanomolar range and a temporal resolution of several milliseconds . The device has a working voltage lower than half of that typically used in a prevalent cyclic voltammetry measurement , and operates continuously in vivo for hours without significant signal drift , which is inaccessible for existing methods . With the OECT-array , we demonstrate simultaneous mapping of evoked dopamine release at multiple striatal brain regions in different physiological scenarios , and reveal a complex cross-talk between the mesolimbic and the nigrostriatal pathways , which is heterogeneously affected by the reciprocal innervation between ventral tegmental area and substantia nigra pars compacta . In the nervous system , neurotransmitters ( NTs ) are released upon arrival of action potentials , and play essential roles in signal transmission for regulating physiological activities ( Kavalali , 2015 ) . Though electrical recording and stimulation of neural activities have offered extremely powerful and widely used toolsets for neuroscience research , electrophysiological methods typically lack cell selectivity and can only be used for stimulatory but not inhibitory manipulation , and their spatial resolution can be limited and highly depend on the size of electrodes ( Tye and Deisseroth , 2012 ) . Alternatively , the methods for the direct measurement of NT release are relatively limited ( Wightman and Robinson , 2002 ) , partially due to the complex chemical dynamics and a substantial variety of NTs . Among the large family of NTs , the category of catecholamine neurotransmitters ( CA-NTs ) consists of dopamine , noradrenaline , and adrenaline , which all share a common chemical structure containing a catechol with two hydroxyl groups and a side-chain amine . These NTs influence a great variety of neural functions ( Fields et al . , 2007 ) , and have been one of the major targets for previous efforts on the detection of NTs in animal brains ( Wightman and Robinson , 2002; Bucher and Wightman , 2015 ) . Currently , a few techniques are available for the detection of NT release in vivo . Microdialysis-based method is a two-step analytical process that involves sample collection from the brain and a later analysis by liquid chromatography ( Fiorino et al . , 1993 ) . It can only resolve averaged biochemical events over minutes and has relatively poor spatial resolution . Electrochemical techniques are developed to resolve transient biochemical processes . Amperometry has the advantage of generally good detection limits but is restricted to electroactive species , and is not suitable for detection in complex environment ( Bucher and Wightman , 2015 ) . Cyclic voltammetry ( CV ) was adopted to monitor biochemical fluctuation in mammalian brains by recording oxidation-induced changes in related current-voltage curves ( Kissinger et al . , 1973; Robinson et al . , 2003 ) . While this technique is efficient and NT selective , the readout from CV is not intuitive , and the technique also involves a complex instrument configuration ( Atcherley et al . , 2015 ) . So far , it is still challenging to have a fully implantable multi-channel device that can detect and monitor NT release with sufficient sensitivity and temporal-spatial resolution . Recently , genetically encoded fluorescent sensor has been developed for detecting dopamine in vivo ( Patriarchi et al . , 2018 ) , which has improved selectivity and temporospatial resolution . But it also requires genetic modifications along with complex optical recording equipments , and is less quantitative compared to other methods . Organic electrochemical transistor ( OECT ) has emerged as a promising transducer for detecting chemical , electrical , and molecular signals ( Rivnay et al . , 2018; Rivnay et al . , 2017; Lin and Yan , 2012; Lin et al . , 2011; Tang et al . , 2011b; Khodagholy et al . , 2013 ) , in virtue of its intrinsic amplification capability , low operating voltage , mechanical flexibility , and efficient ion transport/exchange between the device and electrolyte environment ( Lin and Yan , 2012 ) . So far , many biological molecules ( including NTs ) and cellular activities have been successfully detected ex vivo by using the OECT platforms ( Kergoat et al . , 2014; Rivnay et al . , 2015; Jonsson et al . , 2016; Campana et al . , 2014 ) . However , considering the complex biological environment , recent efforts of in vivo biosensing by OECTs mostly focus on stimulation or detection of electrophysiological signals ( Williamson et al . , 2015; Lee et al . , 2016 ) , while the release and transport of chemical substances in living organisms receive less investigations ( Li et al . , 2015; Xu et al . , 2016 ) . In this study , we describe a fully implantable OECT-array for in vivo detection of NTs . This device is capable of multichannel monitoring of dopamine release as low as 1 nM with a 50 ms sampling rate , and is well suitable for usage in living animals . Using this OECT-array , we successfully demonstrated real-time , multi-site , simultaneous monitoring of CA-NT release at different brain regions , including ventral tegmental area ( VTA ) , nucleus accumbens ( NAc ) , and caudate putamen ( CPu ) , in response to neural stimulation along dopaminergic pathways . Our study provides an electrochemical mapping in the striatal brain and also reveals a complex cross-talk between mesolimbic and nigrostriatal signalling , which relies on a reciprocal innervation between VTA and substantia nigra pars compacta ( SNc ) nuclei ( Margolis et al . , 2008 ) . We found that the connection between the two nuclei can significantly affects dopamine release in NAc and lower CPu , but not upper CPu , suggesting a heterogeneous mapping from SNc to CPu ( Lerner et al . , 2015 ) . Our device is based on a miniaturized microarray of OECTs ( OECT-array ) on a 200 μm thick polyethylene terephthalate ( PET ) substrate . Each functional unit consists of a platinum GATE ( Pt-GATE ) electrode and a conductive poly ( 3 , 4-ethylenedi-oxythiophene ) :poly ( styrene-sulfonate ) ( PEDOT:PSS ) film that bridges the gold SOURCE and DRAIN electrodes ( Figure 1a ) . When an OECT-array was engaged in a biological electrolyte , some molecules ( e . g . , CA-NTs; Figure 1b ) could undergo electro-oxidation reaction on the surface of the Pt-GATE electrode . In this process , the catechol groups of CA-NT molecules are oxidized into quinone with the release of two electrons , which are transferred to the GATE to generate a Faradic current , decrease the potential drop at the GATE/electrolyte interface , and subsequently increase the effective GATE voltage Vg-eff , as described by: ( 1 ) ΔVg−eff∝2 . 30 ( 1+γ ) κT2qlog[C] , where γ is the ratio between the volumetric capacitance of the PEDOT:PSS active channel ( Cvolumetric ) and the capacitance of the GATE-electrolyte interface ( CG-E ) ( Figure 1c ) ; k is the Boltzmann constant; T is the absolute temperature; q is the charge of an electron; [C] is the concentration of the reactive molecules released into the biological solution ( Tang et al . , 2011a ) . Notably , Equation ( 1 ) is normally applicable in a relatively high-concentration regime; For lower concentrations of analyst , we can use an empirical relationship given by Zhang et al . ( 2014 ) : ( 2 ) ΔVg−eff=A×[C]β , where A and β are constants determined by fitting our experimental data . To accommodate an easy implantation in the brain of a living animal , the OECT-array was designed and fabricated into a slim blade shape ( ~1 mm wide , ~15 mm long , and ~200 μm thick ) with a tapering tip . Four sets of functional OECT units were integrated at the top 5 mm of a device , each spaced by 1 . 2 mm ( Figure 2a , also see Figure 2—figure supplement 1 ) . The Pt-GATE and the PEDOT:PSS channel were exposed to the biological environment and all the rest area was insulated by a micropatterned layer of SU-8 photoresist . In the electrical characterization , a dual channel sourcemeter was used to provide the GATE voltage between the Pt-GATE and the SOURCE electrode ( VGS ) , and the drain voltage across the active channel ( VDS , the voltage between the DRAIN and the SOURCE electrode ) ; the channel currents running through the DRAIN and the SOURCE electrode ( IDS ) was monitored ( Figure 2b ) . A transfer curve ( VGS vs . IDS ) and the corresponding transconductance ( gm ) of a functional OECT-unit in phosphate-buffered saline ( PBS ) were firstly acquired ( also see Figure 2—figure supplement 2 ) . The OECT-array was then characterized for its capability to detect the release of CA-NTs . Specific biological molecules were manually added to the electrolyte environment to mimic a pulsed release . In solutions ( e . g . , PBS ) that are clear of any active background molecules , the detection limit can reach as low as 1 nM for pulsed release of dopamine , and was relatively consistent for different types of CA-NTs , including noradrenaline and adrenaline ( Figure 2—figure supplement 2 ) . In the artificial cerebral spinal fluid ( ACSF ) solution containing a high level of interfering background molecules ( e . g . , 1 . 28 mM ascorbate ) that mimics the brain extracellular environment ( Mo and Ogorevc , 2001 ) , the OECT-array still showed obvious responses to the fluctuation of CA-NT concentration at 30 nM ( Figure 2c ) . For quantitative analysis , the recorded IDS fluctuation was converted to the change of the effective GATE voltage ( ΔVg-eff ) using the I-V transfer characterization for each independent device . The ΔVg-eff was further calibrated to the fluctuation of dopamine concentration in ACSF over a wide range of concentration values ( 30 nM ~ 0 . 1 mM ) , suggesting an extremely good dynamic range of the OECT-array for detecting CA-NTs , and the data could be fitted well with Equation ( 2 ) ( Figure 2d ) . We next examined the feasibility of using the OECT-array to detect the release of NTs , especially dopamine , in the brain of a living animal . An OECT-array was implanted to the VTA ( −5 . 6 mm A/P , 0 . 8 mm M/L , 8 mm below dura ) of an anesthetized rat using a stereotactic apparatus ( Figure 3a ) . This region was selected because it is one of the main regions involved in dopaminergic signalling in a mammalian brain , mediating reward- and reinforcement-related behaviors ( Fields et al . , 2007 ) . In parallel , a tungsten electrode was inserted to the medial forebrain bundle ( MFB; −1 . 8 mm A/P , 2 mm M/L , 8 mm below dura ) to evoke somatodendritic release of dopamine in VTA ( Kita et al . , 2009 ) . Another tungsten electrode were bundled with the OECT-array for confirmative electrical recording ( Figure 3b ) . Proper positioning of the OECT-array in VTA was confirmed by the spontaneous firing pattern . After being implanted to a rat brain , the VGS-IDS transfer curve of the OECT-array was firstly acquired , and the derived transconductance curve was found to overlap well with the characterization acquired in ACSF ( Figure 3c , also see Figure 3—figure supplement 1 ) , suggesting a stable device performance in brain tissues and the validity of the ex vivo calibration results ( Figure 2d ) . Upon electrical stimulation ( 200 μA , 2 ms pulse width ) in the MFB , the IDS of the 1st OECT-unit immediately showed a downward fluctuation , indicating the successful detection of a transient dopamine release . The release intensity was observed to linearly associate with the number of electrical pulses delivered to MFB ( Figure 3d , e ) . When the stimulation sequence ramped from 1 to 100 electrical pulses , the amplitude of dopamine release in the VTA ( as a result of transient release ) rose from 36 . 13 ± 11 . 95 to 356 . 05 ± 52 . 07 nM ( Figure 3e ) . When the stimulation electrode was moved out of the MFB to an irrelevant region ( −1 . 8 mm A/P , 2 mm M/L , 3 . 5 mm below dura ) , even a very strong stimulation failed to induce any dopamine release . These results proved the basic feasibility of using the OECT for reliable and real-time detection of NT release in a mammalian brain . In another scenario , NTs are released at axonal terminals originated from the far away somata for signal transmission . We then examined the possibility of using the OECT-array to monitor dopamine release at remote axonal terminals after a long-range projection . Specifically , we focused on the mesolimbic pathway , in which the projection of dopaminergic neurons connects VTA and NAc ( Garris et al . , 1999 ) . Accordingly , the OECT-array was implanted in the NAc ( 1 . 2 mm A/P , 1 . 4 mm M/L , 8 . 4 mm below dura ) , and a tungsten electrode was placed in the VTA for neural stimulation ( Figure 4a , also see Figure 4—figure supplement 1 ) . The surgical tracks in these two regions indicated the precise placement of the devices . Activation of relevant dopamine neurons was later confirmed by immunostaining for tyrosine hydroxylase ( TH ) and c-Fos in these cells ( Figure 4b ) . In VTA , the majority of TH+ staining was observed to be on neuron somata; however , in NAc , the TH+ staining was mostly punctuated , suggesting an enrichment of dopaminergic axonal terminals in this region ( Figure 4b ) . Before any electrochemical measurement , the polysynaptic connection between these two regions was firstly verified by electrical recording , which showed a clear temporal synchronization of spiking activity for neurons in the VTA and the NAc ( Figure 4—figure supplement 2 ) . To evoke dopamine release along the mesolimbic pathway , a 200 μA electrical stimulus ( 50 Hz , 2 ms pulse width , 50 pulses ) was applied in the VTA to activate the dopaminergic neurons . Concurrently , the signals were transmitted to the NAc and monitored by the OECT-array . From the 1st OECT-unit placed in the NAc , a significant downward fluctuation of the IDS was recorded by the OECT-array ( Figure 4c ) , and accordingly the evoked dopamine release was significantly higher than that of the control groups , in which the placement of the stimulation electrode and the recording OECT-unit were not paired on the dopaminergic pathway ( Figure 4d ) . Notably , we found that the evoked dopamine release was frequency dependent , and a 50 Hz stimulation in the VTA induced the most significant changes ( Figure 4e ) , which echoes a previous study that reported varying efficiency to evoke dopamine release by stimulating VTA using electrical signals of different frequencies ( Addy et al . , 2010 ) . We then compared simultaneous recording from multiple OECT-units . As the placement of unit two was already away from the center of NAc , the recorded dopamine release was significantly lower than that from unit 1 ( Figure 4f ) . These results demonstrated sensitive monitoring of NT release by the OECT-array at the remote axon projection terminals , and showed a potential for mapping neuronal electrochemical events by the array of multiple OECT-units on a single device . Activation of VTA/SNc complex simultaneously evokes dopamine release in NAc and CPu via mesolimbic and nigrostriatal pathways , respectively ( Garris and Wightman , 1994 ) . Our OECT-array provides the opportunity to simultaneously characterize the dopamine releasing profile across the two pathways , which cannot be precisely accessed by existing methods ( Schwerdt et al . , 2017 ) . For each OECT-unit , the size of GATE electrode was 0 . 48 mm2 and four of them ( on one OECT-array ) were sufficient to cover brain locations spanning from NAc to different parts of CPu ( Garris et al . , 1999; Phillips et al . , 2003 ) . In this experiment , the OECT-array was inserted into the striatum with an angle of 14o between the device and the dorsal-ventral axis . In this way , the 1st OECT-unit was placed in the NAc region , and the 3rd and 4th OECT units were placed in the lower and upper part of CPu ( Figure 5a , specific coordinates provided in Materials and methods ) . As the 2nd OECT-unit was in the transition region between NAc and CPu , signals were not taken from this unit . To specifically activate the mesolimbic pathway or nigrostriatal pathway , we delivered electrical stimulation to either VTA ( −5 . 3 mm A/P , 0 . 8 mm M/L , 7 . 9 ~ 8 . 4 mm below dura ) or SNc ( −5 . 3 mm A/P , 2 mm M/L , 7 . 5 ~ 7 . 9 mm below dura ) ( Figure 5b ) , and investigated the corresponding dopamine release pattern in the striatum . We found that either VTA- or SNc-stimulation could evoke significant dopamine release in the large striatal region covered by the OECT-array ( Figure 5c ) , which was not observed if the electrical stimulation was delivered to an irrelevant site ( −5 . 3 mm A/P , 0 . 8 mm M/L , 4 . 9 mm below dura ) ( Garris and Wightman , 1994 ) . Switching the stimulation site from VTA to SNc only affected the response in the NAc , but not in other striatal regions , suggesting a primary involvement of NAc in the mesolimbic pathway . However , it is interesting to find that SNc-stimulation also induced some dopamine release in the NAc , suggesting a possible cross-talk between the mesolimbic and the nigrostriatal pathways . Such connection between the two pathways was further evidenced by a similar response in the CPu ( recorded by unit 3 and 4 ) to VTA- or SNc-stimulation . Based on the dopamine mapping in the striatum , we then hypothesized a mutual innervation mechanism between VTA and SNc that contributes to the cross-talk between the mesolimbic and the nigrostriatal pathways . To validate this hypothesis , we firstly conducted electrophysiology recording to probe dopamine neuron activity in the SNc in response to VTA-stimulation or vice versa , and confirmed a reciprocal excitation between these two nuclei ( Figure 6—figure supplement 1 ) . Then , we used a blade ( ~200 μm thick , ~1 . 3 mm wide ) to make a mechanical lesion between the VTA and the SNc , which physically disconnected the two regions ( Figure 6a , b ) . Such disruption has differential effects on dopamine release in different parts of striatum . In the NAc , cutting off VTA-SNc connections lowered the dopamine release in response to VTA-activation , and substantially reduced the response under SNc-stimulation ( Figure 6c ) , suggesting that SNc is involved in mesolimbic signalling via the reciprocal connection between VTA and SNc . Similarly , in the lower part of CPu , we observed a reduction of dopamine release under either VTA- or SNc-stimulation after the physical lesion ( Figure 6d ) , suggesting that VTA also partially affects nigrostriatal pathway through the mutual innervation between VTA and SNc . However , in the upper part of CPu , we found that dissociation of VTA and SNc had almost no effects on either VTA- or SNc-stimulation induced response ( Figure 6e ) . This observation indicated that the VTA-SNc reciprocal connection is not involved in the nigrostriatal signalling from SNc to upper CPu . However , signals resulted from the activation of VTA can still reach upper CPu to induce neurochemical response ( Figure 6e ) , which is a sign for the existence of a cascade transmission circuit from VTA to upper CPu ( Pennartz et al . , 2009; Figure 6f ) . Taken together , the mapping of the electrochemical events by using the OECT-array experimentally demonstrated a cross-talk between the mesolimbic and nigrostriatal dopaminergic pathways , and also supported a heterogeneous projection from SNc to CPu , which is differentially affected by VTA-SNc reciprocal connections . Here , we describe an OECT-array technique for real-time and multi-site monitoring of electrochemical release of CA-NTs in brains of living animals . The device is fabricated from an organic semiconductive PEDOT:PSS layer ( Mo and Ogorevc , 2001 ) , which is integrated with traditional electronic components on a flexible membrane , and configured into a highly sensitive transistor-based array for biosensing applications . As we have demonstrated in this study , the OECT-array is an independent recording system , which does not require a reference electrode as in the traditional CV ( Roberts and Sombers , 2018; Robinson et al . , 2003 ) . The dimensions of each OECT-unit , especially the GATE electrode size , and whole OECT-array were carefully designed based on the stereotaxic coordinates , which covers the striatal brain region and maximizes the recorded NT release signal . Such simple configuration enables each OECT-unit to act as a fully functional sensor , and leads to a greatly easier implantation process and a more convenient recording protocol . Compared to the passive electrodes ( e . g . , carbon fiber ) that are widely used in CV for in vivo electrochemical characterization , the OECT-array is an active sensor with intrinsic amplification capability during signal recording , which is a key feature to eliminate the noise accumulated from surrounding environment over the amplification process ( Rivnay et al . , 2017 ) . Unlike many CV-based measurements , in which signal is usually overwhelmed by noise due to the high-speed voltage scanning , the typical signal-to-noise ratio ( SNR ) of the OECT-device ranges from 5 to 10 ( depending on different recording sites in a brain ) , given an in vivo peak-to-peak noise level of 1 . 16 ± 1 . 52 nA ( n = 26 ) . This feature enables real-time and direct data readout without the need of afterwards background subtraction or complex mathematical fitting as required in traditional CV experiments ( Bucher and Wightman , 2015 ) . As another key advantage , the working voltage of the OECT-array is only ~300 mV , which is almost 80% reduced from the required voltage in CV measurements . Such a low operating voltage causes lower power consumption and also induces less tissue electrolysis during in vivo recording ( Bikson et al . , 2009 ) , which is critical for long-term recordings . As a result , the OECT-array can be used to continuously monitor the electrochemical activity in a brain for hours , which is inaccessible for carbon-fiber measurement that normally continues for minutes because of significant signal drift ( Roberts and Sombers , 2018 ) . Even compared to the genetic approach that was recently reported to measure dopamine release in vivo ( Patriarchi et al . , 2018 ) , the OECT-array is still advantageous in its quantification capability , ease-of-use , and flexibility for detecting different NT species , and great potential for interference-free behavioural experiments . The OECT sensor achieves a detection limit of 30 nM for dopamine in ACSF ( containing 1 . 28 mM ascorbate ) , demonstrating the capability for dopamine detection at low concentration with the existence of high level interferences . Considering that the concentration of ascorbate in vivo is homeostatic while evoked dopamine signals perform fast phasic change , we believe that such high level of electrochemical active interferences does not affect the detection of dopamine in vivo ( Phillips et al . , 2003; Garris et al . , 1999 ) . In this study , the sensor was used to detect dopamine release in brain regions with dominant dopaminergic signalling upon neural stimulation . The Pt-GATE electrode shows a strong electro-oxidation reaction with released dopamine , which can distinguish dopamine from other interference NTs in the brain environment ( e . g . , glutamate and GABA; Figure 2c , also see Figure 2—figure supplement 2 ) . Though 3 , 4-dihydroxyphenylacetic acid ( DOPAC ) has catechol that can be electro-oxidized on Pt , its conversion from dopamine requires monoamine oxidase , and operates in the tonic manner ( on minute timescale ) , which would not affect the sensing of phasic dopamine overflow ( on millisecond timescale ) using the OECT-array ( Phillips et al . , 2003 ) . With further development , the selectivity of the OECT device could be improved by different strategies . For example , surface of the Pt-GATE electrode can be modified with a layer of biocompatible polymer , Nafion ( Liao et al . , 2015 ) , which is a polyelectrolyte with stable Telfon backbone and acidic sulfonic groups , and is negatively charged in neutral solutions . Also , the selectivity could be customized through a multilayer ( Nafion/PANI/enzyme/graphene ) modification of the Pt GATE ( Liao et al . , 2015 ) . A particular NT could be selectively detected by choosing the coating enzyme ( e . g . , dopamine β-hydroxylase ) within the multilayer sandwich . Some of these strategies have been demonstrated in vitro ( Liao et al . , 2015 ) , and can be further optimized for in vivo applications . While OECT-based device has been previously used to detect neuronal action potentials ( Khodagholy et al . , 2013 ) , this study is first to implement an electrochemical sensor for in vivo applications . The differentiation between evoked electrical activity and neurotransmitter release is determined by the configuration of the OECT device . The performance and nature of an OECT are dominated by the ratio of the capacitance of the GATE/electrolyte ( CG-E ) and that of the PEDOT:PSS channel layer ( Cvolumetric ) ( Figure 1c ) , which is reversely proportional to the ratio of potential drop at the two interfaces on a OECT device ( Yaghmazadeh et al . , 2011 ) . For device used to record electrical activity , the GATE/electrolyte capacitance is larger than the capacitance of channel , leading to a larger potential at the channel layer when the total voltage applied across GATE/electrolyte/channel is maintained constant . Change of ion concentration upon neuronal spiking activity influences the amount of cations injected into the channel ( driven by the GATE voltage ) and affects the conductivity through doping and de-doping the channel , which favors the detection of transient electrical signals ( Khodagholy et al . , 2013 ) . For electrochemical sensing , the OECT-device is more complicated because of a trade-off between chemical detection and effective gating of the transistor . When faradic current is generated at the GATE/electrolyte interface as a result of electrochemical reactions , the GATE/electrolyte potential decreases , the channel/electrolyte potential increases , and the channel is de-doped . Subsequently , the channel conductivity is lowered and the current flowing through channel is decreased . In principle , chemical detection on the Pt-GATE demands a smaller GATE/electrolyte capacitance ( compared to the channel/electrolyte capacitance ) to enhance detection sensitivity , whereas effective gating requires vice versa . However , the GATE electrode could not be miniaturized too much , in order to maintain certain potential drop on the channel and to avoid a weak GATE modulation of the device . It is therefore important to tune the capacitance ratio more carefully in this case . The capacitance of Pt-GATE electrode for detection of catecholamine is an electrical double layer ( EDL ) , which is estimated to be around 4 . 8 nF ( 10 μF/cm2 , 0 . 08 × 0 . 06 cm2 ) , considering the EDL of Pt per unit area and the electrode size ( Drüschler et al . , 2010 ) . By tuning the size of the channel , the volumetric capacitance of PEDOT:PSS channel in electrolyte is estimated to be around 0 . 47 nF ( 39 F/cm3 , 0 . 03 × 10 × 40 μm3 ) , which is effective for gating to amplify signal ( Rivnay et al . , 2015 ) , and the oxidation of catecholamine on the GATE electrode is maximized while gating is kept effective , so as to configure the OECT as chemical sensor in favor of electrochemical detection . Using the OECT-based technology , we demonstrated successful detection of evoked dopamine release in different physiological scenarios . Notably , the microarrayed format of the device further enabled a spatial mapping of dopamine release simultaneously at multiple sites over a large brain region . Taking the dopaminergic pathways as a working model , our results from the OECT-array facilitated electrochemical mapping reveal a complex cross-talk between the mesolimbic and nigrostriatal signalling , which relies on a reciprocal innervation between VTA and SNc . These two nuclei are frequently treated as a whole in many studies , though their cell compositions are dramatically different ( Garris and Wightman , 1994 ) . Actually , the VTA/SNc mutual connections has also been shown by other studies . For example , CLARITY with highly cell-specific tracing revealed that SNc-neurons projecting to dorsomedial and dorsolateral striatum receive a small portion of neural input from VTA ( Lerner et al . , 2015 ) . However , in this study , we cannot rule out the possibility that other indirect network or multi-synaptic transmission ( Watabe-Uchida et al . , 2012 ) contributed to results primarily made by the electrical stimulation and OECT recording . In addition , it has recently been reported that NAc not only receives axonal projection from VTA but also projects back to VTA as well as to SNc ( Belin and Everitt , 2008 ) . Therefore , it is not surprising to see that activation of VTA can induce significant dopamine release in different parts of CPu even after a mechanical lesion of the VTA-SNc connection , probably through an indirect circuit involving the feedback projections ( Figure 6f ) . Yet , it is intriguing to observe a different electrochemical response in the upper and lower part of CPu as a result of the lesion , supporting the existence of distinct functional divisions within SNc , which could be heterogeneously associated with VTA via reciprocal innervations . These observations are well in line with the recent report about differential circuit incorporation of subpopulations of SNc dopamine neurons that project to distinct regions of dorsal striatum ( Lerner et al . , 2015; Keiflin and Janak , 2015 ) , and hint the possibility of their specific links between particular subcategory of dopamine neurons in VTA ( Bourdy et al . , 2014; Yetnikoff et al . , 2014 ) , which can be further explored with an array of circuit interrogation tools in more details . In summary , this proof-of-concept study demonstrates the development and application of the OECT-array as a powerful and easy-to-use platform for electrochemical analysis in the nervous system . From an instrumentation perspective , the OECT-array can be highly configurable and expandable for different experimental scenarios . For example , a similar OECT-based device can be forged into a flexible planar format for mapping electrocorticography signals over a large brain surface ( e . g . , cortex ) ( Khodagholy et al . , 2013; Lee et al . , 2017 ) ; more sensing units can be easily integrated to achieve higher spatial resolution; and different organic semiconducting materials can also be tested for the development of more specific and sensitive transistor-based electrochemical sensors ( Rivnay et al . , 2018; Venkatraman et al . , 2018; Pappa et al . , 2018 ) . We believe that the OECT-array can be a readily applicable platform to supplement existing toolsets for brain circuit interrogation . Combined with other neuro-techniques , such as electrophysiology , optogenetics , viral tracing , etc . , it can potentially lead to a wide range of novel neuroscience research that fully utilizes the electrical , optical , and biochemical information from a living animal brain . The fabrication of the OECT-array mainly involves sequential deposition and patterning of the gold ( Au ) /chromium ( Cr ) SOURCE and DRAIN electrodes , the Pt-GATE electrode , the PEDOT:PSS ( Clevios PH-500 , Heraeus ) channel active layer , and the SU-8 ( MicroChem ) insulating layer ( Wang et al . , 2018 ) . First , the 200 μm thick polyethylene terephthalate ( PET ) substrates were cleaned by ultrosonication in chemical solvent ( acetone and isopropanol; Sigma-Aldrich ) and oxygen plasma ( Harrick Plasma ) treatment . Then , the Au ( 100 nm ) /Cr ( 10 nm ) electrodes were deposited on the PET substrate by magnetron sputtering and patterned through a lift-off process . Specifically , a layer of AZ5214 photoresist ( MicroChem ) was spin-coated on the PET substrate , patterned by UV exposure , and then developed using AZ 400K developer ( MicroChem ) . The Au/Cr electrodes was deposited and patterned by a lift-off process , forming the SOURCE and the DRAIN electrode , which were spaced by a channel of 10 μm in length and 40 μm in width . Following the same procedure , a layer of Pt ( 100 nm ) /titanium ( 10 nm ) was deposited to form the Pt-GATE measured at 600 × 800 μm2 . The PEDOT:PSS aqueous dispersion was first mixed with dimethyl sulfoxide ( DMSO; 5% in volume ratio; Sigma-Aldrich ) , glycerin ( 5% in volume ratio; Sigma-Aldrich ) and ( 3-glycidyloxypropyl ) trimethoxysilane ( GOPS; 1% in volume ratio; International Laboratory ) to enhance the conductivity and film stability . Then the PEDOT:PSS was spin-coated and annealed at 150°C in nitrogen for 1 hr . Unwanted PEDOT:PSS was removed along with the photoresist by rinsing in acetone ( Sigma-Aldrich ) . The device was packaged by patterning a layer of photoresist ( 2 μm; SU-8 2002 ) to insulate the metal electrodes from aqueous electrolyte . The devices were cleaned in PBS solution before use . In an electrochemical analysis using the OECT-array , multiple dual-channel sourcemeters ( 2612/2614 , Keithley ) were used . The OECT-units were electrically separated from each other . To acquire the transfer curve for each OECT-unit , the VDS was maintained constant , while VGS was swept from 0 to 1 . 2 V for ex vivo ( or in vivo ) analysis . During signal recording , the applied VGS was determined by referencing to the transfer curve to maximize the transconductance ( gm ) for the best signal amplification . For ex vivo measurements , the neurotransmitter molecules ( e . g . , dopamine , noradrenaline and adrenaline; Sigma-Aldrich ) with designed concentrations were added to PBS solution to mimic a pulsed release , and the fluctuation of IDS as a function of time was recorded . For in vivo measurements , the evoked IDS fluctuation was recorded upon electrical stimulation of the neuronal circuits in the brain of an anesthetized animal . All data was collected using the software TSP-express ( Keithley ) . All experimental procedures involving animals were approved by the university Animal Ethics Committee . Animal licenses , ( 16-97 ) in DH/HA and P/8/2/5 Pt . 5 and ( 18-129 ) in DH/SHS/8/2/5 Pt . 4 , were approved by Department of Health of the Government of Hong Kong Special Administration Region . Both male and female Sprague Dawley rats ( 300 ~ 400 g ) were used . Before the experiments , the animals were firstly anesthetized with intraperitoneal injection of urethane ( 2 g/kg; Sigma-Aldrich ) , and then mounted on a stereotactic frame ( Narishige ) . A heating pad was placed underneath the animals to maintain a temperature of 37°C throughout an experiment . For monitoring somatodendritic dopamine release in VTA ( Figure 3a ) , two holes ( 2 × 2 mm2 ) were drilled on the skull . The stereotactic coordinates for stimulation in MFB was −1 . 8 mm A/P , 2 mm M/L , 8 mm below dura . The stereotactic coordinates for the 1st OECT unit was −5 . 6 mm A/P , 0 . 8 mm M/L , 8 mm below dura . The control experiments were conducted in the same way except that the stimulation is moved out of MFB ( −1 . 8 mm A/P , 2 mm M/L , 3 . 5 mm below dura ) . For monitoring dopamine release remotely along the mesolimbic pathway ( Figure 4a ) , the tungsten electrode was implanted in the VTA ( −5 . 6 mm A/P , 1 mm M/L , 7 . 9 ~ 8 . 4 mm below dura ) for electrical stimulation , and the OECT-array was implanted in the NAc ( 1st OECT unit; 1 . 2 mm A/P , 1 . 4 mm M/L , 8 . 4 mm below dura ) for electrochemical monitoring . In the control experiments , the OECT-array was not moved , and the electrical stimulation was delivered to an irrelevant region ( −5 . 6 mm A/P , 1 mm M/L , 4 . 9 mm D/V below dura ) . For investigations about the cross-talk between the mesolimbic and the nigrostriatal pathway ( Figure 5a ) , a tungsten electrode ( FHC ) was either implanted at VTA ( −5 . 3 mm A/P , 0 . 8 mm M/L , 7 . 9 ~ 8 . 4 mm below dura ) or SNc ( −5 . 3 mm A/P , 2 mm M/L , 7 . 5 ~ 7 . 9 mm below dura ) for electrical stimulation of dopaminergic signalling pathway . In the control experiments , the stimulation was delivered to a region out of VTA or SNc ( −5 . 3 mm A/P , 0 . 8 mm M/L , 4 . 9 mm below dura ) . The OECT-array was inserted to an exposed rat brain ( 1 . 2 mm A/P , 3 mm M/L ) with 14o tilted towards the lateral side , and was allowed to travel 7 . 2 ~ 7 . 4 mm . In this configuration , the 1st unit was placed in NAc and the coordinates were 1 . 2 mm A/P , 1 . 4 mm M/L , 7 . 1 mm below dura; the 3rd unit was placed lower CPu and the coordinates were 1 . 2 mm A/P , 1 . 9 mm M/L , 4 . 7 mm below dura; the 4th unit was placed in upper CPu and the coordinates are 1 . 2 mm A/P , 2 . 2 mm M/L , 3 . 5 mm below dura . After an experiment , the animal was sacrificed and transcardially perfused with chilled PBS and then 4% Paraformaldehyde ( PFA; in PBS ) . The brain was isolated and stored in 4% PFA ( in PBS ) for later anatomical or immunohistochemical analysis . To evoke action potentials in brain tissue , a tungsten metal electrode ( ~200 μm in diameter ) was used to deliver current stimulation to appropriate brain locations . To evoke somatodendritic dopamine release ( Figure 3 ) , the electrical stimuluses ( 50 Hz , 2 ms pulse width , 200 μA amplitude; AMPI ) of different duration ( 1 , 10 , 30 , 50 , and 100 pulses ) were delivered in MFB . To detect the NT release at axonal projection terminals ( Figures 4–6 ) , a pulse sequence ( 50 Hz , 2 ms pulse width , 1 s duration , 200 μA amplitude ) was used . The interval between each stimulation is at least 50 s to avoid NT depletion . An isolated brain was fixed in 4% PFA for 2 days followed by soaking in 30% sucrose until it settled to the bottom . The sample was then frozen-sectioned ( Thermo Scientific ) into coronal slices of 25 or 50 μm thickness for further processing . Before immunostaining , the brain slices were rehydrated in PBS for 2 hr and blocked by 3% bovine serum albumin ( BSA ) in tris-buffered saline with 0 . 25% Triton X-100 ( TBST ) for 2 hr . Next , the slides were incubated with primary antibodies at 4°C overnight followed by a thorough rinse in TBST solution . The slices were further incubated with secondary antibodies at room temperature for 2 hr and afterward DAPI solution ( 10 μg/ml; Abcam ) for 30 min . After rinsing in TBST , the slides were dehydrated in 30% ethanol , 50% ethanol , 75% ethanol , 100% ethanol and 100% xylene sequentially , and mounted for further imaging and storage . Specifically for the horseradish peroxidase/3 , 3'-Diaminobenzidine ( HRP/DAB ) staining , the sectioned brain slices were rehydrated in PBS for 2 hr and were blocked by hydrogen peroxide blocking solution ( Abcam ) for 30 min and by 3% BSA in TBST for 2 hr . Afterwards , the slices were incubated in primary antibody at 4°C overnight . After a thorough wash in TBST solution for 10 min for three times , the slices were further incubated in biotinylated secondary antibody ( Abcam ) for 30 min followed by TBST wash for 10 min for three times . Next , the peroxide-labeled streptavidin solution ( Abcam ) was applied for 30 min and subsequently washed in TBST for 10 min for three times . The DAB chromogen ( Abcam ) was diluted in DAB substrate ( Abcam ) to working concentration , and the diluted chromogen solution was applied directly to the slides . At satisfactory staining level , the slides were rinsed in deionized water to stop the development . The slices were dehydrated and mounted in the same way aforementioned for imaging and storage . The antibodies used in this study are: rabbit anti-tyrosine hydroxylase ( 0 . 3 μg/ml; ab112 , Abcam ) and sheep anti-c-Fos antibody ( 2 μg/ml; ab6167 , Abcam ) . The fluorescently stained brain slides were imaged using a laser scanning confocal microscope equipped with a 40 × water immersion lens ( SP8 , Leica ) . For each slice , the scan was performed with 1 μm z-resolution , and a maximum projection of each scan was then acquired . For overall anatomical evaluation , the HRP/DAB-stained brain slides were imaged using a stereoscope . A MATLAB program was developed to convert the recorded change of IDS to ΔVg-eff by using the transfer curve to quantitatively analyze the level of recorded NT release as recorded by individual OECT-unit . Specifically , the IDS before electrical stimulation was converted to voltage value according to the transfer curved ( as baseline ) ; the peak IDS ( the minimal IDS value evoked by an electrical stimulation before reversing back to baseline ) in response to NT release was also converted to voltage value in the same manner; these two voltage values were further subtracted and absolutized to obtain ΔVg-eff . The calculated ΔVg-eff was further converted to change of molecular concentration by using a calibration curve fitted from the ex vivo experimental results . For statistical analysis , student t-test was performed to determine the statistical significance between the experimental conditions and the control groups , p<0 . 05 indicates a significant difference . At least three independent biological replicates were used if not otherwise specified . For Figure 3 , data from ~20 trials were collected from each animal . For Figure 4 , ~30 trials were performed on each animal . For Figure 5 , the data were collected from 10 animals , and ~30 trials were performed on each animal . For Figure 6 , the data were collected from six animals , and ~30 trials were performed on each animal .

Cells in the nervous system pass messages using a combination of electrical and chemical signals . When an electrical impulse reaches the end of one cell , it triggers the release of chemicals called neurotransmitters , which pass the message along . Neurotransmitters can be either activating or inhibitory , determining whether the next cell fires its own electrical signal or remains silent . Currently , researchers lack effective methods for measuring neurotransmitters directly . Instead , methods mainly focus on electrical recordings , which can only tell when cells are active . One new approach is to use miniature devices called organic electrochemical transistors . Transistors are common circuit board components that can switch or amplify electrical signals . Organic electrochemical transistors combine these standard components with a semi-conductive material and a flexible membrane . When they interact with certain biological molecules , they release electrons , inducing a voltage . This allows organic electrochemical transistors to detect and measure neurotransmitter release . So far , the technology has been shown to work in tissue isolated from a brain , but no-one has used it to detect neurotransmitters inside a living brain . Xie , Wang et al . now present a new device that can detect the release of the neurotransmitter , dopamine , in real-time in living rats . The device is a miniature microarray of transistors fixed to a blade-shaped film . Xie , Wang et al . implanted this device into the brain of an anaesthetised rat and then stimulated nearby brain cells using an electrode . The device was able to detect the release of the neurotransmitter dopamine , despite there being a range of chemicals released inside the brain . It was sensitive to tiny amounts of the neurotransmitter and could distinguish bursts that were only milliseconds apart . Finally , Xie , Wang et al . also implanted the array across two connected brain areas to show that it was possible to watch different brain regions at the same time . This is the first time that transistor arrays have measured neurotransmitter release in a living brain . The new device works at low voltage , so can track brain cell activity for hours , opening the way for brand new neuroscience experiments . In the future , adaptations could extend the technology even further . More sensors could give higher resolution results , different materials could detect different neurotransmitters , and larger arrays could map larger brain areas .

2020

Organic electrochemical transistor arrays for real-time mapping of evoked neurotransmitter release in vivo

Natural signaling circuits could be rewired to reprogram cells with pre-determined procedures . However , it is difficult to link cellular signals at will . Here , we describe signal-connectors—a series of RNA devices—that connect one signal to another signal at the translational level . We use them to either repress or enhance the translation of target genes in response to signals . Application of these devices allows us to construct various logic gates and to incorporate feedback loops into gene networks . They have also been used to rewire a native signaling pathway and even to create novel pathways . Furthermore , logical AND gates based on these devices and integration of multiple signals have been used successfully for identification and redirection of the state of cancer cells . Eventually , the malignant phenotypes of cancers have been reversed by rewiring the oncogenic signaling from promoting to suppressing tumorigenesis . We provide a novel platform for redirecting cellular information . A basic ability of living cells is to sense extracellular signals by translating them into changes in regulation of cell signaling genes . They use the natural signaling network to execute complex physiological functions , such as cell survival , behavior and identity . As an interdisciplinary branch of biology , genetic engineering has developed rapidly during recent years with the objective of reconstituting the signaling network of the cell for therapeutic and biotechnological applications . Genetic devices have been used to construct novel signaling circuits such as genetic switches ( Gardner et al . , 2000; Green et al . , 2014 ) , digital logic circuits ( Moon et al . , 2012; Ausländer et al . , 2012; Siuti et al . , 2013 ) , rewired signaling pathways ( Kiel et al . , 2010; Yuan et al . , 2012; Flock et al . , 2014 ) and feedback loops ( Stricker et al . , 2008; Prindle et al . , 2014 ) . It proved easy to construct networks between synthetic genes using standardized building blocks ( Zhang and Jiang , 2010; Schreiber et al . , 2016 ) . By controlling the gene expression , there are both positive and negative gene connections . The key difference between them is that in positive connection , the regulated gene is activated for expression , while in negative connection , the regulated gene is silenced . The native intracellular communication can be rewired using genetic devices that block or redirect signals , but connecting native input-output signals at will remains a challenge . For example , the previously developed trans-acting ligand-responsive RNA regulators ( Bayer and Smolke , 2005; Win and Smolke , 2007; Ausländer et al . , 2010; Beisel et al . , 2011; Chang et al . , 2012 ) can be used to engineer novel connections by inhibiting native gene expression in response to extracellular molecules . The conversion of signals into specific cellular events has been accomplished via inducible or repressible antisense RNAs or miRNAs . However , they can only build negative gene connections and the antisense or RNAi-based regulation often exhibits relative low efficiency . Our group has developed a newly-engineered class of genetically encoded devices–‘CRISPR signal conductors' ( Liu et al . , 2016 ) –that can sense and respond to cellular signals of interest and in turn activate/repress transcription of specific endogenous genes through a CRISPR interference or activation mechanism . The advantage of these devices is to construct both negative and positive connections between various selected biomolecules and it is only limited by the availability of functional RNA aptamers . However , they require an additional transgene encoding a large protein ( dCas9 ) which further increases the complexity of the system . From an application point of view , the use of a compact RNA-based device is likely to be much more compatible with the limitations of transgene delivery technology than the use of a rather large protein-coding construct . Regretfully , except for miRNAs and siRNAs , no other RNA-based mechanism has been adapted as a wide-spread tool for controlling native gene expression . In this work , we describe the multiple uses of ‘RNA-based signal connectors’ in mammalian cells to modulate translation of mRNAs transcribed from the native genome and from provided plasmids . This new technology acts at the translational stage , apparently promoting or suppressing recruitment of ribosomes to the target mRNA . Without the requirement for an exogenous protein , these small artificial RNAs can establish both negative and positive linkages between input and output signals at will . The work described here is an improvement on the past design of ‘signal conductors’ and shows that they can be used in multiple different applications . Previous studies have demonstrated that insertion of an RNA aptamer into the 5ʹ untranslated region ( 5ʹ-UTR ) of the messenger RNA ( mRNA ) can reduce the rate of translation initiation through blocking ribosome scanning in the presence of ligand ( Werstuck and Green , 1998; Blount and Breaker , 2006 ) . We hypothesized that a designed external complementary sequence can be used to hybridize to the target mRNA and to guide RNA aptamers for trans-regulation of cellular mRNA translation when a specific signal is present . To test this , we engineered RNA devices which use the antisense domain ( a 20 nt antisense RNA ) to recognize the mRNA of interest and the previously developed aptamer domain to control translation ( Figure 1A ) . We used the signal-connector to tether a translational activation domain to enhance translation , or the aptamer domain alone to repress the translation . In principle , these modular devices which we called ‘signal-connectors’ , can be designed to control the translation of any target mRNA in response to a signal-molecule of interest ( Figure 1B ) and thus link desired endogenous signals ( input ) to specific cellular signals ( output ) ( Figure 1C ) . To test whether this approach could cause efficient repression of translation initiation ( Figure 2A ) and elongation ( Figure 2B ) , we designed signal-connectors complementary to 12 different regions of the mRNA sequence of the Renilla luciferase reporter gene , either binding to the 5ʹ-UTR or to the coding sequence ( Figure 2C and Supplementary file 1 ) . Each of these signal-connectors contained two segments: a 20 nt antisense RNA sequence designed to be complementary to the targeted mRNA sequence and two theophylline aptamer copies ( Jenison et al . , 1994 ) . Of these 12 constructs , 11 induced significant decreases in Renilla luciferase expression in the presence of 1000 µM theophylline when they were stably transfected into HEK293 cells expressing Renilla luciferase ( Figure 2D ) . The levels of luciferase activity did not change substantially in cells harboring signal-connectors grown in the absence of theophylline . Nuclear and cytoplasmic fractionation analysis showed that these constructs mainly located in the cytoplasm ( Figure 2—figure supplement 1 ) . In addition , the repression activity seemed to be inversely correlated with the target distance from the 5ʹ cap of mRNA , perhaps indicating that the expression of a target gene could be repressed more effectively at the early stage of translation . We then observed that addition of theophylline inhibited luciferase activity in a dose-dependent fashion ( Figure 2E ) . We speculated that the effects of the signal-connectors should also be affected by the valency of ligands recruited to each mRNA target . To test this possibility , we then introduced one or three theophylline aptamers to the 3ʹ end of signal connectors , constructed stably transfected HEK293 cells , detected the luciferase expression level and compared their effects with those of the 2 × signal connectors ( Figure 2—figure supplement 2 ) . Devices with three aptamers produced repression effects on luciferase expression that were little stronger relative to the analogous 2 × devices , perhaps due to saturation effects . Devices containing only one aptamer , however , only induced a very weak reduction in reporter gene expression . We also performed an in vitro translation reaction using the macromolecular components ( ribosomes , tRNAs , aminoacyl-tRNA synthetases , initiation , elongation and termination factors ) , purified ligand ( theophylline ) , in vitro transcribed mRNA of Renilla luciferase , as well as in vitro transcribed RNA ‘signal-connector’ ( R1 used in Figure 2C ) or the negative controls . The in vitro data suggest that the observed silencing effects for signal-connectors were indeed induced by the ligand-aptamer complex ( Figure 2—figure supplement 3 ) , indicating a roadblock mechanism . A simple mathematical model was then used to better understand the relationship between the various input parameters and the output ( Figure 2—figure supplement 4 ) . The equation described a dose- or concentration-effect relationship and a maximum effect , which are the key features of many biological phenomena . Based on the predictions of this equation and our observed results , we carried out gene knockdown experiments using designed 2 × signal connectors in the presence of sufficient amounts of ligand . To demonstrate the modularity of this approach , we constructed several other signal-connectors to target the Renilla luciferase or the human vascular endothelial growth factor ( VEGF ) gene and replaced the theophylline aptamers with tetracycline aptamers ( Müller et al . , 2006 ) ( Supplementary file 2 and Supplementary file 3 ) . The data on cells stably expressing these devices supported the modularity of the signal-connector to different aptamer domains ( Figure 2F and Figure 2—figure supplement 5 ) . As expected , these devices ( R13 ~R15 and R17 ~R20 ) showed efficient silencing effects only in the presence of 100 µM tetracycline . Relative levels of VEGF mRNA did not change obviously between cells harboring the signal-connectors grown in the absence or presence of tetracycline , indicating that the signal-connectors function through translational inhibition rather than by affecting mRNA levels ( Figure 2—figure supplement 6 ) . These results demonstrated that the signal-connectors could be used as gene switches to down-regulate the expression of a target gene . Using the RNA aptamers ( Miyakawa et al . , 2006 ) for eukaryotic translation initiation factor 4G ( eIF4G ) , we then determined whether signal-connectors could also enhance translation of a target gene by promoting the formation of initiation factor complexes ( Figure 2G and Supplementary file 4 ) . eIF4G recruits the ribosome 40S subunit and activates mRNA translation ( Moore , 2005 ) . We chose the Renilla luciferase gene as the target gene and the results of luciferase reporter assay indicated that the specific signal-connector with two eIF4G aptamers induced a 15-fold increase in activity of luciferase protein relative to controls when they were stably transfected into HEK293 cells ( Figure 2H ) . Elimination of one aptamer copy from the construct dramatically decreased the induced activation efficiency , whereas the fold change value increased minimally with the addition of another copy of aptamer ( Figure 2—figure supplement 7 ) . Nuclear and cytoplasmic fractionation analysis also showed that this construct mainly located in the cytoplasm ( Figure 2—figure supplement 1 ) . We also performed an in vitro translation reaction using the macromolecular components ( ribosomes , tRNAs , aminoacyl-tRNA synthetases , initiation , elongation and termination factors except for eIF4G ) , the purified eIF4G protein , in vitro transcribed uncapped mRNA of Renilla luciferase ( the ORF encoding Rluc was placed downstream of a primary ORF ) , as well as in vitro transcribed RNA ‘signal-connector’ or the negative controls . The data suggest that the observed activating effects for signal-connectors were indeed induced by the eIF4G-aptamer complex ( Figure 2—figure supplement 8 ) , indicating a recruitment mechanism . We also constructed a simple mathematical model to clarify the relationship between the various input parameters and the output and the equation revealed that the relationship is nonlinear and saturable ( Figure 2—figure supplement 9 ) . Based on the predictions of the new equation for gene activation and the observed results , we carried out gene activation experiments with designed 2× signal connectors in the presence of sufficient amounts of ligand . To achieve dynamic regulation of translation initiation , we used a combination of one aptamer recognizing theophylline or tetracycline and two aptamers recognizing eIF4G to regulate gene expression ( Supplementary file 5 and Supplementary file 6 ) , in which the antisense domain was designed to be complementary to the stem sequence of the theophylline ( or tetracycline ) -binding aptamer ( Figure 2G ) . Theophylline or tetracycline binding stabilizes the aptamer and leads to a conformational change that allows the antisense domain to interfere with the mRNA of the target gene . The results of the luciferase reporter assay on HEK293 cells stably expressing these devices indicated that addition of theophylline or tetracycline increased activity of luciferase ( Figure 2I ) . We also observed a dose-dependent effect ( Figure 2—figure supplement 10 ) . These results demonstrated that the signal-connectors could be used as gene switches to up-regulate the expression of a target gene . In the construction of electronic circuits , logical calculations and digital systems can be practically implemented by using logic gates , including NOT , AND , NAND , OR , NOR , XOR and XNOR gates . Many aspects of information processing by biological cells are similar to signal integration of electronic circuits . We then asked the question whether the signal-connectors could be used to construct complex programmable logic gates and circuits . The excellent gene regulatory ability of the signal-connectors inspired us to construct various logic gates that produced output signals in response to multiple input signals through stably transfecting these devices ( Figure 3A ) . We built all the basic types of two-input Boolean logic gates in HEK293 cells stably expressing the 5’ capped or uncapped Renilla luciferase mRNA construct by using the aptamer recognizing exogenous theophylline or tetracycline signal . In the construct expressing uncapped Renilla luciferase mRNA , an open-reading frame ( ORF ) encoding Renilla luciferase was placed downstream of a primary ORF . The primary ORF contained a stop codon at the end . First , we constructed two NOT gates , each of which produced an inverted version of the input at its output . As suggested in Figure 2D and F , the location of the antisense RNA target sequence along the mRNA was important for inhibition efficiency of the signal-connector . We used two devices ( R1 and R13 ) that maximally suppressed translation of 5ʹ capped Renilla luciferase mRNA in the presence of 1000 µM theophylline or 100 µM tetracycline to build these gates . As shown in Figure 3B , each NOT gate exhibited high luciferase output only in the absence of input signal . We then constructed an AND gate that produced high output only if both input signals were high . As shown in Figure 2—figure supplement 10 and Figure 2I , activation was also inversely correlated with the target distance from the 5ʹ end of the mRNA . We used two signal-connectors ( R25 and R29 ) that minimally activated translation of 5ʹ uncapped Renilla luciferase mRNA in the presence of ligand to build this gate . As shown in Figure 3C , although introduction of each individual signal ( 1000 µM theophylline or 100 µM tetracycline ) did not significantly stimulate expression of the target luciferase gene , the two devices acted synergistically to induce robust translational activation in the presence of both the input signals . We also constructed a NAND gate that exhibited high output if any of the inputs were low . We used two signal-connectors ( R12 and R16 ) that minimally suppressed translation of 5ʹ capped Renilla luciferase mRNA in the presence of ligand to build this gate . As shown in Figure 3D , although introduction of either individual signal ( 1 , 000 µM theophylline or 100 µM tetracycline ) did not significantly inhibit expression of the target luciferase gene , the two devices acted synergistically to induce robust translational repression in the presence of both input signals . Next , we constructed an OR gate that gave a high output if one or both of its inputs were high . In the building of this gate , we also used two signal-connectors ( R22 and R26 ) , each of which strongly activated translation of 5ʹ uncapped Renilla luciferase mRNA in the presence of ligand . As shown in Figure 3E , the luciferase could be produced by either of the two signals ( 1000 µM theophylline and 100 µM tetracycline ) . For our next test we constructed a NOR gate which was equivalent to an OR gate followed by a NOT gate . We used two devices ( R1 and R13 ) that maximally inhibited translation of 5ʹ capped Renilla luciferase mRNA in the presence of ligand to build this gate . Since introduction of individual signal significantly suppressed expression of the target gene , the luciferase could be produced only when both of the two signals ( 1000 µM theophylline and 100 µM tetracycline ) were absent ( Figure 3F ) . We also constructed an XOR gate that exhibited high output if either , but not both , of its two inputs were high . We designed signal-connectors ( R22 and R27 ) to target two different regions of the 5ʹ uncapped Renilla luciferase mRNA , which were complementary in their RNA sequence . The results showed that each one of the devices strongly activated expression of luciferase in the presence of corresponding ligand . In contrast , introduction of both devices did not significantly activate expression of the target luciferase gene due to the specific base pairing between their antisense domains ( Figure 3G ) . Finally , we constructed an XNOR gate that exhibited a low output if either , but not both , of its two inputs were high . Using similar design strategies , we used two devices ( R1 and R14 ) that strongly repressed translation of 5ʹ capped Renilla luciferase mRNA in the presence of corresponding ligand . The XNOR gate could have a high luciferase output when both 1000 µM theophylline and 100 µM tetracycline were present or absent ( Figure 3H ) . These results indicated that the signal-connectors could logically link input signals to a desired cellular output signal . In eukaryotic cells , signaling proteins often activate transcription factors to initiate transcription of downstream genes . Because in theory the signal-connectors can link transcription factors to suppression of downstream gene translation , we set out to develop modifiers of a molecular network to rewire the native signaling pathway ( Figure 4A ) . β-catenin is a multifunctional protein and usually accumulates in the nucleus of cancer cells , where it activates the transcription of the oncogenic c-Myc gene ( He et al . , 1998 ) . We synthesized a signal-connector containing β-catenin aptamers ( Culler et al . , 2010 ) to target the region within the 5ʹ-UTR of c-Myc mRNA ( Supplementary file 7 ) . We investigated the effect of stimulating the β-catenin pathway with leukotriene D4 ( LTD4 ) on the HEK-293 cells stably expressing either the signal-connector or the negative control . Both cell lines exhibited increased expression of c-Myc mRNA ( Figure 4B ) , whereas the cells stably expressing the signal-connector showed a strong decrease in expression of c-Myc protein compared with the cells transfected with negative control ( Figure 4C and Figure 4—figure supplement 1 ) . These results demonstrated that our signal-connector could effectively rewire the signaling pathway by establishing a negative connection between the transcription factor and the mRNA of a downstream gene . We also tested whether the signal-connector could create a novel signaling pathway by linking a regulatory factor to the activation of translation of a selected downstream gene ( Figure 4D ) . The specific signal-connector used one aptamer domain to recognize β-catenin signal and the other two aptamer domains to form the initiation factor complexes . In the absence of β-catenin signal , the antisense domain was sequestered by the stem of the β-catenin aptamer . In the presence of β-catenin signal , this signal-connector could interact with the target Renilla luciferase mRNA ( Figure 4E and Supplementary file 8 ) . The effect of leukotriene D4 ( LTD4 ) was investigated by stably transfecting HEK-293 cells with either the signal-connector or the negative control . The results of luciferase assay indicated that the activity of Renilla luciferase in cells expressing the signal-connector was obviously elevated by LTD treatment ( Figure 4F ) , while its activity was not affected by LTD in cells expressing the negative control . These results demonstrated that β-catenin signal could effectively activate the expression of Renilla luciferase with the help of the designed signal-connector . Next , we tested the ability of the signal-connectors to incorporate feedback loops into the gene–gene interaction networks . Positive feedback loops can amplify the cellular signal received from the sender and move a system away from its initial state . We engineered a feedback loop using the signal-connectors in which osteopontin ( OPN ) and VEGF were each other’s activators ( Figure 4G ) . OPN and VEGF are secreted proteins with cytokine properties and regulate cell motility and angiogenesis ( Ferrara et al . , 2003; Lyle et al . , 2014 ) and their RNA aptamers were reported in previous literatures ( Ng et al . , 2006; Mi et al . , 2009 ) . We inserted two copies of eIF4G aptamer into the 3’end of OPN or VEGF riboswitch to construct signal-connector recognizing VEGF or OPN . The results of our over-expression experiments revealed that OPN and VEGF were mutually independently operated in bladder cancer T24 cells stably transfected with the negative control device . We transfected the plasmids over-expressing OPN or VEGF into the T24 cells expressing the signal-connectors ( Supplementary file 9 , 10 and 11 ) and found that expression of the corresponding plasmid effectively increased the level of the regulated gene ( Figure 4G ) . We also investigated whether the signal-connectors could be used to construct negative feedback loops between OPN and VEGF ( Figure 4H ) , which could make the system more stable . Because OPN could induce the expression of VEGF via the signal-connector , we only needed to prove that VEGF could also inhibit the expression of OPN through a similar approach . We inserted two copies of VEGF aptamer into the 3’end of antisense RNA recognizing OPN to construct the signal-connector . Using T24 cells stably expressing this constructed signal-connector , we showed that transient expression of VEGF could decrease the concentration of OPN and that knockdown of VEGF increased the level of OPN again ( Figure 4H ) . These results demonstrated that the signal-connectors were effective tools for constructing regulatory loops and gene–gene networks . To examine whether these devices could be used to identify cell state and to reprogram cellular behavior , we used the human telomerase reverse transcriptase ( hTERT ) promoter to drive the expression of ribozyme-flanked signal-connectors ( Gao and Zhao , 2014 ) that silence survival genes , and chose bladder cancer cells as the target cells ( Figure 5A and Figure 5—figure supplement 1 ) . The hTERT promoter ( hTERTp ) is highly active in over 85% of human cancers , but inactive in most normal cells ( Takakura et al . , 1999 ) . We therefore constructed device-ligand complexes to form a logical AND gate in which the activated hTERTp and the ligand must be combined to suppress the survival gene ( Figure 5B ) . Signal-connectors suppressing the human c-Myc gene ( Sardi et al . , 1998 ) and the BCL2 gene ( Kunze et al . , 2012 ) were generated as before and stably transfected into either bladder cancer cells or normal dermal fibroblasts ( Figure 5C , Supplementary file 12 and 13 ) . In either bladder cancer cell line , the corresponding device was able to display significant decreases in gene expression in the presence of theophylline compared with that in the absence of ligand ( Figure 5D and Figure 5—figure supplement 2 ) . The devices did not lead to inhibitory effects in fibroblasts grown in the absence or presence of theophylline ( Figure 5E and Figure 5—figure supplement 3 ) . The growth curves of these cell lines also demonstrated that the circuit effectively inhibited proliferation of targeted bladder cancer cells without affecting the fibroblasts ( Figure 5F ) . In addition , we then examined whether apoptosis of cancer cells can be induced by these devices . Bladder cancer cells treated with the signal connectors exhibited stronger blue fluorescence , revealing typical apoptotic characteristics . In contrast , the signal connectors had no such effects in the normal cells ( Figure 5G ) . These results indicated that the AND gate circuit based on the signal-connectors could specifically suppress gene expression in the targeted cell lines . The successful application of signal-connector-mediated translational control in human cells opens the way toward a simultaneous ON/OFF multigene translational program in which some genes are activated and others are suppressed . We hypothesized that these devices should have the potential to redirect oncogenic pathway outputs and to control cancer cell fates through simultaneous activation and repression of endogenous genes . NF-kB is an oncogenic signal that is known to be involved in the signaling pathways in cancer development . NF-kB controls cell proliferation by activating several downstream target genes such as cyclin D1 , c-Fos and c-Jun ( Li et al . , 2015 ) . We therefore sought to rewire NF-kB signaling from proliferation pathways to quiescence/death by using the signal-connector . We constructed four signal-connectors ( Supplementary file 14 , 15 , 16 and 17 ) recognizing NF-kB ( p65 ) ( ursterWurster and Maher , 2008 ) to activate two tumor suppressors , Bax ( R37 ) and p21 ( R38 ) , and to repress two tumor promoters , Bcl2 ( R39 ) and c-myc ( R40 ) , in human bladder cancer T24 cells which normally expressed high levels of NF-kB signals ( Figure 6A ) . In detail , two copies of eIF4G aptamer were inserted into the 3’end of NF-kB riboswitch to construct signal-connector activating Bax or p21 , while two copies of NF-kB aptamer were linked with the 3’end of antisense RNA to construct the signal-connector suppressing Bcl2 or c-myc . The results of western blotting showed that the signal-connectors stably transfected in T24 cells could simultaneously enhanced the protein expression levels of Bax and p21 and decreased Bcl2 and c-myc ( Figure 6B and Figure 6—figure supplement 1 ) . Finally , we determined whether the signal-connectors could simultaneously inhibit tumor growth in vivo , since T24 cells stably transfected with these devices showed a slower in vitro growth rate ( Figure 6—figure supplement 2 ) . The cells were then inoculated into male nude mice . Twenty days after injection , we found that the tumors formed in the signal-connectors group were dramatically smaller than those in the negative control group ( Figure 6C–E ) . In addition , the average tumor weight was markedly lower in the signal-connectors group compared to the negative control group at the end of the experiment ( Figure 6F ) . These results indicate that signal-connectors could inhibit tumor growth in vivo by redirecting oncogenic signaling pathways . In this study , we used antisense RNA to target the desired mRNA and the aptamer domain-ligand complex to repress translation of the target gene . Although it has been reported that an aptamer inserted in the 5ʹ-UTR-mRNA can effectively repress translation of a synthetic gene ( expressed in an exogenous vector ) through the road-block mechanism , this is the first work to propose that aptamers could also repress endogenous genes of interest in a trans manner by linking with the antisense RNAs . It should also be noted that some natural non-coding RNAs ( Loh et al . , 2009; Mellin et al . , 2013; Price et al . , 2015; Xu et al . , 2015 ) were found to inhibit their target gene through a similar mechanism , which suggests that the method used in this study is a universal approach for cellular RNAs to interact with genes of interest . More importantly , we showed that the aptamer tethering translation initiation factor could be used to enhance endogenous gene translation . In a similar case , this new mechanism was also proposed by one previous work which reported that the uchl1 gene lncRNA enhances the translation of its target mRNA via base-pairing and by recruiting additional ribosomes via the functional element ( Carrieri et al . , 2012 ) . Therefore , this approach allows simultaneous activation and repression of different target genes , thus enabling robust reprogramming of cellular networks . It is interesting that the signal-connectors mainly located in the cytoplasm . One possible explanation for this phenomenon is that the device binds the mRNA in the nucleus and remains bound to the mRNA during and after export . This observation is consistent with an earlier study which indicated that nuclear-localized sgRNA targeting mRNA can also be exported to the cytoplasm ( Nelles et al . , 2016 ) . In the applications of this methodology , we used two signal connectors targeting luciferase gene , but containing aptamers to different ligands , to construct circuits which are somewhat similar to computer logic gates . We also applied this methodology to couple unrelated signaling pathways and to connect oncogenic signals with an antioncogenic pathway . The signal connectors may provide an alternative approach to traditional cancer gene therapy which usually targets only one single gene . In the construction of activator devices ( with eIF4G ) , some basic thermodynamic/kinetics parameters , such as binding affinity and kinetics behaviors , still require quantitative studies in future works . The potential impact of the secondary structures in the targeted region of mRNA may also not be ignored . It seems that other small RNA regulators that are enzymatically amplified ( such as shRNAs and miRNAs ) exhibit much lower efficiencies than what we report here when expressed from the U6 promoter . We will compare efficacy of this new approach with that of the existing , simpler treatment method in future works . Although conceptually simple , these devices may be included in the biology toolbox that allows for construction of novel signaling circuits and regulatory loops with predetermined properties and may enable development of strategies for treating disease networks . With regard to research applications of this methodology , one clear limitation is the fact that functional aptamers still usually need to be selected from random libraries . With the discovery and development of more and more aptamers , any signal not associated with gene regulation can be directed to inhibit/enhance translation of the targeted gene through the designed signal-connector which is very important to the signal transmission of parts of the gene circuits . Our results showed that these devices could be used to create genetic switches , logic gates , novel signaling and feedback loops , which led to practical applications such as detection of cancer cell state and inhibition of cancer cell survival . Signal-connectors that rewire multiple oncogenic signaling networks may provide an effective network-based strategy to increase the efficiency of current cancer treatment . In addition , mammalian cells harboring digital logic gates can function as living bio-computers and open new avenues for artificial control of future gene- or cell-based therapies in a specific condition-dependent manner . Our novel technique will provide a useful platform for editing the common network structures and their signaling processes and will bring many applications in biology and medicine . We first analyzed the sequences of well-known RNA aptamers in mammalian cells , such as theophylline aptamer ( Jenison et al . , 1994 ) , tetracycline aptamer ( Müller et al . , 2006 ) , eIF4G aptamer ( Miyakawa et al . , 2006 ) , β-catenin aptamer ( Culler et al . , 2010 ) , VEGF aptamer ( Ng et al . , 2006 ) , OPN aptamer ( Mi et al . , 2009 ) and NF-kB ( p65 ) aptamer ( urster et al . , 2008 ) . Then , we truncated and coupled them to the mRNA base pairing regions ( antisense domains ) . Each mRNA base pairing region was perfectly complementary with the 5’ –UTR or the coding region of the target mRNA . Next , the secondary structures of these recombinant RNAs were predicted by MFOLD program . The RNAs which showed exposed antisense domains and maintained the natural secondary structures of aptamers were selected and used in this study . The cDNA sequences for signal-connectors targeting Renilla luciferase/c-Myc/OPN/VEGF/ BCL2/Bax/p21 mRNA were synthesized and inserted into pGPU6/GFP/Puro vector at restriction site of Bam HI/Bbs I , respectively . Using similar approach , the cDNA sequences for ribozyme-flanked signal-connectors targeting c-Myc/BCL2 mRNA were designed , synthesized , and inserted into hTERT-NEO-BAM vector at the restriction site of Sal I/BamH I , respectively . To construct plasmids pcDNA3 . 0-VEGF and pcDNA3 . 0-OPN , cDNA sequences expressing truncated forms of OPN/VEGF that lack the N-terminal signal peptide were inserted into pcDNA3 . 0 digested with BamHI/EcoRI , respectively . To construct plasmids shRNA-NC and shRNA-VEGF , the synthesized shRNA sequences were inserted into pGPU6/GFP/Neo digested with Bam HI/Bbs I , respectively . T24 , 5637 , and HEK-293 cells were purchased from American Type Culture Collection ( ATCC ) by our laboratory and were grown in DMEM medium supplemented with 10% foetal bovine serum ( Invitrogen , Carlsbad , CA ) in the presence of 5% CO2 . Normal human primary fibroblasts derived from the epidermis were primary cultured in the same medium . T24 , 5637 , and HEK-293 cells have been previously authenticated by ATCC with STR profiling and no further authentication was done for these studies . Stable cell lines we re generated from these cell lines as described below . All cell lines used were validated as mycoplasma-free . HEK-293 cells stably expressing Renilla luciferase were obtained by transfecting cells with pcDNA3/Rluc/Neo and selecting positive clones with G418 . In details , stable selections were carried out in 6-well plates seeded with ~2×105 HEK-293 cells per well , where 2 µg of the linearized plasmids were transfected using Lipofectamine 2000 Transfection Reagent ( Invitrogen ) according to the manufacturer's instructions . Cell monolayers were trypsinized 48 hr after transfection and transferred into T25 flasks or 100-mm-diameter culture dishes . A mixed population of stable transfectants was selected by growth in complete medium containing 500 μg of G418/ml . These multiclonal cell lines were expanded and then verified by luciferase reporter gene assay . HEK-293 cells stably expressing either the signal-connector or the negative control were selected after transfection of the pGPU6/GFP/Puro vectors . HEK 293/T24 cells co-expressing multiple signal-connectors were constructed by stably transfecting a single pGPU6/GFP/ Puro vector which simultaneously generated these devices driven by a single U6 promoter . In details , cells were seeded in six-well plates and 2 µg of the linearized plasmids were transfected using Lipofectamine 2000 Transfection Reagent ( Invitrogen ) according to the manufacturer's instructions . After transfection , the cells were grown in the medium supplemented with puromycin at 4 μg/mL for approximately 14 days to select for a mixed population of stable cell lines . Then the multiclonal cells were verified by GFP expression . An inverted fluorescence microscope was used for direct observation of fluorescent cells in the culture plate . HEK-293 cells stably expressing the Renilla luciferase reporter system were seeded in six-well plates ( 5 × 105/ well ) . Forty-eight hours after transfection , the medium was removed and cells were lysed in 500 μl of lysate buffer ( Analytical Luminescence Laboratories ) . Renilla luciferase activity was measured by the Renilla Luciferase Reporter Assay System ( Promega , Madison , WI ) according to manufacturer’s instructions . Renilla luciferase activities were corrected for variation in protein concentrations of the cell extracts ( Bio-Rad ) . The assays were performed in duplicate and the experiments were repeated three times . Nuclear and cytoplasmic RNA were isolated using the Cytoplasmic and Nuclear RNA Purification Kit ( Norgen , Belmont , CA ) according to the provider’s instructions . Purified ligand ( theophylline or eIF4G ) was incubated with 1 µg signal connector and 1 µg Renilla luciferase mRNA and then the mixture was further incubated with the components of the Thermo Scientific 1-Step Human Coupled IVT Kit ( Waltham , MA , USA ) according to the manufacturer’s instructions . The activity of in vitro translated Renilla luciferase was calculated as described above . HEK293 cells were stably transfected with signal-connectors or the control . The concentration of VEGFA/OPN protein was then measured by ELISA assay , which was employed according to the manufacturer's instructions . Briefly , 106/sample cells were harvested and resuspended in 200 μl of lysis buffer . The supernatants of lysates were collected through centrifugation and used for the following procedures . The OD values were then measured by a microplate reader ( Bio-Rad , Hercules , CA ) and converted to protein concentrations using standard calibration curves . Cells were washed in PBS and lysed in RIPA buffer ( 50 mM Tris-HCl pH 7 . 2 , 150 mM NaCl , 1% NP40 , 0 . 1% SDS , 0 . 5% DOC , 1 mM PMSF , 25 mM MgCl2 , and supplemented with a phosphatase inhibitor cocktail ) . The protein concentration was determined using the BCA protein assay . Equal amounts of whole protein extract were electrophoresed onto SDS–polyacrylamide gels and then transferred to PVDF membranes ( Millipore , Billerica , MA ) . Samples were blocked in 5% dry milk and incubated over-night with the primary antibodies ( Abcam , Cambridge , MA ) . Then , the samples was incubated with horseradish peroxidase–conjugated secondary antibody ( Amersham , Piscataway , NJ ) and immunoblots were developed with Super Signal chemiluminescence reagents ( Pierce Chemical Co . ) . The protein bands were quantified using Image J analysis software ( National Institutes of Health , USA ) . Histograms were generated by normalizing the amount of each protein to the GAPDH level detected in the same extracted sample . Each experiment was repeated three times . Cell numbers were calculated by treating the cells with 0 . 25% trypsin ( 15 min , 37°C ) , followed by analysis on an electronic cell counter ( Beckman Coulter ) at 0 , 24 , 48 and 72 hr . The assay was repeated at least three times independently . The Hoechst 33258 staining kit ( Life , Eugene , OR ) was used to observe the apoptotic cells induced by signal-connectors . Briefly , the treated cells were fixed in 4% paraformaldehyde for 10 min and washed twice in PBS . Then , the cells were stained with 0 . 5 ml of Hoechst 33258 staining for 5 min and photos were taken under a fluorescence microscope at a wavelength of 350 nm . Each assay was repeated three times . Total RNA was isolated from cells by using TRIzol ( Invitrogen , Carlsbad , CA ) according to the suggested protocol . The cDNA strand was synthesized from total RNA with RevertAidTM First Strand cDNA Synthesis Kit ( Fermentas , Hanover , MD ) in a 25 μl volume . Real time quantitative PCR was performed with the All-in-OneTM qPCR Mix ( GeneCopoiea Inc , Rockville , MD ) in a 20 μl reaction volume on an ABI PRISM 7000 Fluorescent Quantitative PCR System ( Applied Biosystems , Foster City , CA ) . The PCR cycling parameters were: 95°C for 15 min , followed by 40 cycles of 94°C for 15 s , 55°C for 30 s and 72°C for 30 s . Relative expression fold changes were determined by the 2-ΔΔCt method . All experiments involving animals were approved by Institutional Review Board . Four- week-old female BALB/c nude mice were obtained from Animals Center of the Academy of Sciences . In detail , 107 T24 cells stably expressing signal-connector or negative control were suspended in 100 μl PBS and injected subcutaneously into left or right armpits of three 4-week-old female BALB/c nude mice . Tumor growth was examined every 5 days , and tumor volumes were also calculated using the formula: 0 . 5 × length × width2 . 20 days after injection , mice were euthanized , and the subcutaneous weight of each tumor was measured . No statistical methods were used to pre-determine sample size . The investigators were blinded to allocation during experiments and outcome assessment . Statistical analysis was conducted using Student’s t-test or ANOVA and p<0 . 05 was considered statistically significant . All statistical tests were performed by using SPSS version 17 . 0 software ( SPSS , Chicago , IL ) .

Cells respond to signals from their surrounding environment . External signals activate a sequence of events inside the cell that can change how it behaves . These events are often called signaling pathways and they typically change the cell’s behavior by changing the activity of its genes . A major objective of the field of genetic engineering is to customize or artificially create new signaling pathways to make cells behave in certain ways . The ability to control a cell’s behavior is likely to have a major impact on human health and medicine . For instance , it may be possible to reprogram signaling events in cancer cells so that they die rather than grow rapidly . Researchers are developing artificial genetic devices to manipulate signaling pathways . Molecules of ribonucleic acid ( or RNA ) are widely used to design such devices . In nature , RNA molecules are highly versatile: messenger RNA molecules carry genetic information in a form that can be translated into protein , while other RNA molecules fine-tune gene expression and perform a host of other roles . RNA is apt for artificial devices because it can be tailored to detect signals and convert this information into a predictable outcome , such as turning specific genes on or off . In 2016 , researchers constructed an RNA device to control the expression of genes in response to particular signals . However , this device was too large to deliver efficiently inside cells . Now , Liu , Li , Chen et al . – including some of the researchers involved the 2016 study – design smaller RNA devices to overcome this limitation . Each new device consists of two RNA components: one that recognizes the signal , and another that recognizes the messenger RNA of a target gene . Together the two components trigger the desired change in gene expression in response to a specific signal . The devices were shown to have multiple uses such as making new connections in a signaling pathway and creating new signaling networks . Furthermore , Liu , Li , Chen et al . engineered one device such that it was able to specifically turn off genes in a particular signaling pathway that allows human bladder cancer cells to divide . By silencing these genes , the cancer cells were less able to grow . These newly developed RNA devices should allow other researchers to customize cellular information and may have future therapeutic applications as well .

2018

Synthesizing artificial devices that redirect cellular information at will

Selective relationships are fundamental to humans and many other animals , but relationships between mates , family members , or peers may be mediated differently . We examined connections between social reward and social selectivity , aggression , and oxytocin receptor signaling pathways in rodents that naturally form enduring , selective relationships with mates and peers ( monogamous prairie voles ) or peers ( group-living meadow voles ) . Female prairie and meadow voles worked harder to access familiar versus unfamiliar individuals , regardless of sex , and huddled extensively with familiar subjects . Male prairie voles displayed strongly selective huddling preferences for familiar animals , but only worked harder to repeatedly access females versus males , with no difference in effort by familiarity . This reveals a striking sex difference in pathways underlying social monogamy and demonstrates a fundamental disconnect between motivation and social selectivity in males—a distinction not detected by the partner preference test . Meadow voles exhibited social preferences but low social motivation , consistent with tolerance rather than reward supporting social groups in this species . Natural variation in oxytocin receptor binding predicted individual variation in prosocial and aggressive behaviors . These results provide a basis for understanding species , sex , and individual differences in the mechanisms underlying the role of social reward in social preference . The brain regions and neurochemicals involved in social behaviors show remarkable conservation across species ( O’Connell and Hofmann , 2011 ) . At the same time , social behavior is not a unified construct , with different species exhibiting distinct social structures and behavioral repertoires . The formation of selective social relationships is a particular hallmark of both human and prairie vole societies . Such relationships are difficult to study in traditional lab rodents because mice , rats , and other rodents typically do not form preferences for known peers or mates ( Triana-Del Rio et al . , 2015; Schweinfurth et al . , 2017; Beery et al . , 2018; Cymerblit-Sabba et al . , 2020; Insel et al . , 2020; Beery and Shambaugh , 2021 ) . In species that form specific relationships , selectivity may be based on reward and prosocial motivation toward specific individuals , or on avoidance ( fear , aggression ) of unfamiliar individuals . The role of social motivation and tolerance may also differ by familiarity , sex , and type of relationship ( e . g . same-sex peer versus opposite-sex mate ) . Voles provide an opportunity to probe the role of selectivity and social reward across relationship types and social organization . The reinforcing properties of social interaction have been demonstrated in a variety of rodent species and contexts , often through conditioned place preference for a socially associated environmental cue ( e . g . Panksepp and Lahvis , 2007; Dölen et al . , 2013; Goodwin et al . , 2019 ) . Operant conditioning for access to a social stimulus has been used to more directly measure motivation for specific types of social interaction , particularly access to pups , social play , and sexual opportunities ( reviewed in Trezza et al . , 2011 ) . Social motivation has also been assessed with access to novel same-sex peers ( Martin and Iceberg , 2015; Achterberg et al . , 2016; Borland et al . , 2017 ) . Often social interactions are affiliative , but in some contexts animals will work for access to aggressive interactions ( Azrin et al . , 1965; Falkner et al . , 2016; Golden et al . , 2017 ) . To date , only one study has examined the role of familiarity in social motivation , in novelty-preferring female rats ( Hackenberg et al . , 2021 ) , and none have done so with mate relationships . Prairie voles , Microtus ochrogaster , and meadow voles , Microtus pennsylvanicus , both form selective social relationships but exhibit different social organization and mating systems . Prairie voles are socially monogamous , forming long-term selective relationships between males and females that have been studied for decades ( Carter et al . , 1995; Walum and Young , 2018 ) . Prairie voles also form selective relationships with familiar same-sex cage-mate ‘peers’ ( DeVries et al . , 1997; Beery et al . , 2018; Lee et al . , 2019 ) . Meadow voles are promiscuous breeders that transition to living in social groups and sharing nests during winter ( Getz , 1972; Madison and Mcshea , 1987 ) . Under conditions of short daylength in the laboratory , female ( but not male ) meadow voles exhibit greater social huddling and less aggression than their long daylength counterparts ( Beery et al . , 2008b; Lee et al . , 2019 ) . These vole species thus allow comparison of the properties of peer relationships across species ( prairie vole peers versus meadow vole peers ) and relationship type within species ( prairie vole mates versus prairie vole peers ) . Prairie voles exhibit socially conditioned place preferences ( sCPP ) for familiar opposite-sex mates ( Ulloa et al . , 2018; Goodwin et al . , 2019 ) , and in some circumstances for same-sex peers ( Lee and Beery , 2021 ) . In contrast , meadow voles do not form sCPP and may even condition away from social cues ( Goodwin et al . , 2019 ) . Neurochemical pathways underlying social reward also vary between species and relationship type; dopamine is necessary for the formation of opposite-sex pair bonds in prairie voles ( Aragona and Wang , 2009 ) , but is not necessary for the formation of same-sex peer preferences in meadow or prairie voles ( Beery and Zucker , 2010; Lee and Beery , 2021 ) . These initial findings suggest that social selectivity may result from differential social motivation and tolerance in these species . Voles demonstrate striking preferences for familiar versus novel peers and mates , assessed using the partner preference test ( Williams et al . , 1992b; Beery , 2021 ) . This test quantifies preference , but as no effort is required to access a conspecific , it cannot distinguish between prosocial motivation and avoidance of unfamiliar conspecifics . To examine the role of motivation in relationships , we assessed effort expended by voles of different sexes ( male , female ) , relationship types ( same-sex , opposite-sex ) , and species ( prairie vole , meadow vole ) to reach social targets in an operant conditioning paradigm . Because the seasonal transition from solitary to social is most pronounced in female meadow voles in the field and laboratory ( Madison and Mcshea , 1987; Beery et al . , 2009 ) , only females of this species were used . Subjects underwent >60 active training and testing days ( Figure 1 ) . Responses ( lever presses ) in lightly food-restricted voles were shaped and reinforced using a food reward , followed by 8 days of pressing for a food reward on a progressive ratio 1 ( PR-1 ) schedule . Social testing consisted of 8 consecutive test days in which each reward consisted of 1 min of access to the familiar ( same- or opposite-sex ) partner , and 8 test days for which rewards consisted of access to different sex-matched strangers ( order balanced within groups ) . We assessed effort expended to access familiar and novel social stimuli in four groups of prairie voles ( Figure 1 ) : females lever pressing for a female conspecific ( F➤F ) , females pressing for a male conspecific ( F➤M ) , males pressing for a male conspecific ( M➤M ) , and males pressing for a female conspecific ( M➤F ) . Meadow vole females ( F➤F ) were also trained and tested for 8 days of familiar and 8 days of novel vole exposure , counterbalanced . A subset of voles was used to explore the reward value of an empty chamber , extinguishing timelines , and relationships between oxytocin receptor ( OTR ) density and behavior . Oxytocin is involved in social recognition as well as in preference for familiar individuals ( reviewed in Anacker and Beery , 2013 ) , and in many instances , oxytocin signaling alters the rewarding properties of social stimuli ( Dölen et al . , 2013; Borland et al . , 2018 ) . We conducted receptor autoradiography to assess variation in neural OTR density in female prairie voles . ( OTR was not analyzed in male brains; following early results , later males were used to pilot a two-choice social operant paradigm . ) Together these studies allowed us to examine how the reward value of social contact differs between male and female prairie voles , between opposite-sex and same-sex pairings , and between meadow and prairie vole FF pairings . We found both similarities in and striking differences between social motivation across species , sexes , and pairing types . Detailed examination of social behaviors during social access further underscored the distinction between social motivation and familiarity preference , especially in males . In addition to these group differences in social motivation , individual differences in OTR density were related to aggressive and prosocial behaviors . In order to assess motivation for different kinds of social stimuli across groups , lever pressing responses were quantified on a progressive ratio schedule ( PR-1 ) . Males and females showed qualitatively different response patterns in the social chambers , as well as significant interaction between sex and variables of interest in a model screening for sex differences ( sex*stimulus type ( p = 0 . 01 ) , sex*stimulus familiarity ( p = 0 . 09 ) ) , so responses were further analyzed separately by sex ( Beery , 2018; Beltz et al . , 2019 ) . For each sex , two-way repeated measures ANOVA ( RM-ANOVA ) was performed with familiarity of the tethered stimulus ( partner/stranger ) as the within-subjects/repeated measure , and sex of the tethered stimulus ( opposite-sex/same-sex ) as a between-subjects measure . Female prairie voles pressed more for familiar partners than unfamiliar strangers , with no effect of opposite-sex versus same-sex pairings ( Figure 2A , effect of stimulus familiarity: F ( 1 , 14 ) = 15 . 17 , p = 0 . 0016 , ηp2p20 . 52; no effect of stimulus sex: F ( 1 , 14 ) = 0 . 44 , p = 0 . 51 , ηp2p20 . 03; subject matching: F ( 14 , 14 ) = 4 . 2 , p = 0 . 0057 , ηp2p20 . 81 , no significant interaction ) . Paired t-tests were used for within-group comparisons of responses for the partner or stranger: familiarity preferences were significant in females paired with males ( t ( 7 ) = –2 . 7 , p = 0 . 03 , d = 0 . 96 ) as well as in females paired with females ( t ( 7 ) = –4 . 1 , p = 0 . 0048 , d = 1 . 43 ) . The mapping from response count to the corresponding PR-1 breakpoint ( i . e . the maximum number of responses exhibited to achieve a reward ) is shown in Figure 2A and applies to all response count figures . Male prairie voles pressed at a higher rate for opposite-sex social stimuli regardless of familiarity ( effect of stimulus sex: F ( 1 , 14 ) = 17 . 4 , p = 0 . 0009 , ηp2p20 . 71; no effect of familiarity , F ( 1 , 14 ) = 0 . 013 , p = 0 . 91 , ηp2p20 . 00 , no significant effects of subject matching or interaction ) . Because each vole was tested in eight consecutive sessions of each type , familiarity preference could also be assessed within individuals across days . Significant within-vole familiarity preferences were present in more female pressers ( 6/8 F➤M and 3/8 F➤F ) than males ( 1/8 M➤F and 0/8 M➤M pairs ) ( Figure 2—figure supplement 1; p = 0 . 0059 Fisher’s exact test ) . One male in a M➤F pair exhibited a significant preference for stranger females ( Figure 2—figure supplement 1 ) , and mounted/copulated with strangers in multiple test sessions . In female prairie voles , the familiarity preference for both mates and peers in lever pressing was mirrored in cohabitation time and huddling . Even when these behaviors were scaled relative to lever presses ( and thus access time ) , females spent a significantly higher fraction of the available time in the social chamber ( time in social chamber/access time ) when it was occupied by a familiar vole rather than a novel one ( effect of familiarity F ( 1 , 14 ) =95 . 06 , p < 0 . 0001 , ηp2p20 . 87; subject matching F ( 14 , 14 ) = 2 . 789 , p = 0 . 03 , ηp2p20 . 73; others NS; two-way RM-ANOVA , Figure 3A ) . Females also spent more of the available time huddling ( time spent in immobile side-by-side contact/access time ) with familiar rather than unfamiliar conspecifics of either sex ( effect of familiarity: F ( 1 , 14 ) = 25 . 82 , p = 0 . 0002 , ηp2p20 . 65; others NS; two-way RM-ANOVA , Figure 3C ) . Within-group matched comparisons of time spent with a partner or stranger also revealed that females exhibited significant familiarity preferences in time spent in the social chamber or huddling with the stimulus animal relative to time with access ( time in social chamber/access time: FF: p < 0 . 0001 , d = 3 . 51; FM: p = 0 . 0006 , d = 2 . 12; time huddling/access time: FF: p = 0 . 0090 , d = 1 . 26; FM: p = 0 . 0083 , d = 1 . 29; paired t-tests ) . In contrast , while males exhibited no familiarity preferences in lever pressing responses , they still exhibited strong familiarity preferences in social interaction . Males spent more of the available time in the social chamber when the tethered stimulus was familiar ( effect of familiarity F ( 1 , 14 ) = 6 . 33 , p = 0 . 02 , ηp2p20 . 31; subject matching F ( 14 , 14 ) = 4 . 459 , p = 0 . 0042 , ηp2p20 . 24; others NS; two-way RM-ANOVA , Figure 3B ) , and huddling behavior was even more specific , with a strong effect of stimulus familiarity ( partner versus stranger ) and no effect of stimulus sex ( opposite- versus same-sex ) on the percent of [time huddling]/[time with access to the social chamber] ( effect of familiarity: F ( 1 , 14 ) = 25 . 27 , p = 0 . 0002 , ηp2p20 . 64; all else NS; Figure 3D ) . Within-group matched comparisons also revealed significant familiarity preferences in huddling time relative to access ( huddling/access time: MM: p = 0 . 0177 , d = 1 . 09 , MF p = 0 . 0022 , 1 . 66 ) , with lesser or no familiarity preference in chamber time ( time in social chamber/access time: MM: p = 0 . 0390 , d = 0 . 90 , MF: p = 0 . 56 , d = 0 . 21; paired t-tests ) . There was no apparent sex difference in huddling behavior between male and female prairie voles , confirmed by pooling males and females in a three-way ANOVA ( effect of focal sex NS , p = 0 . 91; significant effect of stimulus familiarity ( F ( 1 , 56 ) = 48 . 03 , p < 0 . 0001 , ηp2p20 . 46 ) ; effect of stimulus sex; NS , no significant interactions ) . Aggressive behavior was exhibited by prairie voles in all groups during social operant sessions and was analyzed by RM-ANOVA on all voles tested with partners and strangers ( between-subjects factors: sex of presser ( M/F ) *pairing type [same/opposite sex]; within-subjects factor: target familiarity ) . Both males and females engaged in far more bouts of aggression with strangers than familiar partners ( F ( 1 , 29 ) = 30 . 22 , p < 0 . 0001 , ηp2p20 . 51 , Figure 3E and F ) . There was no significant effect of sex of the presser ( F ( 1 , 29 ) = 3 . 36 , p = 0 . 077 , ηp2p20 . 10 ) , pairing type ( same-sex or opposite-sex ) , or interactions between these variables . Because aggression was primarily targeted at strangers , we asked whether stranger aggression might be motivating: that is , whether aggression was associated with greater lever pressing for strangers . Correlation of daily stranger lever pressing with bouts of aggression was not significant across females ( R = 0 . 14 , p = 0 . 10 ) , but was significant across males ( R = 0 . 25 , p = 0 . 004 ) . Because more time with access to a stranger provides more opportunity for aggression to occur , aggressive bouts were also scaled relative to access time , as was done for the other social measures . Across groups there were no relationships between stranger-directed daily lever pressing and aggression/access time in either male or females ( males: R = 0 . 04 , p = 0 . 64; females: R = 0 . 12 , p = 0 . 16 , Figure 3G ) . Mounting behavior was present in five prairie voles , all of which were male prairie voles tested with novel ( unfamiliar ) female voles . This distribution was significantly non-random across the eight testing combinations used in prairie voles ( e . g . male with female partner , male with female stranger , etc . ) ( χ2 ( 7 ) = 37 . 97 , p < 0 . 0001 ) . These five voles exhibited an average of 6 bouts of mounting per testing session . OTR density was associated with both motivated and aggressive social behaviors in different brain regions in female prairie voles ( males not assayed ) . There was a strong positive correlation between OTR density and lever presses for same-sex partners in the nucleus accumbens ( NAcc ) core ( R = 0 . 959 , p = 0 . 0098 ) and shell ( R = 0 . 948 , p = 0 . 0141 , Figure 4A ) . There was also a strong positive correlation between mean bouts of stranger-directed aggression and OTR density in the bed nucleus of the stria terminalis ( BNST ) in female prairie voles ( R = 0 . 719 , p = 0 . 0126 ) , again connecting receptor binding to behavior . Binding density in the BNST was not associated with stranger approach or avoidance , operationalized as time spent in the stranger’s social chamber relative to access time ( R = 0 . 350 , p = 0 . 29 ) , or lever presses for the stranger’s chamber ( R = 0 . 264 , p = 0 . 43 ) . OTR density varied with housing condition . Females housed with same-sex cage-mates showed no difference in OTR density in the NAcc or lateral septum ( LS ) , higher OTR density in the BNST ( t ( 8 . 99 ) = 2 . 93 , p = 0 . 0167 , d = 1 . 78 ) , and a non-significant trend in the central amygdala ( t ( 8 . 71 ) = 1 . 92 , p = 0 . 0883 , d = 1 . 17 ) compared to females housed with opposite-sex cage-mates ( Figure 4C ) . Lever pressing responses in prairie voles were compared to those of a related non-monogamous vole species ( the meadow vole ) that exhibits group living during winter months . Female meadow voles are territorial and aggressive in summer or long daylengths in the lab , but socially tolerant in winter or short days . Because male meadow voles do not undergo this transition ( Madison and Mcshea , 1987; Beery et al . , 2009 ) , we focused on comparison of social motivation in female meadow voles relative to female prairie voles . Prior to making this comparison , we assessed whether species and sexes differed in their lever pressing effort in response to a common reward ( food ) . There were no sex or species differences in the number of lever pressing responses for a food reward ( PR-1 schedule; 8 days averaged per subject ) between female prairie voles , male prairie voles , and female meadow voles ( F ( 2 , 40 ) = 1 . 18 , p = 0 . 32 , η2 = 0 . 56; one-way ANOVA; Figure 5A ) . Food responses and social responses were converted to response rates for comparison across trials with different active lever pressing periods: individual response rates for a food reward did not predict response rates during social testing for either the partner ( p = 0 . 78 ) or the stranger ( p = 0 . 98 ) , indicating that responses were not subject-specific across reward types ( Figure 5B ) . These findings validate the specificity of comparisons across species , sexes , and reward types . Female meadow voles pressed significantly more for familiar females than novel females ( t ( 6 ) =3 . 637 , p = 0 . 0109 , d = 1 . 37 , paired t-test , Figure 5C; males not tested ) . This preference was individually significant within four of the seven meadow voles ( Figure 5—figure supplement 1 ) . Comparisons of time spent with a partner or stranger when the door was up also revealed significant familiarity preferences ( P versus S for social chamber/access time: p = 0 . 0351 , d = 1 . 02; P versus S for huddling/access time: p = 0 . 0357 , d = 1 . 02; paired t-tests ) . Despite familiarity preference , meadow vole response rate for both partners and strangers was low . Direct comparison with female prairie voles tested under the same conditions reveals that while both groups pressed more for familiar partners than for strangers , there was significantly less lever pressing in female meadow voles ( two-way ANOVA , effect of target familiarity: F ( 1 , 13 ) = 29 . 51 , p < 0 . 001 , ηp 2 = 0 . 69 , effect of species: F ( 1 , 13 ) = 9 . 71 , p < 0 . 01 , ηp2p20 . 43 , Figure 5C ) . Comparison of lever presses between social conditions and non-social ‘empty control’ conditions indicates that , for female meadow voles , the partner was not more rewarding than the empty chamber control , stranger pressing was significantly lower than empty control , and it was similar to the post-extinction level of pressing ( Figure 5D ) . Aggression was rare in meadow vole trials ( mean 0 . 3 bouts/trial ) , and as in our prior studies ( Lee et al . , 2019 ) it was significantly less frequent than aggression between female prairie voles ( mean 2 . 3 bouts/trial , species difference: t ( 3 . 83 ) , p = 0 . 001 ) . No mounting behavior was observed in meadow vole tests , all of which were conducted in female voles . At the conclusion of social testing , all voles from cohorts 4 to 7 were tested for effort expended to explore an empty chamber without a tethered partner or stranger for 8 days each ( n = 6 meadow females , 10 prairie females , and 14 prairie males ) . Voles were distributed across all housing types . There was no species difference in pressing for the empty chamber ( meadow vole female versus prairie vole female ) . In both male and female prairie voles , the extent of lever pressing for the control chamber was correlated with pressing for the stranger ( females: R = 0 . 75 , p = 0 . 013; males: R = 0 . 71 , p < 0 . 005 ) but not with lever pressing for the partner . The same cohorts were then tested for extinction of lever pressing over 10 days of trials in which the door was closed and the lever did not activate the motor . All groups extinguished lever pressing behavior within ~5 days of testing ( Figure 5 ) . Repeated measures analysis revealed a significant effect of day of testing on pressing ( F ( 9 , 21 ) = 3 . 72 , p = 0 . 0063 , ηp2p20 . 61 ) but no significant effect of the testing group on extinction ( F ( 2 , 29 ) = 0 . 76 , p = 0 . 48 , ηp2p20 . 05 ) . Strong relationships were present between OTR density , housing differences , and behavior , highlighting connections across levels of organization . Variation in OTR density by relationship type has not been previously assessed , although OTR density or mRNA levels differ in response to early-life housing manipulations in prairie voles , such as presence of a father and single versus group housing ( Prounis et al . , 2015 ) as well as chronic social isolation in adulthood ( Pournajafi-Nazarloo et al . , 2013 ) . Oxytocin signaling plays a role in diverse social behaviors in prairie voles , including pair bond formation , consolation behavior , and alloparental care ( Williams et al . , 1992a; Olazábal and Young , 2006; Bales et al . , 2007; Burkett et al . , 2016 ) . Furthermore , oxytocin signaling has been related to social reward in non-selective mice and hamsters ( Dölen et al . , 2013; Song et al . , 2016; Borland et al . , 2018 ) . Strong correlations between NAcc OTR and lever pressing for the partner in the present study provide additional support for the role of NAcc OTR in social reward . Neural OTR was related to aggressive behavior as well as prosocial behavior , underscoring the complexity of oxytocin signaling in different brain regions ( van Anders et al . , 2013; Beery , 2015 ) . Social pressing differed quantitatively but not qualitatively by species in meadow and prairie voles . Females of both species pressed more for partners than for strangers , but responses were lower in meadow voles , indicative of the lack of social reward . This is consistent with prior findings from sCPP tests , in which meadow voles did not condition toward a bedding associated with social contact , and in one setting conditioned away from it ( Goodwin et al . , 2019 ) . These findings are also in line with results from the sole prior study of operant responses in voles . Matthews et al . , 2013 , tested prairie voles and meadow voles housed in long daylengths to determine whether they would learn to lever press for stranger voles . Only prairie voles demonstrated clear learning in this scenario , consistent with low stranger interest in meadow voles housed in the long daylengths that promote territorial behavior in this species ( Beery et al . , 2008b ) . Nonetheless , even under pro-social short daylength conditions used in the present study , social pressing was low in meadow voles . Comparison of short daylength-housed female meadow vole responses for the partner chamber , stranger chamber , and an empty chamber in different trial blocks revealed equivalent levels of pressing for a partner or an empty chamber and less for the stranger . This suggests that decreased pressing for the stranger represents avoidance , but that pressing for the partner may indicate tolerance more than reward . Female ( short daylength-housed ) meadow voles also exhibited lower aggression than female prairie voles , consistent with social tolerance , and with prior descriptions of their behavior ( Lee et al . , 2019 ) . Lever pressing was demonstrated to be an effective metric to compare effort exerted to reach different social stimuli in voles; voles of each species and sex tested pressed at comparable rates for food reward , indicating a lack of major differences in task learning , and thus that social lever pressing can be assessed and compared across groups . Subject response rates were not consistently high or low across reward conditions , indicating that responses are reward-specific . Extinction was effective , with all subjects decreasing lever pressing behavior by more than half their baseline response count . Differences in lever pressing effort between groups could therefore be attributed to reward-specific differences in social motivation . Persistent relationships within specific pairs or groups of conspecifics are present throughout the animal kingdom , including species of invertebrates , fishes , amphibians , reptiles , birds , and mammals ( Bales et al . , 2021 ) . While the nature and extent of these relationships vary considerably , they share in common the specificity of social preferences that leads to repeated association . They may differ , however , in the mechanisms that influence familiar approach and unfamiliar avoidance . In particular , familiar individuals—whether mates or peers—may or may not be socially rewarding , and unfamiliar individuals may or may not be aversive . Even within closely related vole species , we see evidence that only some relationships involve selective social reward , for example , mate relationships in female prairie voles , while others—such as peer relationships in winter phenotype meadow voles—involve selectivity without appreciable reward . Selectivity in the absence of reward may rely instead on changing social anxiety and aggression ( Beery , 2019 ) . For example , when exposed to the short , winter photoperiods associated with the transition from solitary to group living in the wild , meadow voles undergo changes in CRF ( corticotropin-releasing factor ) receptor densities , glucocorticoid secretion , behavioral indicators of anxiety , and aggression ( Ossenkopp et al . , 2005; Beery et al . , 2014; Anacker et al . , 2016 ) . More research is needed to establish causal links between these changes and the transition to group living . More broadly , it remains to be determined to what extent social monogamy and pair bonding with mates shares mechanisms across species ( Goodson , 2013 ) , and to what extent different types of relationships ( e . g . with peers or mates ) share foundations , or differ in their regulation . Ultimately , these studies should help us understand how selective relationships of different types evolve . While other studies have assessed social reward in rodents , few have considered the role of stimulus familiarity , likely because laboratory rodents do not exhibit familiarity preferences under normal conditions ( reviewed in Beery and Shambaugh , 2021 ) . In social choice tests , mice and young rats often prefer social novelty ( Moy et al . , 2004; Smith et al . , 2015 ) , and relative preference for a social stimulus versus a food stimulus is greater when novel rats are presented ( Reppucci et al . , 2020 ) . Indeed , in operant trials in which rats had simultaneous access to familiar and unfamiliar same-sex conspecifics , rats expended more effort to access unfamiliar conspecifics ( Hackenberg et al . , 2021 ) . In the present study , female prairie voles exhibited similar partner preferences but higher social motivation and aggression compared to female meadow voles . Social motivation and selectivity were not linked in male prairie voles , and there was a striking sex difference in the reward value of mates and peers in prairie voles . OTR binding revealed connections between social environment , receptor density , and prosocial behavior , illustrating the importance of this system across levels of biological organization . Better understanding of the interface between social motivation and social selectivity will thus be key to improving our understanding of the nature of social relationships . Prairie voles and meadow voles from in-house colonies were bred in a long photoperiod ( 14 hr light:10 hr dark; lights off at 17:00 EST; described further in Lee et al . , 2019 ) . Meadow voles were weaned into the winter-like short photoperiods associated with group living in this species ( 10:14 light:dark; lights off at 17:00 EST ) . Voles were pair-housed in clear plastic cages with aspen bedding and an opaque plastic hiding tube . Food ( 5015 supplemented with rabbit chow; LabDiet , St Louis , MO ) and water were provided ad libitum , except during food restriction ( described below ) . All procedures adhered to federal and institutional guidelines and were approved by the Institutional Animal Care and Use Committee at Smith College . Training began in adulthood at 62 ± 1 . 3 days of age ( mean ± SEM , range 41–76 ) . Operant conditioning training and testing consisted of multiple phases described briefly here and in greater detail in subsequent sections . Responses ( lever presses ) were shaped and trained using a food reward on a fixed ratio 1 ( FR-1 ) schedule . Animals that met training criteria progressed to the experimental testing sequence , beginning with 8 days of pressing for a food reward on a PR-1 schedule ( Figure 1 ) . Subjects in opposite-sex pairs were placed with either a tubally ligated , hormonally intact female mate , or a castrated and testosterone implanted male mate 5–10 days prior to the start of social habituation and testing . Subjects in same-sex pairs remained with their cage-mate . Social testing consisted of 8 days of PR-1 with rewards yielding access to the familiar ( same- or opposite-sex ) partner , and 8 days with access to a sex-matched stranger ( order balanced within groups ) . Voles were trained and tested over seven cohorts; group membership was distributed across cohorts , and voles were assigned to groups within sex without knowledge of their response rates in the training phase . A subset of voles ( those in cohorts 4–7 ) continued in empty chamber control and/or extinguishing tests as described below . Voles were sacrificed at the conclusion of testing , and brains were stored at –80°C . We tested four groups of prairie voles ( Figure 1 ) : females lever pressing for a female conspecific ( F➤F ) , females pressing for a male conspecific ( F➤M ) , males pressing for a male conspecific ( M➤M ) , and males pressing for a female conspecific ( M➤F ) . Each group consisted of eight focal voles , tested for 8 days with their partner and for 8 days with a series of novel ‘strangers’ , sex-matched to the partner . The order of testing ( partner then stranger or stranger then partner ) was counterbalanced within groups . Some voles did not complete both partner and stranger testing , in which case additional voles were added up to 8/group . Meadow vole females ( F➤F-Mp , n = 7 ) were also trained and tested for 8 days of familiar and 8 days of novel vole exposure , with order counterbalanced within the group . Subjects were weighed for 3 consecutive days to establish baseline body weights , then food-restricted to a target weight of 90% baseline to enhance motivation for the food reward . Weights were recorded daily after training or testing , prior to being returned to their home-cages . Any vole that dropped to or below 85% of the baseline weight was returned to ad libitum food to avoid long-term health consequences . Perforated cage dividers were used during food restriction to ensure each vole had access to its specific ration ( 0 . 3–1 food pellets and ~4 g [half] of a baby carrot ) . Food restriction ended when subjects transitioned to social testing . Operant conditioning was conducted in mouse-sized modular test chambers ( 30 . 5 cm × 24 . 1 cm × 21 . 0 cm ) outfitted with a response lever , clicker , modular pellet dispenser for mouse , and pellet receptacle ( Med Associates Inc , St Albans , VT , Figure 1A ) . Data were acquired using the MED-PC-IV program running training protocols coded by experimenters . Sessions lasted 30 min and took place between 0900 and 1700 . Vole behavior was shaped using manual reinforcement by an experimenter until a subject met the training criterion of 3 days in a row of ≥5 responses without manual reinforcement on an FR-1 schedule . One 20 mg food pellet ( Dustless Precision Pellet Rodent Grain Based Diet; Bio-Serv , Flemington , NJ ) was dispensed as each reward . Animals that did not learn to consistently lever press within ~20 days were used as partners or strangers for future social testing . Subjects that met the training criterion transitioned to a PR-1 schedule with each successive reward requiring an additional response . The progressive ratio has been shown to be a better indicator of motivation than FR programs ( Hodos and Kalman , 1963; Weatherly et al . , 2003 ) . PR-1 testing was conducted for 8 days , at the conclusion of which all focal animals were returned to ad lib food , and cage dividers were removed . Social reward testing was conducted in mouse-sized modular test chambers , custom-equipped with a motorized door ( Med Associates Inc , St Albans , VT ) for access to a second ‘social’ chamber ( Figure 1B ) . This chamber was constructed of clear plastic ( 15 cm × 20 . 5 cm × 13 cm ) and contained an eye-bolt for tethering a stimulus vole ( Figure 1C ) . A clear plastic tunnel ( 2 . 54 cm diameter , 5 . 5 cm long ) connected the operant chamber to the social chamber , and the entire apparatus was fixed to a mounting board . Lever presses were rewarded by door opening and chamber access; the door remained raised for 1 min , after which the experimenter returned the focal vole to the operant chamber . Sessions lasted 30 min and were video-recorded for quantification of additional behaviors . Subjects transitioned to social testing following a habituation session and two FR-1 sessions . Habituation to the social apparatus took place with the door open and the lever covered: voles explored the apparatus for 15 min with an empty social chamber , and 15 min with the partner tethered in the social chamber . Two days of FR-1 pressing for a tethered vole followed habituation to ensure that subjects associated lever pressing with access to the social chamber and a stimulus vole . Social testing took part in two phases: pressing for a partner vole on a PR-1 schedule and pressing for a stranger on a PR-1 schedule . Each phase lasted 8 days . The order of testing was counterbalanced within groups and subjects completed both phases . Social stimulus animals were tethered to the end of the social chamber . During the 8 days of stranger testing , the focal vole was tested against a novel vole each session to prevent familiarity between conspecifics . Empty chamber testing took place after social testing to avoid altering lever pressing for the social stimuli . The empty chamber control was run to assess the value of apparatus exploration: 30 voles ( 10 female prairie voles , 14 male prairie voles , 6 female meadow voles ) pressed the lever for 8 successive days on a PR-1 schedule to access the adjacent chamber when no stimulus vole was present . Sessions lasted 30 min and video was recorded and scored for behavior after testing . For the extinction phase , 31 voles ( 13 female prairie voles , 11 male prairie voles , 7 female meadow voles ) were tested in the social chamber with an unrewarded lever for 10 successive days ( 30 min sessions ) . Counts of responses ( lever presses ) and rewards ( food pellets or door raises ) were automatically recorded during each test . In all social trials ( 16/vole ) and all empty chamber control trials ( 8/vole ) , behavior in the ‘social’ chamber was also filmed with a portable digital video camera . Videos were scored using a custom perl script ( OperantSocialTimer; https://github . com/BeeryLab/Operant/ , Beery , 2017 ) to determine time in the social chamber , time in side-by-side contact with the tethered vole ( huddling ) , and bouts of aggression . These values could also be reported relative to other intervals ( e . g . time huddling/access time when the door was up , or time huddling/time in the social chamber ) . Non-social/empty chamber trials ( 8 days/vole ) were also videotaped and analyzed for time in the social chamber/available time with the door raised . At least 1 week prior to pairing , the future ‘partner’ of each opposite-sex prairie vole pair was surgically altered to prevent pregnancies during testing . Female partners of male focal voles underwent tubal ligation . Dorsal incisions were made over each ovary . Two knots were placed below each ovary at the top of the uterine horn . The wound was closed using a sterile suture . Male partners of female focal voles were castrated and implanted with testosterone capsules . Testes were accessed by midline incision , and the blood supply was cut-off through a tie at the testicular artery . Testes were removed and the muscle wall and skin were closed using sterile suture . A testosterone capsule was implanted subcutaneously between the scapulae . Capsules contained 4 mm of crystalline testosterone ( Sigma-Aldrich , St Louis , MO ) in silastic tubing ( ID 1 . 98 mm , OD 3 . 18 mm; Dow Corning , Midland , MO ) as in Costantini et al . , 2007 . Capsules were sealed with silicone , dried , and soaked in saline for 24 hr prior to insertion . A subset of strangers was also castrated or ligated , with no effect on focal behavior . Surgical procedures were performed under isoflurane anesthesia . Voles received 0 . 05 mg/kg buprenorphine and 1 . 0 mg/kg metacam subcutaneously prior to surgery , and again the following day . Post-operative wound checks continued for up to 10 days post-surgery . OTR binding density was assessed in the brains of 11 female prairie voles at the conclusion of the study ( males were used for an additional pilot study ) . Frozen brains were sectioned coronally at 20 μm , thaw-mounted on Super-frost Plus slides ( Fisher , Inc ) , and stored at –80°C until processing ( as in Beery et al . , 2008a; Beery and Zucker , 2010; Mooney et al . , 2015 ) . Briefly , slides were thawed until dry , then fixed for 2 min in fresh , chilled 0 . 1% paraformaldehyde in 0 . 1 M PBS . Sections were rinsed 2 × 10 min in 50 mM Tris ( pH 7 . 4 ) , and incubated for 60 min at room temperature in a solution ( 50 mM Tris , 10 mM MgCl2 , 0 . 1% BSA , 0 . 05% bacitracin , 50 pM radioligand ) containing the radioactively labeled 125I-ornithine vasotocin analog vasotocin , d ( CH2 ) 5 [Tyr ( Me ) 2 , Thr4 , Orn8 , ( 125I ) Tyr9-NH2] ( 125I-OVTA , PerkinElmer , Inc ) . An adjacent series of slides , processed for non-specific binding , was incubated with an additional 50 nM non-radioactive ligand [Thr4Gly7]-oxytocin ( Bachem ) . All slides were rinsed 3 × 5 min in chilled Tris–MgCl2 ( 50 mM Tris , 10 mM MgCl2 , pH 7 . 4 ) , dipped in cold distilled water , and air-dried . Sections were apposed to Kodak BioMax MR film ( Kodak , Rochester , NY ) for 3 days and subsequently developed . Radioligand binding density in each brain region was quantified in samples of uniform area from three adjacent sections for each brain region and averaged for each brain . Non-specific binding was subtracted from total binding to yield specific binding values . Social data were analyzed for all subjects completing both partner and stranger phases of testing ( n = 8 prairie vole M➤M pairs , 8 prairie vole M➤F pairs , 8 prairie vole F➤F pairs , 8 prairie voles F➤M pairs , and 7 meadow vole F➤F pairs ) . Four additional female prairie voles completed testing with a partner or stranger only: data from these subjects was included in analysis of food responses and food versus social response rates . Group differences in single variables ( e . g . food responses ) were assessed by one-way ANOVA . Two-way RM-ANOVA was used to assess the effects of social factors , with stimulus familiarity [partner , stranger] as a within-subjects repeated measure , stimulus type [same-sex , opposite-sex] as a between-subjects ( non-repeated ) measure , a test for interaction effects [stimulus familiarity*stimulus type] , and for subject matching . Paired t-tests were used within groups for comparison of behavior toward the partner versus stranger . Response count ( i . e . lever presses ) and breakpoint ( i . e . number of rewards achieved ) are highly correlated; detailed results are therefore shown for only one measure ( response count ) . Response rate ( responses/active session time ) was used when comparing food responses to social responses , as the lever was continuously active during food-rewarded testing ( active session time = 30 min ) , but was not capable of raising the door when it was already up ( active session time = 30 min with the door up ) . Autoradiography data were collected in multiple brain regions , and comparisons were performed by two-way ANOVA ( group*brain region ) . Statistical analyses were performed in JMP 15 . 0 ( SAS , Inc ) and Prism 9 ( GraphPad Software Inc ) . Effect sizes were calculated in Excel . Cohen’s d for paired t-tests used the mean of partner-stranger differences/standard deviation of partner-stranger differences . Eta squared ( η2 ) and partial eta squared ( ηp2 ) were reported for one-way and two-way ANOVAs , respectively ( Lakens , 2013 ) . Pearson’s product-moment correlation coefficient was reported for correlations . All tests were two-tailed , and results were deemed significant at p < 0 . 05 . Number of animals: social operant studies in rats have been successful with six subjects ( Tan and Hackenberg , 2016; Hiura et al . , 2018; Hackenberg et al . , 2021 ) . We used 30% more subjects as a buffer ( eight females or males in each condition ) , as operant behavior in voles was not well characterized .

What factors drive the formation of social relationships can vary greatly in animals . While some individuals may be motivated to find social partners , others may just tolerate being around others . A desire to avoid strangers may also lead an individual to seek out acquaintances or friends . Sometimes a mix of these factors shape social behavior . Studying motivation for social relationships in the laboratory is tricky . Traditional laboratory animals like mice and rats do not bond with specific peers or mates . But small burrowing rodents called voles are a more relationship-oriented alternative to mice and rats . Prairie voles form selective and enduring preferences for both their mates and familiar same-sex peers . Meadow voles on the other hand , live alone much of the year but move in with other animals over the winter . Beery et al . show that social motivation in voles varies by relationship type , species and sex . In the experiments , voles were first trained to press a lever to get a food reward . Then , the food reward was swapped with access to familiar or unfamiliar voles . Female prairie voles strived to be with animals they knew rather than to be with strangers , while male prairie voles tried hard to access any female . In contrast , meadow voles did not overly exert themselves to access other animals . Beery et al . then measured oxytocin receptor levels in the brains of prairie voles . Prairie voles that had more receptors for oxytocin in part of their brain known as the nucleus accumbens worked harder to access their familiar partner . But individuals with more oxytocin receptors in the bed nucleus of the stria terminalis were more likely to attack an unfamiliar animal . The meadow voles’ behavior suggests that they are more motivated by tolerance of familiar animals , while the female prairie voles may find it rewarding to be with animals they have bonded with . These differences may help explain why these two species of vole have evolved different social behaviors . The experiments also suggest that oxytocin – which is linked with maternal behavior – plays an important role in social motivation . Learning more about the biological mechanisms that underlie vole social behaviors may help scientists identify fundamental aspects of social behavior that may apply to other species including humans .

2021

Social selectivity and social motivation in voles

Down syndrome ( DS ) results in various degrees of cognitive deficits . In DS mouse models , recovery of behavioral and neurophysiological deficits using GABAAR antagonists led to hypothesize an excessive activity of inhibitory circuits in this condition . Nonetheless , whether over-inhibition is present in DS and whether this is due to specific alterations of distinct GABAergic circuits is unknown . In the prefrontal cortex of Ts65Dn mice ( a well-established DS model ) , we found that the dendritic synaptic inhibitory loop formed by somatostatin-positive Martinotti cells ( MCs ) and pyramidal neurons ( PNs ) was strongly enhanced , with no alteration in their excitability . Conversely , perisomatic inhibition from parvalbumin-positive ( PV ) interneurons was unaltered , but PV cells of DS mice lost their classical fast-spiking phenotype and exhibited increased excitability . These microcircuit alterations resulted in reduced pyramidal-neuron firing and increased phase locking to cognitive-relevant network oscillations in vivo . These results define important synaptic and circuit mechanisms underlying cognitive dysfunctions in DS . Down syndrome ( DS ) is a condition caused by full or partial trisomy of human chromosome 21 , characterized by various physical and neurological features including mild to severe intellectual disability ( Antonarakis et al . , 2020 ) . Individuals with DS present important deficits in cognitive tasks known to depend on the anatomical and functional integrity of the frontal lobe ( Lee et al . , 2015 ) . Moreover , DS is also associated with other CNS-mediated phenotypes , including an ultra-high risk for developing Alzheimer’s disease and high rates of autism for which the mechanisms are unknown ( DiGuiseppi et al . , 2010; Wiseman et al . , 2015 ) . Interventions to ameliorate DS-mediated cognitive dysfunctions are limited . The development of interventions for this vulnerable group of individuals can be achieved through a better understanding of the mechanisms underlying a core feature of DS , such as intellectual disability . Importantly , mouse models of DS recapitulate several cognitive deficits of this condition ( Herault et al . , 2017; Olmos-Serrano et al . , 2016b ) . One of the best-characterized mouse models is the Ts65Dn mouse line ( herein referred to as Ts ) , which carries a partial trisomy of a segment of the mouse chromosome 16 ( Davisson et al . , 1993 ) . Ts mice recapitulate several dysfunctions present in DS individuals , such as reduced birthweight , male sterility , abnormal facial appearance and several cognitive impairments , including executive functions , such as working memory and cognitive flexibility ( Olmos-Serrano et al . , 2016b ) . Recovery of behavioral and neurophysiological deficits underlying cognitive impairments using GABAA receptor blockers led to the hypothesis that intellectual deficits in DS are produced by an excessive activity of inhibitory circuits ( Fernandez et al . , 2007; Zorrilla de San Martin et al . , 2018 ) . Nonetheless , direct evidence for over-inhibition in DS is lacking . Moreover , given the anatomical , molecular and functional diversity of cortical inhibitory neurons ( Tremblay et al . , 2016 ) , the functional implications of this hypothesis at the network level , as well as the involvement of specific GABAergic circuits remain obscure . Executive functions depend on the integrity of the prefrontal cortex ( PFC ) , which plays an essential role in the synchronization of task-relevant , large-scale neuronal activity ( Helfrich and Knight , 2016 ) . An important network correlate of this synchronization is represented by neuronal oscillations: rhythmic fluctuations of the electrical activity of single neurons , local neuronal populations and multiple neuronal assemblies , distributed across different brain regions ( Buzsáki and Wang , 2012 ) . Oscillations are the result of a balanced and coordinated activity of excitatory pyramidal neurons ( PNs ) and a rich diversity of inhibitory neurons that use γ-aminobutiric acid ( GABA ) as neurotransmitter . In particular , parvalbumin ( PV ) -positive inhibitory interneurons form synapses onto the perisomatic region of PNs . PV cells thus tightly control PN spiking activity and drive network oscillations in the γ-frequency range ( 30–100 Hz ) ( Buzsáki and Wang , 2012 ) . γ-Oscillations are necessary for several PFC cognitive functions , such as sustained attention ( Kim et al . , 2016b ) and cognitive flexibility ( Cho et al . , 2015 ) . Conversely , Martinotti cells ( MCs ) are somatostatin ( SST ) -positive interneurons that inhibit distal dendrites of PNs , thereby controlling the integration of distal dendritic glutamatergic synaptic inputs originating from different regions of the brain ( Tremblay et al . , 2016 ) . Dendritic integration of multi-pathway inputs is necessary for working memory ( Abbas et al . , 2018; Kim et al . , 2016a ) . Therefore , PV interneurons and MCs represent two major cortical inhibitory circuits , characterized by a precise division of labor during cortical activity . Both forms of inhibition were shown to be involved in the entrainment of network oscillations ( Cardin et al . , 2009; Chen et al . , 2017; Sohal et al . , 2009; Veit et al . , 2017 ) and in the cognitive performance during medial ( m ) PFC-dependent tasks ( Abbas et al . , 2018; Cho et al . , 2015; Cummings and Clem , 2020 ) . In particular , inhibition from SST interneurons plays a crucial role in mPFC-dependent memory ( Abbas et al . , 2018; Cummings and Clem , 2020 ) . Broad-spectrum GABAAR antagonists are not clinically viable , as they can yield undesired seizure-like activity and/or anxiety . Interestingly , however , treatment of Ts mice with selective and partial negative allosteric modulators of α5-containing GABAARs ( α5 inverse agonist or α5IA ) reverse cognitive behavioral and long-term synaptic plasticity deficits in DS mice ( Braudeau et al . , 2011; Duchon et al . , 2020; Martínez-Cué et al . , 2013; Schulz et al . , 2019 ) . Importantly , neocortical dendritic synaptic inhibition of PNs from MCs relies on α5-containing GABAARs ( Ali and Thomson , 2008 ) . The preference for this specific GABAAR subunit was also recently demonstrated at the equivalent hippocampal dendritic inhibitory circuit ( Schulz et al . , 2019; Schulz et al . , 2018 ) , raising the question of whether dendritic inhibition is specifically altered in DS . Here we found that the dendritic synaptic inhibitory loop formed by MCs and PNs was strongly potentiated in Ts mice , with no alteration of either cell-type excitability . Conversely , the perisomatic synaptic inhibitory loop from PV cells onto PN cell bodies was unaffected in Ts mice . Strikingly , however , PV-cell excitability was strongly altered: these interneurons did not display their typical fast-spiking behavior and exhibited enhanced excitability . At the network level in vivo , these inhibitory microcircuit-specific alterations resulted in significant reduction of putative PN firing , which in turn was more tuned to β- and low γ-oscillations ( 10–60 Hz ) . These results confirm over-inhibition in DS , and reveal unexpected functional alterations of specific GABAergic circuits in this condition . Cognitive and synaptic plasticity deficits in Ts mice can be successfully treated by systemic application of a selective negative allosteric modulator of α5-containing GABAARs , α5IA ( Braudeau et al . , 2011; Duchon et al . , 2020; Martínez-Cué et al . , 2013; Schulz et al . , 2019 ) . α5-GABAARs are expressed at PN synapses originating from dendrite-targeting interneurons: MCs in the neocortex ( Ali and Thomson , 2008 ) and O-LM in the hippocampus ( Schulz et al . , 2018 ) . We therefore tested whether dendritic inhibition of PNs by MCs are affected in Ts mice . We crossed Ts65Dn with GFP-X98 mice , which in the barrel cortex were shown to bias GFP expression in MCs ( Ma et al . , 2006 ) . Accordingly , in the mPFC of these mice , GFP was expressed by a subset of SST-positive interneurons ( Figure 1—figure supplement 1a ) , exhibiting a widely branched axonal plexus in L1 , characteristic of dendrite-targeting inhibitory MCs ( Figure 1—figure supplement 1b , c ) . Using dual whole-cell patch-clamp recordings in acute mPFC slices , we isolated unitary inhibitory postsynaptic currents ( uIPSCs ) in MC-PN connected pairs ( Figure 1a ) . We found that dendritic MC-PN synaptic inhibition relied on α5-containing GABAARs in both Ts and control , euploid ( Eu ) mice . Indeed , in both genotypes , bath application of the selective negative allosteric modulator of α5-containing GABAARs , α5IA ( 100 nM; Sternfeld et al . , 2004 ) , produced a significant reduction of uIPSCs that was close to the maximal potency of the drug ( ~40%; Dawson et al . , 2006; Figure 1b , Table 1 ) . IPSC rise and decay times were not affected by α5IA application ( data not shown ) . Interestingly , MC-mediated uIPSC amplitudes were significantly larger and failure rate of synaptic responses evoked by the first action potential was significantly smaller in Ts compared to Eu mice . Moreover , the total amount of synaptic charge ( Q ) transferred during a train of 5 action potentials was near 5-fold larger in Ts than in Eu ( Figure 1c , Table 1 ) . MC-PN uIPSCs exhibited faster rise time in Ts as compared to Eu mice . Yet , uIPSC decay time-constants were similar in the two genotypes ( Figure 1—figure supplement 2 , Table 1 ) . We then examined glutamatergic recruitment of MCs by PNs and found that it was stronger in Ts than Eu mice . Unitary excitatory postsynaptic currents ( uEPSCs ) in connected PN-MC pairs exhibited larger amplitude , lower failure rate and larger charge transfer in Ts than Eu mice ( Figure 1d , Table 2 ) . Importantly , in both Ts and Eu mice , short-term dynamics of MC-PN GABAergic and PN-MC glutamatergic synaptic transmission were depressing and facilitating , respectively . However , short-term depression of MC-dependent dendritic inhibition onto PNs was more pronounced in Ts than Eu mice ( Figure 1—figure supplement 3 ) . Conversely , glutamatergic recruitment of MCs displayed the classical strong facilitating characteristics ( Silberberg and Markram , 2007 ) with no differences in short-term uEPSC dynamics in the two genotypes ( Figure 1—figure supplement 3 ) . Regardless of whether unitary synaptic responses exhibited significant alterations of short-term plasticity , both MC-PN uIPSCs and PN-MC uEPSCs of Ts mice were characterized by a significant decrease of failure rate in all synaptic responses within the train . These results on short-term plasticity suggest that , in Ts mice , GABAergic and glutamatergic synapses involved in the PN-MC-PN dendritic inhibitory loop increased their efficacy using different pre- and postsynaptic strategies . Overall , these results indicate that the dendritic inhibitory loop involving MCs and PNs is strengthened in Ts mice . Both output GABAergic synapses from MCs and their recruitment by local glutamatergic synapses were stronger and more reliable in Ts mice , as compared to their euploid littermates . Alteration of the MC-PN-MC synaptic loop can be associated to changes in intrinsic excitability and morphological features . We therefore tested whether passive properties , single action potentials and firing dynamics were altered in both PNs and MCs . In addition , we filled neurons with biocytin and we quantified their dendritic and axonal arborizations . Input-output spiking activity of both PNs and MCs was assessed by injecting increasing depolarizing 2 s-long currents . The firing frequency vs . injected current ( f-i ) curve was similar in both cell types in Eu and Ts mice ( Figure 2a , b; Tables 3–4 ) . Furthermore , single-action potential features , and passive properties were similar in both genotypes except for a small but significant increase in action potential threshold of Ts PNs ( Figure 2—figure supplements 1 and 2; Tables 5–6 ) . Importantly , the density of GFP-expressing MCs , and , in general , of SST-positive interneurons was similar in both genotypes ( Figure 2c; Figure 2—figure supplement 3 ) . Increased MC-PN GABAergic transmission synaptic transmission in Ts mice can be attributed to axonal sprouting of MCs and/or increased dendritic branching of PNs . We performed a morphometric analysis of both cell types and found that the spatial distribution and total length of axons and dendrites of MCs were similar in both genotypes ( Figure 2d–f ) . Likewise , both apical and basal dendrite arborizations of PNs were indistinguishable in Eu and Ts mice ( Figure 2g–i ) . These experiments and those illustrated in Figure 1 indicate that the increased dendritic inhibitory loop involving MCs and PNs can be largely attributable to alteration of synaptic transmission between these two cell types . Is the synaptic enhancement of the dendritic inhibitory loop involving MCs a specific alteration or a general feature of glutamatergic and GABAergic synapses in Ts mice ? To address this question , we measured glutamatergic recruitment onto , and synaptic inhibition from , another prominent interneuron class , the PV basket cell . This interneuron class is characterized by its ability to fire high frequency , non-adapting trains of fast action potentials . These properties , along with perisomatic synaptic targeting of PNs , make the PV cell an efficient regulator of PN output . We thus crossed Ts65Dn with PValb-tdTomato mice , a line that expresses TdTomato specifically in PV-positive interneurons ( Kaiser et al . , 2016 ) . We recorded uIPSCs and uEPSCs ( Figure 3a–d ) from pairs of synaptically connected PNs and PV cells . The amplitude , failure rate and charge transfer of uIPSC trains evoked by action potentials in presynaptic PV interneurons were similar in both genotypes ( Figure 3a , b; Table 7 ) . Likewise , the amplitudes , failure rates and charge transfer of uEPSCs elicited by PN firing were indistinguishable in Eu and Ts mice ( Figure 3c , d; Table 8 ) . Consistently , no difference was observed in short-term plasticity of both uIPSCs and uEPSCs in Ts and Eu mice ( Figure 3—figure supplement 1 ) . Surprisingly , however , we found that intrinsic excitability of PV cells of Ts mice was strongly altered . These interneurons required one fourth less current ( rheobase ) necessary to induce firing action potentials . Accordingly , the gain of the f-I curves of Ts mice was dramatically reduced , as compared to their Eu littermates ( Figure 4a–c; Table 9 ) . Moreover , PV cells in Ts mice could not sustain high-frequency firing in response to 2 s-long depolarization , and the maximal spike rate was near half of that reached by PV cells in Eu mice ( Figure 4a–c; Table 9 ) . Notably , in Ts mice , action potential width was 1 . 7-fold wider and input resistance was 1 . 9-fold larger than that observed in Eu mice ( Figure 4d , e , ) . Conversely , action potential threshold and amplitude were not affected ( Figure 4—figure supplement 1; Table 10 ) . Abnormal passive and active properties of PV cells of Ts mice were present throughout all ages under study . Importantly , in Eu mice , PV-cell active and passive properties reached values similar to those reported for this cell type ( Kaiser et al . , 2016; Figure 4—figure supplement 2 ) . Similarly to SST cells , the density of PV-INs in mPFC was similar in the two genotypes ( Figure 4f ) . Altogether , these results indicate that , contrary to dendritic inhibition , the synaptic efficiency of the perisomatic feedback loop mediated by PV-INs was normal in Ts mice . Yet , PV-INs of Ts mice lost their characteristic electrophysiological fast-spiking signature , and their excitability was dramatically increased . Increased dendritic inhibition by the MC-PN loop , altered spiking activity of PV cells and their increased excitability will likely strongly influence the spiking properties and dynamics of mPFC PNs in vivo during spontaneous network activity . In order to assess the activity of the mPFC in vivo , we performed simultaneous local field potential ( LFP ) and loose-patch , juxtacellular recordings from layer 2/3 putative PNs to monitor their spiking dynamics related to overall network activity ( Figure 5a ) . In vivo recordings exhibited typical oscillatory activity consisting of UP and DOWN states ( Ruiz-Mejias et al . , 2011 ) . UP and DOWN states were similar in frequency and duration in Ts mice and their Eu littermates ( Figure 5—figure supplement 1 ) . UP states were enriched in γ-band activity ( 30–100 Hz ) and exhibited increased probability of spiking activity ( Ruiz-Mejias et al . , 2011; Figure 5a ) . Juxtacellular recordings from individual mPFC putative PNs revealed a near 50% decrease in the overall spiking rate in Ts mice as compared to their Eu littermates ( Figure 5b , Table 11 ) . This difference was explained by a significant reduction of spiking rate during UP but not DOWN states ( data not shown ) . Analysis of LFP power spectral density ( PSD ) did not show a significant difference in the two genotypes ( Figure 5c ) . Interestingly , however , when we analyzed LFP waveform specifically during periods of neuronal spiking activity ( spike-triggered LFP or stLFP ) , we found that , in both genotypes , average stLFPs exhibited marked voltage deflections , indicating that spike probability was not randomly distributed but locked to LFP oscillations ( Figure 5d ) . The peak-to-peak amplitude of the stLFP was much larger in Ts than in Eu mice ( Figure 5d , e , Table 11 ) , and , remarkably , the spectral power of the stLFP was largely increased in Ts mice , selectively in the β-γ-frequency band ( Figure 5f ) . In order to quantitatively assess whether PN spikes were differently locked to the phase of network oscillations , we measured the pairwise phase consistency ( PPC ) , which is an unbiased parameter to determine the degree of tuning of single-neuron firing to network rhythmic activity of specific frequencies ( Perrenoud et al . , 2016; Veit et al . , 2017; Vinck et al . , 2010 ) . Ts mice exhibited significantly higher PPC values than Eu littermates for frequency bands ranging between 10 and 60 Hz ( Figure 4g , Table 12 ) , thus revealing stronger phase locking selectively with β- and low γ-oscillations . Altogether , these results indicate that mPFC PNs in Ts mice fire less than their Eu littermates , but their spontaneous spiking activity is more strongly tuned to β- and γ-frequency bands . Reduced firing rate and increased phase locking with fast oscillations are both consistent with increased activity of inhibitory interneurons ( Cardin et al . , 2009; Chen et al . , 2017; Sohal et al . , 2009; Veit et al . , 2017 ) . In this study , we analyzed synaptic and intrinsic properties of two major inhibitory circuits of the prefrontal cortex in a relevant mouse model of DS . We found specific alterations of distinct GABAergic circuits . In particular , we demonstrate that the dendritic inhibitory synaptic loop involving MCs and PNs was strongly potentiated in Ts , as compared to euploid control mice . In contrast , the perisomatic synaptic inhibitory control of PNs by PV cells was not affected in Ts mice . Strikingly , however , the excitability of PV cells was profoundly altered . These GABAergic circuit-specific alterations correlate with reduced PN spiking , and enhanced coupling with network β-γ-activity in Ts mice in vivo . We analyzed MCs , which are a subset of SST interneurons , specialized in inhibiting the distal portion of apical dendrites of PNs , by ascending their axons in L1 ( Kawaguchi and Kubota , 1998 ) , where PN dendrites integrate top-down input originating from distal brain areas . In the hippocampus , dendritic inhibition strongly regulates PN dendrite electrogenesis and supra-linearity ( Lovett-Barron et al . , 2012 ) , likely modulating the emergence of burst firing ( Royer et al . , 2012 ) . Interestingly , α5–mediated dendritic inhibition in the hippocampus , known to strongly control dendritic integration and action potential firing ( Schulz et al . , 2018 ) , is also enhanced in Ts65Dn mice and strongly control NMDAR activation , nonlinear dendritic integration , and AP firing ( Schulz et al . , 2019 ) . In the neocortex , integration of top-down information in L1 and consequent dendrite-dependent generation of burst activity was hypothesized to underlie the encoding of context-rich , salient information ( Larkum , 2013 ) . Our results indicate a specific synaptic strengthening of the dendritic inhibitory loop involving MCs in Ts mice , suggesting a major impact on PN dendritic integration and electrogenesis . Enhanced dendritic inhibition by MCs in DS could underlie the deficits of long-term plasticity of glutamatergic synapses similar to those observed in the hippocampus of Ts mice . Indeed , these LTP deficits can be recovered by treatment of allosteric modulators of α5-GABAARs ( Duchon et al . , 2020; Martínez-Cué et al . , 2013; Schulz et al . , 2019 ) . Likewise , the enhanced MC-PN inhibitory loop in Ts mice shown here can provide a mechanism for the rescue of cognitive deficits in Ts mice , operated by selective pharmacology of α5-GABAARs ( Braudeau et al . , 2011; Duchon et al . , 2020; Martínez-Cué et al . , 2013 ) . This subunit of GABAARs , whose synaptic vs . extrasynaptic expression is still debated ( Ali and Thomson , 2008; Botta et al . , 2015; Glykys and Mody , 2010; Hannan et al . , 2020; Hausrat et al . , 2015; Schulz et al . , 2018; Serwanski et al . , 2006 ) , mediate dendritic synaptic inhibition from neocortical MCs ( Ali and Thomson , 2008 ) and their hippocampal counterparts ( the oriens-lacunosum moleculare or O-LM interneurons , Schulz et al . , 2018 ) . Here , we confirmed that α5-containing GABAARs are majorly responsible for MC-PN dendritic synaptic inhibition , due to the overall block of α5IA , which was close to the max potency of this drug ( Dawson et al . , 2006 ) . The enhancement of the dendritic inhibitory loop involving MCs and PNs in Ts mice could not be attributed to alterations of axonal and/or dendritic arborizations in the mutant mice . On the other hand , it could be due to a combination of pre- and postsynaptic mechanisms , including alterations of release probability , number of release sites or quantal size at either GABAergic or glutamatergic synapses involved in this circuit . The strong increase of synaptic charge at MC-PN GABAergic synapses could be attributed to alterations of uIPSC waveform in Ts mice . Although the small change of uIPSC rise time can in principle account for , at least in part , an increase in synaptic charge , it is unlikely to account for the 5-fold increase of synaptic charge , with no changes in uIPSC decay . The robust increase of synaptic charge in Ts mice is likely due to one or a combination of different non-linearities that dynamically emerge during a spike train . These include the recruitment of peri- or extrasynaptic GABAARs , possible alterations of release probability or quantal size during trains and differences in postsynaptic dendritic filtering in the two genotypes . These complex interplays between these factors occur at distal dendrites; therefore , they are difficult to tease out from somatic recordings as they are characterized by poor dendritic voltage control due to space-clamping constraints . The enhanced short-term depression of MC-PN uIPSCs that we report here suggests that release probability at dendritic GABAergic synapses from MCs is increased in Ts mice . This is consistent with increased short-term depression of inhibition reported in the dentate gyrus ( Kleschevnikov et al . , 2012 ) , but not with unaltered short-term plasticity shown in CA1 ( Mitra et al . , 2012 ) . These discrepancies could be ascribed to differences between brain regions . Moreover , they could also be a consequence of non-specific recruitment of presynaptic axons by global extracellular stimulation , in contrast to the isolation of unitary synaptic responses by dual intracellular recordings as reported here . The lack of short-term plasticity alterations at PN-MC glutamatergic synapses suggests that increased excitatory recruitment of MCs is due to alterations at postsynaptic sites . Future studies will be necessary to pinpoint the exact biophysical and anatomical alterations underlying the prominent increase of dendritic inhibition operated by MCs in DS . Potentiation of glutamatergic synapses in Ts mice seems to be specific for PN-MC connections , as PN-PV synapses were similar in both genotypes , suggesting that presynaptic terminals of local glutamatergic synapses can undergo target-specific modulation of their strength . Intriguingly , it has been recently shown that an increase of the excitatory drive of hippocampal interneurons ( due to triplication of GluR5 kainate receptor expression ) could explain excess of inhibition received by pyramidal neurons in the Ts2Cje Down syndrome mouse model ( Valbuena et al . , 2019 ) . Therefore , a similar gene overdose of kainate receptors can boost the recruitment of specific interneurons in the PFC of Ts mice . The strong increase of membrane resistance of PV cells in Ts mice underlies the augmented intrinsic excitability and can produce early ectopic bouts of activity , thus contributing to network over-inhibition . Enhanced membrane resistance could result from alterations in the expression of TWIK1 and TASK1 leak channels . Indeed , these channels underlie the developmental decrease of membrane resistance in PV cells ( Okaty et al . , 2009 ) . In contrast , the inability of PV cells of Ts mice to sustain high frequency firing could prevent these interneurons from generating high-frequency bursts of action potentials and therefore have a detrimental effect on the temporal coding of these interneurons . Interestingly , however , despite the dramatic alterations of intrinsic excitability in Ts PV cells , their output synaptic perisomatic inhibition was similar in both genotypes . This , despite the widening of single PV-cell action potentials , which could , in principle change the presynaptic Ca2+ dynamics and thus alter release probability . Since action potentials are recorded in the soma , the lack of effect at PV-cell synapses could be due to soma-specific alterations . Alternatively , since neocortical PV-PN GABAergic synapses exhibit high release probability ( Deleuze et al . , 2019; Kawaguchi and Kubota , 1998 ) , alterations of spike width might not be enough to produce additional increases . Increase of action potential width in Ts mice could be due to changes in the expression Kv3 . 1b potassium channels , known to underlie the fast repolarization of action potentials and fast-spiking behavior of PV cells ( Erisir et al . , 1999 ) . Future experiments will be necessary to reveal the exact molecular mechanism underlying the altered firing properties of PV cells , which , in Ts mice , have lost their characteristic fast-spiking signature . The aberrant active and passive properties of PV cells in Ts mice could be due to a delayed development of these interneurons . Indeed , during development , neocortical PV cells display marked accelerations of single action potentials and increased firing frequency , accompanied by decreased input resistance ( Okaty et al . , 2009 ) . Although we cannot rule out the possibility of a delayed development of PV cells in Ts mice , these interneurons showed abnormal passive and active properties within the entire age range studied here . It will be interesting to determine whether these profound alterations of PV-cell firing are present with the same incidence and magnitude along the entire life span of Ts mice . Although previous reports have shown higher number of inhibitory interneurons in the hippocampus ( Hernández-González et al . , 2015 ) and somatosensory cortex ( Aziz et al . , 2018; Chakrabarti et al . , 2010 ) we failed to detect significant differences in the density of both SST- and PV-positive interneurons in the mPFC . This could be due to differences between brain regions . A systematic comparative analysis will be required to better understand the consequences of DS neurodevelopmental alterations of the cellular composition in different brain regions . Both potentiation of the dendritic inhibitory loop and increased PV-cell excitability are consistent with the alterations of PN spiking activity that we recorded in vivo . Indeed , reduced spike rates and increased tuning in the β-γ-frequency range are both consistent with increased activity of inhibitory neurons ( Atallah et al . , 2012; Cardin et al . , 2009; Chen et al . , 2017; Sohal et al . , 2009; Veit et al . , 2017 ) . The overall LFP power spectra observed in mPFC is similar in both Ts and Eu mice , consistently with a recent report ( Chang et al . , 2020 ) . However , close examination of stLFP revealed a strong increase in the power of β- and low γ-frequency bands during periods of spiking activity . This likely reflects the consequences of alterations at the level of local microcircuits . We cannot directly link the inhibitory circuit-specific alterations that we detected in slices with the increased synchronization of β-γ-activity that we measured in vivo . However , a large body of literature indicates that oscillations in this frequency range strongly depends on the activity of PV cells ( Buzsáki and Wang , 2012; Sohal et al . , 2009; Cardin et al . , 2009 ) . More recently , also SST interneurons were shown to control PN phase coupling with low frequency ( 30 Hz ) neocortical γ-oscillations ( Chen et al . , 2017; Veit et al . , 2017 ) . It is therefore tempting to speculate that increased dendritic inhibition from SST-expressing MCs modulates phase coupling of PNs with β- and low γ-oscillations . On the other hand , the increased phase coupling of PN spikes with high-frequency γ-activity could result from the augmented excitability of PV cells . The differential phase coupling of these two interneuron types at distinct frequencies is consistent with the peculiar fast and slower recruitment and biophysical properties of PV interneurons and MCs , respectively . Future experiments involving chemogenetic alterations of PV-interneuron excitability and/or pharmacological manipulations of MC-PN synapses in Ts mice will help decipher the role played by each inhibitory cell subtype in controlling the temporal dynamics of PN firing during different rhythmic cortical network activities . Alternatively , our results could be interpreted as differences in long-range functional connectivity in the two genotypes , possibly due to alterations in myelination and conduction velocity in Ts vs . Eu mice ( Olmos-Serrano et al . , 2016a ) . Nevertheless , a reduction of axonal conduction velocity would produce a temporal shift in spike-to-phase association , rather than increased phase locking . The increase in β-γ-band power and enhanced neural synchronization in these frequency ranges in Ts mice is consistent with recent evidence indicating augmented hippocampal-PFC synchronization and LFP γ-band power during natural non-REM sleep in Ts65Dn mice ( Alemany-González et al . , 2020 ) . This suggests that the alterations of γ-oscillations observed here could play a role in the pathophysiology of sleep disruptions reported in DS children ( Fernandez et al . , 2017 ) and adults ( Giménez et al . , 2018 ) . In sum , here we report direct evidence for over-inhibition of mPFC circuits in a mouse model of DS . However , over-inhibition was not due to a generic increase of GABAergic signaling , but emerged from highly specific synaptic and intrinsic alterations of dendritic and somatic inhibitory circuits , respectively . Future experiments are necessary to reveal whether other inhibitory neuron types are also affected in DS . Likewise , it will be fundamental to assess whether specific dysfunctions of individual GABAergic circuits underlie different aspects of cognitive deficits ( e . g . impaired memory and flexibility , autistic traits ) , which affect individuals with DS . Ts or Eu mice were anesthetized with 15% urethane ( 1 . 5 g/kg in physiological solution ) and placed on a stereotaxic apparatus . The body temperature was constantly monitored and kept at 37°C with a heating blanket . To ensure a deep and constant level of anesthesia , vibrissae movement , eyelid reflex , response to tail , and toe pinching were visually controlled before and during the surgery . A local lidocaine injection was performed over the cranial area of interest and , after a few minutes , a longitudinal incision was performed to expose the skull . Two small cranial windows ( <1 mm diameter ) were opened at at 2 . 5 mm from bregma and ±0 . 5 mm lateral to sagittal sinus ( corresponding to the frontal lobe ) carefully avoiding any damage to the main vessels while keeping the surface of the brain moist with the normal HEPES-buffered artificial cerebrospinal fluid . Pipettes used to record LFP had 1–2 MΩ resistance while those used for juxtacellular patch-clamp recordings typically had 5–7 MΩ resistance . LFP and patch electrodes were pulled from borosilicate glass capillaries . Signals were amplified with a Multiclamp 700B patch-clamp amplifier ( Molecular Devices ) , sampled at 20 KHz and filtered online at 10 KHz . Signals were digitized with a Digidata 1440A and acquired , using the pClamp 10 software package ( Molecular Devices ) . In order to record intrinsic and synaptic properties of L2/3 neurons of mPFC , we prepared acute cortical slices from the described mouse lines . For these experiments , we used slices cut in the coronal plane ( 300–350 μm thick ) . Animals were deeply anesthetized with saturating isofluorane ( Vetflurane , Virbac ) and immediately decapitated . The brain was then quickly removed and immersed in the cutting choline-based solution , containing the following ( in mM ) : 126 choline chloride , 16 glucose , 26 NaHCO3 , 2 . 5 KCl , 1 . 25 NaH2PO4 , 7 MgSO4 , 0 . 5 CaCl2 , cooled to 4°C and equilibrated with a 95–5% O2-CO2 gas mixture . Slices were cut with a vibratome ( Leica VT1200S ) in cutting solution and then incubated in oxygenated artificial cerebrospinal fluid ( aCSF ) composed of ( in mM ) : 126 NaCl , 20 glucose , 26 NaHCO3 , 2 . 5 KCl , 1 . 25 NaH2PO4 , 1 MgSO4 , 2 CaCl2 ( pH 7 . 35 , 310-320mOsm/L ) , initially at 34°C for 30 min , and subsequently at room temperature , before being transferred to the recording chamber where recordings were obtained at 30–32°C . Whole-cell patch-clamp recordings were performed in L2/3 of the medial prefrontal cortex ( mPFC ) neurons . Inhibitory PV-expressing interneurons , labeled with TdTomato in Ts65Dn mice crossed with Pvalb-tdTomato mice and Martinotti cells , labeled with GFP in Ts65Dn crossed with GFP-X98 mice , were identified using LED illumination ( OptoLED , Cairn Research , Faversham , UK ) . Excitatory pyramidal neurons ( PNs ) were visually identified using infrared video microscopy , as cells lacking expression of fluorescent proteins and with somatas exhibiting the classical pyramidal shape . Accordingly , when depolarized with DC current pulses PNs exhibited a typical firing pattern of regular-spiking cells . We used different intracellular solutions depending on the type of experiment and the nature of the responses we wanted to assess . To study passive properties , intrinsic excitability , AP waveform and glutamatergic spontaneous transmission , electrodes were filled with an intracellular solution containing ( in mM ) : 127 K-gluconate , 6 KCl , 10 Hepes , 1 EGTA , 2 MgCl2 , 4 Mg-ATP , 0 . 3 Na-GTP; pH adjusted to 7 . 3 with KOH; 290–300 mOsm . The estimated reversal potential for chloride ( ECl ) was approximately −69 mV based on the Nernst equation . To measure GABAergic currents elicited by perisomatic-targeting interneurons , PNs were patched using an intracellular solution containing ( in mM ) : 65 K-gluconate , 70 KCl , 10 Hepes , 1 EGTA , 2 MgCl2 , 4 Mg-ATP , 0 . 3 Na-GTP; pH adjusted to 7 . 3 with KOH; 290–300 mOsm ( the estimated ECl was approximately −16 mV based on the Nernst equation ) . For distal dendritic uIPSCs , we used a cesium-based solution containing ( in mM ) : 145 CsCl , 10 Hepes , 1 EGTA , 0 . 1 CaCl2 , 2 MgCl2 , 4 . 6 Mg-ATP , 0 . 4 Na-GTP , 5 QX314-Cl; pH adjusted to 7 . 3 with CsOH; 290–300 mOsm . Under these recording conditions , activation of GABAA receptors resulted in inward currents at a holding potential ( Vh ) of −70 mV . Voltage values were not corrected for liquid junction potential . Patch electrodes were pulled from borosilicate glass capillaries and had a typical tip resistance of 2–3 MΩ . Signals were amplified with a Multiclamp 700B patch-clamp amplifier ( Molecular Devices ) , sampled at 20–50 KHz and filtered at 4 KHz ( for voltage-clamp experiments ) and 10 KHz ( for current-clamp experiments ) . Signals were digitized with a Digidata 1440A and acquired , using the pClamp 10 software package ( Molecular Devices ) . For paired recordings , unitary synaptic responses were elicited in voltage-clamp mode by brief somatic depolarizing steps ( −70 to 0 mV , 1–2 ms ) evoking action currents in presynaptic cells . Neurons were held at −70 mV and a train of 5 presynaptic spikes at 50 Hz was applied . 3- ( 5-methylisoxazol-3-yl ) −6-[ ( 1-methyl-1 , 2 , 3-triazol-4-yl ) methyloxy]−1 , 2 , 4-triazolo[3 , 4-a]phthalazine also named L-822179 , IUPAR/BPS 4095 or PubChem CID 6918451 was synthesized by Orga-Link SARL ( Magny-les-Hameaux , France ) , according to Sternfeld et al . , 2004 as in Braudeau et al . , 2011 . Parvalbumin , SST and GFP staining were performed on 20–50 µm-thick slices . Briefly , mice were perfused with 0 . 9% NaCl solution containing Heparin and 4% paraformaldehyde ( PFA ) . Brains were cryo-protected by placing them overnight in 30% sucrose solution and then frozen in Isopentane at a temperature <-50°C . Brains were sliced with a freezing microtome ( ThermoFisher HM450 ) . Permeabilization in a blocking solution of PBT with 0 . 3% Triton and 10% Normal Goat Serum was done at room temperature for 2 hr . Slices were then incubated overnight ( 4°C ) in the same blocking solution containing the primary rabbit anti-PV antibody ( 1:1000; Thermo Scientific ) and mouse anti-SST antibody ( 1:250; Santa Cruz Biotechnologies ) . Slices were then rinsed three times in PBS ( 10 min each ) at room temperature and incubated with goat anti-rabbit and a goat anti-mouse antibody ( 1:500; Jackson IR ) coupled to Alexa-488 or 633 for 3 . 5 hr at room temperature . Slices were then rinsed three times in PBS ( 10 min each ) at room temperature and coverslipped in mounting medium ( Fluoromount , Sigma Aldrich F4680 ) . Immunofluorescence was then observed with a slide scanner ( Zeiss , Axio Scan . Z1 ) . Biocytin Fills: To reliably reconstruct the fine axonal branches of cortical neurons , dedicated experiments were performed following the classical avidin-biotin-peroxidase method . Biocytin ( Sigma ) was added to the intracellular solution at a high concentration ( 5–10 mg/ml ) , which required extensive sonication . At the end of recordings , the patch pipette was removed carefully until obtaining an inside out patch . The slice was then left in the recording chamber for at least further 5–10 min to allow further diffusion . Slices were then fixed with 4% paraformaldehyde in phosphate buffer saline ( PBS , Sigma ) for at least 48 hr . Following fixation , slices were incubated with the avidin-biotin complex ( Vector Labs ) and a high concentration of detergent ( Triton-X100 , 5% ) for at least two days before staining with 3 , 3′Diaminobenzidine ( DAB , AbCam ) . Cells were then reconstructed and cortical layers delimited using Neurolucida 7 ( MBF Bioscience ) and the most up to date mouse atlas ( Allen Institute ) . Electrophysiological and statistical analysis was performed using built-in and custom-written routines made for Igor Pro ( WaveMetrics , Lake Oswego , OR , USA ) , MATLAB R2017b 9 . 3 . 0 . 713579 Natick , Massachusetts: The MathWorks Inc; Origin ( Pro ) 2016 OriginLab Corporation , Northampton , MA , USA; Prism version 7 . 00 for Windows , GraphPad Software , La Jolla California USA; and Python Software Foundation . Python Language Reference , version 3 . 6 , available at http://www . python . org . Traces obtained from juxtacellular recording were high pass filtered ( cutoff: 5 Hz ) and spikes were detected based on threshold = 1 . 5 mV . Spike rate was estimated as the total number of spikes detected divided by the total duration of the recording . The peak time ( tpeak ) corresponding to each detected action potential was used to select the segment of LFP between: [tpeak-100 ms to tpeak+100 ms] . The instantaneous phase was estimated using Hilbert transform on decimated LFP and subsequently used to estimate phase locking . For LFP analysis , extracellular potentials were down-sampled ( 1 kHz ) and low- pass filtered ( cutoff frequency , 100 Hz ) . Power spectra were generated using a Hann window ( window length: 4096 points , 50% overlap ) . Phase locking was determined using pairwise phase consistency ( PPC ) estimation , defined as:PPCf=2 ( ∑i=1N−1∑j=i+1N ( cos ( θi ) cos ( θj ) +sin ( θi ) sin ( θj ) ) ) N . ( N−1 ) Where N is the total number of action potentials and θi is the phase of the ith spike and θj the jth . Cortical states were detected as described elsewhere ( Ruiz-Mejias et al . , 2011 ) . Briefly , filtered and decimated LFP was used to calculate UP and DOWN state likelihood decision ( or evidence ) variable ( Scomb ) based on low frequency ( <4 Hz ) oscillation phase and high frequencies ( 20–100 Hz ) composition of the LFP . To determine the thresholds to detect different cortical states ( UP , Intermediate , DOWN ) states , the distribution of the combined evidence variable , Scomb , was fitted by a mixture of three Gaussians , each representing their corresponding cortical state , UP ( highest level of the signal ) , Intermediate ( Intermediate level ) and DOWN ( lowest level of the signal ) . Periods of the combined signal , Scomb , that were above the UP threshold , μUP−LFP−3*σUP−LFP , were considered the periods of UP states . Similarly , the periods below the DOWN threshold , μDOWN−LFP−3*σDOWN−LFP , were considered the periods of DOWN states ( means and variances of the Gaussians are represented as μUP , μDOWN , and σUP , σDOWN for the up and down cortical states , respectively ) . Periods of UP and DOWN states were refined further by putting constraints on the interval between two states and duration of a state . Minimum interval between two states and duration of a state were set 50 and 70 ms , respectively . Input resistance ( Ri ) was estimated as the slope of the current to voltage relationship obtained with upon the injection of −25 , 0 and 25 pA to a cell kept at resting potential . Membrane time constant was estimated fitting the time course of Vmemb after the injection of a 2 s , −25 pA current step . We used protocols of increasing steps of current injection ( −50 to 500 pA in steps of 25 or 50 pA and 2 s duration ) . Action potentials were detected using a threshold based routine . Threshold was set at 0 mV . Firing dynamics was evaluated fitting AP frequency versus current relationship ( F-I curve ) for somatic current injections from individual cells to a logarithmic function:f ( I ) =gain*ln ( Irheobase ) Where I is the amount of injected current , the parameter gain represents the gain of the system and rheobase represents the minimal amount of current required to trigger an action potential . The first action potential evoked at rheobase was taken to measure amplitude , width and threshold . Threshold was considered as the potential at which dV/dt reached 10 mV/ms; amplitude was the difference between peak amplitude and threshold and half-width was the time interval between rise and decay phase measured at 50% of amplitude . Action potential threshold was defined as the membrane potential value ( Vm ) at which dV/dt becomes larger than 10 mV/ms . Action potential amplitude was defined as:Amp=Amppeak−threshold , where Amppeak was the AP peak potential . Action potential width was measured at 50% of amplitude . Normal distribution of samples was systematically assessed ( Shapiro-Wilkinson normality test ) . Normal distributed samples were statistically compared using two-tailed Student’s t test unless otherwise stated . When data distribution was not normal we used two-tailed Mann Whitney U-test . Compiled data are reported and presented as whisker box plots the upper and lower whiskers representing the 90th and 10th percentiles , respectively , and the upper and lower boxes representing the 75th and 25th percentiles , respectively , and the horizontal line representing the median or the mean ± s . e . m . , with single data points plotted . Differences were considered significant if p<0 . 05 ( *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 ) . Evaluation of synaptic efficiency in the dendritic inhibitory loop composed by MCs and PNs . Evaluation of synaptic efficiency in the perisomatic inhibitory loop composed by PV cells and PNs . Ts65Dn in vivo activity .

Down syndrome is a genetic disorder caused by the presence of a third copy of chromosome 21 . Affected individuals show delayed growth , characteristic facial features , altered brain development; with mild to severe intellectual disability . The exact mechanisms underlying the intellectual disability in Down syndrome are unclear , although studies in mice have provided clues . Drugs that reduce the inhibitory activity in the brain improve cognition in a mouse model of Down syndrome . This suggests that excessive inhibitory activity may contribute to the cognitive impairments . Many different neural circuits generate inhibitory activity in the brain . These circuits contain cells called interneurons . Sub-types of interneurons act via different mechanisms to reduce the activity of neurons . Identifying the interneurons that are affected in Down syndrome would thus improve our understanding of the brain basis of the disorder . Zorrilla de San Martin et al . compared mice with Down syndrome to unaffected control mice . The results revealed an increased activity in two types of inhibitory brain circuits in Down syndrome . The first contains interneurons called Martinotti cells . These help the brain to combine inputs from different sources . The second contains interneurons called parvalbumin-positive basket cells . These help different areas of the brain to synchronize their activity , which in turn makes it easier for those areas to exchange information . By mapping the changes in inhibitory circuits in Down syndrome , Zorrilla de San Martin et al . have provided new insights into the biological basis of the disorder . Future studies should examine whether targeting specific circuits with pharmacological treatments could ultimately help reduce the associated impairments .

2020

Alterations of specific cortical GABAergic circuits underlie abnormal network activity in a mouse model of Down syndrome

N-methyl-d-aspartate receptors ( NMDARs ) are an important receptor in the brain and have been implicated in multiple neurological disorders . Many non-selective NMDAR-targeting drugs are poorly tolerated , leading to efforts to target NMDAR subtypes to improve the therapeutic index . We describe here a series of negative allosteric NMDAR modulators with submaximal inhibition at saturating concentrations . Modest changes to the chemical structure interconvert negative and positive modulation . All modulators share the ability to enhance agonist potency and are use-dependent , requiring the binding of both agonists before modulators act with high potency . Data suggest that these modulators , including both enantiomers , bind to the same site on the receptor and share structural determinants of action . Due to the modulator properties , submaximal negative modulators in this series may spare NMDAR at the synapse , while augmenting the response of NMDAR in extrasynaptic spaces . These modulators could serve as useful tools to probe the role of extrasynaptic NMDARs . N-methyl-d-aspartate receptors ( NMDARs ) are a subtype of ionotropic glutamate receptors that are broadly expressed in the brain and are important for normal development , neuronal plasticity and memory formation ( Traynelis et al . , 2010; Paoletti et al . , 2013 ) . NMDARs contribute a slow , Ca2+ permeable component to fast excitatory neurotransmission that is voltage dependent by virtue of its sensitivity to pore block by extracellular Mg2+ ( Traynelis et al . , 2010 ) . The typical NMDAR is comprised of 2 GluN1 and 2 GluN2 subunits , creating the potential for diversity given that there are eight splice variants of GluN1 and 4 independent genes encoding GluN2 subunits ( A-D ) ( Traynelis et al . , 2010 ) . The regulation and functional properties of the NMDAR are controlled by the subunits incorporated into the receptor ( Monyer et al . , 1994; Vicini et al . , 1998; Vance et al . , 2012 ) . NMDARs are expressed at most excitatory synapses , and are also found in peri- and extrasynaptic locations ( Sans et al . , 2000; Steigerwald et al . , 2000; Groc et al . , 2006; Traynelis et al . , 2010 ) . Although NMDARs have been studied extensively , there are still important questions about the different roles that NMDARs may play given differences in subunit composition and synaptic localization ( Hardingham and Bading , 2010; Gladding and Raymond , 2011; Paoletti et al . , 2013 ) . Pharmacological approaches can be useful for probing these questions , but tool compounds to differentiate these NMDAR subtypes have been limited ( Ogden and Traynelis , 2011; Monaghan et al . , 2012; Santangelo et al . , 2012; Strong et al . , 2014; Zhu and Paoletti , 2015 ) . Whereas NMDAR receptors have been implicated in many neurological diseases , there remains a dearth of clinically-approved drugs that target NMDARs ( Kalia et al . , 2008; Traynelis et al . , 2010; Paoletti et al . , 2013; Strong et al . , 2014 ) . Multiple endogenous and exogenous modulatory sites in the NMDAR recently have been described . In addition , nearly complete NMDAR structures obtained using crystallographic approaches are now available ( Karakas and Furukawa , 2014; Lee et al . , 2014 ) , and these data , together with recent cryo-EM structures in the receptor super family ( Twomey and Sobolevsky , 2018; Twomey et al . , 2017a; Twomey et al . , 2017b ) , illustrate the overall topography of the NMDAR and suggest a mechanism for some allosteric modulators , such as GluN2B-selective ifenprodil ( Tajima et al . , 2016; Zhu et al . , 2016 ) . However , a complete understanding of how other modulatory sites operate and can be exploited has been elusive . Advances in the understanding of specific roles of particular NMDAR subunits have led to renewed interest in targeted therapeutic intervention , and recent work has yielded a growing tool box of novel ligands that act on NMDARs with diverse sites and mechanisms ( Ogden and Traynelis , 2011; Monaghan et al . , 2012; Hackos and Hanson , 2017 ) . To date , there are positive and negative modulators that bind to the amino-terminal domain ( ATD ) , agonist binding domain ( ABD ) , or transmembrane domain ( TMD ) . Each new class of compounds discovered enriches our understanding of the function of NMDAR and how these receptors can be modulated . The information gained from these modulators has the potential to provide insight into both NMDAR function and therapeutic strategies to treat complex neurological diseases . This study describes a series of compounds that includes both positive allosteric modulators ( PAMs ) as well as negative allosteric modulators ( NAMs ) of NMDAR function ( Katzman et al . , 2015 ) . A set of aliphatic substitutions off an ester linkage to the scaffold interconverts these compounds between positive and negative allosteric modulators . Remarkably , the difference between the positive and negative modulation to the same chemical scaffold was influenced by the addition or removal of individual methyl groups . These modulators appear to bind to a shared site to bring about opposing actions , and share mechanistic features , such as agonist dependence and enhancement of agonist potency . The spectrum of properties in this series of modulators could serve as useful tool compounds for probing the role of NMDARs in circuits in both healthy brain and in neuropathological situations . We have previously described the structure-activity relationship ( SAR ) of a series of negative allosteric pan-NMDAR modulators that contained an amidothiophene core , the most potent possessing a tetrahydrobenzothiophene , with different alkyl and aryl substitutions connected via an ester linkage at the 3-position ( Katzman et al . , 2015 ) . These compounds inhibited the response to saturating concentration of co-agonists at NMDARs expressed in Xenopus laevis oocyte experiments , typified by EU1794-2 ( compound 4 in Katzman et al . , 2015 ) . We describe here a new set of closely related analogues that can either potentiate or inhibit responses depending on subtle changes in structure and agonist concentration . As shown in Figure 1 and Table 1 , EU1794-4 contained an ethyl ester similar to EU1794-2 and lacked a methyl substituent on the tetrahydrobenzothiophene core . EU1794-4 potently inhibited GluN1/GluN2D with a substantial residual current remaining at saturating concentrations ( Figure 1B , C , Table 1 ) . Interestingly , ester substitutions that had a larger calculated functional group volume than the ethyl ester in EU1794-4 potentiated NMDAR responses to saturating concentrations of glutamate and glycine . For example , the isopropyl ester ( EU1794-5 ) potentiated GluN1/GluN2D responses to nearly 200% of control ( Figure 1A–C ) . Restoration of the methyl to the tetrahydrobenzothiophene core restored negative allosteric modulation ( EU1794-19 , Figure 1B , C ) . Thus , the direction of modulation ( positive or negative ) could be determined by the size of the alkyl ester and substitution to the tetrahydrobenzothiophene core . The size of the ester-linked substituent controlled the extent and potency of positive modulation . The t-butyl ester EU1794-27 potentiated GluN1/GluN2D NMDARs with greater efficacy and higher potency , increasing responses to maximally effective glutamate and glycine to 250% of control with an EC50 value of 2 . 4 µM ( Figure 1D , E ) . The benzyl ester ( EU1794-25 ) also strongly potentiated responses with a potency similar to EU1794-27 ( Figure 1D , E ) . However , EU1794-29 , which had a longer substitution , n-butyl , reduced both the maximal potentiation and potency , suggesting an optimal substituent size and shape ( Figure 1D , E ) . The tetrahydrobenzothiophene-containing NAMs reported here inhibit all diheteromeric NMDARs without substantial subunit selectivity , similar to those that we previously described ( Katzman et al . , 2015 ) . By contrast , the novel PAMs described here show distinct GluN2 dependence ( Table 1 ) . All PAMs are active at GluN1/GluN2C and GluN1/GluN2D , but do not enhance the response of GluN1/GluN2A to maximally effective concentrations of agonist , in some cases resulted in inhibition of these responses . Most PAMs were also capable of potentiating GluN1/GluN2B ( Table 1 , Figure 2A ) . As seen with other series of NMDAR PAMs ( Malayev et al . , 2002; Horak et al . , 2004; Horak et al . , 2006; Hackos et al . , 2016; Wang et al . , 2017 ) , allosteric modulation can be dependent on agonist concentrations . Thus , we assessed the ability of EU1794-27 to modulate NMDAR responses activated by sub-saturating co-agonist concentrations , clearly exemplified by modulation of GluN1/GluN2C ( Figure 2B , Table 2 ) . In these conditions , EU1794-27 positively modulated all NMDAR subtypes and enhanced previously potentiated subtypes to a greater extent than in saturating agonist ( compare Figure 2A and C ) . We hypothesized that this agonist-dependence was due to EU1794-27 altering the agonist potency . Therefore , we determined the glutamate and glycine EC50 values in the absence and presence of EU1794-27 ( Figure 3A , B , Table 3 , Figure 3—figure supplement 1 ) . EU1794-27 produced modest but significant decreases ( higher potency ) in the EC50 values for both glutamate and glycine at all NMDARs ( Table 3 , Figure 3—figure supplement 1 ) . Given this effect by the PAMs in this series , we subsequently considered whether the NAMs in the series shared this mechanism . We selected EU1794-4 for use in evaluating actions on agonist potency since it retains a large steady-state current even for receptors that have bound EU1794-4 ( saturated inhibition is between 40–70% of control , Figure 2D ) . We co-applied EU1794-4 during responses stimulated by sub-saturating concentrations of agonist ( glutamate and glycine concentrations that resulted approximately in a EC20 response ) , which resulted in positive modulation ( Figure 2E , F , Table 2 ) . Furthermore , the negative modulator EU1794-4 enhanced the glutamate and glycine potencies ( Figure 3C , D , Table 4 , Figure 3—figure supplement 1 ) . Interestingly , positive modulation elicited by 10 µM EU1794-4 is dependent on the sub-saturating agonist concentrations used ( Table 5 ) . EU1794-4 inhibited GluN1/GluN2C responses to saturating concentrations of agonist to 54% of control , and positively modulated equivalent EC30 responses ( equal effective concentrations of glutamate and glycine concentrations that when applied resulted in a 30% of a maximal response ) to 140% of control , and positively modulated average equivalent EC3 responses by 660% ( Table 5 ) . We did not observe augmentation of sub-saturating agonist responses by EU1794-2 , most likely due to its greater extent of inhibition ( Figure 2—figure supplement 1 ) . However , the degree of inhibition produced by EU1794-2 was reduced on sub-saturating responses at GluN1/GluN2C from 16% to 45% of control ( Table 5 ) . We studied NMDARs expressed in HEK293 cells to investigate the time course of modulator action . Concentration-dependent association of EU1794-2 , EU1794-4 and EU1794-27 and concentration-independent disassociation were evaluated by co-applying modulator with glutamate and glycine ( Figure 4A , B ) . These modulators were tested on GluN1/GluN2A and GluN1/GluN2D to see whether there were differences in modulation given the distinct properties of these NMDAR subtypes . We analyzed the concentration-dependence of the exponential time course describing the onset of action during co-application with saturating concentrations of glutamate plus glycine to determine modulator association and dissociation rates . From these we calculated the kinetically-determined affinity constant ( Kd ) , which we found to be similar to the EC50 values determined from concentration-response experiments . EU1794-2 Kd was 1 . 1 µM at GluN1/GluN2D ( Figure 4A , B ) . Complex actions of EU1794-4 and EU1794-27 were observed , with two temporally-distinct phases of modulation evident for the association of these modulators ( Figure 4A ) . Both compounds produced a rapid inhibition followed by a slowly developing potentiating phase for EU1794-27 . The rapid association rate determined during the rapid inhibition produced by EU1794-4 was approximately three times faster than EU1794-2; similarly , the dissociation rate was also faster for EU1794-4 than EU1794-2 . Quantitative analysis of the slower phases was challenging due to its lower signal-to-noise ratio . Likewise , the rapid inhibitory phase was also difficult to measure for EU1794-27 because its rapid time course was convolved with the potentiation time course . Additionally , the depotentiation time course is preceded by a transient enhancement of the current response , followed by a relaxation to the pre-modulation level . The time course for potentiation and depotentiation were best fit with an exponential function summed with an additional linear component after the rapid phase subsided . Kd determined from the association and dissociation rates of EU1794-4 was 4 . 2 µM and EU1794-27 was 4 . 3 µM ( Figure 4B ) . All modulatory effects were independent of voltage ( Figure 4—figure supplement 1A ) . The steady state modulator responses from HEK293 cells approximately match the concentration-response relationship determined from TEVC recordings from X . laevis oocytes ( Figure 4A , inset graphs ) . The time-course of modulator binding of EU1794-2 , EU1794-4 and EU1794-27 was also determined for GluN1/GluN2A ( Figure 4C , D ) . Higher Kd values ( lower potency ) were determined for each molecule at GluN1/GluN2A as compared to GluN1/GluN2D that parallel EC50 values determined using X . laevis oocytes . Similar to oocyte data , robust inhibition was produced by EU1794-2 , modest inhibition was produced by EU1794-4 , and transient inhibition and recovery was observed by the application of EU1794-27 . Interestingly , the extent of steady-state modulation for this series appeared to be dependent on the level of desensitization of the receptors at the time of modulator application . When classifying the cells as either having high or low levels of desensitization ( using 35% steady-state/peak response as a cut-off ) , the extent of modulation by 10 µM was significantly different for EU1794-27 ( p=0 . 02 , unpaired t-test , N = 4 , 2 , respectively ) . The low desensitized group ( 40% average desensitization ) was modulated by 107% , whereas the high desensitized group ( 10% average desensitization ) was modulated to 247% of control by 10 µM EU1794-27 ( Figure 4C and Figure 4—figure supplement 1F ) . Given that other NMDAR modulators have displayed agonist-dependence ( Petrovic et al . , 2005; Acker et al . , 2011; Hansen and Traynelis , 2011; Borovska et al . , 2012; Vyklicky et al . , 2015; Wang et al . , 2017 ) , we performed rapid solution exchange experiments to examine if these modulators had different affinities at agonist-bound and apo receptors ( Figure 5A , B ) . The instantaneous current response to a rapid step into glutamate plus glycine ( from glycine alone ) gives an estimate of whether a modulator was pre-bound to receptors with only glycine ( but not glutamate ) bound . The immediate NMDAR activation after a rapid switch to a solution containing glutamate and glycine should occur faster than modulator binding . We chose to test for agonist-dependence using GluN1/GluN2D since they lack desensitization , which would complicate interpretation of this modulator property . We found that the peak current was similar when cells were preincubated in glycine with or without 3 µM EU1794-2 , suggesting that EU1794-2 does not bind appreciably to the receptor in the absence of glutamate ( Figure 5A , C ) . After the rapid step into glutamate where glycine and modulator were pre-exposed , we observed a relaxation to a new response level that was similar in amplitude to that observed with steady state co-application of glutamate plus glycine and modulator ( Figure 5A , C ) . Interestingly , we also observed a similar effect for the converse experiment , in which we pre-applied glutamate with or without EU1794-2 , followed by a rapid step into glutamate plus glycine and EU1794-2 ( Figure 5B , C ) . We interpret these data to suggest that EU1794-2 associates with the receptor with higher affinity after glutamate and glycine binding . We repeated these use-dependent experiments with EU1794-27 , which yielded a similar result , although a slightly different experimental design was required for consistent responses . The pre-application of glycine with EU1794-27 produced a similar level of immediate activation following glutamate application ( Figure 5D , E ) . This was then followed by the complex actions observed when EU1794-27 was applied to steady-state responses of NMDARs ( Figure 5D , right panel ) . Using the same protocol to pre-apply glutamate became problematic for some cells due the degree to which EU1794-27 enhances agonist potency , a feature exemplified by the prolongation of deactivation . This action of EU1794-27 was able to render nanomolar contaminate levels of glycine ( when present ) more active , which necessitated a different experimental design ( Figure 5—figure supplement 1A ) . When EU1794-2 was used in this experimental paradigm , inhibition of the contaminate level of activity was observed upon modulator application ( Figure 5—figure supplement 1B ) . Nevertheless , we still observed a similar result with pre-application of glycine and EU1794-27 , with the instantaneous response reaching the control level , followed by a slow relaxation to a new potentiated level that reflected the time course for association of EU1794-27 after binding of both glutamate and glycine ( Figure 5E ) . This result illustrates a requirement for glycine to be bound to the receptor for high-affinity binding of EU1794-27 . To circumvent any ambiguity associated with glycine contamination , we utilized another experimental design described by Vyklicky et al . ( 2015 ) . Glutamate was pre-applied with 7-CKA ( 100 µM ) to antagonize the glycine site , blocking any occupancy by contaminant glycine . The receptor was activated by switching to a solution that lacked 7-CKA and contained glycine plus glutamate ( with or without modulator , Figure 5—figure supplement 1C ) . To test the ability of EU1794-27 to bind to the receptor with the GluN1 ABD bound to antagonist , EU1794-27 ( 3 µM ) was added to the solution containing 7-CKA . Upon the switch from the 7-CKA/EU1794-27 solution containing glutamate to a solution just containing glutamate and glycine , no detectable change in the response rise time or peak amplitude was observed . If EU1794-27 bound during the 7-CKA phase prior to agonist binding , we would have expected an increased instantaneous peak current upon switching to glutamate plus glycine given that the dissociation of EU1794-27 is slower than 7-CKA ( compare the steady state response and the 7-CKA response in Figure 5—figure supplement 1C ) . Additionally , comparable experiments were performed with 100 µM APV with similar results , as APV stabilizes the GluN2 agonist binding domain ( ABD ) open-cleft conformation , and thereby prevents the pre-binding of EU1794-27 ( Figure 5—figure supplement 1D ) . When EU1794-27 ( 3 µM ) was included in all solutions , potentiation was observed but APV unbinding was a required prior to modulation ( Figure 5—figure supplement 1D ) . Together these data suggest that the EU1794 series is capable of high affinity binding only when glutamate and glycine are bound to the receptor . To identify the molecular determinants of action for the EU1794 series , we evaluated the ability of GluN2 ATD deletion , GluN1 ATD splice variants , and co-application of known modulators to alter the effects of the EU1794 modulators . The ATD harbors the binding site for the GluN2B-selective negative allosteric modulator ifenprodil . Deletion of the ATD from GluN2A , GluN2B , or GluN2C had no effect on the actions of the PAM EU1794-27; deletion of the ATD from GluN2D reduced but did not eliminate potentiation ( Figure 6—figure supplement 1A ) . Similarly , inclusion of 21 residues in the ATD encoded by alternatively spliced GluN1 exon5 only slightly altered the extent of potentiation of EU1794-27 , but was without effect on EC50 for potentiation of GluN2B- and GluN2D-containing NMDARs activated by saturating agonist ( Figure 6—figure supplement 1B ) . We previously described a similar result for the negative allosteric modulator EU1794-2 ( Katzman et al . , 2015 ) . These data are consistent with minimal involvement of the GluN1 or GluN2 ATD in the actions of EU1794 modulators . We next screened for interaction with known modulators to focus our search for the molecular determinants of EU1794 series modulation ( Mullasseril et al . , 2010; Acker et al . , 2011; Hansen and Traynelis , 2011; Hansen et al . , 2012; Khatri et al . , 2014; Ogden et al . , 2014; Hackos et al . , 2016; Tajima et al . , 2016; Yi et al . , 2016 ) ( Figure 6—figure supplement 2A ) . In these experiments , a known positive or negative modulator was co-applied with either EU1794-2 or EU1794-27 , with each pair always containing one PAM and one NAM . If there is no interaction between paired modulators , their combined activity should be predicted by multiplying the extent of their independent actions . Co-application of the modulator pairs ifenprodil/EU1794-27 , EU1794-2/CIQ and EU1794-2/PYD-106 produced levels of modulation that largely could be predicted from their independent actions ( Figure 6—figure supplement 2B , top row ) . Modest differences from predictions were observed with modulators that bind to the ABD interface , TCN-201/EU1794-27 , EU1794-2/GNE-6901 and EU1794-2/GNE-0723 ( Figure 6—figure supplement 2B , bottom left ) . Co-application of the GluN2C/GluN2D-selective negative allosteric modulators QNZ46 and DQP-1105 paired with EU1794-27 resulted in the greatest divergences from predictions , raising the possibility that these modulators have partially overlapping binding sites ( Figure 6—figure supplement 2B , bottom right ) or similar downstream mechanisms . We also examined the ability of EU1794-2 and EU1794-27 to modulate NMDARs harboring mutations within the structural determinants for other known allosteric modulators . Mutations in GluN1 ( I519A , R755A ) and GluN2A ( L780A , G786A ) that block TCN-201 inhibition were evaluated for effects on EU1794 series of modulators ( Hansen et al . , 2012 ) . NMDARs that contained GluN1/GluN2A ( L780A ) , GluN1/GluN2A ( G786A ) , GluN1 ( R755A ) /GluN2D , GluN1 ( I519A ) /GluN2D were equally sensitive as wild type receptors to inhibition by EU1794-2 or potentiation by EU1794-27 ( Figure 6—figure supplement 3A ) . GluN1/GluN2A ( E530A ) , GluN1/GluN2A ( V783W ) , GluN1 ( Y535W ) /GluN2A , GluN1 ( Y535V ) /GluN2D and GluN1 ( Y535W ) /GluN2D , which reduce the actions of GNE-6901 and GNE-0723 ( Hackos et al . , 2016 ) , produced no significant effects on inhibition by EU1794-2 or potentiation by EU1794-27 ( Figure 6—figure supplement 3B ) . Inhibition by EU1794-2 of GluN1/GluN2C ( K470G ) and GluN1/GluN2C ( S472T ) , which block PYD-106 potentiation of GluN1/GluN2C ( Khatri et al . , 2014 ) , was similar to wild type NMDARs ( Figure 6—figure supplement 3C ) . Interestingly , inhibition by EU1794-2 and potentiation by EU1794-27 was not significantly changed by GluN1/GluN2D ( Q701Y ) and GluN1/GluN2D ( L705F ) , mutations that appear to confer subunit selectivity for GluN2C/D over GluN2A/B for QNZ-46 and DQP-1105 ( Figure 6—figure supplement 3D ) ( Acker et al . , 2011; Hansen and Traynelis , 2011 ) . The result obtained with the QNZ-46/DQP-1105 interaction test suggested that the two residues we evaluated for QNZ-46 and DQP-1105 insufficiently probed the structural determinants of action for these compounds . We therefore examined a GluA2 AMPA receptor structure bound to CP-465 , 022 , which shares a core scaffold with QNZ modulators ( Yelshanskaya et al . , 2016 ) . Residues identified as being important for CP-465 , 022 binding in Yelshanskaya et al . , 2016 were aligned to the GluN1 and GluN2D subunits to map this modulatory site onto the NMDAR subunits , in addition to critical residues for GNE-9278 in this same region ( Figure 6A , Wang et al . , 2017 ) . This broader range of residues were located on the pre-M1 , M3 , and pre-M4 regions of both GluN1 and GluN2D , which have previously been suggested to cooperate to control gating ( Ogden and Traynelis , 2013; Alsaloum et al . , 2016; Chen et al . , 2017; Ogden et al . , 2017; Yelshanskaya et al . , 2017 ) . These residues were suggested by Yelshanskaya et al . , 2016 to constitute a binding site in homomeric GluA2 AMPARs . However , mapping the homologous residues onto NMDAR structures yields two pockets given the multimeric subunit architecture , one of which consists of the residues of GluN1 and the other of residues of GluN2 . Certain residues of M3 , depending on their position on the helix , could point towards either pocket , rendering the two pockets to be lined by a mixture of GluN1 and GluN2 residues . 15 GluN1 and 15 GluN2D residues in these two regions that probed these pockets were identified and mutated to allow a test of the contribution of each residue to EU1794-2 inhibition and EU1794-27 potentiation . Inhibition by EU1794-2 was significantly altered by substitutions at six residues in GluN1 and 2 GluN2D residues ( Figure 6B ) , which were found on pre-M1 and M3 regions of both GluN1 and GluN2D . Potentiation by EU1794-27 was more labile , being altered in mutations at 9 GluN1 residues and 6 GluN2D residues ( Figure 6B ) . Residues that perturbed the actions of EU1794-27 were spread across all regions tested except for the GluN2D pre-M1 . Modulation was observed to be altered in three different ways by the mutations studied here: activity could be reduced , increased , or inverted . The distributions of residues that altered modulation by EU1794-2 and EU1794-27 shows clear overlap ( Figure 6C ) . Inhibition by EU1794-2 was altered primarily by mutations in GluN1 , whereas potentiation by EU1794-27 was perturbed by residues both in GluN1 and in the M3 helix of GluN2D . Interestingly , the mutations that invert activity of EU1794-2 and EU1794-27 were distinct , but in some cases were in close proximity . For example , EU1794-2 was inverted by GluN1-F654A , whereas mutation of the adjacent residue GluN1-L655A/Y inverted the modulatory action of EU1794-27 ( Figure 6D ) . EU1794-27 potentiation was also converted to inhibition by four mutations at residues residing on the 3 areas of GluN1 that were investigated ( pre-M1 , M1 , M3 ) and also on the M3 GluN2D helix , which were in close proximity to each other ( Figure 6D ) . EU1794-27 actions on GluN1-L655A/Y and GluN2D-M678Y , which are homologous residues immediately downstream of the SYTANLAAF motif , resulted in opposite effects ( GluN1-L655A/Y converts EU1794-27 to an inhibitor and GluN2D-M678Y increases the potentiation of EU1794-27 ) . We interpret these results to suggest that the activity of both PAMs and NAMs of the EU1794 series is dependent on multiple residues in the GluN1 subunit , some of which are overlapping . Furthermore , potentiation by EU1794-27 is dependent on a wider range of GluN1 and GluN2 residues . One possible way to account for this would be if both modulators bound near the GluN1 pre-M1 helix , with potentiator actions dependent on the nearby GluN2 M3 residues associated with the GluN1 pocket ( Figure 6A , C ) . To assess whether mutated residues impacted EU1794 series modulation through altering receptor open probability , we estimated mutant open probability using MTSEA modification of an introduced cysteine residue ( Jones et al . , 2002; Yuan et al . , 2005 ) . The open probability of GluN2D-containing receptors is very low and may not allow for precise determination of the effects of mutations , thus we used GluN2B-containing receptors for comparison . GluN1 mutations were expressed with a GluN2B A7C construct ( See Figure 6—figure supplement 4 legend ) and GluN2 mutations were expressed with a GluN1 A7C construct . Several residues were assayed that resulted in reversed , enhanced , or blocked modulator activity ( Figure 6—figure supplement 4A ) . GluN1-P557A , which blocked activity of both EU1794-2 and EU1794-27 , had an increased open probability as compared to wild type . GluN1-F654A , which converted EU1794-2 into a potentiator and enhanced EU1794-27 potentiation , had a similar open probability as compared to wild type ( Figure 6—figure supplement 4A ) . Interestingly , mutation of the homologous GluN2 residue ( GluN2-F653A ) , which reduced EU1794-27 potentiation , also lowered open probability as compared to wild type ( Figure 6—figure supplement 4B ) . Figure 6—figure supplement 4 shows that GluN1-L665A and GluN1-N803A , which both converted EU1794-27 into an inhibitor , had significantly lower open probabilities as compared to wild type NMDARs ( Figure 6—figure supplement 4A ) . These data suggest that altered EU1794 modulator action by mutated residues is not directly dependent on alterations in open probability . Given the similarity in chemical structure between positive and negative modulators in this series , we hypothesized that they might have overlapping binding sites . In order to conduct a detailed functional analysis of competition between allosteric modulators , we first used modeling to examine a receptor’s functional response to co-application of positive and negative modulators acting at the same or different sites . We evaluated two models of modulator binding built into a scheme first proposed by ( Lester and Jahr , 1992 ) . Model 1 describes receptors with a single binding site that can accommodate either PAM or NAM , but not both . Model 2 describes a receptor to which both modulators can bind at the same time in different sites ( Figure 7A , Figure 7—figure supplement 1 ) . We evaluated hypothetical responses from these models ( see Materials and methods , Figure 7—figure supplement 1A–C ) . Responses to maximally effective concentrations of agonist were simulated at various concentrations of the positive and negative modulators . When the modulators compete for the same binding site , they act similarly to stated theory about competitive antagonists ( Arunlakshana and Schild , 1959; Christopoulos and Kenakin , 2002 ) , with the PAM causing a rightward shift of the NAM IC50 and generating a linear relationship in dose ratio analysis ( Figure 7B , Figure 7—figure supplement 1D , E , Model 1 ) . By contrast , there is no apparent EC50 shift when the two modulators are capable of binding simultaneously ( Figure 7C , Figure 7—figure supplement 1D , E , Model 2 ) . Additional simulations showed the reciprocal effect of NAM on PAM EC50 ( Figure 7—figure supplement 1F , G ) . We subsequently performed competition experiments to test the hypothesis that this series of PAMs and NAMs compete for a mutually exclusive modulatory pocket ( Figure 7D ) . We observed that fixed concentrations of the PAM EU1794-27 caused parallel shifts in the concentration-response curve of the NAM EU1794-2 ( Figure 7E , Table 6 ) . A similar phenomenon was shown for the reverse , as fixed concentrations of EU1794-2 caused parallel shifts in the EU1794-27 concentration-response curves ( Table 6 ) . These results closely match Model 1 , where PAMs and NAMs bind in a mutually exclusive fashion . This suggests that the opposing modulators in the EU1794 series share overlapping binding sites instead of the coincidence that the subtle chemical differences confer unique binding sites . The enantiomers of EU1794-27 were separated to determine whether the racemic activity reflected the actions of only one enantiomer ( see Supplemental methods ) . ( - ) -EU1794-27 potentiated NMDAR responses similarly to the racemic mixture ( Figure 8—figure supplement 1A , Table 7 ) . By contrast , ( + ) -EU1794-27 exhibited only weak potentiation of GluN1/GluN2C and GluN1/GluN2D , which may be due to the purity achieved via chiral separation ( Table 7 ) . We previously reported ( Katzman et al . , 2015 ) that purified enantiomers were likely to racemize in aqueous solutions , which should proceed by a first-order reaction dependent on multiple factors such as ionic strength , pH , buffer , etc . ( Smith et al . , 1978 ) . Thus , to quantitatively determine if racemization would impact our experiments using the enantiomers , we performed a functional assessment of this property in our standard experimental solution . We observed racemization of the enantiomers of EU1794-27 with a half-life of 197 min ( Figure 8—figure supplement 1B , Table 7 ) . Both enantiomers of EU1794-4 inhibited NMDAR responses , but with different potencies ( Figure 8—figure supplement 1C , Supplementary File 2 ) . A similar rate of racemization was observed for the enantiomers of EU1794-4 ( half-life of 196 min , Figure 8—figure supplement 1D ) . All enantiomer studies , other than the racemization time-course , were performed rapidly to minimize any racemization , being completed in less than 80 min after making the aqueous solution of modulator . Similar to results in oocytes , when the enantiomers of EU1794-27 were applied to steady-state GluN1/GluN2D responses in HEK cells , ( - ) -EU1794-27 potentiated responses whereas ( + ) -EU1794-27 had a slight inhibitory effect ( Figure 8A ) . In additional , 10 µM of both enantiomers of EU1794-4 had similar inhibitory actions on GluN1/GluN2D responses to maximally effective concentrations of glutamate and glycine ( Figure 8B ) . To assess whether both enantiomers bound to similar or distinct sites on the NMDAR , we performed single concentration competition experiments . Co-application of each enantiomer of EU1794-4 with the ( - ) -EU1794-27 resulted in a degree of modulation that was significantly different than that predicted for independent sites of action at both GluN1/GluN2A and GluN1/GluN2D ( Figure 8C ) . Interestingly , the competition by EU1794-4 was not dependent on the direction of modulation by ( - ) -EU1794-27 , which had a potentiating action at GluN1/GluN2D ( Figure 8C , top panels ) and an inhibitory effect at GluN1/GluN2A ( Figure 8C , bottom panels ) . Although there remain potential caveats , the evidence suggests that the co-application of ( - ) -EU1794-27 with either enantiomer of EU1794-4 displays mutual exclusivity in modulator binding . In the evaluation of the kinetic properties of modulator action , we observed complex actions of EU1794-4 and EU1794-27 , which could arise from multiple binding sites , enantiomers of these compounds , or could reflect distinct modulator-dependent mechanistic actions from occupancy of a single binding site . The modulation by EU1794-4 and EU1794-27 is voltage independent , eliminating potential channel block within the ion channel pore by potential cationic species as a confounding site that contributes to the observed effects . Given that there is high homology between GluN1 and GluN2 , especially in the ABD and the TMD , a reasonable hypothesis is that multiple binding sites exist for EU1794 series modulators in homologous regions on GluN1 and GluN2 subunits . However , the mutagenesis data argue against this idea , given that the residues at which mutations perturbed the actions of EU1794-2 and EU1794-27 overlapped and clustered around the pre-M1 , M3 , and M4 of the GluN1 subunit . Whereas there were a few residues of the GluN2 M3 helix that influenced EU1794-27 modulation , the structural NMDAR models suggest that these residues face the GluN1 M3 helix , and thus could interact with the pocket adjacent to the GluN1 pre-M1 helix . Additionally , it’s likely the M3 helixes of both GluN1 and GluN2 , which are in close contact , act in concert with one another to control rapid pore opening or closing . If these residues identified by mutagenesis controlled conformational changes downstream of the EU1794 binding site ( s ) , it would limit potential binding site candidates to the interface between the GluN1 and the GluN2 ABDs , which would then have only two identical binding sites in diheteromeric NMDARs . Therefore there would be less potential for non-identical binding sites that contribute to the mixed actions that are observed for EU1794-27 and EU1794-4 . For these reasons , the idea that the positive and negative allosteric actions reflect modulator specific mechanisms from occupancy of a single site seems the most plausible interpretation . Evaluation of the effects of enantiomers provides further insight into the binding site of the EU1794 series . EU1794-27 exhibits strong stereoselectivity , with the ( - ) enantiomer showing a typical potentiation time course and the ( + ) enantiomer producing inhibition . The enantiomers of EU1794-4 are both capable of producing inhibition but with the ( - ) enantiomer being more potent ( 3–10 fold ) than the ( + ) enantiomer . Additionally , both enantiomers of EU1794-4 appear to compete with EU1794-27 for access to the binding site , suggesting that enantiomers may interact differently with the same binding site . An alternative explanation for lack of additivity of the effects of the two compounds could be that there are shared residues downstream of the PAM and NAM binding sites that mediate their effects; additional studies are required to evaluate this hypothesis . However , the idea that the enantiomers of the EU1794 series act at the same site suggests their complex actions on NMDARs expressed in HEK is a mixture of receptors bound to one or the other enantiomer of racemic EU1794-27 and EU1794-4 . The available enantiomeric data further support the idea that positive and negative modulators within the EU1794 series share a single or overlapping binding site . An increasing number of NMDAR and AMPAR modulators have been identified with structural determinants of action that reside in transmembrane linker regions and extracellular portions of the transmembrane domain ( Mullasseril et al . , 2010; Acker et al . , 2011; Hansen and Traynelis , 2011; Ogden and Traynelis , 2013; Yelshanskaya et al . , 2016; Swanger et al . , 2018; Wang et al . , 2017 ) . Other cell surface receptor families have bi-directional modulator pockets , including multiple GPCRs as well as the benzodiazepine binding site in GABA-A receptors ( Barnard et al . , 1998; Rudolph and Knoflach , 2011; Wootten et al . , 2013 ) . Among all ionotropic glutamate receptor modulators interacting with this region of the receptor , there are positive ( CIQ , GNE-9278 ) and negative ( DQP-1105 , QNZ-46 , CP-465 , 022 , GYKI 52466 ) modulators with diverse scaffolds . However , the EU1794 compounds represent the first bidirectional NMDAR modulator series with structural determinants of activity within this region , low micromolar potency and with clear rules to control modulation . The similarity in structure of positive and negative modulators illustrates how subtle differences in the ligand can interconvert functional actions between inhibition and potentiation . It is unclear precisely how the different sized ester alkyl chain substitutions of these analogs bring about opposing actions , further investigation of the structure activity relationship may elucidate these details . The ability to potentiate NMDAR seems to be unique to ( - ) -EU1794-27 , which may have a specific interaction with the receptor achieved only by its stereoselective active pose in the binding pocket . Work with this series may lead to an understanding of the mechanistic link between the PAMs and NAMs that interact in this portion of the receptor . The EU1794 series has a property of agonist-dependence , requiring both glutamate and glycine to be bound before the modulator binding site adopts a high potency orientation for members of the series . Relatedly , use-dependence is a property of open channel blockers ( e . g . memantine , MK-801 , ketamine , etc . ) that inhibit the receptor through interactions within the ion permeation path , and thus rely on pore opening ( Traynelis et al . , 2010 ) . Given that the EU1794 series are allosteric modulators are not voltage-dependent , the mechanism of their agonist-dependence is unclear . Previously , the NAMs QNZ-46 , DQP-1105 , and NAB-14 have been reported to show varying degrees of glutamate- but not glycine-dependence ( Acker et al . , 2011; Hansen and Traynelis , 2011; Swanger et al . , 2018 ) . Moreover , neurosteroid derivatives with NAM activity have also been shown to have glutamate- and glycine-dependence ( Vyklicky et al . , 2015 ) , and the PAM GNE-9278 was reported to be glutamate-dependent ( Wang et al . , 2017 ) . Thus , the EU1794 series is the first series of positive and negative modulators that has been shown to possess the property of being both glutamate- and glycine-dependent . Our working hypothesis is that the EU1794 series of modulators requires conformational changes in both GluN1 and GluN2 subunits that reflect pre-gating or gating transitions , after which its affinity for its binding site is increased . However , we cannot rule out at this time the possibility that the pore must open to increase modulator binding . Resolving the specific mechanism of the EU1794 may lead to a more complete understanding of the activation transitions NMDAR . We believe that EU1794-4 highlights a novel sub-class of NMDAR modulators that have the capability to selectively act at extrasynaptic NMDARs based on the combinations of its properties . Extrasynaptic NMDAR are hypothesized to respond to glutamate spillover or glial release of glutamate ( Rusakov and Kullmann , 1998; Haydon and Carmignoto , 2006; Sahlender et al . , 2014 ) . The potential ability to preferentially act at these non-synaptic sites arises from three mechanistic features of the allosteric mechanism: ( 1 ) the submaximal inhibitory effects at saturating concentrations of modulator and agonist , ( 2 ) the agonist-dependent and slow association rate , which might limit activity at synaptic receptors , and ( 3 ) the ability to enhance NMDAR responses to low agonist concentrations . A similar property has been described for the GluN2B-selective agent ifenprodil ( Kew et al . , 1996 ) , although the extent to which EU1794-4 can enhance the response to low concentrations of agonist is amplified by the large degree of residual current at saturating levels of EU1794-4 ( 30–60% ) compared to ifenprodil ( ~10% ) . There are numerous studies that suggest the importance of extrasynaptic NMDARs in normal biology but study of them requires complex experimental paradigms ( Harris and Pettit , 2008; Paoletti et al . , 2013; Papouin and Oliet , 2014 ) . Alternatively , the agonist-dependence of EU1794-4 may alter this capability in instances of repeated stimulation . In either case , EU1794-4 is a unique compound that may act as a tool that could be used to probe the contribution of distinct types of NMDARs or their activity in circuit function . Further work is required to fully understand the utility of this modulator as a probe for extrasynaptic NMDARs . cDNAs for rat wild type NMDAR subunits GluN1-1a ( GenBank U11418 , U08261; hereafter GluN1 or GluN1a ) , GluN1-1b ( U08263 , hereafter GluN1b ) , GluN2A ( D13211 ) , GluN2B ( U11419 ) , GluN2C ( M91563 ) , and GluN2D ( L31611 , modified as described in Monyer et al . , 1994 ) were provided by Drs . S . Heinemann ( Salk Institute ) , S . Nakanishi ( Kyoto University ) , and P . Seeburg ( University of Heidelberg ) . Site-directed mutagenesis was conducted using the QuikChange kit ( Agilent Technologies , Santa Clara , CA ) following the recommended protocol . DNA sequencing was used to verify all mutations . The amino acid numbering system started with the initiating methionine as residue number one . For expression in Xenopus laevis oocytes , cDNA constructs were linearized by restriction enzymes , and subsequently used to synthesize in vitro cRNAs following the manufacturer’s protocol ( mMessage mMachine , Ambion , ThermoFisher Scientific , Waltham , MA ) . Xenopus laevis stage VI oocytes ( Ecocyte Biosciences , Austin , TX ) were injected as previously described ( Hansen et al . , 2013 ) 2–7 days prior to recording with cRNA encoding the NMDAR subunits and stored at 15°C in media containing ( in mM ) 88 NaCl , 2 . 4 NaHCO3 , 1 KCl , 0 . 33 Ca ( NO3 ) 2 , 0 . 41 CaCl2 , 0 . 82 MgSO4 , 5 Tris-HCl; pH was adjusted to 7 . 4 with NaOH and the solution supplemented with 1 U/mL penicillin , 0 . 1 mg/mL gentamicin sulfate , and 1 μg/mL streptomycin . Two-electrode voltage-clamp recordings were performed at room temperature ( 21–23°C ) using different internal solutions for voltage and current electrodes ( 0 . 3 M and 3 . 0 M KCl , respectively ) . All extracellular solutions were made from a solution containing ( in mM ) 90 NaCl , 1 KCl , 10 HEPES , 0 . 5 BaCl2 , 0 . 01 EDTA ( pH 7 . 4 with NaOH ) . Experiments were performed using an eight port Modular Valve Positioner ( Hamilton Company , Reno , NV ) and controlled by custom software ( Easy Oocyte , Emory University , Atlanta , GA ) to exchange solutions . Oocyte currents were recorded at a holding potential of −40 mV using a two-electrode voltage-clamp amplifier ( OC-725B or OC-725C , Warner Instruments , Hamden , CT ) . For some recordings , 1 mM 2- ( hydroxypropyl ) -β-cyclodextrin was added to wash solutions to ensure modulators did not adhere to the tubing , valves , or recording chamber . In MTSEA experiments , fresh MTSEA solutions ( 0 . 2 mM ) were prepared and utilized in less than 30 min to limit compound degradation . HEK293 cells ( CRL-1573 , ATCC , Manassas , VA ) were maintained in a tissue culture room solely reserved for their culturing and transfected using a CaPO4 protocol to express the diheteromeric NMDARs as previously described ( Hansen et al . , 2013 ) . Cell lines were obtained directly from the supplier , were expanded , frozen and kept in liquid nitrogen in a dewar . Mycoplasma testing was not performed . cDNA ratios for transfection were 1:1:1 ( GluN2D ) or 1:1:5 ( GluN2A ) for GluN1:GluN2:GFP . Cells were maintained in 5% CO2 at 37°C for 12–36 hr post-transfection . Prior to and during recording , cells were bathed in a solution containing ( in mM ) 150 NaCl , 10 HEPES , 3 KCl , 0 . 5 CaCl2 , 0 . 01 EDTA ( pH 7 . 4 ) . Cells were patched with borosilicate glass micropipettes ( 3–4 MΩ ) that were filled with internal recording solution that contained ( in mM ) 95 CsGluconate , 5 CsCl , 40 HEPES , 8 NaCl , 5 MgCl2 , 10 BAPTA , 0 . 6 EGTA , 2 Na2ATP , and 0 . 3 NaGTP ( pH 7 . 35 ) . The whole cell recording conformation was achieved , and the cell lifted into the flow of solution from a two barreled theta glass or a triple barrel square glass perfusion system to perform rapid solution exchange experiments . The theta/triple barreled glass was translated using a piezoelectric manipulator to exchange the solution around the cell . Calibration of the perfusion manifold was performed each day to ensure the 10–90% rise time of the solution exchange ( switching between 0/100% and 50/50% H2O/external solution ) around an open tip was less than 1 . 5 ms for a theta tube and 4 ms for a transition into an outside lane and 2 ms for a transition between lanes for a triple barreled manifold . For some recordings , 1 mM 2- ( hydroxypropyl ) -β-cyclodextrin was added to wash solutions to ensure modulators did not adhere to the tubing , valves , or recording chamber . Concentration-response curves for both positive and negative modulators were fitted by the Hill equation:IIM=0=1+ExtentMhMh+EC50hwhere I is the measured response of receptor activation , Extent is the maximal predicted modulation of the glutamate/glycine response , [M] is the concentration of the modulator being used , h is the Hill slope , and EC50 is the half-maximal effective concentration of the modulator . Modulator concentration response data were plotted as a percentage of unmodulated response and displayed fitted curves were obtained by fitting all data simultaneously . Glutamate and glycine concentration-response curves were fit with the Hill equation as follows:IIMax=AhAh+EC50hwhere I is the measured response of receptor activation , [A] is the concentration of the agonist , h is the Hill slope , and EC50 is the half-maximally effective concentration of the agonist . Agonist concentration-response data , unless otherwise stated , were plotted as a percentage of maximal response and displayed fitted curves were obtained by fitting all data simultaneously with the appropriate Hill equation . The time course for the onset and offset of positive and negative modulation was fitted using a single exponential function:∆It=A* ( 1-e-tτ ) +cwhere t is time , It is the difference in current at a given time point as compared to the response at t = 0 , A is the amplitude of the exponential , τ is the time constant for the exponential and c is a constant . Additionally , some of the complex modulator time-courses were fitted with the sum of a single exponential equation and a linear component∆It=A* ( 1-e-tτ ) +mt+cwhere m is the slope . Measurements are given as mean ± SEM unless otherwise indicated . The number of replicate experiments used was chosen to ensure a power level of at least 0 . 80 when α = 0 . 05 and when detecting appropriate effect sizes . EC50 and IC50 values and confidence intervals reported were calculated by averaging the log ( EC50 ) or log ( IC50 ) value , determining confidence intervals for mean log ( EC50 ) or log ( IC50 ) , and converting back to units of molarity . Statistical significance evaluations ( α set to 0 . 05 ) were performed using a one-way ANOVA with Dunnett’s multiple comparison test , a paired t–test , or other appropriate tests as described . Matlab ( Mathworks , Natick , MA ) scripts were written based on previously described approaches to modelling channel function ( Colquhoun and Hawkes , 1977; Colquhoun and Hawkes , 1995 ) . Briefly , a Q matrix was constructed for a model that accounts for macroscopic responses ( Lester and Jahr , 1992 ) . The glutamate association and dissociation rates were from ( Erreger et al . , 2005 ) , and unidirectional rate constants for gating and modulator binding were chosen to allow the model to be used for modulator competition . The differential equations derived by the Q matrix were solved using an ordinary differential equation solver ( ode23s , Matlab ) and occupancy of each state was determined at equilibrium given the concentrations of the theoretical ligands .

The neurons in the brain form networks that can change in response to experience , causing new connections to form between certain neurons and breaking the connections between others . This remodeling process underlies learning and memory . However , in certain neurological disorders , such as schizophrenia and epilepsy , these networks are disrupted and no longer work correctly . A receptor protein called the NMDA receptor plays an important role in reshaping the networks of neurons . Chemicals called neurotransmitters that are released by one neuron bind to and activate NMDA receptors on a neighbouring neuron to communicate with it . This activation encourages new connections to form between neurons . Drugs that alter the activity of the NMDA receptor could potentially act as treatments for neurological conditions that disrupt how the networks of neurons work . However , few have been approved for use in patients because most of the potential drug compounds investigated so far produce severe side effects . Perszyk et al . have now identified a new group of compounds that can potentially alter the activity of NMDA receptors without fully blocking the response , potentially eliminating the unwanted side effects . The compounds were tested on frog egg cells and human embryonic kidney cells that had been engineered to produce NMDA receptors . Small changes to the chemical structure of these compounds could switch their effect from increasing to decreasing the activity of the receptor . The compounds only interact with active receptors that have a neurotransmitter bound to them , and compete with each other to bind to the receptors . Perszyk et al . also found that some compounds behaved differently depending on how active the NMDA receptor was . These compounds could potentially be used to sense the activity in a network of neurons , which opens up new options for treating neurological conditions that affect the networks . Further experiments are now required to see how these compounds affect the activity of NMDA receptors in neurons and living animals . The range of effects produced by the compounds studied by Perszyk et al . suggests that other related compounds may have different effects on receptor activity . Future work could investigate the properties of these compounds to see if they could treat a different set of neurological disorders .

2018

An NMDAR positive and negative allosteric modulator series share a binding site and are interconverted by methyl groups

Lamin B receptor ( LBR ) is a polytopic membrane protein residing in the inner nuclear membrane in association with the nuclear lamina . We demonstrate that human LBR is essential for cholesterol synthesis . LBR mutant derivatives implicated in Greenberg skeletal dysplasia or Pelger-Huët anomaly fail to rescue the cholesterol auxotrophy of a LBR-deficient human cell line , consistent with a loss-of-function mechanism for these congenital disorders . These disease-causing variants fall into two classes: point mutations in the sterol reductase domain perturb enzymatic activity by reducing the affinity for the essential cofactor NADPH , while LBR truncations render the mutant protein metabolically unstable , leading to its rapid degradation at the inner nuclear membrane . Thus , metabolically unstable LBR variants may serve as long-sought-after model substrates enabling previously impossible investigations of poorly understood protein turnover mechanisms at the inner nuclear membrane of higher eukaryotes . Lamin B receptor ( LBR ) is an evolutionary conserved , multifunctional protein ( Olins et al . , 2010 ) . The N-terminal moiety of LBR , which resides in the nucleoplasm , contains a chromatin-binding TUDOR domain and associates with the nuclear lamina ( Makatsori et al . , 2004; Pyrpasopoulou et al . , 1996; Worman et al . , 1988 ) , while the polytopic C-terminal domain anchors LBR in the inner nuclear membrane ( INM ) and has sequence homology to sterol C14 reductases ( Li et al . , 2015; Silve et al . , 1998; Worman et al . , 1990 ) . Sterol C14 reductases are widely conserved throughout evolution and are responsible for the reduction of a specific carbon-carbon double bond as part of the tightly controlled enzymatic cascade which results in the production of cholesterol and cholesterol-related compounds ( Benveniste , 2004; Holmer et al . , 1998 ) ( Figure 1 ) . 10 . 7554/eLife . 16011 . 003Figure 1 . Cholesterol biosynthesis pathway . Simplified schematic of cholesterol biosynthesis starting from Acetyl-CoA . After 13 enzymatic steps , the intermediate Lanosterol can enter one of two parallel pathways designated Bloch and Kandutsch-Russel pathways , respectively , both of which employ an NADPH-dependent reduction step which can be catalyzed by sterol C14 reductases LBR or TM7SF2 ( highlighted in magenta ) . Adapted , with modifications , from ( Sharpe and Brown , 2013 ) . The inset on the upper right depicts distinct subcellular localizations of the human sterol reductases LBR and TM7SF2 localizing to the inner nuclear membrane and ER , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 16011 . 003 There is currently no known functional connection between the chromatin-binding N-terminus of LBR and the sterol reductase C-terminus of the protein . The function of the sterol reductase domain ( SRD ) of LBR is further obfuscated by the fact that human cells have a second C14 sterol reductase enzyme called TM7SF2 , which is conserved in evolution and localizes to the endoplasmic reticulum ( ER ) , where other enzymes responsible for cholesterol biogenesis are typically found ( Bennati et al . , 2006 ) . The TM7SF2 promoter harbors a sterol response element ( SRE ) common to most , if not all enzymes implicated in cholesterol synthesis , allowing for the tight regulation of their transcription in response to cholesterol availability ( Brown and Goldstein , 1999; Sharpe and Brown , 2013 ) . However , the LBR gene lacks an SRE consensus sequence and is constitutively expressed ( Bennati et al . , 2006; Cohen et al . , 2008; Sharpe and Brown , 2013 ) , drawing into question whether LBR has a significant role in cholesterol synthesis . While congenital diseases associated with defects in cholesterol homeostasis have been extensively investigated ( Goldstein and Brown , 2015 ) , much less is known about the possible involvement of LBR mutations in cholesterol metabolism . Two congenital disorders are known to be associated with mutations in LBR: Pelger-Huët anomaly and Greenberg skeletal dysplasia ( Oosterwijk et al . , 2003; Shultz et al . , 2003; Wassif et al . , 2007; Waterham et al . , 2003 ) ( see Table 1 ) . Pelger-Huët anomaly is an autosomal dominant disorder in which a single mutation in one LBR allele results in abnormal hypolobulation of granulocyte nuclei ( Best et al . , 2003; Hoffmann et al . , 2002; Shultz et al . , 2003 ) . The other human disease associated with LBR , Greenberg skeletal dysplasia , is a perinatally lethal , autosomal recessive condition that results in abnormal bone development , fetal hydrops , and the ultimate nonviability of the fetus ( Chitayat et al . , 1993; Greenberg et al . , 1988; Horn et al . , 2000; Konstantinidou et al . , 2008; Trajkovski et al . , 2002 ) . Interestingly , mounting evidence indicates that Greenberg skeletal dysplasia results from the inheritance of two mutant LBR alleles that when heterozygous cause Pelger-Huët anomaly ( Konstantinidou et al . , 2008; Oosterwijk et al . , 2003 ) , indicating that the two diseases represent different allelic states of the same chromosomal lesion . However , it is unclear whether these diseases are caused by structural changes in the nuclear lamina , or whether they are diseases of cholesterol metabolism ( Clayton et al . , 2010; Olins et al . , 2010; Wassif et al . , 2007; Waterham et al . , 2003; Worman and Bonne , 2007 ) . 10 . 7554/eLife . 16011 . 004Table 1 . Diseases-associated LBR mutations used in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 16011 . 004LBR variantMutationPhenotypeReferenceN547Dc . 1639A>GHeterozygous - No PhenotypeClayton et al . , 2010p . N547DHomozygous - Greenberg DysplasiaKonstantinidou et al . , 2008R583Qc . 1748G>AHeterozygous - No PhenotypeClayton et al . , 2010p . R583QHomozygous - Greenberg Dysplasia1402TΔc . 1402delTHeterozygous - Phenotype UnknownClayton et al . , 2010p . Y468TfsX475Homozygous - Greenberg Dysplasia1600*c . 1599-1605TCTTCTA→CTAGAAGHeterozygous - Pelger-Huët AnomalyWaterham et al . , 2003p . X534Homozygous - Greenberg Dysplasia In this study , we show that LBR is essential for cholesterol synthesis . Using a human cell culture model , we demonstrate that it is this function that is perturbed by LBR mutations associated with Pelger-Huët anomaly and Greenberg skeletal dysplasia , suggesting a loss-of-function mechanism for these congenital disorders . Unexpectedly , disease-causing mutations involving C-terminal truncations of LBR lead to their rapid degradation in the nuclear envelope ( NE ) . Such LBR mutants appear to be dislocated from the INM directly into the nucleoplasm , unlike traditional substrates of the ER-associated degradation ( ERAD ) machinery , which are eliminated in the cytosol after their dislocation from the ER ( Claessen et al . , 2012; Vembar and Brodsky , 2008 ) . Metabolically unstable LBR mutant proteins will therefore be informative for future studies aimed at elucidating mechanisms of protein quality control at the nuclear envelope of mammalian cells , a site that was previously experimentally inaccessible due to the absence of suitable model substrates . In order to clarify the cellular function of LBR both in cholesterol metabolism and as a structural component of the nuclear lamina , we used the CRISPR/Cas9 system ( Mali et al . , 2013 ) to generate LBR knockout HeLa cell lines ( Figure 2—figure supplement 1A ) . CRISPR/Cas9 editing of LBR alleles was performed in a recombination-competent HeLa FlpIn cell line ( hereafter designated wild type or WT cells ) , allowing for rapid and efficient introduction of WT rescue and disease-specific LBR alleles into the LBR knockout cell background via site-specific recombination ( Turner et al . , 2015 ) . CRISPR/Cas9-treated WT cells were screened for the absence of full-length LBR protein by immunoblotting using antibodies against both the N and C termini of the protein ( Figure 2—figure supplement 1B ) , and via genotyping using PCR primers flanking the CRISPR target site ( Figure 2—figure supplement 1A , arrows ) . A clone was obtained that yielded no detectable LBR protein as judged by immunoblotting , corresponding to the absence of a PCR product of the size predicted by the wild-type LBR allele ( Figure 2—figure supplement 1C ) , indicating that all LBR alleles had been effectively targeted . To exclude the presence of hypomorphic alleles , we performed deep sequencing on the genetic locus encompassing the LBR CRISPR/Cas9 target site . Since HeLa cells are aneuploid , including three complete copies of chromosome 1 where the LBR gene is located , any LBR knockout should have three distinct genome 'edits' . Indeed , sequence analysis revealed three distinct mutant alleles , all containing frame-shift mutations or premature stop codons within the 5' region of the LBR open reading frame , showing that no more than 12 amino acids of LBR WT sequence can be produced from any of the three mutant alleles ( Figure 2—figure supplement 2 ) . As indicated by its name , LBR has long been implicated in NE integrity and NE anchoring to the nuclear lamina ( Appelbaum et al . , 1990; Worman et al . , 1990 , 1988; Ye and Worman , 1994 ) , prompting us to investigate if removing LBR perturbs the structure and composition of the nuclear lamina . We performed immunofluorescence microscopy analysis of known INM proteins and components of the nuclear lamina in both LBR knockout ( KO ) and WT cells . No differences in overall cell morphology or growth were observed between WT and LBR KO cells under normal growth conditions ( Figure 2A ) . Surprisingly , we found no change in the localization of Lamin B1 , Lamin A/C or Emerin in LBR KO cells compared to control cells ( Figure 2A ) . Similarly , we found that the absence of LBR also had no effect on the localization of other structural proteins of the NE such as Sun1 or Sun2 , which serve as the INM components of the LINC ( linker of nucleoskeleton and cytoskeleton ) complex ( Crisp et al . , 2006 ) ( Figure 2—figure supplement 3A and B ) . Similar results were obtained for other NE , nuclear and ER markers ( Figure 2—figure supplement 3 ) . 10 . 7554/eLife . 16011 . 005Figure 2 . LBR deficient cells display normal nuclear envelope morphology but are sensitive to cholesterol restriction . ( A ) HeLa WT and HeLa LBR KO cells were stained with antibodies against LBR using an antibody recognizing the N-terminal domain ( ND ) , Lamin B1 , Lamin A/C , and Emerin and then imaged by immunofluorescence microscopy . Scale bar = 10 μm . ( B ) EM images of WT and LBR KO HeLa cells . The nucleus ( N ) and cytosol ( C ) are labeled and nuclear pores are indicated by arrowheads . Scale bar = 500 nm . ( C ) Indicated cell lines were cultured under cholesterol-restrictive growth conditions for two days , imaged by bright-field microscopy , and cultured for 2 more days under cholesterol restrictive conditions , and imaged again . Cells were exposed either to vehicle , free cholesterol or LDL ( 10 µM ) for four days as indicated . Scale bar = 50 µm . ( D ) HeLa WT and LBR KO cells in LPDS-containing medium were metabolically labeled with [14C]-acetate . Lipids were extracted and separated by TLC and visualized via autoradiography . [14C]-cholesterol was included as a standard ( std ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16011 . 00510 . 7554/eLife . 16011 . 006Figure 2—figure supplement 1 . LBR knockout cells were generated using the CRISPR/Cas9 genome editing system . ( A ) Schematic diagram encompassing the human LBR locus . Shown are the CRISPR guide RNA sequence used to create the knockout and the flanking genotyping PCR primers used to amplify the targeted region . Exons are black boxes , introns are black lines , and PCR primers are half arrows . ( B ) Parental and LBR knockout cells were harvested and analyzed for the presence of LBR protein by SDS-PAGE followed by immunoblotting ( IB ) using antibodies raised against the N-terminal domain ( top ) and C-terminal domain ( bottom ) of LBR . Blots were stripped and probed for β-actin as a loading control . ( C ) Genotyping PCR of the targeted LBR locus using primers shown in panel A for parental and LBR knockout cells . GAPDH was used as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 16011 . 00610 . 7554/eLife . 16011 . 007Figure 2—figure supplement 2 . The genomic LBR CRISPR target site of LBR KO HeLa cells was analyzed using Illumina MiSeq . ( A ) At the nucleotide level , three different mutations , designated Edit 1 , Edit 2 , and Edit 3 , each corresponding to one of the three copies of LBR found in HeLa cells , were detected . The first mutation completely removes the LBR start codon , while the second two mutations are large out-of-frame deletions , the second of which removes the exon 1 5’ splice site . ( B ) At the protein level , Edit 1 results in the translation of a 34 amino acid out-of-frame peptide from an internal start codon located at p . 251 . Edit 2 results in the truncation of LBR at position p . 17 , and Edit 3 results in the correct translation of the first 11 amino acids of LBR , followed by 38 amino acids translated from LBR intron 1 , with the introduction of a premature stop codon at LBR p . 50 . DOI: http://dx . doi . org/10 . 7554/eLife . 16011 . 00710 . 7554/eLife . 16011 . 008Figure 2—figure supplement 3 . LBR deficient cells display normal nuclear envelope morphology . ( A–G ) HeLa WT and HeLa LBR KO cells were fixed and processed for immunofluorescence using the indicated antibodies . Scale bar = 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 16011 . 008 Lastly , we utilized electron microscopy on both LBR KO and WT cells to obtain high-resolution images of the NE . We found that in both WT and LBR KO cells , the inner and outer nuclear membranes were regularly spaced and featured a nuclear lamina and nuclear pores of normal morphology ( Figure 2B , arrowheads ) . These data , together with our previous observations by immunofluorescence microscopy , indicate that in these cells , LBR does not play a significant role in maintaining NE integrity . We cannot , however , exclude that LBR plays a role in the structural integrity of the nuclear lamina in specific cell types , under conditions of mechanical stress , or developmental stages found only in the context of the living organism . Next , we set out to investigate the role of LBR in cholesterol synthesis . In order to determine if LBR is required for cell proliferation under cholesterol-restrictive growth conditions , HeLa WT cells and LBR KO cells were cultured in medium containing lipoprotein-depleted fetal bovine serum . We found that after 4 days under cholesterol-restrictive growth conditions LBR KO , but not WT HeLa cells exhibited slow growth , cell rounding , and detachment , followed quickly by cell death on days 5–7 ( Figure 2C ) . Addition of 10 µM exogenous cholesterol to the cell culture medium effectively rescued the observed sensitivity of LBR KO cells to low-cholesterol growth conditions ( Figure 2C ) , indicating that the observed phenotype is in fact due to a deficiency in cholesterol production . The observed growth defect of LBR KO cells was also rescued by the addition of low density lipoprotein ( LDL ) particles , a physiologically relevant cholesterol carrier ( Lodish , 2013 ) , to the cell culture medium ( Figure 2C ) . These results suggest that the primary function of LBR in our tissue culture model is to sustain cholesterol biogenesis when the extracellular supply of cholesterol is scarce . Given that LBR is endowed with sterol C14 reductase activity ( Bennati et al . , 2006; Silve et al . , 1998 ) , we tested directly whether LBR-deficient cells are compromised in de novo cholesterol synthesis . To this end , HeLa WT and LBR KO cells were cultured in lipid-deprived medium for 48 hr , followed by the addition of 14C acetate to the culture medium . After four hours , cells were harvested , lysed and the extracted lipids were separated via thin layer chromatography ( TLC ) . We readily detected newly synthesized cholesterol in WT cells , as judged by its co-migration with purified 14C cholesterol , which was included as a standard ( Figure 2D ) . Notably , we observed a near-complete loss of cholesterol synthesis in LBR KO cells ( Figure 2D ) , validating our previous assumption that LBR KO cells cannot effectively sustain cholesterol synthesis . Finally , we asked whether these findings can be reproduced in other human cell types . We chose human foreskin fibroblasts ( HFFs ) to include non-transformed primary cells , as well as commonly used HEK293T cells . In either case , we observed a strong reduction of viability in LBR-silenced cells under cholesterol-restrictive conditions , whereas control cells transfected with non-targeting siRNAs displayed normal cell morphology under these conditions ( Figure 3 A , B , D , E ) . To validate our RNA interference approach , we subjected the corresponding cell lysates to immunoblotting and observed a robust LBR knockdown efficacy in both cell types ( Figure 3C , F ) . These findings rule out that the cholesterol auxotrophy observed in HeLa LBR KO cells is attributable to their degeneracy or their transformed nature . 10 . 7554/eLife . 16011 . 009Figure 3 . Cells with reduced levels of LBR are sensitive to cholesterol restriction despite the presence of TM7SF2 . ( A ) 293T cells were transfected with control siRNA or LBR siRNA , incubated for 48 hr , cultured in cholesterol-restrictive condition for 1 day and imaged via bright-field microscopy . Scale bar = 50 µm ( B ) Surviving adherent cells were quantified using crystal violet staining , and data were standardized relative to the level of control siRNA . The error bars represent mean ± SD from three independent experiments ( N = 3 ) , and the statistical analysis was performed using paired t-test ( *p value < 0 . 05 , **p value < 0 . 01 ) . ( C ) An additional parallel well of 293T cells with same treatment as described above was lysed and analyzed using immunoblotting . ( D–F ) HFF cells were treated with siRNA and cultured under cholesterol-restrictive conditions as described in ( A ) with an exception that HFF cells are cultured in the LPDS medium for 3 days . ( G and H ) HFF , 293T , and HeLa cells were cultured in normal medium or cholesterol-restrictive medium for 2 days , and cells were harvested and split into two aliquots . Total RNAs were extracted from one aliquot , reverse transcribed into cDNA , and analyzed using real-time PCR with gene specific primers as indicated . The data was represented as a relative level to the normal condition ( i . e . –LPDS was set to one in each cell line ) . ( I ) The other aliquot of cells was lysed with sample buffer and subjected to immunoblotting analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 16011 . 00910 . 7554/eLife . 16011 . 010Figure 3—source data 1 . TM7SF2 sequence cloned using RT-PCR from HeLa cells . The TM7SF2 cDNA sequence cloned from HeLa cells ( top strand ) was aligned with the coding sequence of the published human TM7SF2 transcript isoform1 ( NM_003273 ) ( lower strand ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16011 . 01010 . 7554/eLife . 16011 . 011Figure 3—figure supplement 1 . LBR transcription does not change under cholesterol-restrictive condition . ( A–D ) Wild-type or LBR KO HeLa cells were transfected with control siRNA or SMARTpool siRNA targeting to SREBP2 for 48 hr , and cultured in normal medium or cholesterol-restrictive medium for another 48 hr . Total RNAs were then isolated , reversed transcribed into cDNA , and analyzed using real-time PCR with specific primers as indicated . Data were normalized relative to wild-type HeLa cells transfected with control siRNA , cultured in normal conditions . Error bars represent mean ± SD from triplicates . DOI: http://dx . doi . org/10 . 7554/eLife . 16011 . 011 Our assignment of LBR to an essential role in cholesterol synthesis contrast earlier findings in mice , which reported redundant functions for LBR and TM7SF2 in cholesterol synthesis ( Wassif et al . , 2007 ) . We therefore asked whether HeLa , HEK293T and HFF cells expressed both TM7SF2 and LBR under normal and cholesterol-restrictive conditions . For this analysis , we isolated total RNA and prepared protein extracts from these three cell types grown under normal and cholesterol-restrictive conditions . TM7SF2 and LBR transcript levels were quantified via qPCR , and the observed fold change under restrictive conditions was normalized relative to the transcript abundance under normal growth conditions , which was set to one . As expected ( Bennati et al . , 2006 ) , we found that HFF and HeLa cells upregulated TM7SF2 under starvation conditions ( Figure 3G ) , whereas LBR was constitutively expressed and unresponsive to cholesterol starvation ( Figure 3H ) . Interestingly , HEK293T cells did not materially up-regulate TM7SF2 on the transcript level ( Figure 3G ) . Similar results were obtained when we monitored the LBR and TM7SF2 protein levels via immunoblotting . The abundance of the TM7SF2 protein in HFF and HeLa cells increased under cholesterol-restrictive conditions , whereas HEK293T cells displayed somewhat higher levels of TM7SF2 even under normal conditions but did not upregulate TM7SF2 to the same degree as HFF or HeLa cells ( Figure 3I ) . Lastly , we wanted to ascertain that LBR KO cells are capable of mounting a sterol regulatory element-binding protein-2 ( SREBP2 ) -dependent transcriptional response under cholesterol starvation conditions ( Brown and Goldstein , 1997 ) . We did not observe significant differences between HeLa WT and HeLa LBR KO cells in their ability to up-regulate TM7SF2 on the transcript level , as judged by qPCR ( Figure 3—figure supplement 1A ) . Similar results were obtained for HMG-CoA reductase ( Figure 3—figure supplement 1C ) , which was included as a control since the HMG-CoA reductase gene is an established target of SREBP2 ( Brown and Goldstein , 1997 ) . In accordance with previous findings ( Bennati et al . , 2006 ) , the observed upregulation was dependent on SREBP2 , as judged by a profound reduction of TM7SF2 and HMG-CoA reductase on the transcript levels in cells depleted of SREBP2 via RNA interference . Finally , to exclude the formal possibility that coding mutations in TM7SF2 are responsible for the observed essential role of LBR in HeLa cells through perturbation of TM7SF2 sterol C14 reductase activity , we cloned and sequenced the corresponding cDNA . This isolated cDNA ( 1401 bp ) contains the entire coding sequence ( CDS ) of the published transcript variant 1 ( NM_003273 ) , with no apparent mutations as evidenced by a sequence alignment to the CDS of NM_003273 ( Figure 3—source data 1 ) . This isoform appears to be the major isoform expressed in HeLa , HFF , and HEK 293T cells , as judged by the observed co-migration of the corresponding TM7SF2 proteins in SDS-PAGE/Immunoblots ( Figure 3I ) . Based on the foregoing , we arrive at the conclusion that LBR is essential for cholesterol synthesis in several human cell lines despite the presence of TM7SF2 . Having demonstrated that LBR is essential for cell viability under cholesterol-depleted conditions , we asked whether the mutant variants of LBR found in Pelger-Huët anomaly and Greenberg skeletal dysplasia can sustain cholesterol biogenesis in human cells . To this end , we used the FlpIn gene integration system to generate HeLa cells that express either WT LBR or LBR disease alleles under doxycycline control from the FlpIn locus ( Turner et al . , 2015 ) . We focused on two LBR point mutations , one frameshift mutation , and one nonsense mutation , the latter of which results in a truncated LBR C-terminus ( Table 1 ) ( Clayton et al . , 2010; Konstantinidou et al . , 2008; Waterham et al . , 2003 ) . The two point mutations , LBR N547D and LBR R583Q ( Figure 4A ) , are both associated with Greenberg skeletal dysplasia when both LBR alleles are mutated ( Clayton et al . , 2010; Konstantinidou et al . , 2008 ) . Both positions are highly conserved in homologous C14 sterol reductases ( Figure 5—figure supplement 1 ) and are not associated with Pelger-Huët anomaly in heterozygous individuals ( Table 1 ) ( Clayton et al . , 2010 ) . For the LBR truncations , the first frameshift mutation results from a single nucleotide deletion in the C-terminal sterol reductase domain of LBR , c . 1402delT , which results in the alteration of downstream codons 468–475 and the formation of a premature stop codon at position 475 ( p . Y468TfsX475 ) ( Clayton et al . , 2010 ) , which we designated LBR 1402TΔ . This mutation results in a C-terminally truncated LBR protein that is missing the final three membrane spanning helices that form the sterol-reductase domain of LBR ( Figure 4A ) and results in Greenberg skeletal dysplasia in homozygous individuals . It is unknown whether individuals heterozygous for the LBR c . 1402delT mutation exhibit Pelger-Huët anomaly . The second frameshift mutation causing a distinct C-terminal truncation , LBR c . 1599-1605TCTTCTA->CTAGAAG ( LBR p . X534 ) , which we have designated as LBR 1600* , has been shown to cause Pelger-Huët anomaly in heterozygous individuals as well as Greenberg skeletal dysplasia when both LBR alleles are mutated ( Waterham et al . , 2003 ) . This mutation is a seven nucleotide substitution beginning at LBR position 1599 that directly causes the introduction of a premature stop codon at position LBR p . 534 , resulting in a truncated LBR C-terminus lacking the final two transmembrane helices of the protein ( Figure 4A ) . To determine if these mutations are loss-of-function alleles , we expressed each of these as well as wild-type LBR under doxycycline control in our LBR KO HeLa cell line . Cells expressing various LBR alleles in an LBR knockout background were grown under cholesterol-restrictive culture conditions for 7 days . We then imaged and counted the cells and measured total cholesterol content ( see Materials and methods ) . We found that expression of LBR WT from the FlpIn locus completely rescued the growth defect observed for LBR KO cells in cholesterol starvation growth medium ( Figure 4B and C ) . This excluded off-target effects as a cause for the observed phenotype . Importantly , all four disease-associated LBR alleles failed to rescue the observed growth defect ( Figure 4B and C ) . LBR KO cells were found to have approximately 40% less total cholesterol content than LBR WT cells after 7 days of cholesterol starvation ( Figure 4E ) . Cellular cholesterol content was fully restored to that of WT cells by expression of LBR WT from the FlpIn locus even at somewhat lower LBR expression levels compared to WT cells ( Figure 4D and E ) , but not by expression of any of the four disease-associated LBR alleles ( Figure 4E ) . From these data we conclude that LBR KO HeLa cells are strongly compromised for cholesterol production . Neither cell growth nor cellular cholesterol content are restored by the expression of LBR N547D , LBR R583Q , LBR 1402TΔ or LBR 1600* in a LBR KO background , indicating that these genetic lesions ultimately result in a failure to sustain cell growth under cholesterol-restrictive conditions . 10 . 7554/eLife . 16011 . 012Figure 4 . Cholesterol auxotrophy of LBR KO cells is rescued by wild-type but not disease-mutant LBR . ( A ) Domain structure of LBR . Locations of disease-associated point mutations are indicated as black circles and asterisks demark disease-associated frameshift/truncation mutations . ( B ) Parental WT , LBR KO , or LBR KO cells expressing either WT LBR or mutant LBR from the FlpIn locus were cultured for 7 days in cholesterol-restrictive growth medium and then imaged by bright field microscopy . ( C ) The cell lines described above were grown in triplicate for 7 days in cholesterol-restrictive growth conditions , trypsinized , and counted . Values represent a mean of three independent experiments with error bars indicating the standard deviation ( D ) Immunoblot analysis of cell lysates harvested on day 7 of the experiment showing LBR expression level in each cell line relative to wild-type . ( E ) The above described cell lines were treated exactly as in ( C ) , harvested , and the total cholesterol concentration was determined using a fluorometric assay ( see Materials and methods ) . Values represent a mean of three independent experiments with error bars indicating the standard deviation . A cholesterol standard curve is shown on the right . DOI: http://dx . doi . org/10 . 7554/eLife . 16011 . 012 We next investigated the functional relationship between the N-terminal nuclear lamin B/chromatin binding domain of LBR and the C-terminal sterol reductase domain ( SRD ) . We generated cell lines in an LBR KO background that expressed either LBR WT , the isolated LBR SRD in an LBR knockout cell background ( see Figure 4A for LBR domain structure ) , or the fusion protein Sun2-LBR , which contains the nuclear domain of Sun2 containing a nuclear targeting signal ( Turgay et al . , 2010 ) , fused to the LBR SRD . Additionally , we generated a cell line that expressed an LBR construct that encompasses the nuclear domain of LBR together with the first transmembrane helix of the SRD ( LBR TM1 ) , a construct that is competent for INM targeting ( Smith and Blobel , 1993; Soullam and Worman , 1993 ) . LBR KO-derived cell lines expressing constructs described above were cultured in triplicate in lipoprotein-depleted growth medium for 7 days . Cells were then trypsinized and stained with Trypan blue to exclude non-viable cells and counted . As expected , very few live cells remained in the LBR KO cell sample , while expression of full-length LBR in an LBR knockout cell background restored the cells to near-WT levels of growth ( Figure 5A ) . Conversely , LBR TM1 failed to rescue cell growth . Finally , both the isolated LBR SRD and the fusion protein SUN2-LBR rescued the LBR KO phenotype . Given that all constructs were expressed at levels higher than the rescuing WT allele ( Figure 5B ) , our results indicate that the SRD is necessary and sufficient for survival of cholesterol-starved cells . 10 . 7554/eLife . 16011 . 013Figure 5 . The C-terminal sterol reductase domain of LBR Is necessary and sufficient for cell viability under cholesterol restrictive growth conditions . ( A ) WT HeLa cells , LBR knockout HeLa cells , or LBR knockout cells expressing either WT LBR , the LBR nuclear domain plus the first transmembrane helix ( LBR-TM1 ) , the LBR sterol reductase domain ( LBR-SRD ) , or the nuclear domain of Sun2 fused to the sterol reductase domain of LBR ( Sun2-LBR ) were grown in triplicate under cholesterol restrictive growth conditions for 7 days . The cells were then trypsinized and counted and the results were plotted with standard deviations shown . ( B ) Anti-LBR immunoblots of lysates from the above treated cells show LBR expression level relative to WT . Membranes were probed with two distinct anti-LBR antibodies recognizing an N- and C-terminal LBR epitope , respectively . ( C ) Electrostatic surface potential representation of themaSR1 ( crystal structure ( PDB: 4QUV ) with kT/e ± 1 . NADPH and residues N359 and R395 corresponding to LBR disease-associated residues N547 and R583 ( cf . Figure 5—figure supplement 1 ) are shown as sticks . ( D ) Disease-associated LBR point mutants LBR N547D and LBR R583Q show a decreased affinity for NADPH compared to wild-type LBR . Intrinsic tryptophan fluorescence of purified LBR WT and mutants upon NADPH binding was plotted against NADPH concentration and non-linear regressions were fitted in GraphPad Prism . All measurements were performed in triplicate . ( E ) HeLa LBR KO cells stably expressing LBR WT , LBR mutant N54D or R583Q were cultured in LPDS containing medium for 48 hr prior to metabolically labeling with [14C]-acetate . Lipids were extracted and separated by TLC and visualized via autoradiography . Bands corresponding to [14C]-cholesterol are marked by an arrowhead . DOI: http://dx . doi . org/10 . 7554/eLife . 16011 . 01310 . 7554/eLife . 16011 . 014Figure 5—figure supplement 1 . Sequence alignment of LBR and related sterol reductases . LBR sequences from human , rat , and mouse together with maSR1 ( Methylomicrobium alcaliphilum delta ( 14 ) -sterol reductase ) were aligned with Clustal Omega . Secondary structures of maSR1 are labeled above the sequences . The boxes indicate helices , the arrows indicate β-sheets , and the solid lines indicate random structural regions . The asterisks indicate stop codons of LBR 1402TΔ and LBR 1600* mutants . The triangles indicate N547D and R583Q mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 16011 . 014 The observations that ( i ) the LBR N547D and LBR R583Q transgenes are defective in rescuing the cholesterol auxotrophy and that ( ii ) both mutations map to the SRD , which is necessary and sufficient to complement the LBR KO phenotype , strongly suggests that these mutations affect sterol reductase activity . Indeed , both affected residues are widely conserved in related reductases , including the sterol reductase from Methylomicrobium alcaliphilum ( maSR1 ) , the structure of which was recently determined ( Li et al . , 2015 ) . Based on a sequence alignment ( Figure 5—figure supplement 1 ) , the LBR N547 and R583 residues correspond to maSR1 N359 and R395 , respectively , both of which map to the NADPH binding pocket , with N359 being implicated in a hydrogen bond formation with a phosphate oxygen of NADPH ( Li et al . , 2015 ) ( Figure 5C ) . We therefore tested whether these mutants displayed a reduced affinity for NADPH , the cofactor that contributes the electrons for the reduction reaction . WT LBR , LBR N547D , and LBR 583Q were purified and their NADPH binding affinities measured using a spectroscopic approach ( see materials and methods ) . WT LBR displayed hyperbolic saturation for the cofactor with a KD of 24 . 2 µM . Both mutants had a severely decreased binding affinity to NADPH , with N547D having the strongest effect ( Figure 5D ) . We attribute this lack of affinity to a possible charge repulsion between the introduced Asp side chain and the negatively charged phosphate of the cofactor . Given that the physiological concentration of NADPH is in the range of ~ 100 µM–a concentration range at which the discrepancy of cofactor occupancy of LBR and LBR N547D/LBR R583Q would be maximal– our measurements provide a direct rationale for the inability of those mutants to efficiently rescue the cholesterol auxotrophy ( cf . Figure 4B ) . As expected , neither LBR N547D nor LBR 583Q transgenes efficiently restored de novo cholesterol synthesis in LBR KO cells , which was monitored by metabolic incorporation of 14C acetate into cholesterol ( Figure 5E ) . Neither of the LBR C-terminal truncation mutant proteins ( 1402TΔ or 1600* ) were detected via immunoblotting using an LBR antibody that recognizes the N-terminal ( nuclear ) domain of LBR ( see Figure 4D ) . Given that these cell lines were engineered to express their WT , LBR N547D , or LBR 583Q transgenes to comparable levels , we speculate that the LBR truncation either leads to nonsense-mediated decay ( NMD ) of the encoding mRNA ( Kurosaki and Maquat , 2016 ) , or that the proteins themselves are recognized as aberrant and degraded by a cellular protein quality-control system ( Labbadia and Morimoto , 2015 ) . To distinguish between these possibilities , we monitored the metabolic stability of LBR and its mutant derivatives by pulse-chase analysis as described previously ( Rose et al . , 2014 ) . We found that LBR WT , LBR N547D and LBR R583Q are extremely stable ( Figure 6A and B ) , indicating that the lack of cofactor binding does not lead to LBR instability . In contrast , both truncated LBR variants were produced at levels comparable to WT LBR at the beginning of the chase period but were rapidly degraded ( Figure 6A and B ) . These observations are consistent with a post-translational effect on protein stability and argue against a major contribution of NMD . 10 . 7554/eLife . 16011 . 015Figure 6 . C-terminally truncated LBR mutants associated with Pelger-Huët anomaly and Greenberg skeletal dysplasia are rapidly degraded via the proteasome . ( A ) , ( B ) LBR KO cells expressing either WT LBR or the disease-associate LBR mutants were metabolically labeled with 35S and then chased with an excess of unlabeled cysteine/methionine . LBR was then retrieved at the indicated time points via immunoprecipitation , resolved by SDS-PAGE and imaged via autoradiography . ( C ) , ( D ) Turnover of LBR 1402TΔ and LBR 1600* was measured on a shorter time scale in the absence or the presence of MG132 . DOI: http://dx . doi . org/10 . 7554/eLife . 16011 . 015 To investigate the degradation mechanisms of LBR 1402TΔ and LBR 1600* , we repeated the pulse-chase experiments on a shorter time scale and in the absence or presence of MG132 , a potent cell-permeable proteasome inhibitor ( Rock et al . , 1994 ) . We found that both LBR 1402TΔ and LBR 1600* proteins were degraded extremely rapidly , with little or no protein remaining after 30 min ( Figure 6C and D ) . Both mutants were significantly stabilized by MG132 , with minimal degradation taking place in 60 min after synthesis . We conclude that the degradation of truncated LBR variants depends on the proteasome . Notably , the kinetics of LBR 1402TΔ and LBR 1600* degradation were remarkably rapid , with half-lives of 10–15 min , especially considering that these mutant derivatives are polytopic membrane proteins . As a standard of comparison , the half-life of CFTRΔ506 , a mutant variant of the polytopic chloride channel responsible for cystic fibrosis ( Turnbull et al . , 2007 ) which is widely used as model substrate to study protein turnover , is four hours ( Heda et al . , 2001 ) . Thus , LBR disease variants have significant potential as novel model substrates to study protein turnover . Since protein ubiquitylation is a key step in proteasome-mediated protein turnover of misfolded membrane proteins ( Claessen et al . , 2012; Raasi and Wolf , 2007; Vembar and Brodsky , 2008 ) , we next determined if the rapidly degraded truncated LBR proteins LBR 1402TΔ and LBR 1600* are ubiquitylated . We expressed either FLAG-LBR WT , FLAG-LBR 1402TΔ , or FLAG-LBR 1600* together with HA-Ubiquitin ( Ub ) in a LBR knockout cell background . Cells were then treated with either MG132 or vehicle for 2 hr and subjected to denaturing detergent lysis to disrupt non-covalent protein-protein interactions . Following dilution with SDS-free buffer , extracts were immunoprecipitated with an anti-FLAG resin to retrieve tagged LBR protein . Anti-HA immunoblotting of input samples demonstrated that HA-Ub was expressed in all HA-Ub transfected cells , with the expected increase of HA-Ub conjugates in the higher molecular mass range in response to proteasomal inhibition ( Figure 7A , top panel ) . Identical samples were subjected to immunoblotting with anti-FLAG antibodies to detect LBR or its mutant derivatives . In agreement with our pulse-chase data , both LBR 1402TΔ and LBR 1600* are expressed at extremely low levels under steady-state conditions when compared to LBR WT ( Figure 7A , bottom panel ) . 10 . 7554/eLife . 16011 . 016Figure 7 . LBR 1402TΔ and LBR 1600* proteins are polyubiquitylated . LBR KO cells were co-transfected with plasmids encoding HA-tagged ubiquitin and FLAG-tagged LBR WT , LBR 1600* , or LBR1402TΔ . Sixteen-hours-post transfection , cells were treated with MG132 or DMSO for another 4 hr before harvesting and denaturing SDS lysis . ( A ) Five percent of cell lysates used per immunoprecipitation ( B ) were separated by SDS-PAGE and subjected to immunoblotting using the indicated antibodies ( B ) Lysates were diluted in SDS-free buffer and subjected to immunoprecipitation using anti-FLAG antibody , followed by SDS-PAGE and immunoblotting analysis using the indicated antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 16011 . 016 Next , the anti-FLAG immunoprecipitates were subjected to SDS-PAGE and immunoblotting . As expected , HA-Ub conjugated species were significantly more abundant for both LBR 1402TΔ and LBR 1600* than for LBR WT in both untreated and MG132-treated samples , with the latter condition leading to an additional increase ( Figure 7B , upper panel ) . Since the levels of unmodified LBR WT far exceed those of LBR 1402TΔ and LBR 1600* ( Figure 7B , lower panel ) , these data reflect a disproportionate increase in abundance of ubiquitylated LBR species for the disease-associated variants relative to LBR WT . Together , the data argue that the metabolic instability of LBR 1402TΔ and LBR 1600* is caused by degradation via the Ub/proteasome system ( UPS ) . Given our finding that C-terminally truncated LBR is degraded via a UPS-dependent pathway , LBR 1402T and LBR 1600* could represent typical substrates for the pathway responsible for the degradation of ER-resident proteins referred to as ER-associated degradation ( ERAD ) ( Claessen et al . , 2012; Raasi and Wolf , 2007; Vembar and Brodsky , 2008 ) . The ERAD machinery can act on LBR 1402TΔ and LBR 1600* before they arrive at the INM . Alternatively , a distinct , INM resident pathway might mediate LBR 1402TΔ and LBR 1600* turnover . As a first step towards resolving this question , we monitored the cellular localization of LBR 1402TΔ and LBR 1600* in the absence or presence of MG132 using confocal fluorescence microscopy . In the absence of MG132 , LBR 1402TΔ is partitioned between the ER and the NE , as judged by a co-staining with anti-LBR and anti-Lamin A/C antibodies ( Figure 8A , left panel ) . LBR 1600* , on the other hand , displays nuclear rim staining that is indistinguishable from LBR WT ( Figure 8B , left panel; cf . Figure 2A , upper left panel ) ( Worman et al . , 1988 ) . 10 . 7554/eLife . 16011 . 017Figure 8 . LBR 1402TΔ and LBR 1600* proteins accumulate in the nucleus after MG132 treatment . ( A , B ) LBR knockout cells expressing either LBR 1402TΔ or LBR 1600* were treated with MG132 or DMSO vehicle for 4 hr and then fixed , stained , and imaged by confocal fluorescence microscopy . The cells were stained with antibodies against LBR ( green ) and Lamin A/C ( red ) . ( C ) Cells were treated and imaged as above ( A , B ) , and the nuclear fluorescence obtained from 20 complete z-stack series for each condition was quantified using ImageJ , summed up , and standardized relative to the sum of total fluorescence . The ratio of nuclear to total cellular fluorescence is given as arithmetic mean value . Error bar represents mean ± SD . The statistical significance is determined by unpaired t-test . ***p<0 . 001 ( D ) Pulse-chase analysis of LBR KO HeLa cells co-transfected with FLAG-tagged LBR 1600* mutant and p97 WT or QQ mutant . ( E ) Densitometric quantification of pulse-chase data . ( F ) LBR-knockout HeLa cells were co-transfected with LBR mutant 1600* and with p97 WT or QQ mutant and treated with 10 µM of MG132 or DMSO for 4 hr . Cells are stained with anti-LBR ( green ) and anti-p97 ( red ) , and imaged with a confocal microscope . Scale bar = 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 16011 . 01710 . 7554/eLife . 16011 . 018Figure 8—figure supplement 1 . Complete confocal z-stack series corresponding to Figure 8A , B . LBR knockout cells expressing either LBR 1402TΔ or LBR 1600* were treated with MG132 or DMSO vehicle for 4 hr and then fixed , stained , and imaged by confocal fluorescence microscopy . The cells were stained with antibodies against LBR ( green ) and Lamin A/C ( red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16011 . 018 Against all expectations , both truncated LBR proteins appear to accumulate in the nucleoplasm upon proteasomal inhibition ( Figure 8A and B , right panels ) . In addition , both LBR mutants are clearly excluded from nucleoli in the presence of MG132 . Quantification of these results , based on a total projection average of 20 independent z-stack series for each condition , is shown in Figure 8C ( see Figure 8—figure supplement 1 for a corresponding z-stack series ) . We observe that with both mutants , MG132 treatment resulted in a shift from a non-nuclear to a nuclear LBR immunofluorescence signal , with LBR 1402TΔ showing a slightly larger shift in localization than LBR 1600* , which resides primarily in the nuclear compartment even in the absence of MG132 ( Figure 8A , B ) . These unexpected results indicate that disease-associated , C-terminally truncated LBR mutants surprisingly accumulate in the nuclear compartment , rather than the ER or cytosol , after MG132 treatment . Given that LBR1600* is extremely short-lived but mainly localizes to the INM under steady-state conditions even in the absence of MG132 , we suggest that it is here that membrane dislocation occurs . The requirement of the AAA+ ATPase p97 for the extraction or dislocation of membrane proteins from the ER membrane is firmly established ( Ye et al . , 2003 ) . Given that p97 has additional functions in the nuclear compartment ( Dantuma and Hoppe , 2012 ) , we asked whether p97 is implicated in the extraction of LBR 1600* from the INM . LBR KO cells were transfected either with FLAG-LBR 1600* alone or in combination with either p97 WT or p97 QQ , a dominant-negative mutant of p97 that potently blocks p97-dependent functions ( Ernst et al . , 2009; Ye et al . , 2003 ) . We then performed pulse-chase analyses to monitor the stability of LBR 1600* . As expected , p97 WT had no effect on LBR 1600* degradation ( Figure 8D and E ) . However , p97 QQ inhibited LBR 1600* turnover to an extent that is comparable to proteasomal inhibition ( Figure 8D and E ) . We next asked whether p97 QQ can block LBR 1600* turnover in the nuclear compartment using confocal microscopy . Cells were again co-transfected with LBR 1600* and either p97 QQ or p97 WT and then treated with MG132 or carrier . This experimental setup allowed us to explore the possibility of an epistatic relationship between p97 QQ and MG132 . If p97 is indeed required for the dislocation of LBR 1600* from the INM , we would expect p97 QQ to prevent the accumulation of LBR 1600* in the nucleoplasm upon MG132 treatment . Using double-staining with anti-p97 and anti-LBR antibodies , we indeed found this to be the case . In the presence of both p97 WT and MG132 , we observed the expected LBR 1600* accumulation in the nucleoplasm and to a lesser degree , the ER ( Figure 8F ) , whereas LBR 1600* accumulates nearly quantitatively at the nuclear rim in the presence of p97 QQ even in the presence of MG132 . The data are consistent with a blockage of the membrane dislocation step at the INM ( Figure 8F ) . In conclusion , these results are consistent with a mechanism in which only a small percentage of LBR 1600* is degraded by the canonical ERAD pathway . The majority of the truncated protein is localized to the INM , where it appears that a second quality control pathway operates to extract and destroy LBR 1600* in a series of reactions employing both p97 and the UPS ( Figure 9 ) . 10 . 7554/eLife . 16011 . 019Figure 9 . Model for partitioning of metabolically unstable LBR variants between ER- and INM-resident protein turnover pathways . C-terminal truncation of LBR ( e . g . in LBR 1600* or 1400TΔ ) causes the sterol reductase domain ( SRD ) to misfold , leading to LBR ubiquitylation , membrane dislocation and its subsequent degradation via the Ub/proteasome pathway . A minor portion of LBR1600* is degraded by the canonical ERAD pathway in the ER , whereas the majority will enter the nucleus by virtue of the correctly folded N-terminal domain and be retained by binding to the nuclear lamina . Here , a presumably ERAD-independent pathway is operative , involving both p97 and the Ub/proteasome . Turnover at the INM can be inhibited at the dislocation or degradation step by a dominant negative p97 variant ( p97 QQ ) or the proteasome inhibitor MG132 , respectively . Note that several ERAD components as well as p97 cofactors are omitted for clarity . ER , endoplasmic reticulum; INM , inner nuclear membrane; ONM , outer nuclear membrane; Ub , ubiquitin . DOI: http://dx . doi . org/10 . 7554/eLife . 16011 . 019 In this study , we assign human LBR to an essential role in cholesterol synthesis . We found LBR to be essential for survival under cholesterol-restrictive growth conditions in three different human cell types ( Figures 2C , 3A–D ) . Correspondingly , LBR-deficient HeLa cells cannot efficiently synthesize cholesterol ( Figure 2D ) , but are readily rescued by the addition of LDL particles or free cholesterol to the cholesterol starvation medium ( Figure 2C ) . These results are unexpected since previous studies in mice reported that the sterol reductase activities of LBR and TM7SF2 are functionally redundant , suggesting that HEM dysplasia is likely a laminopathy that is unrelated to the sterol reductase activity of LBR ( Wassif et al . , 2007 ) . In cultured HeLa cells , however , LBR does not play a significant role in NE organization despite being a constitutively expressed , abundant INM component ( Figure 2A , B ) , possibly due to functional redundancy with some of the multitude of other lamina-associated membrane proteins ( Hetzer and Wente , 2009; Schirmer and Gerace , 2005 ) . In our opinion , a major role for TM7SF2 in generic cholesterol synthesis is difficult to reconcile with the absence of an overt phenotype upon deletion of TM7SF2 in mice , which are not compromised in cholesterol synthesis ( Bennati et al . , 2008 ) . In fact , TM7SF2 is expressed under cholesterol-restrictive conditions in all tested human cell lines ( Figure 3I ) , but cannot compensate for the absence of LBR . Nevertheless , it is striking that mice lacking a fully functional LBR allele due to homozygous mutations at the ichthyosis locus ( ic/ic ) are viable , although these animals display numerous phenotypic abnormalities including alopecia , syndactyly and hydrocephalus as well as an increase in embryonic lethality ( Shultz et al . , 2003 ) . Growth defects were also observed in primary cells isolated from mouse models with mutations in the LBR gene ( Subramanian et al . , 2012; Verhagen et al . , 2012 ) . Of note , defects in neutrophil maturation in ic/ic animals can be recued in vitro by expression of the LBR SRD ( Subramanian et al . , 2012 ) . It is likely that LBR function is subject to diversification in the course of evolution . For example , human LBR can rescue sterol reductase deficiency in yeast ( Silve et al . , 1998 ) , while LBR from Droshophila melanogaster does not complement this phenotype ( Wagner et al . , 2004 ) . Since D . melanogaster is a cholesterol-auxotrophic organism , we speculate that additional LBR functions could involve the N-terminal Tudor domain . While we found this domain to be dispensable for cholesterol synthesis ( Figure 5A ) , a possible role could involve heterochromatin organiziation ( Solovei et al . , 2013 ) , which may be related to the nuclear abnormalities observed in e . g . Pelger-Huet anomaly ( Hoffmann et al . , 2002 ) and mouse models of ichtyosis ( Shultz et al . , 2003 ) . Since the LBR SRD is neccesary and sufficient to restore cholesterol synthesis ( Figure 5A ) and rescues a defect in neutrophil maturation observed in ic/ic cells in vitro ( Subramanian et al . , 2012 ) , a knock-in of either the SRD or the Tudor domain ( including the first transmembrane domain ) into the LBR locus could help to deconvolute distinct functions of the SRD and Tudor domains in mammalian development . Given that our study establishes LBR as the major sterol reductase required for cholesterol synthesis in human cells , the question arises why a second , functionally equivalent enzyme is encoded in the human genome . Since cholesterol synthesis can proceed through differential , tissue-specific usage of distinct yet interchangeable segments of the established synthesis pathways ( Mitsche et al . , 2015 ) , we speculate that TM7SF2 could be required for cholesterol synthesis in a specific physiological context . Interestingly , a recent study suggested that TM7SF2 may participate in healing of burn wounds ( Lei et al . , 2016 ) , possibly representing a case of a more specialized physiological role . Based on our finding that all tested disease-causing mutations in LBR fail to complement the cholesterol auxotrophy imposed by an LBR deficiency in our tissue culture model ( Figure 4 ) , we believe that animal models of LBR malfunction warrant additional scrutiny from the perspective of cholesterol metabolism . The administration of a low cholesterol diet or inhibitors of cholesterol synthesis in animal models with LBR mutations could reveal previously unknown phenotypes and thus provide new insight into disease etiology . Similarly , it would be interesting to explore states in which specific cell types are exposed to conditions of low environmental cholesterol in a physiological context , as for example in the course of embryogenesis , which could help to rationalize the embryonic-lethal phenotypes in Greenberg skeletal dysplasia . Another unexpected outcome of this study is that C-terminally truncated LBR proteins found in both Pelger-Huët anomaly and Greenberg skeletal dysplasia are rapidly degraded by a proteasome-dependent protein quality-control pathway that appears to be distinct from the canonical ERAD pathway ( Figures 8 , 9 ) . Given that INM-resident proteins are synthesized in the ER prior to their targeting to the INM ( Lusk et al . , 2007; Ungricht and Kutay , 2015 ) , we propose that the correctly folded N-terminal moiety of LBR containing INM targeting information ( Smith and Blobel , 1993; Soullam and Worman , 1993 ) leads to the rapid trafficking of LBR 1600* to the INM before elements of the ERAD pathway can act ( Figure 8 ) . Considering that near-quantitative nuclear accumulation of an HA-tagged LBR variant is achieved in ~20 min ( Ungricht et al . , 2015 ) , and that the half life of LBR 1600* is ~10–15 min , it follows that a significant fraction of LBR 1600* escapes the ERAD system leading to localization to the INM mediated by the intact N-terminal domain , which confers lamin binding ( see . Figure 8B and Figure 8—figure supplement 1 demonstrating INM localization of LBR 1600* under steady-state conditions ) . Upon arrival of LBR 1600* at the INM , a system analogous to but distinct from the known ERAD system is responsible for the ubiquitylation of LBR 1600* and its p97-dependent extraction from the INM ( Figures 8 , 9 ) . Based on the unusually rapid kinetics and substantial protein accumulation in a dislocated nuclear state in presence of MG132 , we believe that the INM degradative system is different from the established ERAD machinery . In fact , it is still unclear which Ub ligases account for protein turnover at the INM of mammalian cells . In general , the mechanisms of protein quality control operative at the NE of budding yeast are far better understood than analogous systems in mammalian cells ( Deng and Hochstrasser , 2006; Foresti et al . , 2014; Khmelinskii et al . , 2014; Rose and Schlieker , 2012; Webster et al . , 2014 ) . The absence of suitable model substrates in higher eukaryotes has been the major limitation . This study establishes the long sought-after methodological framework to investigate NE-directed quality control mechanisms in human cells . We propose that the combination of ( i ) the extremely short half-life of LBR1600* , ( ii ) its near-exclusive localization to the INM as well as ( iii ) and ease with which LBR1600* can be arrested at distinct stages of dislocation and turnover make this LBR variant an ideal tool to pursue mechanistic studies aimed at the elucidation of protein quality control and turnover at the INM . All cell lines and primary cells were purchased from ATCC ( HFF-1: ATCC cat# SCRC-1041; HEK293T: ATCC cat# CRL-11268; HeLa: ATCC act# CCL-2 ) and regularly tested to be Mycoplasma-negative as judged by the absence of extranuclear DAPI staining . LBR knockout HeLa cells were generated using the CRISPR/Cas9 genome editing system as described previously ( Mali et al . , 2013; Turner et al . , 2015 ) . The CRISPR guide sequence LBR 5’-GACTCCCTCGGCGTCTGGAAGGG-3’ targeting the first exon of LBR was chosen from a published index of human exon gRNA targets ( Mali et al . , 2013 ) . Potential knockout colonies were harvested , expanded , and screened both by genotyping PCR and by immunoblotting . The genotyping primers used were LBR gt-F: 5’-TTCAAGCTCTGTTCC-3’ and LBR gt-R: 5’-TGTGTATGTATTGACTC-3’; GAPDH-F: 5’-CGACCGGAGTCAACGGATTTGGTCG-3’ and GAPDH-R: 5’-GGCAACAATATCCACTTTACCAGA-3’ . Illumina MiSeq of LBR knockout cell CRISPR target sites was performed in collaboration with the Yale Center for Genomic Analysis . Genomic DNA from LBR knockout FlpIn HeLa cells was harvested using QuickExtract DNA Extraction Solution ( Epicentre , Madison , WI ) according to manufacturer instructions . A 500 nucleotide region centered around the CRISPR target site of the LBR gene ( see Figure 2—figure supplements 1 and 2 ) was then PCR amplified using the primers LBR FW 5’-TAGTGTCACATAGATAACGCAGTGGCT-3’ and LBR RV 5’-CAAGAGCTCAATCCTCTGCCTTCA-3’ . The resulting mixture was then gel purified and submitted for a single lane of Illumina MiSeq sequencing , obtaining several million reads of the target region . Reads with complete sequence coverage of the target area were binned according to the mutation detected , resulting in the delineation of three separate LBR gene edits , corresponding to the three copies of the LBR gene found in HeLa cells . Electron microscopy was carried out by the Yale Biological Electron Microscopy facility as described previously ( Rose et al . , 2014 ) . Stable cell lines expressing the gene of interest under doxycycline control were generated using the FlpIn T-REx cell system ( Invitrogen ) as described previously based on HeLa cell obtained from ATCC CCL-2 ( Turner et al . , 2015 ) . LBR N547D , LBR R583Q , and LBR 1402TΔ constructs were generated via quickchange mutagenesis according to standard protocols . LBR 1600* , LBR ND , LBR TM1 , and LBR SRD were generated by PCR amplification of the relevant cDNA region of LBR as follows: LBR 1600* amino acids 1–534 , LBR ND amino acids 1–209 , LBR TM1 amino acids 1–246 , and LBR SRD amino acids 197–616 . The Sun2-LBR fusion construct was generated via fusion PCR and encompasses amino acids 1–177 of Sun2 fused to amino acids 197–616 of LBR . HeLa cells were grown on a 6-well plate and starved for 48 hrs in DMEM medium containing lipoprotein-deficient serum ( LPDS ) . Cell were metabolically labeled with 2 uCi/well [14C]-acetate ( Perkin Elmer ) for 4 hr at 37°C as described previously ( Zelcer et al . , 2014 ) . Cells were lysed and saponified . The lipids were extracted three times with 2 ml hexane and dried under nitrogen stream . Extracts were re-dissolved in 60 μL hexane and aliquots were separated on a Silica Gel 60 F254 plate ( Merck ) with a mobile phase of hexane: diethyl ether: glacial acetic acid ( 60:40:1 , v/v/v ) as described by Gill et al . ( 2011 ) . The TLC plate was exposed to an imaging plate ( Fujifilm ) and visualized with a Storm Scanner ( GE Healthcare ) . Wild-type or LBR KO HeLa cells were seeded in a 12-well plate at a density of 1x105 cells/well and transfected with 50 nM of control siRNA or SMARTpool siRNA targeting to SREBP2 . After 48h , the cells were split 1:2 into two 12-well plates in normal medium or LPDS medium and incubated for another 2 days . Total RNAs were then isolated and transcribed into cDNA using SuperScript II reverse transcriptase ( ThermoFisher Scientific ) . The RT reactions were diluted 1:5 with water , and 1 . 25 μL were used in real-time PCR which is carried out using iQ SYBR Green mix and CFX Real-Time PCR Detection System ( Bio-Rad ) . Data was analyzed using △△Ct method , in which the △Ct was calculated first as Ct of internal control ( RPL32 ) was subtracted from each sample , and the △△Ct was further calculated by subtracting △Ct of control group from △Ct of each treated group , and final results were represented as 2 ( -△△Ct ) . Primer sequences used in qPCR are listed as follows ( 5’ to 3’ ) : RPL32 ( Forward: CGGCGTGCAACAAATCTTACTGTGCCG; Reverse: CCAGTTGGGCAGCTCTTTCC ) , SREBP2 ( Forward: CCGGGCGCAACGCAAAC; Reverse: CGCCCATGACACCCGACAA ) , LBR ( Forward: AGTATAGCCTTCGTCCAAGAAGA; Reverse: CAAAGGTTCTCACTGCCAGTT ) , TM7SF2 ( Forward: AACTCAGGCAATCCGATTTACG; Reverse: GGGTCGCAGTTCACAGAAATA ) , HMGCR ( Forward: AGGGGATGCCATGGGGATGA; Reverse: ACGGCTAGAATCTGCATTTCAGGG ) The TM7SF2 cDNA was cloned from HeLa cells by RT-PCR . Total RNAs were isolated from HeLa cells cultured in cholesterol-restrictive medium and reversed transcribed into cDNA using a cDNA amplification kit ( SMARTer RACE kit , Clontech ) . The TM7SF2 cDNA was amplified using a universal forward primer ( 5’-CTAATACGACTCACTATAGGGC-3’ ) and a gene-specific reverse primer ( 5’-TCAGTAGATGTAGGGCATGATGCG-3’ ) . The resulting PCR product was subjected to Sanger sequencing . Immunoblotting was performed according to standard protocols in 5% ( wt/vol ) skim milk in Tris-buffered saline and 0 . 1% ( vol/vol ) Tween 20 ( TBS-T ) using Western Lightning plus ECL reagent ( Perkin Elmer ) . The antibodies used in this study were the following ( numbering according to http://antibodyregistry . org ) : anti-LBR N-terminal domain ( AB_775968 , Abcam ) at 1:2 , 000 , anti-LBR C-terminal domain ( AB_10712378 , Abcam ) at 1:2 , 000 , anti-Tubulin ( AB_477583 , Sigma ) at 1:2000 , anti- β-actin ( AB_306371 , Abcam ) at 1:2000 , and anti-TM7SF2 ( Covance custom antiserum ) . The specificity of the anti-TM7SF2 antibody was confirmed by inclusion of a TM7SF2 knockout cell lysate as a reference ( see Figure 3I ) . Immunofluorescence microscopy was performed as described previously ( Rose et al . , 2014 ) . The primary antibodies used were the following ( numbering according to http://antibodyregistry . org ) : anti-LBR N-terminal domain ( AB_775968 , Abcam ) at 1:500 , anti-Lamin B1 ( AB_10107828 , Abcam ) at 1:500 , anti-Lamin A+C ( AB_306913 , Abcam ) at 1:500 , anti-β-actin ( AB_306371 , Abcam ) at 1:1 , 000 , anti-Sun1 ( AB_1080462 , Sigma ) at 1:500 , anti-Sun2 ( Covance custom antiserum ) at 1:1 , 000 , anti-Lap1 ( Covance custom antiserum ) ( Turner et al . , 2015 ) ( Zhao et al . , 2013 ) at 1:1 , 000 , anti-Mab414 ( AB_448181 , Abcam ) at 1:500 , anti-hnRNP A1 ( AB_305145 , Abcam ) at 1:500 , anti-hnRNP A2B1 ( AB_732978 , Abcam ) at 1:500 , and anti-calnexin ( AB_1310022 , Abcam ) at 1:500 . 200 , 000 HeLa cells were counted in triplicate , washed with 1mL PBS , and then resuspended in lipoprotein-depleted growth medium ( DMEM + 10% lipoprotein depleted fetal bovine serum ) in 24-well plates . The cells were then cultured using standard mammalian tissue culture conditions for 7 days . All samples were split 1:3 on days 2 and 4 . If exogenous cholesterol or LDL was used in the experiment , it was introduced on day 2 and was continued through the end of the experiment . If expression of LBR from the FlpIn locus was used ( e . g . for LBR rescue ) , all cell samples were treated with doxycycline ( 500 ng/mL ) beginning on day 1 and continuing through the end of the experiment . After day 4 , the cells were grown until day 7 , at which time they were trypsinized , treated with 1:1 trypan blue to exclude nonviable cells , and then counted in triplicate . Human foreskin fibroblast cell ( HFF ) and 293T were seeded in a 12-well plate at a density of 1x105 cells/well one day prior to transfection . Cells were transfected with 50 nM of control siRNA or ON-TARGETplus SMARTpool targeting to LBR ( GE Dharmacon ) using Lipofectamine RNAiMax ( ThermoFisher Scientific ) . After 48h , cells were trypsinized , washed with PBS , and split 1:3 ( 293T ) or 1:5 ( HFF ) into 24-well plates in medium containing LPDS , and incubated for another 1 day ( 293T ) or 3 days ( HFF ) , respectively . The surviving adherent cells were quantified using crystal violet staining as described ( Zivadinovic et al . , 2005 ) . Total cholesterol measurements of cell extracts were performed using a fluorometric cholesterol + cholesterol ester detection kit ( Abcam ab65359 ) . Triplicate sets of cells treated for 7 days as described above and samples were subjected to fluorometric total cholesterol detection according to manufacturer instructions . Flag-tagged LBR and mutants were expressed in Expi293 cells ( Life technologies ) for 3 days as described by the manufacturer and purified as reported previously ( Zhao et al . , 2013 ) . The NADPH binding measurement was performed using intrinsic tryptophan fluorescence of LBR . In short , 150 µl of 10 µM purified LBR was titrated by increasing concentration of NADPH in elution buffer ( 20 mM HEPES , 150 mM NaCl , 5 mM MgCl2 , 5 mM KCl , 0 . 01% ( w/v ) n-dodecyl-D-maltoside ( DDM ) pH 7 . 5 ) . Fluorescence decrease ( Ex:295 nm / Em:335 nm ) upon NADPH addition was recorded on a spectrofluorometer ( Photon Technology International ) at 20°C and further corrected for the inner filter effect ( Woodyer et al . , 2005 ) . The fluorescence change was plotted against NADPH concentration and non-linearly fitted using GraphPad Prism .

In humans , mutations in the gene that encodes a protein called Lamin B receptor can lead to diseases ranging from harmless anomalies of blood cells to fatal developmental defects . The severity of the disease depends on the nature of the specific mutation , and whether one or both copies of the gene are affected . Lamin B receptor – or LBR for short – is found at the envelope that surrounds the cell’s nucleus and was previously proposed to anchor this envelope to an underlying scaffold to provide it with support . LBR can also catalyze a chemical reaction involved in producing cholesterol – an essential component of cell membranes . However , this enzymatic activity was assumed to be less important because a second enzyme named TM7SF2 can perform the same reaction . Thus , it was not clear – at the molecular level – why the mutations in this gene lead to a variety of diseases . All disease-causing mutations map to the part of LBR that is responsible for its enzymatic activity . This fact motivated Tsai , Zhao et al . to reassess the importance of LBR for the production of cholesterol . The experiments revealed that many human cells that can be grown in the laboratory strictly depend on LBR to produce cholesterol . As such , these findings challenge the previous assumption that TM7SF2 can compensate for the loss of LBR’s activity and sustain cholesterol synthesis . Tsai , Zhao et al . also discovered that all known disease-causing mutations strongly perturb LBR’s ability to engage in cholesterol synthesis , albeit through different mechanisms . Some mutations interfered with the enzyme ability to bind with an essential molecule or cofactor that is required to catalysis; others led to LBR rapidly degrading at the nuclear envelope . It was previously not known that proteins could be degraded at the inner membrane of the nuclear envelope of mammalian cells , and LBR mutants may turn out to be useful tools to investigate how this happens in future . Further studies could also test if other diseases caused by mutations in proteins found in the nuclear envelope act in similar ways , or if mutations in these proteins inhibit the nucleus’s protein disposal machinery .

2016

The Lamin B receptor is essential for cholesterol synthesis and perturbed by disease-causing mutations

Acidotoxicity is common among neurological disorders , such as ischemic stroke . Traditionally , Ca2+ influx via homomeric acid-sensing ion channel 1a ( ASIC1a ) was considered to be the leading cause of ischemic acidotoxicity . Here we show that extracellular protons trigger a novel form of neuronal necroptosis via ASIC1a , but independent of its ion-conducting function . We identified serine/threonine kinase receptor interaction protein 1 ( RIP1 ) as a critical component of this form of neuronal necroptosis . Acid stimulation recruits RIP1 to the ASIC1a C-terminus , causing RIP1 phosphorylation and subsequent neuronal death . In a mouse model of focal ischemia , middle cerebral artery occlusion causes ASIC1a-RIP1 association and RIP1 phosphorylation in affected brain areas . Deletion of the Asic1a gene significantly prevents RIP1 phosphorylation and brain damage , suggesting ASIC1a-mediated RIP1 activation has an important role in ischemic neuronal injury . Our findings indicate that extracellular protons function as a novel endogenous ligand that triggers neuronal necroptosis during ischemia via ASIC1a independent of its channel function . Acidosis occurs commonly in a variety of neurological disorders and is a main contributing factor to neural injury ( Xiong et al . , 2004; Wemmie et al . , 2013; Friese et al . , 2007; Vergo et al . , 2011 ) . For example , ischemic stroke causes pronounced brain acidosis ( ∼pH 6 . 0 ) , the treatment of which with NaHCO3 resulted in a significant reduction in the infract volume ( Pignataro et al . , 2007 ) . Interestingly , a similar neuroprotective effect was achieved via either pharmacological intervention or genetic deletion of acid-sensing ion channel 1a ( ASIC1a ) , suggesting a critical role for ASIC1a in mediating ischemic acidotoxicity ( Xiong et al . , 2004; Gao et al . , 2005; Pignataro et al . , 2007 ) . ASIC1a belongs to the H+-gated subgroup of the degenerin/epithelial Na+ channel ( DEG/ENaC ) family of non-selective cation channels , widely expressed in central ( CNS ) and peripheral ( PNS ) nervous systems ( Wemmie et al . , 2006 , 2013 ) . To date , besides ischemic stroke , accumulating evidence from cell/animal models shows that ASIC1a is an effective molecular target for mitigating acid-induced neural damage in many other diseases including multiple sclerosis , Huntington's disease , and Parkinson's disease ( Wemmie et al . , 2013 ) . Thus , these previous findings strongly suggest that ASIC1a is the key extracellular proton receptor in neurons and the main mediator of acid-induced neuronal death . As such , ASIC1a may be a potential broad-spectrum therapeutic target in many neurological disorders ( Xiong et al . , 2004; Wemmie et al . , 2006 , 2013 ) . However , despite the strong appreciation of its importance , the mechanism underlying ASIC1a-mediated acidic neuronal death remains poorly understood . Traditionally , Ca2+ influx through homomeric ASIC1a channels has been considered to be the main cause of acidotoxicity ( Xiong et al . , 2004; Yermolaieva et al . , 2004 ) . Although intracellular Ca2+ elevation is generally responsible for cell death mediated by Ca2+-permeable channels , for example , NMDA receptors ( Dong et al . , 2006; Lai et al . , 2014 ) , this hypothesis is incompatible with some intrinsic properties of ASIC1a channels . First , homomeric ASIC1a channels are completely desensitized after just a few seconds of continued exposure to extracellular acid , a phenomenon termed steady-state desensitization ( Krishtal , 2003; Duan et al . , 2011 ) . Second , compared to many other Ca2+-permeable channels , such as NMDA receptors , the Ca2+ permeability of ASIC1a channels is relatively small ( Samways et al . , 2009; Wang and Xu , 2011 ) . Thus , under pathological conditions with persistent acidosis ( e . g . , for hours ) , Ca2+ influx through ASIC1a channels should only occur in the first few seconds at the onset of acidosis , which would generate a negligible increase in intracellular Ca2+ ( Samways et al . , 2009 ) . It is unlikely that such a small change in the intracellular Ca2+ level can fully account for the dramatic neuronal damage mediated by ASIC1a under pathological conditions . Thus , additional mechanism ( s ) must be at work for ASIC1a-mediated neuronal death . Necrosis is an important form of cell death in development and diseases ( Syntichaki and Tavernarakis , 2003; Vandenabeele et al . , 2010 ) . Traditionally , necrosis was considered to be an accidental , uncontrolled form of cell death ( Syntichaki and Tavernarakis , 2003 ) . However , accumulating evidence now suggests that necrotic cell death may be accomplished by a set of signal transduction pathways and execution mechanisms ( thus termed necroptosis ) ( Vandenabeele et al . , 2010; Linkermann and Green , 2014 ) . Recently , the serine/threonine kinase receptor interaction protein 1 ( RIP1 ) was identified as the crucial mediator of this process , with RIP1 phosphorylation being the key step in necroptosis ( Degterev et al . , 2005; Christofferson et al . , 2014 ) . Blockade of RIP1 phosphorylation by the specific inhibitor , necrostatin-1 ( Nec-1 ) , inhibited necroptosis ( Degterev et al . , 2005; Christofferson et al . , 2014 ) . Necroptosis plays a critical role in many pathophysiological processes , such as ischemic injury and viral infection ( Vandenabeele et al . , 2010; Christofferson et al . , 2014; Linkermann and Green , 2014 ) . In most studies , necroptosis was mainly induced by the activation of death receptors ( DRs ) , for example , tumor necrosis factor-α ( TNF-α ) receptor , in the absence of caspase activation ( He et al . , 2009; Vandenabeele et al . , 2010 ) . Other endogenous initiators of necroptosis remain largely unknown . Here we report that extracellular protons trigger a novel form of necroptosis in neurons via ASIC1a , but independent of its ion-conducting function . Using RNA interference and pharmacological blockade , we identified RIP1 as a critical component in acid-induced neuronal death . RIP1 is recruited to ASIC1a within 30 min of acid stimulation , which causes RIP1 phosphorylation and triggers the downstream death events . Similarly , RIP1 became physically associated with ASIC1a and hyperphosphorylated in response to middle cerebral artery occlusion ( MCAO ) in mice . The enhanced RIP1 phosphorylation in response to either ischemia or acidosis was undetected in neurons from Asic1a−/− mice , demonstrating the involvement of ASIC1a-mediated RIP1 activation in ischemic brain injury . According to morphological appearance , cell death can be divided into apoptotic and necrotic death ( Kroemer et al . , 2009 ) . Either death form represents a specific set of signaling pathways and biochemical/cellular processes ( Kroemer et al . , 2009 ) . In order to classify acid-induced neuronal death , we first examined the morphological changes of cultured mouse cortical neurons exposed to acidosis using electron microscopy ( EM ) . While most neurons treated with a pH 7 . 4 solution ( Figure 1—figure supplement 1A , upper panel ) showed normal cellular morphology ( Figure 1A1 , left panel; Figure 1A2 , upper panel ) , those treated with a pH 6 . 0 solution ( 1 hr treatment and 24 hr recovery in normal culture medium , Figure 1—figure supplement 1A , middle panel ) displayed a typical necrotic phenotype ( Kroemer et al . , 2009 ) , including plasma membrane rupture , organelle swelling , and cell lysis ( Figure 1A1 , middle and right panels; Figure 1A2 , lower panel ) . No obvious apoptotic morphological change was observed , based on comparison with staurosporine-treated neurons ( data not shown ) . 10 . 7554/eLife . 05682 . 003Figure 1 . Acid ( pH 6 . 0 ) induces RIP1-dependent necrotic cell death in cultured mouse cortical neurons . ( A1 ) Electron microscopy images of neurons treated with pH 7 . 4 ( left ) or pH 6 . 0 solution ( middle and right ) . Of 57 cells counted in the pH 6 . 0-treated samples , 47 showed morphology similar to that shown in the middle and right panels . For pH 7 . 4-treated samples , the majority of cells had a similar morphology to that shown in the left panel; only 3 out of the 41 cells examined showed morphology that resembled that in the middle panel . ( A2 ) Enlarged images from the white boxes in A1 showing swelling of organelles in pH 6 . 0- but not pH 7 . 4-treated neurons . ( B ) PcTX1 ( 10 nM ) and Nec-1 ( 20 μM ) , but not BHA ( 100 μM ) , BEL ( 30 μM ) , DPI ( 15 μM ) , RTO ( 25 μM ) , CHX ( 100 μM ) , or z-VAD-fmk ( 10 μM ) , rescued cells from acid-induced neuronal death ( indicated by the dashed line ) ( n=4–12 , ***p<0 . 001; NS , no statistical significance , vs vehicle ( Veh ) at pH 6 . 0 ) . Inset: dose-dependence of the rescue by Nec-1 ( CTB assay , n=3–4 ) . ( C1 ) Rescue from acid-induced neuronal death by 20 μM Nec-1 ( propidium iodide [PI] staining assay ) . ( C2 ) Summary data for C1 . At least 200 neurons were counted for each condition ( ***p<0 . 001; NS , no statistical significance , vs Veh at pH 7 . 4 ) . ( D ) Knockdown efficiency of RIP1 shRNA as determined by Western blotting ( ***p<0 . 001 , vs β-Gal ) . ( E ) Rescue of acid-induced neuronal death by RIP1 shRNA ( CTB assay , n=3 , ***p<0 . 001; NS , no statistical significance , vs β-Gal at pH 7 . 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05682 . 00310 . 7554/eLife . 05682 . 004Figure 1—figure supplement 1 . Acid ( pH 6 . 0 ) treatment does not induce caspase 3/7 activation in cultured mouse cortical neurons . ( A ) Scheme of acid treatment and assay protocols . Unless indicated otherwise , all interventional drugs were applied 30 min before and were present during the pH 6 . 0 treatment . ( B ) Moderate activation of caspase 8 after pH 6 . 0 treatment ( n=3 , **p<0 . 01; NS , no statistical significance , vs pH 7 . 4 ) . ( C ) Caspase 3/7 were not activated after pH 6 . 0 treatment ( n=3 , *p<0 . 05; NS , no statistical significance , vs pH 7 . 4 ) . ( D ) Caspase 3/7 activity and neuronal death were tested 8 hr after 1 hr of pH 6 . 0 treatment from the same batch of neurons . Staurosporine ( STS , 100 nM ) strongly increased caspase 3/7 activity . However , no increase in caspase 3/7 activity was observed even though severe neuronal death occurred ( n=3 , for caspase 3/7 activity , ***p<0 . 001; NS , no statistical significance , vs pH 7 . 4; for neuronal death , ##p<0 . 01; ###p<0 . 001 , vs pH 7 . 4 ) . The dashed line indicates caspase 3/7 activity and neuronal viability at pH 7 . 4 . ( E ) No cleavage of caspase 3 ( C-caspase 3 ) was detected after treatment with the pH 6 . 0 solution . Cleavage was seen with the STS treatment . ( F ) z-VAD-fmk significantly inhibited STS-induced apoptotic neuronal death ( n=3 , ***p<0 . 001; NS , no statistical significance , vs vehicle ( Veh ) ; ###p<0 . 001 , vs STS+Veh ) , demonstrating that z-VAD-fmk was effective ( control for Figure 1B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05682 . 00410 . 7554/eLife . 05682 . 005Figure 1—figure supplement 2 . RIP1 mediates ASIC1a-dependent acid-induced neuronal death in cultured mouse cortical neurons . ( A ) No inhibitory effect of Nec-1 ( 40 μM , pre+co-treatment ) on I6 . 0 in cultured mouse cortical neurons ( n=5 ) was observed . Neurons were stimulated with the pH 6 . 0 solution and then treated with Nec-1 for 5 , 10 , and 30 min at pH 7 . 4 before the pH was switched to 6 . 0 in the presence of Nec-1 . ( B ) Pre+co-application of 7-Cl-O-Nec-1 ( Nec-1s , 20 μM ) , Nec-1 ( 20 µM ) , or PcTX1 ( 10 nM ) with the pH 6 . 0 solution suppressed acidic neuronal death ( LDH assay , n=3–5 , ##p<0 . 001 , vs vehicle ( Veh ) at pH 7 . 4; ***p<0 . 001 , vs Veh at pH 6 . 0; n . s . , no statistical significance , vs PcTX1 ) . ( C ) No neuroprotection by Nec-1 and Nec-1s when co-administered with the pH 6 . 0 solution was observed ( CTB assay , n=3–5 , ***p<0 . 001 , vs Veh at pH 7 . 4; n . s . , no statistical significance , vs Veh at pH 6 . 0 ) . ( D ) Geldanamycin ( GA ) treatment ( 1 mg/ml , 18 hr ) decreased the RIP1 expression level and acid-induced neuronal death . Note: GA also caused cell death by apoptosis , which was abolished by z-VAD-fmk ( CTB assay , n=3–5 , ***p<0 . 001; NS , no statistical significance , vs Veh at pH 7 . 4; ###p<0 . 001; n . s . , no statistical significance , vs Veh at pH 6 . 0 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05682 . 00510 . 7554/eLife . 05682 . 006Figure 1—figure supplement 3 . Acid ( pH 6 . 0 ) treatment does not induce reactive oxygen species ( ROS ) production in cultured mouse cortical neurons . ( A ) H2O2 but not the pH 6 . 0 solution caused a rapid increase in intracellular ROS ( stained with CM-H2DCFDA ) . ( B ) No change in the ROS level was observed during 1 hr treatment with the pH 6 . 0 solution ( n=25 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05682 . 006 Because our culture system lacks phagocytes , we measured caspase activities to evaluate the possible involvement of secondary necrosis following apoptosis ( Berghe et al . , 2010 ) . Although the activity of caspase 8 was moderately increased 4 and 8 hr after the pH 6 . 0 treatment ( Figure 1—figure supplement 1B ) , that of executive apoptosis effectors , caspase 3/7 , did not increase ( Figure 1—figure supplement 1C-E ) . Rather , caspase 3/7 activities decreased slightly at 4 hr ( Figure 1—figure supplement 1C ) and the cleavage of caspase 3 ( activated caspase 3 ) ( Figure 1—figure supplement 1E ) was undetectable by Western blotting . These data , together with the result that pan caspase inhibitor z-VAD-fmk failed to block pH 6 . 0-induced neuronal death ( Figure 1B , Figure 1—figure supplement 1F ) , suggest that acid triggers necrotic neuronal death but not apoptosis under the current experimental conditions . We then adopted a modulatory profiling strategy ( Wolpaw et al . , 2011 ) to explore the molecular mechanism ( s ) underlying acid-induced neuronal death . Based on the literature ( Berghe et al . , 2010; Wolpaw et al . , 2011 ) , we tested the effect of eight death modulators , including scavengers of reactive oxygen species ( ROS ) and inhibitors of Ca2+ signaling , protein synthesis , proteases , ASIC1a , and the main necroptosis mediator , RIP1 , all with 30 min pretreatment and then co-incubation with the pH 6 . 0 solution ( Figure 1—figure supplement 1A , lower panel ) . Cell viability was assessed by the Cell Titer Blue ( CTB ) assay . As reported previously ( Xiong et al . , 2004 ) , the ASIC1a specific inhibitor , psalmotoxin ( PcTX1 ) , attenuated acid-induced neuronal death . To our surprise , however , inhibiting RIP1 phosphorylation ( Figure 2A , B ) by Nec-1 ( Degterev et al . , 2005 ) also resulted in a similar protective effect ( IC50 , 12 . 6 μM , Figure 1B; CTB assay ) . Unlike PcTX1 , up to 30 min pretreatment with Nec-1 did not affect the acid ( pH 6 . 0 ) -evoked currents ( I6 . 0 ) in the neurons ( Figure 1—figure supplement 2A ) . 10 . 7554/eLife . 05682 . 007Figure 2 . Acid ( pH 6 . 0 ) induces RIP1 phosphorylation and physical association between RIP1 and ASIC1a . ( A ) Acid-induced phosphorylation of RIP1 and its inhibition by 40 μM Nec-1 , as measured by 32P incorporation . ( B ) Acid-induced phosphorylation of RIP1 and its inhibition by Nec-1 ( 40 μM ) and PcTX1 ( 50 nM ) , detected using the anti-phospho S/T antibody . ( C ) Acid failed to induce RIP1 phosphorylation in neurons from Asic1a−/− mice . Five-fold more proteins were loaded for Asic1a−/− samples than for Asic1a+/+ samples . ( D1 , D2 ) Time courses ( D1 ) and representative ratio images ( D2 ) of intracellular acidification of cultured mouse cortical neurons from Asic1a+/+ and Asic1a−/− mice in response to extracellular pH decrease from 7 . 4 to 6 . 0 , monitored using BCECF ( n=30 for each genotype , peak changes summarized in D1 ) . ( E ) In vitro RIP1 phosphorylation assay in pH 7 . 4 and pH 6 . 0 reaction solutions . Bands for phosphorylated RIP1 are indicated by the arrowheads . ( F1 , F2 ) Acid altered RIP1 expression levels with time . Shown are representative blots ( F1 ) and summary data ( F2 ) ( n=5 , **p<0 . 01 vs pH 7 . 4 , by paired t test ) . ( G1 , G2 ) pH 6 . 0 treatment caused association of RIP1 with ASIC1a ( G1 , IP ( immunoprecipitation ) RIP1 , IB ( immunoblotting ) ASIC1a; G2 , IP ASIC1a , IB RIP1 ) . ( H ) PcTX1 disrupted acid-induced ASIC1a–RIP1 association . ( I ) TNF-α neutralizing antibody ( 1 μg/μl and 2 μg/μl ) failed to rescue pH 6 . 0 solution-induced neuronal death ( CTB assay , n=3 , ***p<0 . 001 vs vehicle ( Veh ) at pH 7 . 4; n . s . , no statistical significance , vs Veh at pH 6 . 0 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05682 . 00710 . 7554/eLife . 05682 . 008Figure 2—figure supplement 1 . Controls for Figure 2G , H , I . ( A1 , A2 ) No ASIC1a or RIP1 was immunoprecipitated by mouse IgG ( Ms-IgG ) or goat IgG ( Gt-IgG ) . ( B ) TNF-α neutralizing antibody ( 1 μg/μl ) rescued TNF-α ( 10 ng/ml , added in culture medium ) -induced neuronal death ( CTB assay , n=3 , **p<0 . 01 , vs vehicle ( Veh ) without anti-TNF-α ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05682 . 008 As an alternative method for examining necrotic neuronal death , neurons were stained with propidium iodide ( PI ) ; the pH 6 . 0 treatment dramatically increased the number of PI-positive neurons . This effect was also suppressed by Nec-1 ( Figure 1C1 , C2 ) ; acid-treated neurons retained their healthy appearance of cell bodies and neurites in the presence of Nec-1 ( Figure 1C1 , bottom panel , DIC images ) , suggesting that neuronal functions may be preserved . Additionally , acid ( pH 6 . 0 ) -induced neuronal damage was detected using the lactate dehydrogenase ( LDH ) assay , which , unlike the CTB assay , is independent of mitochondrial metabolism , and the effect was inhibited by PcTX1 , Nec-1 , and 7-Cl-O-Nec-1 ( Nec-1s ) , a derivative with improved specificity for RIP1 ( Takahashi et al . , 2012 ) ( Figure 1—figure supplement 2B ) . Importantly , drug pretreatment was necessary for neuronal protection , as co-applied Nec-1 or Nec-1s failed to inhibit acidic neuronal death ( Figure 1—figure supplement 2C ) , suggesting a key role for RIP1 at the onset of acidotoxicity . Consistent with the pharmacological intervention , RNA interference of endogenous RIP1 protected neurons from acid-induced death ( Figure 1D , E ) . In addition , geldanamycin ( GA ) treatment , which greatly reduced RIP1 expression level ( Vanden Berghe et al . , 2003 ) , also significantly suppressed acid-induced cell death ( Figure 1—figure supplement 2D ) . A previous report showed that inhibitors of acid-induced neuronal death also suppressed ROS generation ( Liu et al . , 2009 ) , implying a potential pro-necrotic effect of ROS ( Vandenabeele et al . , 2010 ) on acid-induced necrosis . However , the pH 6 . 0 treatment resulted in no obvious change in ROS levels in the neurons ( Figure 1—figure supplement 3 ) and neither scavenging ROS by anti-oxidant butylated hydroxyanisole ( BHA ) nor blockade of ROS production by rotenone ( RTO ) or diphenylene iodonium ( DPI ) rescued neurons from acid-induced death ( Figure 1B ) . These results indicate that ROS production is not necessary for acid-induced cell death , consistent with the previous finding ( Xiong et al . , 2004 ) . Also , earlier reports showed that extracellular protons led to intracellular Ca2+ elevation ( Xiong et al . , 2004; Yermolaieva et al . , 2004 ) , which is one of the major causes of conventional necrotic cell death . Besides enhancing mitochondrial ROS production through activation of key enzymes of the Krebs cycle , Ca2+ also triggers cytosolic phospholipase A2 ( cPLA2 ) -mediated necrosis ( Vandenabeele et al . , 2010 ) . However , a cPLA2 inhibitor , bromoenol lactone ( BEL ) , failed to protect neurons from acid-induced death ( Figure 1B ) . Removing Ca2+ from the treatment solution also had no clear protective effect in the present study ( Figure 3D ) . The above results , thus , suggest that acid-induced neuronal death involves a form of RIP1-mediated necrosis which is independent of ROS and Ca2+ . Consistent with this idea , recent evidence showed that RIP1 was not involved in certain forms of intrinsic necrosis induced by ROS generators or Ca2+ ionophores ( Sun et al . , 2012; Wang et al . , 2012 ) . 10 . 7554/eLife . 05682 . 009Figure 3 . Ion-conducting function is not necessary for ASIC1a-mediated neuronal death . ( A1 , A2 ) Two-minute pretreatment with mild acidification greatly suppressed I6 . 0 in cultured mouse cortical neurons . Shown are representative current traces at −60 mV ( A1 ) and summary data for peak currents ( A2 ) ( n=4–6 , ***p<0 . 001; NS , no statistical significance , vs pH 7 . 4 ) . ( B ) Pretreatment with pH 6 . 8 ( Pre-6 . 8 ) for 10 min failed to prevent acid-induced neuronal death ( CTB assay , n=4–6 , ***p<0 . 001 , NS , no statistical significance , vs pH 7 . 4; n . s . , no statistical significance , vs pH 6 . 0 ) . Note: treatment with the pH 6 . 8 solution for 1 hr only did not alter neuronal viability . ( C ) Summary of propidium iodide ( PI ) -positive neurons for Asic1a+/+ and Asic1a−/− cultures treated with pH 7 . 4 ( CTRL ) or pH 6 . 0 solutions for 15 , 30 , and 60 min . Representative images are shown in Figure 3—figure supplement 1B . At least 400 neurons were counted for each condition ( ***p<0 . 001; NS , no statistical significance , vs corresponding Asic1a−/− cultures ) . ( D ) Neuronal death induced by 1 hr treatment with the pH 6 . 0 solution in normal ( SS , standard external solution ) , Na+- , K+- , Ca2+-free ( NMDG replacement ) , and Ca2+-free conditions ( CTB assay , n=3 , ##p<0 . 01 vs Na+- , K+- , Ca2+-free in pH 6 . 0; ***p<0 . 001 vs pH 7 . 4 under the same cation conditions ) . Dashed line , pH 6 . 0 solution-induced neuronal death under normal SS condition . ( E ) Representative traces of I6 . 0 for cortical neurons in normal ( SS ) , Na+- , K+- , Ca2+-free and Ca2+-free conditions . ( F ) Representative traces of I6 . 0 for wild type ( WT ) ASIC1a and its HIF and RC mutants expressed in CHO cells . ( G ) Current–voltage relationship of WT-ASIC1a ( filled squares ) and HIF-ASIC1a ( open circles ) in response to the pH 6 . 0 solution . Note: no current was induced by pH 6 . 0 in CHO cells that expressed HIF-ASIC1a at a broad range of holding voltages . ( H ) Summary data of cell death induced by 1 hr pH 6 . 0 solution treatment in Asic1a−/− neurons expressing GFP vector , WT-ASIC1a , and its HIF and RC mutants , based on PI staining of GFP-labeled neurons ( see Figure 3—figure supplement 6 for representative images , n=100 for each condition , ***p<0 . 001; NS , no statistical significance , vs pH 7 . 4 of the corresponding transfection; n . s . , no statistical significance , vs WT-ASIC1a ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05682 . 00910 . 7554/eLife . 05682 . 010Figure 3—figure supplement 1 . Time-dependence of acid-induced neuronal death . ( A ) Experimental scheme for Figure 3B , where mild acid pretreatment was used to induce ASIC1a channel steady-state desensitization before pH 6 . 0 treatment and subsequent death assay . ( B ) Dependence of neuronal death ( propidium iodide [PI] staining assay ) on the duration of exposure to pH 6 . 0 solution . Neurons were identified by DIC . DOI: http://dx . doi . org/10 . 7554/eLife . 05682 . 01010 . 7554/eLife . 05682 . 011Figure 3—figure supplement 2 . Fast perfusion is necessary for acid to induce [Ca2+]i elevation in ASIC1a-expressing neurons . ( A1 ) Acid ( pH 6 . 0 ) -induced changes in Fura-2 ratio in cultured wild type ( WT ) mouse cortical neurons in the absence and presence of a cocktail of inhibitors ( 20 μM AP5 , 20 μM CNQX , 1 μM TTX , and 5 μM nimodipine ) without or with 50 nM PcTX1 . Representative ratio images at indicated time points are shown above the trace . Fast perfusion ( ∼50 µl/min ) was used . ( A2 ) Summary data of normalized ratio changes after baseline subtraction for A1 ( n>10 , ***p<0 . 001 , vs SS without cocktail; ##p<0 . 001; n . s . , no statistical significance , vs SS with cocktail ) . ( B1–B4 ) pH 6 . 0-induced Fura-2 ratio changes in cultured cortical neurons from Asic1a−/− mice transfected with control vector ( GFP , B1 ) and GFP-tagged WT-ASIC1a ( B2 ) , HIF-ASIC1a ( B3 ) , or RC-ASIC1a ( B4 ) via fast perfusion ( ∼50 µl/min ) . Note: there was no change in control ( B1 ) and HIF-ASIC1a ( B3 ) -transfected neurons . ( B5 , B6 ) pH 6 . 0-induced Fura-2 ratio changes in cultured cortical neurons from Asic1a−/− mice transfected with GFP-tagged WT-ASIC1a ( B5 ) and RC-ASIC1a ( B6 ) via slow perfusion ( ∼15 µl/min ) . Similar results were obtained for at least 10 neurons each for B1–B6 . DOI: http://dx . doi . org/10 . 7554/eLife . 05682 . 01110 . 7554/eLife . 05682 . 012Figure 3—figure supplement 3 . Validation of the Y-tube apparatus at fast and slow perfusion rates . CHO cells were transiently transfected with the cDNA for rat TRPV1 ( A , B , EGFPC1-TrpV1 ) or mouse ASIC1a ( C , pEGFPC3-ASIC1a ) and used for whole-cell patch clamp recordings after 24–30 hr . The cell was held at −60 mV while an acidic solution ( pH 5 . 0 for A , pH 6 . 0 for B , C ) was applied either at a fast ( 50 μl/min ) or slow ( 15 μl/min ) rate as indicated . The same Y-tube apparatus was used for all experiments and in the same manner as used for the Ca2+ imaging experiments shown in Figure 3—figure supplement 2 . As summarized in D , the perfusion rate does not alter proton activation of the non-desensitizing TRPV1 currents , suggesting that the acidic solution effectively reached the cell under both conditions . Note: because the pH 6 . 0-evoked TRPV1 currents were small , we also used pH 5 . 0 to elicit larger and more reliable TRPV1 currents . On the other hand , the development of fast desensitizing ASIC1a currents was strongly affected by the perfusion rate of the pH 6 . 0 solution , giving slower and smaller currents at 15 μl/min than at 50 μl/min . DOI: http://dx . doi . org/10 . 7554/eLife . 05682 . 01210 . 7554/eLife . 05682 . 013Figure 3—figure supplement 4 . Acid-induced neuronal death is independent of Ca2+-flux and ionic currents via ASIC1a . ( A ) PcTX1 and Nec-1s rescued acid-induced neuronal death in the Ca2+-free solution ( CTB assay , n=3–5 , ***p<0 . 001 , vs vehicle ( Veh ) at pH 7 . 4; ##p<0 . 001 , vs Veh at pH 6 . 0 ) . ( B1 , B2 ) Inhibition of I6 . 0 by AMI ( 100 µM ) and PcTX1 ( 10 nM ) in cultured mouse cortical neurons . AMI was applied together with acid . PcTX1 was applied 60 min before acid . Both greatly suppressed I6 . 0 . Shown are representative current traces at −60 mV ( B1 ) and summary data for peak currents ( B2 ) ( n=4–6 , ***p<0 . 001 vs pH 6 . 0 only ) . ( C ) Upper , scheme of AMI , PcTX1 , and acid ( pH 6 . 0 ) treatment of cortical neurons . Lower , summary data for neuronal death under the conditions indicated ( CTB assay , n=4–6 , **p<0 . 01 , ***p<0 . 001 vs pH 7 . 4 ) . AMI pre+co: AMI pretreatment plus AMI co-application with acid ( pH 6 . 0 ) ; AMI co: AMI co-application with acid ( pH 6 . 0 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05682 . 01310 . 7554/eLife . 05682 . 014Figure 3—figure supplement 5 . HIF and RC mutants of ASIC1a are normally expressed on the plasma membrane . ( A ) Diagrams of the wild-type ( WT ) and mutant ASIC1a constructs . ( B ) Membrane expression of WT-ASIC1a and its HIF and RC mutants as determined by surface biotinylation experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 05682 . 01410 . 7554/eLife . 05682 . 015Figure 3—figure supplement 6 . Expression of WT-ASIC1a and HIF-ASIC1a , but not RC-ASIC1a , in Asic1a−/− neurons resulted in acid-induced death . Representative images for data shown in Figure 3H . Asic1a−/− neurons were transfected with GFP vector ( A ) , GFP-tagged WT-ASIC1a ( B ) , and its HIF ( C ) and RC ( D ) mutants . Transfected cells were identified by the green fluorescence signal . Cell death induced by 1 hr treatment with the pH 6 . 0 solution was assessed by propidium iodide ( PI ) staining ( red ) . Note the nuclear PI labeling of green cells ( yellow areas ) in enlarged images from white boxes in B and C of pH 6 . 0-treated samples . These cells were counted as PI-positive . DOI: http://dx . doi . org/10 . 7554/eLife . 05682 . 01510 . 7554/eLife . 05682 . 016Figure 3—figure supplement 7 . CHO cell death induced by acid ( pH 6 . 0 ) is dependent on ASIC1 and involves recruitment of RIP1 . ( A1 , A2 ) Acid ( pH 6 . 0 ) induced no detectable [Ca2+]i elevation ( by Fluo-4 ) in CHO cells transfected with mCherry-ASIC1a . Transfected CHO cells were identified by mCherry ( A2 , red image ) . Ionomycin ( 2 μM ) was added at the end of the experiment to show that cells were properly loaded and viable . ( B ) Cell death induced by 1 hr treatment with the pH 6 . 0 solution in CHO cells stably expressing GFP vector , WT-ASIC1a , and its HIF and RC mutants ( CTB assay , n=3 , ***p<0 . 001; NS , no statistical significance , vs pH 7 . 4; n . s . , no statistical significance , vs WT-ASIC1a ) . ( C ) Acid caused ASIC1a–RIP1 association in CHO cells that expressed WT-ASIC1a . Cells were treated or not with the pH 6 . 0 solution for 90 min . Cell lysates were subjected to immunoprecipitation by anti-ASIC1a and then Western blotting by anti-RIP1 . ( D ) Cell death induced by 1 hr treatment with the pH 6 . 0 solution in CHO cells transfected with pEGFP vector , ASIC1a , ASIC1b , ASIC2a , and ASIC3 ( CTB assay , n=3 , ***p<0 . 001; NS , no statistical significance , vs pH 7 . 4; n . s . , no statistical significance , vs ASIC1a ) . Both ASIC1a and ASIC1b , but not ASIC2a and ASIC3 , mediated acid-induced cell death . DOI: http://dx . doi . org/10 . 7554/eLife . 05682 . 016 RIP1 is a crucial mediator in many forms of necrotic cell death and its inhibition by Nec-1 ( Degterev et al . , 2005 , 2008 ) is neuroprotective in rodent disease models including ischemic stroke ( Cho et al . , 2009; He et al . , 2009; Zhang et al . , 2009 ) . To test whether acid causes RIP1 activation in neurons , we measured RIP1 phosphorylation , a common signature for RIP1-mediated necrotic death , using two methods: 32P incorporation and direct detection of phosphorylated RIP1 by an anti-phospho Ser/Thr antibody from anti-RIP1 immunoprecipitated samples . Both methods showed significant increases in RIP1 phosphorylation upon 30 min treatment with the pH 6 . 0 solution , which were suppressed by Nec-1 ( Figure 2A , B ) . Consistent with ASIC1a being critical for necrotic cell death and upstream of RIP1 activation , acid-induced RIP1 phosphorylation was inhibited by PcTX1 ( Figure 2B ) and was undetectable in neurons from Asic1a−/− mice ( Figure 2C ) . Intracellular acidification , which can be induced by extracellular acidosis ( Figure 2D1 , D2 ) , modulates many biochemical functions including kinase activities ( Kraut and Madias , 2010 ) . To rule out the possibility that acid-induced RIP1 phosphorylation resulted from intracellular acidification , we performed an in vitro cell-free phosphorylation assay and found that reducing the pH of the reaction solution to 6 . 0 did not alter RIP1 phosphorylation ( Figure 2E ) . Moreover , despite similar extracellular acid-induced intracellular acidification between neurons from Asic1a+/+ ( wild type , WT ) and Asic1a−/− mice ( Figure 2D1 , D2 ) , only WT , but not Asic1a−/− , neurons showed enhanced RIP1 phosphorylation in response to the extracellular pH reduction ( Figure 2C ) , suggesting that the presence of ASIC1a proteins rather than intracellular acidification was necessary for such a response . Interestingly , the total protein level of RIP1 was also increased in the first 30 min of acidosis treatment and then declined to ∼80% of the basal level at 60 min ( Figure 2F1 , F2 ) . Although a decrease in RIP1 level was previously observed in other forms of cell death ( Lin et al . , 1999; Van de Walle et al . , 2010 ) , its role in acid-induced neuronal death requires further exploration . Furthermore and possibly related to the acid-induced increase in RIP1 phosphorylation , we found that RIP1 was recruited to ASIC1a in mouse cortical neurons after 30 min treatment with the pH 6 . 0 solution ( Figure 2G1 , G2 , Figure 2—figure supplement 1A1 , A2 ) and that this effect was suppressed by PcTX1 ( Figure 2H ) . Taken together , these results suggest that acid-induced neuronal death is a programmed form of RIP1-dependent necrosis , or necroptosis ( Degterev et al . , 2005; Linkermann and Green , 2014 ) , which is dependent on the physical association of RIP1 with ASIC1a . Nearly all forms of previously reported necroptosis are induced by death receptor ( DR ) ligands , such as TNF-α , particularly in the presence of caspase inhibitors ( Vandenabeele et al . , 2010 ) . Recent studies showed that , under certain conditions , autocrine production of TNF-α could lead to RIP1-mediated necroptosis ( Biton and Ashkenazi , 2011; Wu et al . , 2011 ) . To evaluate the possibility that acid-induced necroptosis arose from autocrine production of cytokines , we first tested whether de novo protein synthesis was required for cell death and found that the protein synthesis inhibitor cycloheximide ( CHX ) failed to protect neurons from acid-induced death ( Figure 1B ) . Second , we used the TNF-α neutralizing antibody but found that it failed to rescue neurons from acid-induced death , while it was effective in protecting neurons from TNF-α-induced death ( Figure 2I , Figure 2—figure supplement 1B ) . Therefore , it is unlikely that any pro-necrotic protein production is involved in acid-induced necrotic cell death . Interestingly , Nec-1 significantly reduced acid-induced neuronal death without affecting I6 . 0 ( Figure 1—figure supplement 2A ) , suggesting that the ion-conducting function of ASIC1a was perhaps unrelated to RIP1-dependent acidotoxicity . To test this hypothesis , we examined the steady-state desensitization of proton-evoked currents in cortical neurons by moderate pH decreases , that is , to pH 7 . 2 or 6 . 8 . Just 2 min superfusion with the pH 7 . 2 or pH 6 . 8 solution drastically reduced I6 . 0 by approximately 60% or 90% , respectively ( Figure 3A1 , A2 ) . Therefore , ASIC1a channels in these neurons desensitized , leaving nearly no ionic flux during continued exposure to acid ( >2 min ) . In contrast , pretreatment of neurons with pH 6 . 8 solution failed to inhibit acid-induced neuronal death ( Figure 3B , Figure 3—figure supplement 1A ) , indicating that ASIC1a channels were able to mediate acidotoxicity even under steady-state desensitized conditions . As it is an ion channel , the ion-conducting function is generally considered a main consequence of ASIC1a activation . However , it is rather paradoxical that whereas the severity of acid-induced neuronal death correlates positively with the duration of acid treatment ( Xiong et al . , 2004; Duan et al . , 2011 ) , the ionic currents mediated by homomeric ASIC1a channels only last for seconds due to the complete steady-state channel desensitization at acidic pH , at least in in vitro conditions ( Krishtal , 2003; Duan et al . , 2011 ) . To examine the time dependence on acidosis of neuronal necrosis , we treated the cortical neurons with the pH 6 . 0 solution for different durations ( 0 , 15 , 30 , and 60 min ) and then returned them to the normal culture medium for 24 hr before staining with PI . As shown in Figure 3C and Figure 3—figure supplement 1B , the number of PI-positive neurons continued to increase with increasing durations of the pH 6 . 0 treatment in cultures prepared from Asic1a+/+ mice , indicating that despite the nearly complete channel desensitization throughout the major part ( except for the first few seconds ) of the 1 hr treatment , acid continued to exert an effect on neuronal death . On the other hand , consistent with previous findings ( Xiong et al . , 2004; Duan et al . , 2011 ) , acid treatment failed to induce significant neuronal death in cultures prepared from Asic1a−/− mice for all treatment durations ( Figure 3C , Figure 3—figure supplement 1B ) , indicating that despite the lack of proton-induced current , the neuronal death induced by persistent acidosis was still dependent on the presence of ASIC1a . Therefore , although ASIC1a expression was necessary , continued ASIC1a currents appeared not critical for acid-induced cell death . Ca2+ influx via homomeric ASIC1a channels has been considered to play a key role in acidic neuronal death ( Xiong et al . , 2004; Yermolaieva et al . , 2004; Wang and Xu , 2011 ) . Consistent with previous studies , application of the pH 6 . 0 solution to the cultured mouse cortical neurons elicited a rise in intracellular Ca2+ concentration ( [Ca2+]i ) , which was significantly reduced following inhibition of ionotropic glutamate receptors and voltage-gated Na+ and Ca2+ channels , while the remaining response was largely blocked by PcTX1 ( Figure 3—figure supplement 2A1 , A2 ) . The Ca2+ response was not detected in neurons from Asic1a−/− mice ( Figure 3—figure supplement 2B1 ) , but restored with the transient expression of ASIC1a cDNA in these neurons ( Figure 3—figure supplement 2B2 , B4 ) . However , the Ca2+ response was only seen with fast focal application of the pH 6 . 0 solution ( ∼50 µl/min ) . When the perfusion rate was decreased by approximately threefold ( ∼15 µl/min ) , the [Ca2+]i rise became very shallow or nearly undetectable ( Figure 3—figure supplement 2B5 , B6 ) . This sensitivity to the rate of acidification is likely related to the fast desensitization of acid-induced ASIC1a current , because the acid-evoked activation of TRPV1 ( either pH 6 . 0 or 5 . 0 ) , which did not desensitize , was unaffected by the perfusion rate ( Figure 3—figure supplement 3A , B , D ) . For fast desensitizing channels , it requires simultaneous activation of the majority of channels in the entire cell in order to give rise to robust whole-cell currents ( Figure 3—figure supplement 3C ) and [Ca2+]i elevation ( Figure 3—figure supplement 2 ) . Given that tissue acidosis occurs slowly under most pathological conditions , we reasoned that ASIC1a-mediated [Ca2+]i elevation during this process must be quite small and unlikely a major cause of the acid-induced neuronal death ( Wang and Xu , 2011 ) . Indeed , a Ca2+-free pH 6 . 0 solution ( with no Ca2+ added and the addition of 10 mM EGTA ) , which would not support Ca2+ influx , induced neuronal death to a similar level as the normal Ca2+-containing pH 6 . 0 solution ( Figure 3D ) . Importantly , acid-induced neuronal death in the Ca2+-free condition was also inhibited by PcTX1 and Nec-1s ( Figure 3—figure supplement 4A ) , indicating the involvement of the ASIC1a-RIP pathway despite the lack of Ca2+ influx . Notably , unlike I6 . 0 in ASIC1a-transfected CHO cells ( Duan et al . , 2011 ) , depletion of extracellular Ca2+ did not reduce I6 . 0 in cultured mouse cortical neurons ( Figure 3E ) , suggestive of a different modulatory mechanism exerted on native ASIC1a by extracellular Ca2+ under the present experimental conditions . To further examine the contribution of ion influx to acid-induced neuronal death , we replaced Na+ , K+ , and Ca2+ in the treatment solution with an impermanent cation , N-methyl-D-glucamine ( NMDG ) , keeping osmolarity unchanged . Under these conditions , no I6 . 0 was detected because of the lack of permeant ions ( Figure 3E ) , but acid still resulted in marked neuronal death ( Figure 3D ) . We also tested the non-specific ASIC blocker , amiloride ( AMI ) , which acts by blocking the channel pore . Although co-administration of AMI with acid significantly inhibited I6 . 0 ( Figure 3—figure supplement 4B1 , B2 ) , it did not show neuroprotection ( Figure 3—figure supplement 4C ) . These results further indicated that the ion fluxes or conducting function of ASIC1a might not be required for acid-induced neuronal death . Interestingly , if neurons were pretreated with AMI for 1 hr before acid stimulation , the acidotoxicity was largely prevented ( Figure 3—figure supplement 4C ) , suggesting that pretreatment with AMI may modulate the ASIC1a channel with a different mechanism beyond the channel blockade . Being an extracellular proton sensor , ASIC1a channels undergo conformational changes irrespective of the ion-conducting outcome . As such , the death pathway may only require the proton sensor function rather than ionic conduction . To test this possibility , we created two ASIC1a mutants: HIF-ASIC1a and RC-ASIC1a ( Figure 3—figure supplement 5A ) . While both were expressed normally on the plasma membrane ( Figure 3—figure supplement 5B ) , HIF-ASIC1a ( 32HIF34 mutated to 32AAA34 ) was ion non-conducting due to pore dysfunction ( Pfister et al . , 2006 ) ( Figure 3F , G ) . RC-ASIC1a was electrophysiologically functional as the WT channel ( Figure 3F ) , but its C-terminus ( R462-C526 ) was replaced by a shortened scrambled amino acid sequence , KLRILQSTVPRARDDPDLDN ( Figure 3—figure supplement 5A ) . By expressing WT-ASIC1a , HIF-ASIC1a , or RC-ASIC1a in cortical neurons from Asic1a−/− mice , we found that both WT-ASIC1a and the ion non-conducting HIF mutant , but not the conducting RC mutant , restored the acid-induced death in Asic1a−/− neurons ( Figure 3H , Figure 3—figure supplement 6 ) , although both WT- and RC-ASIC1a , but not the HIF mutant , restored the Ca2+ response to fast focal perfusion with the pH 6 . 0 solution ( Figure 3—figure supplement 2B2–B4 ) . Furthermore , although CHO cells expressing homomeric ASIC1a failed to yield a [Ca2+]i rise in response to acid ( Figure 3—figure supplement 7A1 , A2 ) , the stable expression of WT-ASIC1a and HIF-ASIC1a , but not RC-ASIC1a , in CHO cells also led to significant increases in acid-induced cell death ( Figure 3—figure supplement 7B ) . Supporting a similar ASIC1a/RIP1-mediated death mechanism in CHO cells as in neurons , pH 6 . 0 solution treatment also caused ASIC1a–RIP1 association in CHO cells that expressed ASIC1a ( Figure 3—figure supplement 7C ) . Further supporting the importance of the ASIC1a C-terminus in acid-induced necroptosis , we found that ASIC1b , a splice variant differing from ASIC1a only at the N-terminus , also mediated acid-induced death when expressed in CHO cells ( Figure 3—figure supplement 7D ) . In contrast , the expression of ASIC2a and ASIC3 , which have very different C-terminal sequences from ASIC1a ( Figure 4—figure supplement 1A ) , did not restore the acid-induced death of CHO cells ( Figure 3—figure supplement 7D ) . Because homomeric ASIC1b is Ca2+ impermeable ( Bassler et al . , 2001 ) , this finding also supports the argument that Ca2+ entry through ASIC1a is not critical for acid-induced cell death under the present experimental conditions . The above data , thus , strongly support the notion that ASIC1a-mediated neuronal death does not require the ion permeation ability of the ASIC1a channel . Rather , a death signal presumably located at the C-terminus of the channel protein and activated upon stimulation by extracellular protons might be responsible for the acid-induced cell death . Because conservation at the C-termini of ASIC isoforms is very low , the sequence alignment ( Figure 4—figure supplement 1A ) was uninformative about potential key amino acids for acidic neuronal death . We reasoned that synthetic peptides representing the critical motif ( s ) involved in acid-induced necroptosis might be able to mimic the action of the ASIC1a C-terminus on cell death . Therefore , we synthesized four peptides based on the mouse ASIC1a C-terminus , designated as CP-1 , 2 , 3 , and 4 ( see Figure 4A , Figure 4—figure supplement 1A ) . CP-1 covered a relatively less conserved region among ASIC C-termini than CP-2 , CP-3 , and CP-4 ( Figure 4—figure supplement 1A ) . All four peptides were tagged with the TAT sequence to facilitate penetration through the plasma membrane ( Figure 4A ) . Interestingly , incubation of mouse cortical neurons with CP-1 ( 10 µM , 24 hr ) , but not CP-2 , CP-3 , or CP-4 , at the physiological pH 7 . 4 , induced cell death ( Figure 4B ) and enhanced RIP1 phosphorylation ( Figure 4C ) , which were both rescued , at least partially , by Nec-1 ( Figure 4D , E ) . However , none of the peptides affected acid-induced neuronal death ( data not shown ) and in the presence of CP-1 , the pH 6 . 0 solution still induced further loss of neuronal viability ( Figure 4—figure supplement 1B ) . Moreover , the CP-1 peptide caused the death of CHO cells ( Figure 4—figure supplement 1C ) , which do not endogenously express ASIC1a , suggesting that its toxicity does not require full-length ASIC1a . These data suggest that the proximal C-terminal region of ASIC1a included in the CP-1 peptide can induce neuronal death via activation of RIP1 , mimicking acid-induced necroptosis . 10 . 7554/eLife . 05682 . 017Figure 4 . A peptide representing the proximal C-terminal region of ASIC1a induces RIP1 phosphorylation and neuronal death . ( A ) Relative positions and amino acid sequences of four peptides representing different regions of the mouse ASIC1a ( mASIC1a ) C-terminus . ( B ) TAT-tagged CP-1 , but not CP-2 , CP-3 , or CP-4 , peptide ( 10 μM , 24 hr ) induced neuronal death at pH 7 . 4 ( CTB assay , n=3 , ***p<0 . 001; NS , no statistical significance , vs control [CTRL , the TAT peptide alone] ) . ( C , D ) CP-1 , but not CP-2 , CP-3 , or CP-4 , enhanced RIP1 phosphorylation ( C ) and the effect was blocked by Nec-1 ( 20 µM ) ( D ) . ( E ) Nec-1 ( 20 μM ) partially rescued CP-1-induced neuronal death ( CTB assay , n=3 , ***p<0 . 001 vs CTRL; ###p<0 . 001 vs CP-1 alone ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05682 . 01710 . 7554/eLife . 05682 . 018Figure 4—figure supplement 1 . Design and characterization of ASIC1a-derived peptides . ( A ) Alignment of ASIC C-termini of different isoforms from various species . Starting positions are indicated in parentheses . Locations of the four mouse ASIC1a C-terminal peptides are shown by the boxes . c , chicken; f , toad fish; h , human; m , mouse; r , rat; s , shark; z , zebrafish . ( B ) CP-1 failed to inhibit acid-induced neuronal death ( CTB assay , n=3 , ***p<0 . 001 vs pH 7 . 4; ##p<0 . 001 vs pH 6 . 0+CP-1 ) . ( C ) CP-1 , but not CP-3 , caused the death of CHO cells in the absence of ASIC1a ( CTB assay , n=3 , ***p<0 . 001; NS , no statistical significance , vs CTRL , the TAT peptide alone ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05682 . 018 The above experiments firmly established the role of ASIC1a–RIP1 coupling in the acid-induced death of cultured cortical neurons . To verify the involvement of this pathway in ischemic brain injury , which is accompanied by severe tissue acidosis , with pH values typically falling to 6 . 5–6 . 0 ( Bassler et al . , 2001; Xiong et al . , 2004; Linkermann and Green , 2014 ) , we used a transient MCAO model to mimic ischemic stroke in mice ( Figure 5A ) . Supporting the role of ASIC1a in mediating RIP1 activation in ischemic brain damage , we detected a physical association between RIP1 and ASIC1a only in the ischemic hemisphere but not the control hemisphere of the same brain ( Figure 5B ) . Importantly , 30 min MCAO , which was too short to cause obvious brain damage as shown by TTL staining ( Figure 5C ) , was sufficient to induce the ASIC1a-RIP1 association . Such association persisted for up to at least 2 hr ( Figure 5D ) , the longest duration of MCAO tested ( Figure 5C ) . The long time window of ASIC1a-RIP1 complex formation is consistent with the notion that acidosis occurs slowly and progressively in the ischemic brain and both acidosis and necroptosis contribute mainly to delayed ischemic brain injury ( Degterev et al . , 2005; Pignataro et al . , 2007 ) , a phase of neuronal damage with high clinical relevance because of the need for post-ischemic neuroprotection following stroke . Interestingly , the total protein level of RIP1 tended to decrease in the ischemic side of the brain ( Figure 5D ) , similar to that observed in cultured cortical neurons subjected to acid treatment ( Figure 2F1 , F2 ) . This lends additional support to the involvement of a similar regulatory mechanism between acid-induced neuronal death in vitro and ischemic brain damage in vivo . The phosphorylation of RIP1 is a critical event in brain ischemia and the blockade of RIP1 function with Nec-1 can significantly prevent ischemic brain injury ( Degterev et al . , 2005 , 2008 ) . Consistent with previous reports ( Degterev et al . , 2005 , 2008 ) , we observed increased RIP1 phosphorylation in ischemic brain from WT mice , which was successfully removed by treating the RIP1 immunoprecipitant for 1 hr with a bovine intestinal alkaline phosphatase ( Figure 5E , upper ) . By contrast , the level of RIP1 phosphorylation in the ischemic hemisphere of the Asic1a−/− mice remained unchanged as compared to the control hemisphere ( Figure 5E , lower ) , demonstrating the critical role of ASIC1a in ischemia-induced RIP1 activation . 10 . 7554/eLife . 05682 . 019Figure 5 . RIP1 is recruited to ASIC1a in ischemic brain . ( A ) Cerebral blood flow ( CBF ) of ischemic brain hemisphere before and during middle cerebral artery occlusion ( MCAO , arrow ) was monitored by transcranial laser Doppler . C , control hemisphere; M , MCAO hemisphere . ( B ) One-hour MCAO treatment caused association of RIP1 with ASIC1a ( upper , IP: ASIC1a , IB: RIP1; lower , IP: RIP1 , IB: ASIC1a ) . Both bands in IB: RIP1 represent RIP1; the upper band may represent phospho-RIP1 or ubiquitinated-RIP1 as shown in previous studies ( Cho et al . , 2009; He et al . , 2009 ) . ( C ) TTC-staining of ischemic brain slices after 0 . 5 , 1 , and 1 . 5 hr of MCAO treatment followed by 24 hr reperfusion . The 2 hr MCAO was lethal . ( D ) RIP1-ASIC1a association in control ( C ) and MCAO hemisphere ( M ) following MCAO with durations as indicated ( IP: ASIC1a; IB: RIP1 ) . The levels of ASIC1a and RIP1 were assessed by IB . Note: total levels of RIP1 were reduced in the MCAO hemisphere compared to control , despite the increased association with ASIC1a . ( E ) Ischemia-induced RIP1 phosphorylation was abolished in Asic1a−/− brain . Upper , phosphorylation of RIP1 in ischemic brain of wild type ( WT ) mice , which was largely removed by 1 hr phosphatase treatment . Lower , Asic1a gene deletion prevented the ischemia-induced increase in RIP1 phosphorylation . ( F ) Schematic of a possible mechanism of acid-induced necroptosis . Upper , under normal physiological pH , the C-terminus of ASIC1a is protected by being buried inside or bound by an unknown protein; lower , acid stimulation exposes the CP-1 region of the ASIC1a C-terminus , allowing for association with and activation of RIP1 , which in turn leads to necroptosis . DOI: http://dx . doi . org/10 . 7554/eLife . 05682 . 019 In the present study , we describe a novel form of neuronal necroptosis induced by extracellular acidosis and mediated by ASIC1a . Compared to conventional DR-dependent necroptosis , ASIC1a-dependent necroptosis did not require de novo synthesized ligands and extracellular protons may serve as the fast ‘extrinsic death signal’ . ASIC1a-dependent necroptosis does not require intrinsic ROS generation ( Figure 1B , Figure 1—figure supplement 3 ) , suggesting a different molecular mechanism from TNF-α-induced necroptosis . It has been shown previously that some forms of necroptosis are autophagy-dependent and do not rely on ROS ( Bonapace et al . , 2010 ) . Notably , acidosis was reported to induce cell autophagy ( Wojtkowiak and Gillies , 2012 ) , suggesting a possible autophagy mechanism underlying ASIC1a-dependent necroptosis . Thus , further studies are needed to elucidate the detailed underlying mechanism ( s ) and clarify the similarities/differences between DR-dependent and ASIC1a-dependent necroptosis . Furthermore , as mentioned in the Introduction , deletion of the Asic1a gene strongly protects against ischemic neuronal death in a mouse model of focal ischemia ( Xiong et al . , 2004 ) . A similar neuroprotective effect of Nec-1 was also observed in the same model ( Degterev et al . , 2005 ) . Additionally , both ASIC1a and RIP1 were reported as potential therapeutic targets in traumatic brain injury ( You et al . , 2008; Yin et al . , 2013 ) . Importantly , we show here that RIP1 was recruited to ASIC1a in ischemic brains and the loss of ASIC1a prevented ischemia-induced RIP1 phosphorylation ( Figure 5A–E ) , suggesting that ASIC1a-mediated RIP1 activation is a key step in ischemic brain injury . The inability to activate RIP1 probably explains the resistance of Asic1a−/− mice to ischemic brain damage . These findings , altogether , strongly suggest ASIC1a is an important up-stream factor regulating RIP1 in vivo . Because both ASIC1a and RIP1 are widely expressed in nervous systems ( Degterev et al . , 2005; Wemmie et al . , 2006 ) and tissue acidosis is a common feature of many neurological diseases , this novel DR-independent necroptosis probably contributes to neuronal injury in a broad range of neurological disorders . We recently demonstrated the expression of ASIC1a in mitochondria ( mtASIC1a ) and found that it plays an important role in ROS-induced and mitochondrial permeability transition ( MPT ) -dependent neuronal death ( Wang et al . , 2013 ) . The mtASIC1a functions quite differently from the plasma membrane ASIC1a . Although ASIC1a null neurons resisted H2O2-induced neuronal death , this effect was not reproduced with treatment with PcTX1 , which cannot gain access to mtASIC1a localized in the inner mitochondrial membrane ( Wang et al . , 2013 ) . However , PcTX1 effectively inhibited acid-induced neuronal death through blocking plasma membrane ASIC1a . Furthermore , whereas mtASIC1a mainly contributed to ROS-induced neuronal death , ROS production was not involved in acid-induced neuronal death ( Figure 1B , Figure 1—figure supplement 3 ) . Therefore , the death mechanism examined in the current study does not appear to involve mtASIC1a . Here we show the indispensible role of ASIC1a channels in mediating neuronal necroptosis in response to extracellular acidification , but the ion-conducting function of the ASIC1a channels appears to not be essential for this process . Although it cannot be ruled out that ionic fluxes through the channel pore may play a modulatory role , our data suggest that RIP1 activation but not ionic conduction per se is necessary for acid-induced necroptotic cell death ( Figure 5F ) . Since ion channel activation reflects conformational changes induced by ligands and other gating factors , it is plausible that the acid-induced conformational change in ASIC1a channels exposes its proximal C-terminal region , represented by the CP-1 peptide ( Figure 4 ) , to trigger RIP1 phosphorylation ( Figure 5F ) and the consequent neuronal necroptosis in response to tissue acidosis . Ironically , steady-state desensitization of the ASIC1a channel by moderate pH decreases , although it suppressed subsequent activation of ASIC1a current , did not prevent acid-induced neuronal death ( Figure 3B ) . This differed from the neuroprotective effect of PcTX1 , which is believed to inhibit ASIC1a activation by causing steady-state channel desensitization at neutral and alkaline pH ( Chen et al . , 2005 , 2006 ) . Presumably , extracellular protons cause at least two steps in conformational changes in ASIC1a: one that exposes the C-terminal RIP1 activation domain and the other that mediates channel gating . Only the latter is blocked by the steady-state desensitization induced by the moderated pH decrease , whereas both are inhibited by PcTX1 , especially because of the presence of PcTX1 throughout the entire period of pH 6 . 0 treatment . This could not be the case for pH 6 . 8 pretreatment as the switch to the pH 6 . 0 solution necessarily eliminated the condition ( pH 6 . 8 ) that caused steady-state desensitization in the first place and the C-terminal RIP1 activation domain eventually adopts the acid-induced conformation . Consistent with the two-step conformational changes , AMI was protective against neuronal death when applied before , but not during acidosis , which differed from its action in blocking channel conductance ( Figure 3—figure supplement 4B , C ) . Conduction-independent functions have been shown to contribute to various physiological and pathological processes for several other ion channels ( Kaczmarek , 2006; Levitan , 2006 ) . For example , the ether-à-go-go ( EAG ) K+ channel modulates cell proliferation via a mechanism independent of K+ flux conducted by the channel ( Hegle et al . , 2006 ) and certain Na+ channel β subunits contribute to cell adhesion without requiring Na+ permeability ( Kaczmarek , 2006 ) . Some channels are tightly associated with signaling molecules such as enzymes ( Kaczmarek , 2006; Levitan , 2006 ) . A typical example is the EAG channel , which regulates the activity of the mitogen-activated protein kinase ( MAPK ) pathway ( Hegle et al . , 2006 ) . In another example , GluN2B-containing NMDA receptors are directly bound to phosphatase and tensin homolog deleted on chromosome ten ( PTEN ) , cyclin-dependent kinase 5 ( cdk5 ) , and death-associated protein kinase 1 ( DAPK1 ) , to contribute crucially to excitotoxicity in ischemic brain ( Lai et al . , 2014 ) . Others are multifunctional proteins containing clearly separate enzymatic and channel domains ( Kaczmarek , 2006 ) . For example , the melastatin-related transient receptor potential 7 ( TRPM7 ) channel contains a kinase domain at its cytoplasmic C-terminus ( Runnels et al . , 2001 ) . For ASIC1a channels , whether this occurs directly via auto-phosphorylation of RIP1 with its own kinase activity or indirectly by other associated kinase ( s ) requires further experimentation . Although non-conducting functions have been reported for nearly every major class of ion channel ( Kaczmarek , 2006 ) , it remains unclear whether they represent exceptional cases in a few channel proteins or a general , but unrecognized , property of most ion channels . It is not uncommon for a protein to be multifunctional , but for an ion channel , the ion-conducting function probably attracts the most attention because of the well-developed methodologies and the rich information that can be acquired to characterize the channel . As a result , the non-conducting functions of ion channels have frequently been overlooked . This appears to be the case for ASIC1a . Ever since the protective role of Asic1a gene deletion on ischemic brain damage was reported , efforts have been made to understand the underlying mechanism ( s ) and nearly all reported studies focused on the ionic conduction , especially Ca2+ influx , mediated by these channels ( Xiong et al . , 2004; Yermolaieva et al . , 2004 ) . However , as mentioned above and in accordance with the new evidence shown in the current study , the ionic conductance and Ca2+ toxicity hypothesis is incompatible with the observed time dependence of severity of acid-induced neuronal death , at least under in vitro conditions . Interestingly , the acid-induced [Ca2+]i rise and whole-cell currents in neurons , although highly correlated with ASIC1a expression , only occurred with fast focal application of the acidic solution ( Figure 3—figure supplement 2 , 3 ) , a condition that rarely happens during the development of tissue acidosis . Previously , partial protection was attained by reducing extracellular [Ca2+] to 0 . 2 mM , leading to the conclusion that Ca2+ influx was involved in acidic neuronal death ( Xiong et al . , 2004 ) . However , the reduced extracellular [Ca2+] could cause other complications that compromised necroptotic death . Therefore , we removed extracellular Ca2+ completely and observed no neuroprotection in the present study . We further show that the ability of ASIC1a to mediate acidotoxicity is independent of its ion-conducting function , but requires a C-terminal region of the channel protein . The sequence of CP-1 peptide that mimics acidosis in causing RIP1 phosphorylation and neuronal death represents a non-conserved region of ASIC C-termini ( Figure 4—figure supplement 1A ) , raising the possibility that the conduction-independent necroptotic effect of ASIC1a channels arose later in evolution than the channel conducting function , which is universal for nearly all ASICs . For neurons , a popular hypothesis is that protons co-released with neurotransmitters from acidic synaptic vesicles could activate postsynaptic ASIC1a , regulating physiological functions such as synaptic transmission , neuronal excitability , and learning/memory ( Wemmie et al . , 2006 , 2013 ) . However , although Asic1a−/− mice exhibit abnormal synaptic activity and learning/memory deficits ( Wemmie et al . , 2002 , 2006; Cho and Askwith , 2008; Urbano et al . , 2014 ) , the non-substantial contribution of ASIC1a channels to synaptic events , such as excitatory postsynaptic potentials ( EPSPs ) , is insufficient to account for the functions of these channels ( Alvarez de la Rosa et al . , 2003; Cho and Askwith , 2008; Kreple et al . , 2014 ) . It might be possible that a conduction-independent mechanism was also involved in the synaptic function of ASIC1a channels . A knock-in mouse model carrying the non-conducting HIF-ASIC1a mutant would be a useful tool to examine this possibility in future studies . The experimental protocols ( ethics protocol number: 2014022 ) were approved by the Animal Care and Use Committee of Shanghai Jiao Tong University School of Medicine , Shanghai , China . A transient focal ischemia model was prepared as described previously ( Xiong et al . , 2004; Duan et al . , 2011 ) . Briefly , animals ( male C57BL/6 mice , ∼25 g ) were anesthetized using 10% chloral hydrate with intubation and ventilation . Rectal and temporalis muscle temperature was maintained at 37 ± 0 . 5°C with a thermostatically controlled heating pad and lamp . A suture occlusion was made to the middle cerebral artery while cerebral blood flow ( CBF ) was monitored by transcranial laser Doppler . Animals whose blood flow did not reduce below 20% were excluded . For co-IP experiments , animals were killed and brains were removed immediately after MCAO of various durations of 0 . 5 , 1 , 1 . 5 and 2 hr . In other animals the suture was removed to allow reperfusion and the mice were euthanized 24 hr later . Brains were removed , sectioned coronally at 1 mm intervals , and stained with the vital dye 2 , 3 , 5-triphenyltetrazolium hydrochloride ( TTC ) . Under the present experimental conditions , 2 hr MCAO was lethal to the animals . Postnatal day 1 C57BL/6 WT or Asic1a−/− mice ( with a congenic C57BL/6 background ) were anesthetized with halothane . Brains were removed rapidly and placed in ice-cold Ca2+- and Mg2+-free phosphate-buffered saline ( PBS ) . Tissues were dissected and incubated with 0 . 05% trypsin-EDTA for 15 min at 37°C , followed by trituration with fire-polished glass pipettes , and plated in poly-D-lysine-coated 100 mm culture dishes ( 1 × 107 cells per dish ) or 24-well plates ( 1 . 5 × 106 cells per well ) . Neurons were cultured with Neurobasal medium supplemented with B27 and maintained at 37°C in a humidified 5% CO2 atmosphere incubator . Cultures were fed twice a week and used for all the assays 14–16 days after plating . Glial growth was suppressed by the addition of 5-fluoro-2-deoxyuridine ( 20 μg/ml; Sigma–Aldrich , St . Louis , MO ) and uridine ( 20 μg/ml; Sigma–Aldrich , St . Louis , MO ) . The virus-based RIP1 shRNA and negative control plasmids which had been successfully used in a previous study ( Zhang et al . , 2009 ) were kindly provided by Dr JH Han ( School of Life Science , Xiamen University , Xiamen , China ) . Briefly , RIP1 shRNA was designed to target mouse RIP1 with the sequence GCATTGTCCTTTGGGCAAT , and its effectiveness was tested by Western blotting . shRNA targeting an irrelevant gene β-galactosidase was used as a negative control with the sequence TTGGATCCAA . The cultured mouse cortical neurons were infected with lentivirus for RIP1 shRNA or the negative control shRNA at DIV 7 . Assays were performed 7 days after virus infection . Cortical neurons from Asic1a−/− mice were cultured in no . 0 glass bottom dishes coated with poly-D-lysine for 5 days and a half volume of medium removed for later use before transfection . CHO cells were grown in 35 mm dishes , 24-well plates , or glass coverslips for 1 day . Transfection was carried out using HilyMax ( Dojindo , Japan ) according to the standard protocol . Briefly , neurons ( in 1 ml of medium per culture ) or CHO cells ( in 2 ml of medium per culture ) were transfected with 0 . 5–1 μg of the desired plasmid: EGFP-vector , EGFP-tagged or untagged WT-ASIC1a ( human , EU078959 . 1 ) , HIF-ASIC1a , RC-ASIC1a , ASIC1b ( rat , EDL86977 . 1 ) , ASIC2a ( rat , NM_001034014 . 1 ) , ASIC3 ( rat , NM_173135 . 1 ) , or TRPV1 ( rat , NM_031982 ) and 2 µl HilyMax . For neurons , the transfection medium was replaced after 6 hr by a 1:1 mixture of the medium removed before the transfection and fresh culture medium . Cells were used at 48 hr after transfection . The GFP signal was used for the identification of transfected cells . Cultured mouse cortical neurons or brain tissues were collected and re-suspended in a lysis buffer [20 mM Tris-Cl , pH 7 . 4 , 150 mM NaCl , 1% Triton X-100 , 1 mM EDTA , 3 mM NaF , 1 mM β-glycerophosphate , 1 mM sodium orthovanadate , 2 mM N-ethylmaleimide and 10% glycerol , complete protease inhibitor set ( Sigma–Aldrich , St . Louis , MO ) , and phosphatase inhibitor set ( Roche , Switzerland ) ] . The re-suspended lysates were vortexed , incubated on ice for 40 min , and centrifuged at 13 , 000 × g for 15 min . The supernatant was incubated with 4 μg antibody overnight at 4°C . The following day , 20 μl protein G agarose beads were added to the sample and incubated for 2 hr at 4°C . Then , the beads were washed three times with the lysis buffer and the immunoprecipitants eluted with 2× loading buffer and subjected to Western blot analysis . ASIC currents were recorded using whole-cell patch-clamp techniques at room temperature ( 22–25°C ) . For voltage-clamp recordings , the membrane voltage was held at −60 mV . The standard external solution ( SS ) contained: 150 mM NaCl , 5 mM KCl , 1 mM MgCl2 , 2 mM CaCl2 , and 10 mM glucose , buffered to various pH values with 10 mM HEPES . The osmolarity of all solutions was kept at 300–330 mOsm/l . In NMDG-replacement and Ca2+-free experiments ( Figure 3D , E ) , external solutions were adjusted accordingly . The patch pipette solution contained: 120 mM KCl , 30 mM NaCl , 1 mM MgCl2 , 0 . 5 mM CaCl2 , 5 mM EGTA , 4 mM Mg-ATP , and 10 mM HEPES , pH 7 . 4 . All drugs for electrophysiological experiments were purchased from Sigma–Aldrich ( St . Louis , MO ) . A Y-tube apparatus was used for drug administration in electrophysiological and Ca2+-imaging ( see below ) experiments . The inner diameter of the open end of the Y-tube was ∼100 μm . The flow rate was adjusted by changing the height of the solution reservoir . The tip of the Y-tube was placed ∼500 μm away from the cell body in order to ensure complete exposure to the perfusion solution by the cells being recorded without distorting the cell shape due to solution flush . Cultured mouse cortical neurons grown on no . 0 glass bottom dishes were washed three times with SS ( pH 7 . 4 ) and incubated with 1 µM Fura-2 AM for 30 min at 37°C , followed by washing three times with SS . The dish was mounted on the stage of an inverted fluorescence microscope ( Nikon Eclipse TI , Japan ) and neurons were observed with a 20× objective lens . Fura-2 fluorescence images were acquired with alternating excitation wavelengths of 340 and 380 nm and an emission wavelength of 510 nm at 0 . 5 Hz while cells were continuously perfused with a Y-tube placed approximately 500 µm from the cells at a rate of ∼50 µl/min ( fast perfusion ) or ∼15 µl/min ( slow perfusion ) . The acidic ( pH 6 . 0 ) solution was applied through the Y-tube at the same rate . To block secondary activation of glutamate receptors and voltage-gated Na+ and Ca2+ channels , 20 μM AP5 , 20 μM CNQX , 1 μM TTX , and 5 μM nimodipine were included in the perfusate . PcTX1 ( 50 nM ) was included to inhibit ASIC1a . The Fura-2 ratio ( 340/380 ) was used to represent [Ca2+]i changes . CHO cells grown on glass coverslips transfected with mCherry-ASIC1a were washed twice with SS ( pH 7 . 4 ) and then incubated with 2 μM Fluo4-AM ( Dojindo , Japan ) in the presence of 0 . 02% Pluronic F-127 at 22°C for 60 min , followed by washing twice with SS ( pH 7 . 4 ) . The coverslip was transferred to a perfusion chamber , which was mounted on the stage of a Nikon Eclipse TI ( Japan ) . Fluo4 fluorescence images were taken at 0 . 3 Hz with excitation and emission wavelengths of 488 and 520 nm , respectively . Solution changes were achieved with the use of a Y-tube at ∼50 µl/min . Cultured mouse cortical neurons grown on glass coverslips were incubated with 5 μM BCECF-AM at 37°C for 30 min and then washed twice with SS ( pH 7 . 4 ) . The coverslip was transferred to a perfusion chamber , which was mounted on the stage of a Nikon Eclipse TI ( Japan ) . BCECF fluorescence images were taken with alternating excitation wavelengths of 490 and 440 nm and an emission wavelength of 535 nm at 0 . 1 Hz . Solution changes were achieved with the use of a Y-tube at ∼50 µl/min . Acid-induced neuronal death was achieved as described previously ( Xiong et al . , 2004 ) . First , cells were washed three times with the treatment solution ( 150 mM NaCl , 5 mM KCl , 1 mM MgCl2 , 2 mM CaCl2 , and 10 mM glucose , buffered to the desired pH value with 10 mM HEPES ) within 5 min at room temperature ( 22–25°C ) , and then incubated at 37°C for different time periods depending on experimental purposes . At the end of treatment , the solution was replaced with the normal pH culture medium and the culture resumed at 37°C for 24 hr . Cell viability was assessed by propidium iodide ( PI ) staining , lactate dehydrogenase ( LDH ) measurement , and the Cell Titer Blue ( CTB ) assay . Briefly , cells were stained with 10 μg/ml PI for 10 min at room temperature and then examined by fluorescence microscopy . NeuN-staining or DIC was used to distinguish neurons from glia . For the LDH assay , neurons were washed three times with the external solution and randomly divided into treatment groups . Neurons were washed and incubated in the normal culture medium at 37°C for 24 hr . The LDH level in the culture medium , indicative of cell death , was measured using the LDH assay kit ( Roche Molecular Biochemicals , Switzerland ) . An aliquot of the medium ( 100 μl ) was transferred from the culture wells to the wells of a 96-well plate and mixed with 100 μl of the reaction solution provided by the kit . Optical density was measured at 492 nm 45 min later using the SpectraMax Paradigm Multimode Microplate Reader ( Molecular Devices , Sunnyvale , CA ) . Background absorbance at 620 nm was subtracted . The maximal releasable LDH in each well was then obtained by a 15 min incubation with 1% Triton X-100 at the end of each experiment . For the CTB assay , neurons were cultured in the wells of 24-well plates . The amount of culture medium was adjusted to the same in each well ( 0 . 5 ml ) and pH 6 . 0 treatment was performed in the absence and presence of different drugs . After a return to normal culture for 24 hr , 0 . 1 ml CTB solution ( Promega , Madison , WI ) was added to each well and the plate incubated for 2 hr at 37°C . The fluorescence intensities ( excitation , 560 nm; emission , 590 nm ) , indicative of the amounts of viable cells , were measured using the SpectraMax Microplate Reader . All death assays were performed with four to eight repeats each time . Mouse cortices were collected and resuspended in a lysis buffer [20 mM Tris-Cl , pH 7 . 4 , 150 mM NaCl , 1% Triton X-100 , 5 mM EDTA , 3 mM NaF , 1 mM sodium orthovanadate , 10% glycerol , and complete protease inhibitor set ( Sigma–Aldrich , St . Louis , MO ) ] . The lysates were vortexed for 20 s and then incubated on ice for 40 min and centrifuged at 13 , 000 rpm for 15 min . The supernatant was incubated with 4 μg antibody against RIP1 ( BD Biosciences , San Jose , CA , 610458 ) overnight at 4°C . The next day , 20 μl protein G agarose beads were added to the sample and incubation continued for 2 hr at 4°C . Then , the beads were washed three times with the lysis buffer before being resuspended in 20 μl of the kinase reaction buffer ( 20 mM HEPES , pH 7 . 3 , 5 mM MgCl2 , and 5 mM MnCl2 ) . This was followed by the addition of 20 μM of cold ATP to initiate the kinase reaction , which lasted for 1 hr at 30°C and was terminated by the addition of 20 μl of 2× SDS loading buffer . The samples were vortexed , centrifuged for 1 min at 13 , 000 rpm , heated to 95–100°C for 5 min , and then cooled on ice for 1 min . After another centrifugation for 5 min at 13 , 000 rpm , the supernatants were subjected to Western blot analysis using the anti-phospho-S/T antibody ( Cell Signaling , Danvers , MA , phospho-PKA substrate , clone100 G7E ) to detect phosphorylated RIP1 . Immunoprecipitation of RIP1 was performed as described above . The immunoprecipitants were washed three times with a dephosphorylation buffer ( DB; 100 mM NaCl , 50 mM Tri-HCl , 10 mM MgCl2 , 1 mM dithiothreitol , pH 7 . 9 , and Sigma complete protease inhibitor set ) , then incubated in DB with 10 units of bovine intestinal alkaline phosphatase ( Sigma–Aldrich , St . Louis , MO , P0114 ) at 37°C for 1 hr before Western blot analysis for phosphorylated RIP1 was performed as described above . Cultured mouse cortical neurons were washed three times with pre-warmed SS ( pH 7 . 4 or pH 6 . 0 ) . Then , 2 ml pre-warmed SS ( pH 7 . 4 or pH 6 . 0 ) containing 1 mCi/ml 32P was added and cells were incubated for 30 min at 37°C . After incubation , the labeling medium was removed and the dish was washed with cold SS three times . Cells were then lysed in a lysis buffer [20 mM Tris-Cl , pH 7 . 4 , 150 mM NaCl , 1% Triton X-100 , 1 mM EDTA , 3 mM NaF , 1 mM β-glycerophosphate , 1 mM sodium orthovanadate , 2 mM N-ethylmaleimide and 10% glycerol , complete protease inhibitor set ( Sigma–Aldrich , St . Louis , MO ) , and phosphatase inhibitor set ( Roche , Switzerland ) ] , vortexed , incubated on ice for 40 min , and centrifuged at 13 , 000 × g for 15 min . The supernatant was incubated with 4 μg anti-RIP1 antibody overnight at 4°C . The next day , 20 μl protein G agarose beads were added to the sample , which was then incubated for 2 hr at 4°C . The beads were washed three times with the lysis buffer , the immunoprecipitants eluted with 2× loading buffer and were subjected to Western blot analysis . The gel was exposed to a phosphorimaging screen overnight to detect phosphorylated RIP1 . The screen was scanned using the STORM imaging system . Statistical comparisons were performed using unpaired or paired Student's t tests ( for data with non-normal distribution , the Kolmogorov–Smirnov test was used ) where values of p<0 . 05 are considered significant .

What happens in the minutes and hours after a stroke can determine how much brain damage occurs . In some types of stroke , a blood clot cuts off the blood supply to part of the brain , depriving the brain cells of oxygen and other nutrients , including glucose . One of the consequences is that the blood-starved brain becomes more acidic , which triggers cell death . Protecting brain cells from acidity-induced death could therefore reduce the damage caused by a stroke , and may also be an effective treatment for other brain disorders that involve increased brain acidity , like multiple sclerosis and Huntington's disease . To create such treatments , researchers must first understand how increased acidity in the brain triggers cell death . A protein called the acid-sensing ion channel 1a ( ASIC1a ) is thought to contribute to acid-induced cell death by allowing calcium to flow into cells . However , this increased flow of calcium occurs only briefly ( for seconds ) in response to increased acidity , which cannot explain why the severity of cell death strongly depends on the length of increased brain acidity that lasts for hours during stroke . Wang , Wang et al . now show that while ASIC1a is essential for acid-induced brain cell death , this is not because it allows calcium to enter cells . Instead , when acid levels increase , a protein called RIP1 comes to bind to one end of the ASIC1a protein . This causes the addition of a phosphate tag to RIP1 , an important cellular process well known to cause the cell to die . Wang , Wang et al . found that in mice genetically engineered to lack ASIC1a , the phosphate tag is not added to RIP1 , and the brain cells survive the increased acidity caused by stroke . This suggests that preventing ASIC1a and RIP1 from interacting could be a new way to protect brain cells from the increased acidity caused by brain diseases .

2015

Tissue acidosis induces neuronal necroptosis via ASIC1a channel independent of its ionic conduction

Measuring nascent macromolecular synthesis in vivo is key to understanding how cells and tissues progress through development and respond to external cues . Here we perform in vivo injection of alkyne- or azide-modified analogs of thymidine , uridine , methionine , and glucosamine to label nascent synthesis of DNA , RNA , protein , and glycosylation . Three-dimensional volumetric imaging of nascent macromolecule synthesis was performed in axolotl salamander tissue using whole-mount click chemistry-based fluorescent staining followed by light sheet fluorescent microscopy . We also developed an image processing pipeline for segmentation and classification of morphological regions of interest and individual cells , and we apply this pipeline to the regenerating humerus . We demonstrate our approach is sensitive to biological perturbations by measuring changes in DNA synthesis after limb denervation . This method provides a powerful means to quantitatively interrogate macromolecule synthesis in heterogenous tissues at the organ , cellular , and molecular levels of organization . The measurement of nascent DNA , RNA , and protein synthesis in animals provides critical information on the state of cells ( dividing , growing ) in relation to their surrounding cells . Traditionally , radiolabeled or brominated nucleosides/amino acids are introduced to live animals , where they are incorporated into macromolecules during DNA synthesis ( Sidman et al . , 1959; Gratzner , 1982 ) , transcription ( Uddin et al . , 1984; Wansink et al . , 1993 ) , and translation ( Garlick et al . , 1980 ) . Performing such approaches have allowed for the characterization of cells actively undergoing macromolecular synthesis as well as quantification of synthesis rates , which facilitates the study of cell behavior during tissue remodeling , proliferation , stress , or disease ( Yoshizawa et al . , 1997; Rombouts et al . , 2016; O'Brien and Lis , 1993; Rose et al . , 1975 ) . In the past decade , bio-orthogonal macromolecule precursor analogs including 5-ethynyl-2′-deoxyuridine ( EdU ) , 5-ethynyl-uridine ( 5-EU ) , and L-azidohomoalanine ( AHA ) have become commercially available for analysis of DNA synthesis , transcription , and translation , respectively . After injection of macromolecule precursor analogs into animals , precursors can later be detected in nascent DNA , RNA , and protein macromolecules with fluorescently labeled azides or alkynes through highly selective copper-catalyzed azide-alkyne cycloaddition ( ‘click’ ) chemistry ( Kolb et al . , 2001; Salic and Mitchison , 2008; Best , 2009 ) . These powerful new analogs provide an alternative to the use of dangerous isotopes and the challenges associated with brominated precursors such as the requirement of large secondary antibodies ( ~150 kd ) and harsh tissue retrieval methods that limit their use in whole tissue samples . Advantages of the click-chemistry labeling approach include the inert nature of macromolecule precursor analogs that have minimal impact on the animal , the small size of fluorescently labeled alkynes and azides , and the high selectivity of click-chemistry . These advantages have enabled whole-mount fluorescent labeling of DNA synthesis ( Salic and Mitchison , 2008 ) , RNA transcription ( Jao and Salic , 2008 ) , protein translation ( Hinz et al . , 2012 ) , and glycans ( Sawa et al . , 2006; Laughlin and Bertozzi , 2009 ) in animals . These pioneering proof-of-principle experiments have demonstrated that imaging macromolecular synthesis is possible , but the fact that most model organisms are large , optically opaque , and consist of heterogenous tissues has made it a challenge to image biological phenomena in deep tissues . Challenges such as photon penetration , differences in refractive indices among different cellular components , light-induced photodamage , and background fluorescence have limited the use of whole-mount imaging of macromolecular synthesis . Advances in light sheet fluorescence microscopes ( LSFMs ) have recently enabled the imaging of large biological specimens from millimeters to several centimeters in size ( Power and Huisken , 2017; Dodt et al . , 2007; Keller and Dodt , 2012 ) . LSFMs have been utilized for volumetric imaging of many varieties including visualization of mRNA in whole-mount fluorescence in situ hybridization experiments ( Mano et al . , 2018 ) and in vivo interrogation of deep tissue dynamics in transgenic reporter animals ( Tomer et al . , 2012 ) , among others ( Power and Huisken , 2017 ) . Clearing methods like CLARITY ( Chung et al . , 2013 ) , CUBIC ( Susaki et al . , 2014 ) , and 3DISCO ( Ertürk et al . , 2012 ) have further enabled volumetric imaging by decreasing refractive index mismatches and tissue clearing to advance large specimen imaging even further . Together , the rise of three-dimensional imaging , new staining techniques , and tissue clearing has demanded new means for cell counting , segmentation , and fluorescence quantification . Here we present a click-chemistry based method to visualize DNA synthesis , transcription , translation , and protein glycosylation in whole-mount samples using LSFM ( Figure 1A ) . We demonstrate the utility of this technique by imaging macromolecular synthesis in the regenerating axolotl salamander limb . Following limb amputation , the axolotl regenerates its limb by generating a mass of proliferating cells at the limb stump called a blastema ( Stocum , 2017 ) . The blastema is an ideal environment to test our method because it is an accessible , heterogenous tissue that increases DNA synthesis , transcriptional output , and translation rates compared to uninjured tissue . Furthermore , DNA , RNA , and protein synthesis decrease after denervation of the regenerating limb ( Singer and Caston , 1972; Dresden , 1969 ) . We use the regenerating limb to demonstrate that two click-it ready precursors can be administered and subsequently visualized simultaneously in a single sample . We also show that optical resolution of images can be improved with the clearing agent 2 , 2’-thiodiethanol ( TDE ) , and that this method works in a number of tissues . We outline an image analysis pipeline for three dimensional ( 3D ) morphology segmentation , cell counting , and fluorescence quantification of stained tissues . We apply this pipeline to the regenerating humerus to demonstrate the multiscale quantitative analysis capabilities of our method . Finally , we show that our method is sensitive enough to detect and quantify changes in DNA synthesis rates in whole-mount innervated and denervated regenerating limbs . Taken together , our method provides a unique approach to simultaneously interrogate cell state at the organ , cellular , and molecular levels of organization . To visualize macromolecule synthesis , we injected click-it compatible monomer analogs ( Table 1 ) intraperitonially 3 hr before sample collection . During this time , analogs metabolically incorporated into nascently synthesized macromolecules , resulting in in vivo labeling of macromolecule synthesis . Our group has found that collection of tissue after 3 hr provides sufficiently strong metabolic labeling of DNA synthesis ( EdU ) , transcription ( EU ) , translation ( AHA ) , and protein glycosylation ( GlcNAz ) . These labeled macromolecules contain either azide- or alkyne-modified monomers that can be detected with click-it compatible fluorescent molecules , enabling imaging of nascent macromolecules in whole-mount tissues with LSFM ( Figure 1B–C ) . We show that our whole-mount method can be used at the organismal level to visualize the whole torso of a stage 52 axolotl larvae , where we observe proliferating cells in the developing limbs and hematopoiesis within the spleen ( Figure 1B ) . To demonstrate the multiplexing capabilities of the approach , modified monomer analogs with disparate functional groups ( EdU/AHA , 5-EU/AHA , EdU/GlcNAz , 5-EU/GlcNAz ) were co-injected and visualized with LSFM ( Figure 2A–D , Videos 1–3 . For color blind accessible images , see Figure 2—figure supplement 1 ) . whole-mount samples were comparable to 2D longitudinal tissue sections of the same stains ( Figure 2I–L ) , showing that our method generates similar results in both whole-mount and tissue sections ( Figure 2—figure supplement 2 ) with subcellular resolution ( Figure 2—figure supplement 3 ) . Furthermore , the number of EdU+ cells per blastema area in a single Z slice obtained by LSFM ( EdU-AHA: 1954 cells/mm2 , EdU-GlcNAz: 1458 cells/mm2 ) is similar to that of a tissue section obtained by confocal microscopy ( EdU-AHA: 1272 cells/mm2 , EdU-GlcNAz: 1349 cells/mm2 ) . We finally demonstrate the specificity of GlcNAz incorporation by pretreating a GlcNAz-specific antibody on tissue sections collected from GlcNAz injected animals . By doing so , the subsequent click-it reaction was prevented due to the antibodies likely sterically hindering alkyne fluorophores from accessing the azide functional groups ( Figure 2—figure supplement 4 ) . An advantage of our method is that staining whole-mount tissues eliminates the need for sectioning , reducing the potential inconsistencies that arise as a result of the sectioning process ( uneven tissue , different cutting planes , etc . ) . Additionally , traditional methods of obtaining 3D images of thick tissues with confocal microscopy are impractically slow when imaging hundreds of images in a single stack . LSFM allows for more rapid imaging of whole samples , requiring only minutes to image each sample . However , several considerations exist when imaging whole tissue in 3D with LSFM . Stain penetrates slower in whole 3D tissues compared to 10–20 µm thick tissue sections , requiring longer staining times . An advantage of our method is that the click reaction requires molecularly diminutive reagents that readily pass through cell and nuclear membranes , ensuring stain penetration in the center of dense tissues . Different refractive indices between disparate cellular components and the imaging media cause light to scatter , which can reduce the resolution and brightness of 3D images ( Fadero and Maddox , 2017 ) . To improve image resolution and light penetration , overnight refractive index matching with 67% 2 , 2’-thiodiethanol ( TDE ) can sufficiently clear axolotl limbs for imaging with LSFM . TDE was chosen due to its simplicity , cost , safety , and compatibility with common imaging modalities . This clearing method rapidly and effectively improves the signal-to-noise ratio of stained samples compared to imaging in PBS ( Figure 2—figure supplement 5 ) . Tissue morphology was minimally disturbed after clearing , with only mild shrinkage of the blastema epithelium observed . With careful attention to these challenges , our method provides a means to obtain high-quality 3D images from tissues 1 mm in depth in less than 10 min per sample with clear , consistent staining ( Figure 2A–H , Figure 2—figure supplement 3 ) . To demonstrate that our whole-mount click-it method can obtain quantifiable data on the organ , cellular , and molecular levels of organization , we applied our technique to the regenerating humerus . After regenerating for 35 d , axolotls with mid-humeral amputations were injected with EdU/AHA to identify cells within the humerus undergoing DNA synthesis ( EdU ) and protein translation ( AHA ) . We observed EdU staining in chondrocytes distal to the amputation plane and AHA staining in the humerus perichondrium ( Figure 3A ) . We outline a multiscale , quantitative pipeline that leverages the staining patterns of these macromolecules for analysis of 3D humerus morphology and 3D macromolecule synthesis . This workflow combines available plugins in Fiji ( Schindelin et al . , 2012 ) and scripts developed in Fiji and Matlab ( The MathWorks , 2019 ) for the data analysis process ( see supplementary information for a detailed description ) . For 3D organ level analysis of macromolecule synthesis , AHA staining provided an adequate outline of the humerus , allowing us to segment its 3D morphology ( Figure 3A–B ) . We quantified organ shape by assessing cross-sectional area ( Figure 4A’ ) and circularity ( Figure 4A” ) of the segmented humerus along the proximodistal axis with BoneJ ( Doube et al . , 2010; Figure 4A ) . These measurements of 3D organ size can be obtained with other methods such as microCT and focused ion beam scanning electron microscopy ( FIB-SEM ) . However , microCT is unable to image every tissue due to stain limitations and uses hazardous radiation , while FIB-SEM can capture 3D surface topography but not the entire organ morphology . Our method has the capability to fully image the 3D structure of entire organs ( size permitting ) without the need for radiation while simultaneously capturing information from both the cellular and molecular levels of organization . On the cellular level , we segmented the 3D morphology of proliferating chondrocytes based on EdU staining using the Trainable Weka Segmentation 3D plugin ( Arganda-Carreras et al . , 2017; Figure 4B” ) . With this segmentation , we identify highly condensed regions of cells undergoing rapid rates of DNA synthesis ( Figure 4B ) . From these data , we observe cells synthesizing DNA most abundantly distal to the plane of amputation , as expected for dividing chondrocytes . Traditionally , cell quantification as such is conducted on 2D tissue sections . In heterogenous tissues , however , cells are distributed non-uniformly; 2D sampling may not accurately capture the cellular distribution and cannot be used to determine cell volume or shape . Our whole-mount staining method allows quantification of cells within an entire 3D tissue , resulting in a more accurate assessment of cell density within heterogenous tissues . To demonstrate the quantitative molecular analysis of 3D macromolecule synthesis , we assessed the staining intensity of EdU in slices along the proximodistal axis of the regenerating humerus ( Figure 4C ) . EdU intensity represents the rate at which a cell undergoes DNA synthesis , which is one of the first steps of cell division and proliferation . Proliferating cells rapidly synthesize DNA , which enables these cells to integrate EdU into nascent DNA strands . This provides ample opportunities for covalent linkage of fluorescent molecules to the DNA strand . Thus , higher pixel values within EdU+ cells are representative of more DNA synthesis in the cell . These data show that EdU intensity is strongest in the regions distal to the amputation plane of the humerus ( Figure 4C’ ) , providing a quantitative measure to assess macromolecule synthesis on the molecular level . Within a tissue , cells synthesize macromolecules heterogeneously , reflected by different fluorescence values between cells . Thus , quantifying macromolecule synthesis based on the presence/absence of signal instead of fluorescence does not account for variability in macromolecule synthesis rates among cells . To compound this issue , quantifying fluorescence in tissue sections only represents the rate of macromolecule synthesis from a fraction of cells in larger , heterogenous tissues . Our method provides a means to capture this molecular heterogeneity in 3D samples , allowing us to observe whole 3D regions in the regenerating humerus that synthesize DNA more rapidly than others , which further demonstrates the utility of our whole-mount click staining method . Taken together , these results demonstrate that our method can provide quantifiable data on the organ , cellular , and molecular levels of organization . This highlights the novelty of our method , as we have not found previous examples of multiscale analysis as outlined here . We foresee this multiscale , quantitative analysis having broad applications in the examination of dynamic cell processes in 3D , such as in cancer metabolism and mammalian neurogenesis or other fields where macromolecule synthesis is traditionally studied in tissue sections . To demonstrate that our method is sensitive enough to detect subtle changes in macromolecule synthesis in vivo , we quantified the difference in EdU intensity between limbs regenerating with and without a nerve supply . Blastema cells are thought to have a cell cycle length of 40–50 hr , with S phase approximately 30 hr ( Tassava et al . , 1987 ) . In newts , it has been shown that amino acid , RNA , and DNA analog incorporation decreases approximately 30% by 24 hr after denervation ( Singer and Caston , 1972 ) . Based upon these estimates , we amputated both forelimbs at the mid-humerus and denervated the left limb at the brachial plexus 24 hr before collection . This timepoint for denervation was chosen because it should be sufficiently long to have an impact on DNA , RNA , and protein synthesis rates . At 6 , 9 , 12 , 15 , 18 , and 25 d post-amputation ( dpa ) animals were pulsed with EdU for 3 hr before collection to label proliferating blastema cells ( Figure 5A ) . We chose 3 hr of EdU incorporation to provide a snapshot of DNA synthesis in the 30 hr total S phase of blastema cells . LSFM was used to image samples ( Figure 5B ) , ensuring pixel resolution was consistent between samples . We quantified DNA synthesis in denervated limbs compared to innervated limbs by creating a 175 × 175 × 175 µm cube 250 µm from the distal most tip of the blastema ( Figure 6A ) . The size and location of the quantification cube was chosen to maximize the number of EdU+ mesenchymal cells and to exclude any epithelial cells . Although we limited the cube to the size of the smallest blastema , these parameters can be customized depending on the size of the blastemas . For this study , we estimate 50–150 blastema cells are found within the cube . From these results , we observed a marked decrease in blastema EdU incorporation due to denervation at 9 , 12 , and 15dpa ( Figure 6B , Figure 6—figure supplement 1 ) , demonstrating that our whole-mount staining approach is capable of detecting changes in macromolecule synthesis after biological perturbations . One potential limitation of our method in the blastema is the inability to perform single cell segmentation . This is due to the density and abundance of cells within the blastema . We predict that higher magnification imaging and deep learning segmentation techniques may overcome this limitation but will significantly increase imaging time and file size . For comparison , we have included estimates of imaging time and file sizes for confocal microscopy with a 20X objective and LSFM with a 5× objective ( Supplementary file 1 ) . The work presented here provides a fast , simple pipeline for visualizing macromolecule turnover in the 3D space of whole tissues . While tissue sections were previously the standard for studying these cellular processes , a new standard in the field must be expected where dynamic processes like macromolecule synthesis are visualized in 3D to obtain a more complete understanding of how these processes occur in a larger tissue context . Few modalities of imaging exist to provide this level of analysis . Here , we outline a method to study macromolecule synthesis at the organ , cellular , and molecular levels of organization , which is important in understanding cell state and how cell state affects neighboring cells and tissues . To this end , we show high levels of DNA synthesis , transcription , translation , and protein glycosylation in the entire 3D space of the regenerating blastema after limb amputation and that these processes can be visualized concurrently . Our lab has also demonstrated this whole-mount technique in other axolotl tissues including the lung ( Jensen , 2018 ) . We outline a multiscale pipeline for analysis and quantification of heterogeneous tissues at the organ , cellular , and molecular levels of organization . Further , we demonstrate that our method is sensitive to detect biological perturbations by showing a decrease in DNA synthesis in the blastema following limb denervation . We foresee our method being used to similarly readdress other classical questions with modern techniques for a more exhaustive understanding of biological processes; traditional questions within the fields of cancer biology and neurobiology may especially benefit from technology as such . Additionally , as more click-it ready macromolecules monomer analogs are generated , our method will provide a means to study more biological processes in whole tissues . Finally , we expect that this method should be amenable both with other staining techniques such as whole-mount immunohistochemistry and to a number of animal models , including mouse and zebrafish . Axolotls were either bred in captivity at Northeastern University or purchased from the Ambystoma Genetic Stock Center at the University of Kentucky . Experiments were performed in accordance with the Northeastern University Institutional Animal Care and Use Committee . Animals were grown to 4–6 cm ( Mean 5 . 3 cm , SD 0 . 36 ) and 1–1 . 5 g ( Mean 1 . 3 g , SD 0 . 19 g ) for use in all studies . For all experiments , animals were anesthetized by treatment of 0 . 01% benzocaine until visually immobilized . Limbs were amputated either at the distal end of the zeugopod or midway through the stylopod , and bones were trimmed below the amputation plane to allow for uniform growth . At the date of collection , animals were reanesthetized and injected with either 5-ethynyl-2′-deoxyuridine ( EdU ) to identify proliferating cells ( 8 . 0 µg/g animal ) , 5-Ethynyl Uridine ( 5-EU ) to label RNA ( 270 . 0 µg/g animal ) , L-Azidohomoalanine ( AHA ) to label protein ( 180 . 59 µg/g animal ) , or N-azidoacetylglucosamine-tetraacylated ( GlcNAz ) to label glycosylated proteins ( 430 µg/g animal ) alone or simultaneously in the following combinations: EdU/AHA , EdU/GlcNAz , 5-EU/AHA , 5-EU/GlcNAz ( Table 1 ) . All monomer analogs were purchased from www . clickchemistrytools . com and resuspended in DMSO at the following concentrations: EdU- 300 mM , 5-EU- 100 mM , AHA- 100 mM , GlcNAz- 100 mM . Stocks were further diluted in 1× phosphate buffered saline ( PBS ) for injection . After 3hr of analog incorporation , limbs were collected from the upper stylopod and fixed in 4% paraformaldehyde ( PFA ) ( diluted in 1× PBS ) at 4°C overnight . If limbs were denervated , the nerve supply was severed at the brachial plexus 24 hr before tissue collection . Following fixation in 4% PFA , samples were washed three times with 1X PBS at room temperature ( ~23°C ) for 5 min . Samples were dehydrated in an increasing methanol series at room temperature starting with 25% methanol ( diluted in 1× PBS ) , 50% methanol , 75% methanol , and 100% methanol for 5 min at each step . Samples could then be stored in 100% methanol indefinitely at −20°C . For staining , samples were rehydrated in a decreasing methanol series starting with 75% methanol ( diluted in 1× PBS ) , 50% methanol , 25% methanol , and finally placed in 100% 1× PBS for 5 min at each step . Samples were then washed three times with 1× PBST ( 1× PBS with 0 . 1% Triton ) for 5 min at room temperature . To aid in clearing , samples were washed in 0 . 5% trypsin ( diluted in 1× PBS ) for 30–90 min on a rocker at room temperature , or until the sample appeared translucent . Samples were washed three times at room temperature for 5 min with deionized water , then washed in 100% acetone for 20 min at −20°C and washed with deionized water again for 10 min . Samples were washed in 1× PBST three times at room temperature for 5 min before applying click-it cocktail for overnight at room temperature . The click-it cocktail was made in 500 µl of 1× TRIS buffered saline as follows: 50 µL 1M sodium ascorbate ( 100 mM final ) , 20 µL 100 mM CuSO4 ( 4 mM final ) , and 2 µL 500 µM azide- or alkyne-modified Alexa Flour ( 2 µM final ) , combined in order as listed . After the first round of staining , samples were washed at room temperature six times for 30 min with rocking . For double-labelling , samples were again placed in the click-it cocktail with a different fluorescent dye to stain for the second analog at room temperature overnight . Both rounds of staining were conducted in the dark to prevent photodegradation of fluorescent molecules . For staining with DAPI , samples were washed three times for 5 min with 1× PBS , then placed in 2 . 86 µM DAPI for 4 d at room temperature . Samples were washed three times for 20 min with 1× PBS and left in 1× PBS at 4°C for short-term storage before imaging with LSFM . whole-mount samples were cleared with 67% TDE ( diluted in 1× PBS ) overnight at room temperature in the dark . Following fixation in 4% PFA , samples were washed three times in 1× PBS each for 5 min , and cryoprotected in 30% sucrose on a rocker until the tissue fully sank . Samples were removed form sucrose and briefly washed in optimal cutting temperature ( OCT ) compound before mounting in OCT compound and frozen at −80°C . A cryostat was used to obtain 10 µm sections , and slides were baked at 65°C for 15 min . Slides were then washed with water for 30 min at room temperature to remove residual OCT . Slides were washed once with 1× PBS for 5 min at room temperature . The click-it cocktail ( same as above ) was applied to the slides and incubated at room temperature in the dark for 30 min . If staining for a second macromolecule , slides were washed five times for 5 min with 1× PBS at room temperature . The samples were then stained for 30 min at room temperature in the dark using the above click-it cocktail with a different a fluorophore dye . Following the final click-it reaction , slides were washed once with 1× PBS at room temperature for 5 min , then stained with 2 . 86 µM DAPI for 5 min at room temperature . Slides were washed again with 1× PBS for 5 min at room temperature and water for 5 min at room temperature and mounted with SlowFade Gold Antifade Mountant . Slides were imaged using a Zeiss LSM800 confocal microscope . All 3D images were acquired using a Zeiss light sheet Z . 1 microscope paired with Zen software . Unless otherwise indicated , samples were cleared and imaged in 67% TDE . Post-processing for visualization purposes was performed with Arivis Vision4D v3 . 1 . 4 on a workstation with a 64-bit Windows Embedded Standard operating system , and an Intel ( R ) Xeon ( R ) CPU E5-2620 v3 @ 2 . 40 GhZ ( two processors ) , 128 GB RAM , and NVIDIA Quadro K2200 GPU . Sub-volumes were stitched together with the Tile Sorter in Arivis , using the manual projection option . Volume fusion was performed through automatic landmark registration of manually selected points for alignment . For visualization , background intensity was corrected using the automatic functionality . All data were processed on desktop computers with the aid of Fiji ( Schindelin et al . , 2012 ) and Matlab ( The MathWorks , 2019 ) . The custom Fiji scripts and Matlab codes used are available in the supplementary material . We performed all analyses on unprocessed . czi files acquired directly from the light sheet microscope .

Cells often respond to changes in their environment by producing new molecules and building new cell components , such as proteins , which perform most tasks in the cell , or DNA and RNA , which carry genetic information . Complex tissues – such as limbs , which are made up of muscles , tendons , bones and cartilage – are difficult to see through , so studying when and where cells in these tissues produce different types of molecules is challenging . New approaches combining advanced three-dimensional microscopy and fluorescent labelling of molecules could provide a way to study these processes within whole animal tissues . One application for this is studying how salamanders regrow lost limbs . When salamanders such as axolotls regrow a limb , some cells in the limb stump form a group called the blastema . The blastema contains cells that are specialized to different purposes . Each cell in the blastema produces many new proteins as well as new DNA and RNA molecules . Fluorescently labeling particular molecules and taking images of the regenerating limb at different times can help to reveal how these new molecules control and coordinate limb regrowth . Duerr et al . developed a three-dimensional microscopy technique to study the production of new molecules in regenerating axolotl limbs . The method labeled molecules of different types with fluorescent markers . As a result , new proteins , RNA and DNA glowed under different colored lights . Duerr et al . used their method to show that nerve damage , which hinders limb regrowth in salamanders , reduces DNA production in the blastema . There are many possible applications of this microscopy method . Since the technique allows the spatial arrangement of the cells and molecules studied to be preserved , it makes it possible to investigate which molecules each cell is making and how they interact across a tissue . Not only does the technique have the potential to reveal much more about limb regrowth at all stages , but the fluorescent markers used can also be easily adapted to many other applications .

2020

3D visualization of macromolecule synthesis

During mitosis , transcription is shut off , chromatin condenses , and most transcription factors ( TFs ) are reported to be excluded from chromosomes . How do daughter cells re-establish the original transcription program ? Recent discoveries that a select set of TFs remain bound on mitotic chromosomes suggest a potential mechanism for maintaining transcriptional programs through the cell cycle termed mitotic bookmarking . Here we report instead that many TFs remain associated with chromosomes in mouse embryonic stem cells , and that the exclusion previously described is largely a fixation artifact . In particular , most TFs we tested are significantly enriched on mitotic chromosomes . Studies with Sox2 reveal that this mitotic interaction is more dynamic than in interphase and is facilitated by both DNA binding and nuclear import . Furthermore , this dynamic mode results from lack of transcriptional activation rather than decreased accessibility of underlying DNA sequences in mitosis . The nature of the cross-linking artifact prompts careful re-examination of the role of TFs in mitotic bookmarking . A key component of cell identity is the epigenetic maintenance of cell-type specific transcription programs . However , this maintenance is challenged during each cell cycle . In mitosis , the global transcriptional machinery is inactivated via a cell cycle-dependent phosphorylation cascade ( Prescott and Bender , 1962; Rhind and Russell , 2012; Taylor , 1960 ) . Interphase chromatin organizes into highly condensed mitotic chromosomes ( Koshland and Strunnikov , 1995 ) , and the nuclear envelope is disassembled ( Terasaki et al . , 2001 ) . Furthermore , most transcription factors ( TFs ) have been shown to be excluded from mitotic chromosomes ( Gottesfeld and Forbes , 1997; John and Workman , 1998; Martínez-Balbás et al . , 1995; Rizkallah and Hurt , 2009 ) , leading to the conclusion that mitotic chromosomes may be inaccessible to DNA binding and exclude most TFs . Following mitosis , how then do the new daughter cells faithfully re-establish the cell-type specific transcription program ? Several mechanisms have been proposed to play important roles in re-establishing transcription following mitosis ( Lodhi et al . , 2016 ) . These include maintenance of DNA methylation patterns for heritable silencing and the propagation of histone modifications , although a model to account for targeted modifications in the absence of TFs has been elusive . Moreover , there are indications that DNA methylation and histone modifications are not sufficient to maintain transcription profiles through the cell cycle . For example , DNase I hypersensitive sites on the human hsp70 locus were shown to be maintained in mitotic chromosomes ( Martínez-Balbás et al . , 1995 ) implying the presence of a ‘bookmarker’ to keep the region accessible to nuclease digestion . Similarly , the transcription start sites ( TSSs ) of certain genes scheduled for reactivation following mitosis were shown to remain sensitive to permanganate oxidation in mitosis , suggesting a conformationally privileged structure at the TSSs of these genes ( Michelotti et al . , 1997 ) . It was thus proposed that some unknown factors must escape the exclusion from mitotic chromosomes and bookmark these regions , yet none have been shown to remain bound on chromosomes . It was therefore a significant step in resolving this conundrum when HSF2 was shown to bind at the hsp70i locus during mitosis ( Xing et al . , 2005 ) . Since then , and coincident with the advent of live-cell microscopy , a few other TFs have been discovered to associate with mitotic chromosomes ( Caravaca et al . , 2013; Kadauke et al . , 2012; Lodhi et al . , 2016 ) , beginning a re-emergence of an appreciation for TFs in propagating transcription programs through mitosis . For instance , GATA1 , a major regulator of the erythroid lineage , has previously been reported to be excluded from mitotic chromosomes by immunofluorescence ( Xin et al . , 2007 ) . Subsequently , the Blobel group has shown , by live-cell imaging and chromatin immunoprecipitation analysis , that GATA1 actually remained bound on its target regions during mitosis ( Kadauke et al . , 2012 ) . TFs such as GATA1 seem to act as the elusive ‘bookmark’ that maintain chromatin architecture at regulatory regions , and thus have been termed mitotic bookmarkers . Despite several recent examples of TFs that have been identified as potential mitotic bookmarkers ( Lodhi et al . , 2016 ) , these have generally been regarded as special cases while most of the literature document robust eviction of TFs from chromosomes during mitosis . Using a combination of in vitro biochemical assays , genome editing , and fixed versus live-cell imaging , we report that contrary to decades of published literature , most TFs we tested remain associated with mitotic chromosomes . The widely observed exclusion of TFs from mitotic chromosomes is due primarily to a formaldehyde-based cross-linking artifact . Sox2 , for example , appears excluded from chromosomes after chemical fixation , but is highly enriched on mitotic chromosomes as determined by live-cell imaging . This enrichment of TFs at mitotic chromosomes is facilitated by both the DNA binding domain of Sox2 and by active nuclear import . Using orthogonal imaging approaches such as single particle tracking and fluorescence recovery after photobleaching , we show that Sox2 binds dynamically to mitotic chromosomes , and that this dynamic behavior relates to the absence of transcriptional activation rather than a global inaccessibility of DNA in condensed chromosomes . These findings led us to investigate how chemical fixation may alter the localization of TFs in mitotic cells . We present a model for the mechanistic action of formaldehyde-based cross-linkers on transcription factor localization , and consider the overarching implications of this cell fixation artifact on interpreting experiments designed to study many biological processes and particularly transcriptional bookmarking . We initially hypothesized that Sox2 , one of the key pluripotency TFs in embryonic stem cells , may function as a mitotic bookmarker to maintain the ES cell state . To examine whether Sox2 binds to mitotic chromosomes , we synchronized cells at various stages of the cell cycle and obtained about 95% pure mitotic population . ( Figure 1—figure supplement 1 ) . We then performed biochemical fractionation to assess the chromatin-bound fraction on the asynchronous ( A ) , mitotic ( M ) , G2- and S- phase cells ( Figure 1—figure supplement 2 ) . We detected Sox2 on chromatin fractions from synchronized populations , including mitotic cells ( Figure 1A ) , providing initial evidence that Sox2 may associate with mitotic chromosomes . Similarly , TBP fractionated with mitotic chromosomes whereas Pol II did not ( Figure 1—figure supplement 2 ) . To biochemically assess the strength of this association , we performed salt fractionation on asynchronous and mitotic cells ( Figure 1—figure supplement 2 ) . Nuclear transcription factors elute from chromatin at the salt concentration that overcomes their binding strength to DNA . In the asynchronous population , the majority of Sox2 fractionated at high salt and in the micrococcal nuclease-digested chromatin , suggesting a strong interaction of Sox2 with chromatin ( Figure 1B ) . In contrast , Oct4 fractionated with much lower salt concentrations ( Figure 1B ) , consistent with its more dynamic association with chromatin ( Chen et al . , 2014 ) . Sox2 displayed a similar salt fractionation profile in synchronized mitotic cells , albeit with a somewhat reduced signal in the digested chromatin ( Figure 1B ) . TBP also showed a strong association with mitotic chromosomes whereas Pol II was primarily cytoplasmic ( Figure 1—figure supplement 2 ) . Taken together with the biochemical fractionation assay , these data suggest that Sox2 likely associates with mitotic chromosomes in a manner that is qualitatively weaker than its association with interphase chromatin . 10 . 7554/eLife . 22280 . 003Figure 1 . Transcription factors are not excluded from mitotic chromosomes . ( A ) Biochemical fractionation of asynchronous ( A ) mouse ES cells and synchronized populations at mitosis ( M ) , G2 , and S phases was performed to isolate the chromatin-associated fraction . Sox2 and H3 were detected by Western blot analysis . ( B ) Salt fractionation of asynchronous ( A ) and mitotic ( M ) mouse ES cells . Sox2 and Oct4 were detected by Western blot analysis . Cyt , Cytoplasmic fraction . Chr , Chromatin fraction . ( C ) Immunofluorescence with α-Sox2 of mouse ES cells stably expressing H2B-GFP using confocal microscopy showing exclusion of Sox2 from mitotic chromosomes ( D ) Live-cell imaging using confocal microscopy of mouse ES cells stably expressing H2B-GFP with overexpressed Halo-Sox2 ( Halo-Sox2 OE , top ) and endogenously tagged Halo-Sox2 ( Halo-Sox2 KI , bottom ) ( E ) Epi-fluorescence time-lapse imaging of mouse ES cells stably expressing H2B-GFP and endogenously-tagged Halo-Sox2 KI ( F ) Live cells with overexpressed Halo-Sox2 ( top ) or endogenously-tagged Halo-Sox2 ( bottom ) were labeled with JF549 dye and subjected to standard immunofluorescence by fixation with 4% PFA and detection with α-Sox2 . ( G ) Strategy for quantifying TF chromosome enrichment . ( H ) Chromosome enrichment levels for overexpressed and endogenously-tagged Halo-Sox2 . n = 40 cells ( I ) . Chromosome enrichment levels for indicated Halo-tagged transcription factors . n = 40 cells . Data are represented as mean ± SEM . Scale bars , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 22280 . 00310 . 7554/eLife . 22280 . 004Figure 1—figure supplement 1 . Synchronization of mouse ES cells . ( A ) Time-course strategy for synchronizing mouse ES cells . Thymidine at 2 mM is added to cells for 6 hr followed by release with fresh media for 6 hr . This cycle of thymidine block and release is repeated , and followed by Nocodazole treatment ( 100 ng/µL ) for 6 hr . Mitotic cells are collected by shake off . ( B ) Flow cytometry analysis of asynchronous ( A ) cells and synchronized populations as S , G2 , and mitosis ( M ) phases . DAPI is used to mark DNA content , and an antibody against phosphorylated H3 is used to mark mitotic cells . This analysis shows fairly pure populations of synchronized cells . ( C ) Asynchronous and Mitotic cells are stained with DAPI and immunostained with antibody against phosphorylated H3 , and visualized on a confocal microscope . By counting the H3-phosphorylated positive cells over total cells as marked by DAPI , we can measure the efficiency of mitotic enrichment by the synchronization strategy to be about 95% , providing an orthogonal measurement for the purity of mitotic cells . DOI: http://dx . doi . org/10 . 7554/eLife . 22280 . 00410 . 7554/eLife . 22280 . 005Figure 1—figure supplement 2 . Biochemical and Salt fractionation strategies . ( A ) For biochemical fractionation , asynchronous mouse ES cells or synchronized populations are swelled with hypotonic buffer and lysed by gentle shearing to separate nuclei from cytoplasmic fraction . The nuclei are washed repeatedly with EDTA-containing no-salt buffer to lyse the nuclei and remove the soluble nuclear fraction . The remaining chromatin-bound fraction is collected . ( B ) For salt fractionation , asynchronous mouse ES cells or synchronized mitotic cells are swollen with hypotonic buffer and lysed by gentle shearing to separate nuclei from cytoplasmic fraction as before . The nuclei are then incubated in low salt buffer containing 75 mM NaCl for 1 hr at 4°C . The nuclei are pelleted and the supernatant containing 75 mM soluble fraction is collected . The nuclei are then incubated in 150 mM salt buffer for 1 hr at 4°C . The nuclei are similarly pelleted and the supernatant containing the 150 mM soluble fraction is collected . This regime is repeated for 300 mM NaCl salt buffer . The final insoluble chromatin is resuspended in 300 mM NaCl buffer and incubated overnight at 37°C with 1 U of Micrococcal nuclease ( MNase ) to digest and solubilize chromatin . After overnight incubation , the insoluble material is pelleted by high speed centrifugation , and the soluble chromatin fraction is collected . ( C ) Biochemical fractionation of asynchronous ( A ) mouse ES cells and synchronized populations at mitosis ( M ) , G2 , and S phases was performed to isolate the chromatin-associated fraction . TBP ( Abcam #ab51841 ) and Pol II ( 8WG16 ) were detected by Western blot analysis . ( D ) Salt fractionation of asynchronous ( A ) and mitotic ( M ) mouse ES cells . TBP and Pol II were detected by Western blot analysis . Cyt , Cytoplasmic fraction . Chr , Chromatin fraction . DOI: http://dx . doi . org/10 . 7554/eLife . 22280 . 00510 . 7554/eLife . 22280 . 006Figure 1—figure supplement 3 . Endogenous knock-in of HaloTag to Sox2 locus using CRISPR/Cas9 . ( A ) We obtained three independent clones with homozygous knock-in of HaloTag to the Sox2 locus . We performed Western blot analysis of cell lysates from wild type ES cells ( WT ) and each individual knock-in clone for α-Sox2 and α-TBP for loading control . All of Sox2-detected molecules for each clone are migrating higher than the wild-type Sox2 suggesting that all molecules contain the HaloTag . Quantification of Western blot analysis is shown on the right . ( B ) Live-cell imaging of the three homozygous knock-in clones with cells labeled with Halo-dye JF549 . ( C ) Halo-Sox2 KI cells were injected into SCID-Beige mice . Thirty days post inoculation kidney and testis tumors were harvested . Histological analyses of tumors show small areas of differentiated cells , representing endoderm ( gland ) , mesoderm ( blood vessel ) , and ectoderm ( epidermal tissue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22280 . 00610 . 7554/eLife . 22280 . 007Figure 1—figure supplement 4 . Live versus fixed images of Halo-tagged TFs in mouse ES cells . Mouse ES cells stably expressing H2B-GFP are transfected with the indicated Halo-tagged TF . Cells were labeled with Halo dye JF549 and imaged using a confocal microscope under live conditions or after 15 min of fixation with 4% PFA . Quantification of chromosome enrichment is shown on Figure 1I . DOI: http://dx . doi . org/10 . 7554/eLife . 22280 . 00710 . 7554/eLife . 22280 . 008Figure 1—figure supplement 5 . Controls for live versus fixed imaging . ( A ) To ensure that the results were not biased by the HaloTag , we expressed mCherry-tagged Sox2 , Oct4 , or NLS in mouse ES cells stably expressing H2B-GFP and performed live and fixed imaging as before . mCherry-tagged factors are enriched on mitotic chromosomes in a similar manner as Halo-tagged factors . Quantification for chromosome enrichment as described in Figure 1G is shown . ( B ) We also performed fixation using 1% formaldehyde and 100% methanol on endogenously-tagged Halo-Sox2 KI cells . Quantification of chromosome enrichment is shown . Data is represented as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 22280 . 008 To visualize Sox2 in mitotic cells , we performed immunofluorescence analysis of Sox2 in standard formaldehyde-fixed mouse ES cells stably expressing H2B-GFP . In contrast to our biochemical data , we found that Sox2 is largely excluded from mitotic chromosomes ( Figure 1C ) . However , when we over-expressed Halo-tagged Sox2 ( Halo-Sox2 OE ) in mouse ES cells stably expressing H2B-GFP and imaged mitotic cells under live-cell conditions , we observed that Halo-Sox2 is highly enriched on mitotic chromosomes ( Figure 1D ) . To resolve the discrepancy between the fixed immunofluorescence data versus the over-expressed live-cell imaging , we sought to endogenously knock-in the HaloTag at the Sox2 locus using the CRISPR/Cas9 system such that all endogenous Sox2 molecules could be visualized . We obtained three independent clones that were homozygously tagged ( Halo-Sox2 KI ) , and confirmed that the tagging has no detectable effect on Sox2 function as Halo-Sox2 KI ES cells maintain pluripotency ( Figure 1—figure supplement 3 ) . Imaging under live-cell conditions , we confirmed that Halo-Sox2 KI is indeed highly enriched on mitotic chromosomes ( Figure 1D ) . Furthermore , using time-lapse microscopy , we detected strong enrichment of Halo-Sox2 KI to chromosomes throughout all stages of mitosis ( Figure 1E ) . To validate that the signal is indeed from the endogenous Sox2 and not an artifact of the knock-in , we first labeled the Halo-Sox2 KI in live cells followed by fixation and standard immunofluorescence using α-Sox2 . The signal for the HaloTag and α-Sox2 co-localized , confirming that the HaloTag is fused correctly with Sox2 ( Figure 1F ) . Surprisingly , both the HaloTag and the α-Sox2 signals in these fixed cell preparations were found to be excluded from mitotic chromosomes as marked by H2B-GFP ( Figure 1F ) in contrast to the live-cell imaging results . We performed the same fixation experiment on ES cells over-expressing Halo-Sox2 and observed the same chromosome exclusion phenomenon typically reported for TFs ( Figure 1F ) . We quantified the change in enrichment on mitotic chromosomes by taking the log2 ratio of the mean intensity on chromosomes over the whole cell intensity , using the H2B-GFP as a mask for chromosomes ( Figure 1G ) . Positive values correspond to enrichment on mitotic chromosomes whereas negative values signify exclusion from chromosomes . With this metric , both Halo-Sox2 OE and Halo-Sox2 KI switched from positive enrichment under live-cell imaging conditions to exclusion after fixation ( Figure 1H ) , suggesting that fixation by paraformaldehyde results in the exclusion of Sox2 from mitotic chromosomes . This result is reminiscent of a previous study showing that HMGB proteins become excluded from mitotic chromosomes after fixation ( Pallier et al . , 2003 ) . The authors argue that formaldehyde may uniquely alter the structure of HMGB proteins and thus result in exclusion , which may be consistent with Sox2 as it contains an HMG domain . However , because most of the evidence for TF exclusion derives from fixed cell immunofluorescence analysis , we wondered how general this artifact might be . We generated stable cell lines expressing a variety of Halo-tagged TFs in mouse ES cells , including Oct4 , Esrrb , Klf4 , Sp1 , Foxo1 , Foxo3a , Stat3 , and Hsf1 , and performed the same quantitative analysis of live versus fixed images . In all cases , the level of enrichment on mitotic chromosomes dramatically decreased after fixation , resulting in an apparent exclusion of all TFs examined ( Figure 1I and Figure 1—figure supplement 4 ) . When imaged under live-cell conditions , however , the majority of these factors exhibited varying levels on mitotic chromosomes , from highly enriched to uniform levels . This is consistent with a study showing a large number of TFs associated with mitotic chromosomes in chicken DT40 cells ( Ohta et al . , 2010 ) . We found two TFs that are excluded from mitotic chromosomes under live-cell conditions , Stat3 and Hsf1 ( Figure 1I ) , both of which are known to require external signals to translocate into the nucleus during interphase . We have also tested other commonly used fixation methods and have varied the fused tag , and all the results suggest that chemical fixation somehow evicts TFs from mitotic chromosomes ( Figure 1—figure supplement 5 ) . This surprising finding suggests that a large body of literature spanning several decades of studying TFs during mitosis , from their widely accepted exclusion from mitotic chromosomes to the apparent uniqueness of certain select TFs as specialized ‘mitotic bookmarkers’ , may have been founded in large part on a fixation artifact . Given the varying levels of TFs on mitotic chromosomes , we next investigated the intrinsic TF properties that might determine the association with mitotic chromosomes . Sox2 , for example , is composed primarily of two domains , the N-terminal high mobility group ( HMG ) DNA binding domain and the C-terminal trans-activation domain ( TAD ) ( Figure 2A ) . The HMG domain contains sequence-specific DNA binding residues as well as nuclear localization signals ( NLS ) , whereas the TAD is integral to protein-protein interactions with Sox2 partner proteins , including Oct4 and p300 ( Cox et al . , 2010 ) . To address which regions are important for mitotic enrichment , we expressed truncations of Sox2 as HaloTag fusions in ES cells stably expressing H2B-GFP . Halo-Sox2 HMG became highly enriched on mitotic chromosomes whereas Halo-Sox2 TAD was mostly cytoplasmic ( Figure 2B ) . When we quantified the chromosome enrichment of the truncations , we observed that the HMG domain is sufficient to render mitotic chromosome localization ( Figure 2B ) . We next tested whether sequence-specific binding and/or the NLS within the HMG domain is necessary for mitotic enrichment . We mutated five amino acid residues that have been shown to contact the DNA minor groove ( N48A , N70A , S71A , S74A , Y112A ) ( Reményi et al . , 2003 ) , and fused this to the HaloTag ( Halo-Sox2 DBD5M ) . Expressed in ES cells , mutations of DNA binding residues resulted in exclusion from mitotic chromosomes under live-cell imaging conditions ( Figure 2B ) . A separate study has shown that mutating three distinct residues ( M47G , F50G , M51G ) also abolished DNA binding ( Chen et al . , 2014 ) . Expressing this construct ( Halo-Sox2 DBD3M ) also showed exclusion from mitotic chromosomes ( Figure 2B ) , confirming that DNA binding is necessary for mitotic enrichment . We next tested what role Sox2 NLS might play on mitotic enrichment . Sox2 contains two NLS , a bipartite signal located at residues 43–46 ( VKRP ) and 57–60 ( QRRK ) , and a monopartite signal located at residues 114–117 ( PRRK ) ( Polakova et al . , 2014 ) . To abolish the NLS completely , we introduced mutations at both NLS elements , specifically K44A , R45A , R58A , R59A , R115A , R116A , and K117A . Surprisingly , mutations to the NLS also resulted in exclusion from mitotic chromosomes ( Figure 2B ) . 10 . 7554/eLife . 22280 . 009Figure 2 . Mitotic enrichment of Sox2 requires DNA binding and nuclear import . ( A ) Schematic of Sox2 domains . HMG , High Mobility Group domain . TAD , Transactivation Domain . NLS , Nuclear Localization Signal . ( B ) Live-cell imaging of mouse ES cells stably expressing H2B-GFB and various Halo-tagged Sox2 truncations or mutations . Bottom , chromosome enrichment quantification for the various Halo-tagged Sox2 over-expressing ( OE ) constructs . Halo-Sox2 HMG construct is a truncation of Halo-Sox2 with the TAD region deleted . Halo-Sox2 TAD is a truncation of Halo-Sox2 with the HMG domain deleted . Halo-Sox2-DBD5M and Halo-Sox2 DBD3M are the full length Halo-Sox2 with 5 and 3 point mutations to abrogate DNA binding , respectively . Halo-Sox2 NLSM is the full length Halo-Sox2 with point mutations to abolish NLS function . n = 40 cells . ( C ) Live-cell imaging of mouse ES cells stably expressing H2B-GFP and HaloTag fused to SV40 NLS , plant-specific NLS , or by itself . Right , chromosome enrichment levels for the indicated HaloTag constructs . n = 40 cells . Data are represented as mean ± SEM . Scale bars , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 22280 . 009 Although the Sox2 NLS is in close proximity to the DNA binding residues and as such , mutations to the NLS may affect DNA binding , some indication for the role of nuclear import in the mitotic enrichment of TFs has recently been shown . For instance , mutants of HNF1B become enriched on mitotic chromosomes following cold shock , and that this enrichment is dependent on nuclear import ( Lerner et al . , 2016 ) . We therefore wondered if an NLS is sufficient to confer enrichment on mitotic chromosomes . We expressed a fusion of the HaloTag and the SV-40 NLS in mouse ES cells stably expressing H2B-GFP . When imaged under live conditions , we discovered that the NLS enriched the HaloTag protein in mitotic chromosomes ( Figure 2C ) . In contrast , HaloTag protein alone is excluded from mitotic chromosomes ( Figure 2C ) . There are two potential mechanisms that could enrich an NLS-containing protein on mitotic chromosomes . The first is that highly positive residues on NLSs , generally lysines and arginines , interact non-specifically with the negatively charged chromosomes . In the case of the SV-40 NLS , the sequence is PKKKRKV . The second potential mechanism is that the nuclear import machinery actively enriches nuclear factors on mitotic chromosomes . To distinguish between these two possibilities , we expressed the fusion of a plant-specific NLS with the HaloTag in mouse ES cells stably expressing H2B-GFP . The plant NLS contains positively charged residues ( SVLGKRKFA ) , but is functional only in plants ( Kosugi et al . , 2009 ) . When imaged under live-cell conditions , we observed no enrichment of Halo-Plant NLS on mitotic chromosomes ( Figure 2C ) . These results suggest that both sequence-specific binding and active nuclear import likely contribute to the enrichment of TFs on mitotic chromosomes . We next examined the interaction dynamics of TFs on mitotic chromosomes using Sox2 as a model through fluorescence recovery after photobleaching ( FRAP ) ( Figure 3A ) . Quantifying intensities at the bleach spot over time shows that Halo-Sox2 recovery is faster in mitotic cells than during interphase , but significantly slower than Halo-NLS ( Figure 3B ) . The average time for Halo-Sox2 to reach 90% recovery in interphase is 19 . 7 s , whereas 90% recovery in mitosis is 4 . 3 s on average ( Figure 3C ) . In contrast , Halo-NLS reaches 90% recovery in 0 . 9 s for both interphase and mitotic cells ( Figure 3C ) . These results suggest that the interaction of Sox2 with mitotic chromosomes is more dynamic than its interaction with interphase chromatin . As an orthogonal approach to FRAP , we performed single particle tracking ( SPT ) with long exposure times ( 500 ms ) to determine residence times of Halo-Sox2 in mitosis relative to interphase ( Videos 1 and 2 ) . The long exposure times allow for a ‘blurring out’ of fast moving molecules while immobile , 'stably' bound molecules appear as bright diffraction-limited spots as previously reported for Sox2 ( Chen et al . , 2014 ) . We plotted the semi-log histogram of Halo-Sox2 dwell times , the amount of time each molecule remains detected , for interphase and mitotic cells ( Figure 3D ) . A two-component exponential decay model , representing specific versus non-specific binding events , was fit to the dwell time histograms as previously reported to extract the residence times of bound molecules ( Chen et al . , 2014 ) . After correction for photobleaching , the residence time of specific Halo-Sox2 binding events during mitosis is 54% relative to interphase ( Figure 3D inset ) . The decreased residence times for specific Sox2 interactions in mitosis relative to interphase is comparable with the faster FRAP recovery of Sox2 during mitosis . We next investigated whether there is also a change in the total fraction of molecules that are bound versus freely diffusing by performing SPT at faster imaging frequencies ( 223 Hz ) ( Videos 3 and 4 ) . For each tracked molecule , we measured the displacement of individual molecules between frames ( jump length ) , and plotted the histogram of jump lengths for interphase and mitotic cells ( Figure 3E ) . We then fitted a 2-state model to the probability distribution of jump lengths as previously reported ( Mazza et al . , 2012 ) to extract the fraction of molecules that are bound versus freely diffusing . We performed the same analysis for H2B-Halo and Halo-NLS to distinguish bound ( short jump lengths ) and freely diffusing ( large jump lengths ) states , respectively ( Figure 3—figure supplement 1 ) . In interphase , 30 . 9% of Halo-Sox2 molecules are bound , whereas 18 . 3% of Halo-Sox2 molecules are bound in mitosis ( Figure 3E inset ) , an almost two-fold decrease in bound population . This total bound fraction includes both short- and long-lived binding events . Taken together with the slow tracking SPT analysis , these data suggest that there are fewer Sox2 molecules bound during mitosis , and that these bound molecules experience a faster off-rate in mitosis . 10 . 7554/eLife . 22280 . 010Figure 3 . Sox2 interaction with mitotic chromosomes is highly dynamic . ( A ) FRAP analysis of HaloSox2 KI and HaloSox2 HMG cells for interphase and mitosis . ( B ) Quantification of fluorescence recovery at the bleach spot for the indicated Halo-tagged construct in interphase and mitosis . n = 30 cells . ( C ) From ( B ) , the average time to reach 90% recovery for the indicated ( color-coded ) Halo-tagged construct . ( D ) Dwell time histogram of the fraction of endogenously-tagged Halo-Sox2 molecules remaining bound for interphase ( gray ) and mitotic ( red ) cells . Representative images are shown . Inset , quantification of the relative Sox2 residence time as percentage of interphase cells . n = 30 cells . ( E ) Jump length histogram for three consecutive images ( Δt = 13 . 5 ms ) of the endogenously-tagged Halo-Sox2 molecules for interphase ( gray ) and mitotic ( red ) cells . A 2-state model is used to fit the histogram ( solid line ) , and the fraction bound is calculated ( inset ) . n = 24 cells . Data are represented as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 22280 . 01010 . 7554/eLife . 22280 . 011Figure 3—figure supplement 1 . Controls for single particle tracking experiments . ( A ) For photobleaching corrections , we performed slow-tracking SPT for H2B-Halo in interphase ( gray ) and mitosis ( red ) and plotted the semi-log histogram of H2B-Halo dwell times . These results show little difference in H2B-Halo residence time between interphase and mitotic cells . ( B ) Probability distribution of jump lengths for H2B-Halo , Halo-Sox2 KI , and Halo-NLS in interphase ( gray ) and mitosis ( red ) at △t = 13 . 5 ms . DOI: http://dx . doi . org/10 . 7554/eLife . 22280 . 01110 . 7554/eLife . 22280 . 012Video 1 . SPT for residence time analysis of Halo-Sox2 KI cells in interphase . Related to Figure 3 . Imaging immobile Halo-Sox2 molecules in interphase ES cells at 2 Hz . Movie fps = 20 . one pixel = 160 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 22280 . 01210 . 7554/eLife . 22280 . 013Video 2 . SPT for residence time analysis of Halo-Sox2 KI cells in mitosis . Related to Figure 3 . Imaging immobile Halo-Sox2 molecules in mitotic ES cells at 2 Hz . Movie fps = 20 . one pixel = 160 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 22280 . 01310 . 7554/eLife . 22280 . 014Video 3 . SPT for fraction bound measurements of Halo-Sox2 KI cells in interphase . Related to Figure 3 . Imaging fast Halo-Sox2 molecules in interphase ES cells at 223 Hz . Movie fps = 20 . one pixel = 160 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 22280 . 01410 . 7554/eLife . 22280 . 015Video 4 . SPT for fraction bound measurements of Halo-Sox2 KI cells in mitosis . Related to Figure 3 . Imaging fast Halo-Sox2 molecules in mitotic ES cells at 223 Hz . Movie fps = 20 . one pixel = 160 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 22280 . 015 Why is Sox2 interaction with mitotic chromosomes more dynamic than with interphase chromatin ? By examining truncations of Sox2 , we have discovered that the HMG domain of Sox2 is important for mitotic enrichment , but what are the contributions of the TAD on the dynamics of this interaction ? To answer this question , we performed FRAP analysis on mouse ES cells expressing Halo-Sox2 HMG in interphase and mitotic cells . Compared to full length Halo-Sox2 , the Halo-Sox2 HMG showed faster FRAP recovery both in interphase and mitosis ( Figure 3B–C ) . Quantification of fluorescence intensities over time at the bleach spot shows that Halo-Sox2 HMG in interphase and mitosis resemble that of Halo-Sox2 in mitosis ( Figure 3C ) . Indeed , the average time for Halo-Sox2 HMG to reach 90% recovery in both interphase and mitosis is 6 . 6 s , comparable to Halo-Sox2 in mitosis ( 4 . 3 s ) ( Figure 3C ) . These results suggest that the TAD plays a significant role in stabilizing Sox2 interactions with interphase chromatin . The TAD region participates in interactions with partner proteins , including Oct4 and p300 , to activate transcription ( Cox et al . , 2010 ) . Given these results , one possible model is that the HMG domain of Sox2 allows for initial contact with target DNA sites , and transcriptional activation by the TAD stabilizes Sox2 at the main target sites via protein-protein interactions with partner factors , perhaps in the assembled pre-initiation complex ( PIC ) . During mitosis , transcription is shut off globally , and therefore , the full length Sox2 now interacts with mitotic chromosomes primarily through its DNA binding domain , rendering the TAD inactive . The lack of transcriptional activation during mitosis may contribute to the more dynamic interaction of Sox2 with mitotic chromosomes , but another possibility is that the highly condensed nature of mitotic chromosomes also decreases the ability of Sox2 to access its binding sites . To test if mitotic chromosomes have decreased accessibility , we utilized Assay for Transposase-Accessible Chromatin using sequencing ( ATAC-seq ) analysis where the integration of sequencing-compatible adapters by the Tn5 transposase is directly correlated with the accessibility of genomic regions ( Buenrostro et al . , 2013 ) . We performed ATAC-seq in asynchronous cells as well as in Nocodazole-synchronized mitotic cells in two biological replicates each . Synchronized mitotic cells remain mitotic throughout the ATAC-seq protocol , and biological replicates for both asynchronous and mitotic cells were highly correlated with each other and were therefore combined in subsequent analyses ( Figure 4—figure supplement 1 ) . The length distribution profile of sequenced fragments from asynchronous cell populations and synchronized mitotic cells revealed near super-imposable patterns in Tn5 integration ( Figure 4A ) , including the 10 bp periodicity patterns that mark DNA helical pitch ( Figure 4A inset ) . These nearly identical integration patterns and sequencing read counts ( Figure 4—figure supplement 1 ) suggest that Tn5 transposase may access mitotic chromosomes as equally well as interphase chromatin . Indeed , when we mapped the sequenced reads and visualized them on the genome , we observed nearly identical patterns and intensities of peaks ( Figure 4B ) . We further parsed the reads computationally based on length , with short reads ( under 100 bp ) representing sub-nucleosomal DNA regions , and reads between 180–247 bp representing mono-nucleosome sized fragments ( Figure 4A inset ) . We then mapped these size classes separately and obtained distinct genomic profiles ( Figure 4C ) . Similar to the total mapped reads , there is little qualitative difference in peak patterns and intensities between asynchronous and mitotic cells when we analyze both the short reads and the mono-nucleosome sized fragments ( Figure 4C ) . To quantitatively assess the level of concordance between asynchronous and mitotic cells , we called peaks individually for both samples and combined unique peaks . We then measured the intensities of each called peak for both asynchronous and mitotic samples and plotted the values as a log-scale scatter plot heatmap ( Figure 4D ) . This analysis shows a near perfect symmetry along the diagonal , suggesting that the peak intensities between asynchronous and mitotic cells are concordant . Indeed , linear regression analysis shows the slope of the fit at 0 . 965 , with an R2 value of 0 . 991 ( Figure 4D ) . Performing the same analysis on the short reads ( Figure 4E ) and mono-nucleosome sized fragments ( Figure 4F ) also yields symmetrical patterns along the diagonal , with linear fits close to one . These quantitative analyses indicate that mitotic chromosomes are accessed by the Tn5 transposase in nearly the same manner as it accesses interphase chromatin , suggesting that the massive condensation of mitotic chromosomes has little effect on DNA accessibility . 10 . 7554/eLife . 22280 . 016Figure 4 . Global accessibility is maintained in mitotic chromosomes . ( A ) Fragment length distribution of ATAC-seq reads for asynchronous ( blue ) and mitotic ( red ) cells . Inset , magnification of fragment length distribution under 300 bp showing size cut-offs for short reads ( under 100 bp ) and mono-nucleosome sized fragments ( 180–247 bp ) . ( B ) Asynchronous and mitotic ATAC-seq profiles for an 800 kb region in chromosome 6 . ( C ) Comparison of total , short , and mono-nucleosome sized reads for asynchronous and mitotic samples in a 40 kb region in chromosome 6 . ( D–F ) Heatmap scatter plots of peak intensities for asynchronous vs mitotic samples in total reads ( D ) , short reads ( E ) , and mono-nucleosome sized reads ( F ) . Linear regression fit and R2 values are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 22280 . 01610 . 7554/eLife . 22280 . 017Figure 4—figure supplement 1 . ATAC-seq replicates for asynchronous and mitotic samples . ( A ) Time lapse imaging of Nocodazole-arrested mouse ES cells expressing H2B-GFP during the 30 min incubation with Tn5 transposase of ATAC-seq experiment . The cells remain mitotic throughout the tagmentation procedure . ( B ) ATAC-seq profiles at a locus in chromosome six for replicates 1 and 2 of asynchronous and mitotic samples show close concordance of each replicates . ( C ) Heatmap scatter plot of peak intensities for replicates of asynchronous ( left ) and mitotic ( right ) samples . DOI: http://dx . doi . org/10 . 7554/eLife . 22280 . 017 Global analysis of DNA accessibility shows no marked difference between interphase chromatin and mitotic chromosomes , but what about at specific TF binding sites such as those of Sox2 ? We examined total ATAC-seq reads as well as short and mono-nucleosome sized fragments on the well-characterized Oct4 distal enhancer ( DE ) that contains binding sites for Sox2 ( Figure 5A ) . We observed a prominent peak in short reads centered at the DE that is flanked by well-defined nucleosomes in asynchronous cells . Importantly , the peak of ATAC-seq reads in between the two well-positioned nucleosomes is maintained in mitosis ( Figure 5A ) . To assess accessibility for all bound Sox2 sites , we collated significantly bound Sox2 regions from published ChIP-seq experiments for Sox2 in mouse ES cells ( Chen et al . , 2008 ) , and analyzed ATAC-seq data at and surrounding these binding sites in asynchronous and mitotic samples . The heatmap of ATAC-seq signal on a 4 kb region centered at the Sox2 peaks shows a prominent region of accessibility at the peak center in asynchronous cells that is maintained albeit at somewhat reduced levels in mitotic cells ( Figure 5B , left ) . Averaging the integration density at all Sox2 bound sites shows that mitosis integration density is roughly 60% of the integration density in asynchronous cells ( Figure 5B , right ) . This significant decrease in integration density at Sox2 bound regions may reflect the reduced residence time of Sox2 at these sites during mitosis ( Figure 2D ) , and is consistent with a previous study showing decreased accessibility at enhancer regions ( Hsiung et al . , 2015 ) . However , this result may not reflect the ability of Sox2 to physically access the full repertoire of binding regions during mitosis . To assess whether Sox2 can sample its binding sites during mitosis , we collated regions of the genome containing the Sox2 binding motif . We averaged the integration density at single base resolution in an 80 bp region centered at the Sox2 motif to precisely define the footprint of Sox2 in asynchronous and mitotic cells ( Figure 5C ) . This footprinting analysis shows nearly identical integration patterns and footprinting depth for asynchronous and mitotic cells that is absent in matching random genomic sites , suggesting that Sox2 can sample binding sites as equally well in mitotic as in asynchronous cells . This interpretation is consistent with the similar binding dynamics of the Sox2 HMG domain in interphase and in mitosis ( Figure 3B ) , lending further support for the model that Sox2 scans its binding sites relatively quickly through its DNA binding domain and is stabilized through the TAD when transcription is activated . 10 . 7554/eLife . 22280 . 018Figure 5 . Accessibility of Sox2 binding sites in mitosis . ( A ) Comparison of total , short , and mononucleosome sized reads for asynchronous and mitotic samples at the Pou5f1 gene . The boxed region centered at the distal enhancer ( DE ) is shown in greater detail on the right . ( B ) Heatmaps using the short reads were for asynchronous and mitotic ATAC-seq samples for all Sox2 bound sites ( Chen et al . , 2008 ) GEO Accession number GSE11431 . Right , the average integration density for all Sox2 binding sites is plotted for asynchronous ( blue ) and mitotic ( red ) samples . ( C ) Aggregate ATAC-seq footprint for Sox2 ( left ) and for matched random genomic regions ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22280 . 018 Our surprising discovery of a formaldehyde fixation artifact prompted us to examine a potential mechanism for the artificial eviction of TFs . To capture this artifact in action , we performed time-lapse two-color imaging at 2 Hz of mitotic Halo-Sox2 KI cells stably expressing H2B-GFP as we added 1% paraformaldehyde ( PFA ) to the cells ( Figure 6A and Video 5 ) . At 2 s before 1% PFA is added , Halo-Sox2 is highly concentrated on mitotic chromosomes as marked by H2B-GFP . Within 10 s after PFA addition , Halo-Sox2 levels at the chromosomes are visibly reduced , and becomes almost indistinguishable from the cytoplasmic signal by 60 s after PFA addition ( Figure 6A ) . These results point to a robust effect of PFA in inducing the apparent eviction of TFs from mitotic chromosomes . A potential mechanistic model through which PFA could produce such an artifact is as follows ( Figure 7 ) . As PFA molecules cross the cell membrane , they rapidly cross-link to the nearest protein available . This would result in a steep gradient of cross-linking that moves inward as more and more PFA molecules cross the membrane . The non-uniform rate and directionality of cross-linking would likely deplete the cytoplasmic pool of TFs that could associate with chromosomes . Furthermore , the initial cross-linking of the cytoplasmic pool would result in an effective decrease in kon of TFs on mitotic chromosomes . We note that recent studies have shown , using live-cell single molecule tracking and FRAP experiments , that most TFs have a residence time of under 20 s and are thus quite dynamic even during interphase ( Chen et al . , 2014; Elf et al . , 2007; Gebhardt et al . , 2013; Mueller et al . , 2008; Swinstead et al . , 2016 ) . At least in the case of Sox2 , residence time is also decreased in mitosis . This increase in effective koff of TFs superimposed on this moving gradient of cross-linking would result in the apparent exclusion of TFs from mitotic chromosomes . This model predicts that the rate of artifact manifestation would increase as a function of PFA concentration . To test this , we performed time-lapse imaging on Halo-Sox2 KI cells while adding varying concentrations of PFA , from 0 . 25% to 4% , and quantified over time the chromosome enrichment of Halo-Sox2 using H2B-GFP as chromosomal mask . At 60 s post PFA addition , chromosome enrichment of Halo-Sox2 decreased as the concentration of PFA added is increased ( Figure 6B ) , confirming that the fixation artifact is dose-dependent . A second prediction based on the model is that TFs with lower koff ( more stably bound ) would be more resistant to PFA-induced mis-localization . One TF that has been shown to be stably bound to its target sites is TBP ( Chen et al . , 2002 ) . We imaged Halo-TBP expressing cells before and after fixation and quantified the level of chromosome enrichment as before . Compared to the highly dynamic Halo-NLS , Halo-TBP is resistant to PFA mis-localization ( Figure 6C ) . Furthermore , whereas active nuclear import is required for chromosome enrichment , it had no effect on PFA-induced mis-localization as Halo-Plant NLS also becomes excluded from mitotic chromosomes after fixation , in contrast to Halo-only control ( Figure 6C ) . A corollary of the second prediction based on the proposed mechanistic model is that if diffusion is reduced while allowing for PFA to equilibrate throughout the cell , the artifact should be reduced . We tested this hypothesis by performing high pressure freezing ( HPF ) followed by freeze substitution ( FS ) , a method commonly used for preparing samples for electron microscopy . First , the cells are rapidly chilled to liquid nitrogen temperatures in sub-millisecond time scale and under high pressure ( 2100 bar ) to better preserve molecular structures and localization . The cells are then immersed in PFA dissolved in an organic solvent at low temperatures over time to allow for a gradual substitution of water molecules with PFA-containing organic solvent . As temperatures warm , the equilibrated PFA molecules begin to crosslink molecules in the cell . We performed this experiment on Halo-Sox2 KI cells labeled with dye immediately before HPF-FS and imaged the cells after HPF-FS ( Figure 6D ) . Under these conditions , we partially rescued the localization of Halo-Sox2 on mitotic chromosomes , suggesting that the HPF-FS method can somewhat counteract the PFA-induced artifact . Taken together , these results point to a robust artifact caused by formaldehyde based cross-linking , and support a model whereby highly dynamic TFs are more susceptible to PFA-induced mis-localization . 10 . 7554/eLife . 22280 . 019Figure 6 . Mechanism of formaldehyde-based mis-localization of TFs . ( A ) Time-lapse two color imaging of endogenously tagged Halo-Sox2 mouse ES cells stably expressing H2B-GFP after adding 1% PFA . ( B ) Quantification of chromosome enrichment at 60 s after PFA addition with the indicated concentrations of PFA . n = 10 cells . ( C ) Quantification of chromosome enrichment of indicated HaloTag-fused constructs and HaloTag only in live and fixed conditions . n = 30 cells . ( D ) High Pressure Freezing and Freeze Substitution was performed on Halo-Sox2 KI cells stably expressing H2B-GFP . Comparison of chromosome enrichment quantification for HPF samples with live and fixed cells are shown . n = 10 cells . Data are represented as mean ± SEM . Scale bars , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 22280 . 01910 . 7554/eLife . 22280 . 020Video 5 . Time-lapse imaging with PFA addition . Related to Figure 6 . Mitotic Halo-Sox2 KI cells were imaged at 2 Hz . 1% PFA was added after 10 s of imaging . Movie fps = 20 . one pixel = 160 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 22280 . 02010 . 7554/eLife . 22280 . 021Figure 7 . Model for formaldehyde-based mis-localization of transcription factors . In live mitotic cells , TFs interact dynamically with mitotic chromosomes with intrinsic kon and koff rates , but can also sample the entire cellular space . During fixation , formaldehyde molecules enter the cell membrane and immediately cross-link with the nearest protein , resulting in a wave of cross-linking gradient that moves from the cell membrane inward . Such a gradient would cross-link cytoplasmic TFs first and result in an effective decrease in kon rates . As the gradient moves to the center , the result is an apparent exclusion of TFs from mitotic chromosomes . DOI: http://dx . doi . org/10 . 7554/eLife . 22280 . 021 We initiated our studies by testing whether Sox2 acts as a mitotic bookmarker to maintain the ES cell state . Indeed , we have found that Sox2 is dynamically enriched on mitotic chromosomes and that this interaction is mediated by both the DNA binding domain and the NLS . Furthermore , mitotic chromosomes remain highly accessible , ensuring that TFs such as Sox2 can efficiently and directly sample its binding sites . With these results alone , we might have concluded that Sox2 is one of the few privileged TFs that act as mitotic bookmarkers . In the process of tracking the behavior of Sox2 during mitosis , however , we unexpectedly discovered that most TFs we sampled continue to associate with mitotic chromosomes in varying levels , in direct contrast to the established literature documenting the eviction of most transcription factors from mitotic chromosomes . Surprisingly , we have found that the previously reported exclusion of TFs from mitotic chromosomes is caused by a chemical crosslinking-based artifact . By combining the dynamic properties of TFs with the chemical properties of formaldehyde , we have described a potential mechanistic model for how the artifact might be produced . Formaldehyde molecules rapidly cross-link to the nearest protein as they cross the cell membrane , resulting in a cross-linking gradient that moves from the cell membrane inward . In mitotic cells , TFs have intrinsic kon and koff rates with respect to mitotic chromosomes , but with the breakdown of the nuclear envelope , can also sample the entire cellular space . The advancing gradient of cross-linking would result in depletion of cytoplasmic TF molecules as they are cross-linked first , reducing the effective kon rates . As the cross-linking gradient moves inward , this effect , combined with the mitotic-specific increase in koff , would result in an apparent exclusion of TFs from mitotic chromosomes ( Figure 7 ) . We have presented evidence that support specific predictions based on this model . In particular , the rate of artifact manifestation is correlated with PFA concentration . Furthermore , highly dynamic TFs are more susceptible to PFA-induced mis-localization compared to stably bound factors . Indeed , some indications of this latter point exist in the literature . For example , factors such as CTCF , the cohesin complex , and DNA topoisomerases , all of which exhibit stable binding , have been shown by fixed immunofluorescence to remain bound on mitotic chromosomes ( Burke et al . , 2005; Christensen et al . , 2002; Losada and Hirano , 2001; Nakahashi et al . , 2013 ) . The existence of a formaldehyde based artifact , and the potential mode of its action , have several implications . The most direct implication is that it questions decades of research built upon the model that most TFs are evicted and excluded from mitotic chromosomes . Although TFs display varying levels on mitotic chromosomes , from highly enriched to uniform levels , the majority of TFs tested do not exhibit the reported exclusion from mitotic chromosomes , suggesting as yet unknown functions for most TFs during mitosis . The reported TF exclusion has also been implicated in studies of cellular differentiation and nuclear reprogramming ( Egli et al . , 2008 ) , which , in light of our findings reported here , may benefit from reconsideration . A far-reaching implication of the formaldehyde artifact is that many molecular biology techniques aimed at studying dynamics that are dependent on formaldehyde-based fixation may have to be revisited . Chromatin immunoprecipitation ( ChIP ) studies reliant on formaldehyde cross-linking may underestimate the total binding space of any given TF . Furthermore , studies designed to measure the dynamics of TFs through time-course cross-linked chromatin immunoprecipitation experiments ( Lickwar et al . , 2012 ) may be susceptible to formaldehyde-based artifacts . Indeed , previous examples of this limitation in ChIP have been documented ( Schmiedeberg et al . , 2009 ) . As technological advances in live-cell and single molecule techniques increase , we can begin to study dynamic processes more accurately . The visual compaction of chromatin into condensed mitotic chromosomes has led to the general perception that the underlying DNA in mitotic chromosomes is largely inaccessible . However , several experiments point to the idea that DNA remains accessible to some extent in the highly condensed mitotic chromosomes . For example , TFs of the RNA Polymerase I machinery have been shown to rapidly exchange on and off ribosomal DNA clusters during mitosis through FRAP analysis ( Chen et al . , 2005 ) . Diffusion of EGFP molecules is delayed but not hindered through mitotic chromosomes as measured by pair correlation function , suggesting that molecules can freely cross chromosomes during mitosis ( Hinde et al . , 2011 ) . More recently , the Blobel group has performed DNase I hypersensitivity assay coupled with sequencing in mitotic and asynchronous mouse erythroblast cells and has shown that , although there is a global decrease in DNase-seq signals during mitosis , overall patterns of DNase I HS patterns are preserved ( Hsiung et al . , 2015 ) . Our ATAC-seq analysis shows that the levels of integration globally are quantitatively similar for asynchronous and mitotic cells . Taken together , these studies suggest that DNA in mitotic chromosomes is as accessible as in interphase chromatin . The altered dynamic behavior of Sox2 binding with mitotic chromosomes relative to interphase chromatin implies that TF interactions with target DNA sites are likely cell cycle-dependent . During interphase , Sox2 target search and binding is dominated by three factors: 3D diffusion , protein-DNA contacts , and protein-protein interactions . Our fast-tracking SPT experiments have revealed that , at any given time during interphase , roughly 70% of Sox2 molecules are undergoing diffusion while 30% are bound . Previous studies have shown that bound molecules of Sox2 can be classified further as engaged in fast , non-specific interactions , or stable , site-specific binding ( Chen et al . , 2014 ) . Here , we have confirmed that the interphase long-lived stable binding events are highly dependent on the TAD-mediated protein-protein interactions that occur during transcriptional activation ( Chen et al . , 2014 ) . During mitosis when the nuclear envelope is disassembled and DNA becomes highly condensed , we have observed an increase in the fraction of freely diffusing Sox2 molecules to an average of 82% . Although this greater proportion of freely diffusing molecules could be attributed to an effective increase in total explorable space during mitosis , this effect would have resulted in a cytoplasmic rather than chromosome accumulation of Sox2 during mitosis . Live-cell imaging clearly shows that Sox2 molecules are concentrated at mitotic chromosomes , and that only a small proportion of molecules explore the whole cytoplasmic region . Thus , the majority of Sox2 molecules that are observed on mitotic chromosomes are actually sampling cognate binding sites dynamically . Interestingly , the 18% of Sox2 molecules that are bound on mitotic chromosomes exhibit decreased residence times relative to bound Sox2 molecules in interphase . Indeed , Sox2 dynamics in mitosis mimics the behavior of interphase Sox2 without its transcriptional activation domain , suggesting that Sox2 protein-protein interactions dependent on transcriptional activation are severely abrogated in mitosis . Therefore , in contrast to interphase cells , Sox2 behavior during mitosis appears to be dominated by two main factors , 3D diffusion and protein-DNA contacts . Similarly , FoxA1 , another TF that has been described as a mitotic bookmarker , has also been shown to bind to mitotic chromosomes in a more dynamic manner than to interphase chromatin ( Caravaca et al . , 2013 ) . Such cell cycle dependent behavior , as dictated by the changes in global transcriptional status , may be common for many TFs . We have shown that TF enrichment on mitotic chromosomes is facilitated by a functional NLS , which was surprising given the nuclear envelope breakdown in mitosis . How might the active nuclear import mechanism function to enrich TFs on mitotic chromosomes ? During interphase , the canonical nuclear import mechanism employs three main proteins , importins α and β , and the small regulatory GTPase Ran ( Freitas and Cunha , 2009 ) . Importin α recognizes the NLS-containing protein and binds to importin β to form a cargo complex . Upon import , nuclear Ran-GTP binds to importin β and induces cargo dissociation and the release of the NLS-containing protein . To maintain the directionality of the import mechanism , Ran-GTP must be present at higher concentrations in the nucleus while Ran-GDP must be predominantly cytoplasmic ( Harel and Forbes , 2004 ) . This compartmentalization is established by the chromatin associated Ran GTP-exchange factor , RCC1 , and by the Ran GTPase-activating protein ( Ran-GAP ) , which is localized to the cytoplasmic periphery of the nuclear envelope . Despite nuclear envelope breakdown during mitosis , RCC1 remains localized at mitotic chromosomes , thereby maintaining the Ran-GTP gradient in mitosis ( Moore et al . , 2002 ) . This Ran-GTP gradient has recently been shown to be vital in accurate formation of the spindle assembly during mitosis through the re-purposing of the canonical import proteins ( Forbes et al . , 2015 ) . One possible mechanism for mediating the NLS-dependent enrichment of TFs on mitotic chromosomes is that the maintained Ran-GTP gradient allows importins to retain their ability to transport NLS-containing proteins to mitotic chromosomes . Indeed , the accumulation of NLS containing proteins on chromosomes was hypothesized and modeled previously ( Caudron et al . , 2005 ) and a drug-dependent block of nuclear import resulted in exclusion of a TF mutant from mitotic chromosomes ( Lerner et al . , 2016 ) . This postulated active shuttling of nuclear proteins to the chromosomes serves two potential functions . First , such a mechanism would facilitate the accumulation of high local concentrations of nuclear factors near the chromosomes to ensure an efficient re-establishment of the nuclear environment immediately following mitosis when the nuclear envelope reforms . Secondly , the increased local concentration of TFs on mitotic chromosomes may contribute to maintaining the accessibility of DNA regulatory regions , the original definition of mitotic bookmarking , even under the highly condensed state of mitotic chromosomes by facilitating increased TF-chromosome interactions . This model is supported by the near perfect concordance of ATAC-seq profiles for asynchronous and mitotic samples , indicating that mitotic chromosomes are as accessible to TFs as interphase chromatin . Therefore , most TFs can and do interact with mitotic chromosomes and thus function as general mitotic bookmarkers . For all experiments , we used the mouse ES cell line JM8 . N4 ( RRID: CVCL_J962 ) obtained from the KOMP repository ( https://www . komp . org/pdf . php ? cloneID=8669 ) and tested negative for mycoplasma . ES cells were cultured on gelatin-coated plates in ESC media Knockout D-MEM ( Invitrogen , Waltham , MA ) with 15% FBS , 0 . 1 mM MEMnon-essential amino acids , 2 mM GlutaMAX , 0 . 1 mM 2-mercaptoethanol ( Sigma ) and 1000 units/ml of ESGRO ( Chem- icon ) . ES cells are fed daily and passaged every two days by trypsinization . Mitotic cells were synchronized by adding 100 ng/mL of Nocodazole for 6 hr followed by shake-off . Synchronization for other cell cycle stages were performed using the following regime . Cells were incubated in 2 mM Thymidine for 6 hr followed by 6 hr of fresh media . This cycle was repeated twice and cells synchronized in S phase were collected . The double thymidine block was followed by incubation with 100 ng/mL of Nocodazole for 6 hr . Mitotic cells were collected by shake-off , and remaining adherent cells were collected as G2-phase cells . To generate HaloTag-fused TFs , coding sequence of each TF were cloned into a Piggybac vector containing 3X-Flag-Halo-TEV construct upstream of a multiple cloning site . Stably expressing cell-lines were generated by transfecting the Halo-TF Piggybac construct together with the Super Piggybac transposase using Lipofectamine 3000 . After 24 hr , stable integration of construct was selected with 1 mg/mL of G418 . A similar method was used to generate stable expression of H2B-GFP under puromycin selection at 5 µg/mL . The design of the guide RNA was performed using the CRISPR Design Tool ( http://crispr . mit . edu ) ( Hsu et al . , 2013 ) . The following oligonucleotide pair ( 5’ CACCGCTGTGGGGGCCGCGCTCG and 5’ AAACCGAGCGCGGCCCCCACAGC ) was cloned into a modified pX330 vector . The modification involves the inclusion of the Venus coding sequence upstream of the gRNA , driven under the PGK promoter . Venus expression can then be used to sort transfection positive cells . The final vector encoded the gRNA , Venus , and the Cas9 nuclease . The left homology arm ( LHA ) , HaloTag , and right homology arm ( RHA ) DNA sequence was constructed using overlapping PCR . LHA and RHA were amplified from genomic DNA using the following primer pairs . LHA: 5’ GCGCGGTACCGCCAATATTCCGTAGCATGG and 5’ CTTTTGCCTTGTTCTCAGCGCTAGCCATGCGGGCGCTGGGCGGGCG . RHA: 5’ GAAAGTCCCTGTGTGCGAACCAAGTGGTACGCGTTATAACATGATGGAGACGGAG and 5’ GATATCTAGACCTCGGACTTGACCACAGAG . We introduced silent mutations within the Cas9 nuclease binding region of the right homology arm . The HaloTag was amplified using the following primers to create overlap with LHA and RHA: 5’ CGCCCGCCCAGCGCCCGCATGGCTAGCGCTGAGAACAAGGCAAAAG and 5’ CTCCGTCTCCATCATGTTATAACGCGTACCACTTGGTTCGCACACAGGGACTTTC . The final 3-way overlapping PCR was performed with all 3 PCR products as a template and using the following primer pair: 5’ GCGCGGTACCGCCAATATTCCGTAGCATGG and 5’ GATATCTAGACCTCGGACTTGACCACAGAG . The final PCR product was cloned into the pUC19 vector . Mouse ES cells were transfected with of nuclease ( 0 . 5 µg ) and donor ( 1 µg ) vectors using Lipofectamine 3000 . After 24 hr , cells were trypsinized into single-cell suspension and sorted for Venus expression . Sorted cells were grown on gelatin-coated plates under dilute conditions to obtain individual clones . After one week , individual clones were isolated and were used for direct cell lysis PCR using Viagen DirectPCR solution ( cat #302 C ) and the following primers: 5’ ACCACGTCCGCTTCATGGAT and 5’ GCCGTAAGGCATCATTGGAC . PCR positive clones were further grown on gelatin-coated plates and cell lysates were obtained for Western blot analysis using α-Sox2 ( Millipore Cat# AB5603 RRID:AB_2286686 ) . Three independent homozygous knock-in clones were obtained and further verified by immunofluorescence using α-Sox2 . Halo-Sox2 KI C3 ( clone 3 ) was further tested by teratoma assay and was performed by Applied Stem Cell . For all live-cell imaging experiments including FRAP , cells were grown on gelatin-coated glass bottom microwell dishes ( MatTek #P35G-1 . 5–14-C ) and labeled with the Halo-ligand dye JF549 at 100 nM concentration for 30 min . Cells were washed 3x with fresh media for 5 min each to remove unbound ligand . Live-cell imaging was performed using a Zeiss LSM 710 confocal microscope equipped with temperature and CO2 control . For fixed imaging , labeled cells were fixed with 4% PFA for 10 min and were washed with 1x PBS prior to imaging . Standard immunofluorescence was performed on labeled cells using 4% PFA . Quantification of chromosome enrichment was performed using Fiji . Epi-fluorescence time lapse imaging was performed on Nikon Biostation IM-Q equipped with a 40x/0 . 8 NA objective , temperature , humidity and CO2 control , and an external mercury illuminator . Images were collected every 2 min for 12 hr . FRAP was performed on Zeiss LSM 710 confocal microscope with a 40x/1 . 3 NA oil-immersion objective and a 561 nm laser . Bleaching was performed using 100% laser power and images were collected at 1 Hz for the indicated time . FRAP data analysis was performed as previously described ( Mueller et al . , 2008 ) . For each cell line , we collected 10 cells for technical replicates in one experiment , which was repeated for a total of three biological replicates ( 30 cells total ) . For all single molecule tracking experiments , indicated cells were grown on gelatin-coated glass bottom microwell dishes ( MatTek #P35G-1 . 5–14-C ) . For experiments imaging at slow frame rates , cells were labeled with JF549 at 10 pM for 30 min and washed as before . Cells were imaged in ESC media without phenol-red . A total of 600 frames were collected for imaging experiments and were repeated 3x for biological replicates , with each experiment consisting of 10 cells for technical replicates each . Data are represented as mean over experimental replicates ( 30 cells total ) ± SEM . For experiments imaging at fast frame rates , cells were labeled with JF646 at 25 nM as before . Cells were imaged in ESC media without phenol-red . A total of 20 , 000 frames were collected for fast-tracking imaging experiments and were repeated 3x for biological replicates , with each experiment consisting of eight cells of technical replicates . Data are represented as mean over experimental replicates ( 24 cells total ) ± standard error of means . Single particle tracking experiments were performed at either a slow frame rate ( 2 Hz; dye: JF549 ) to measure residence times or at a high frame-rate ( 225 Hz; dye: PA-JF646 ) to measure the displacement distribution and the fraction bound . Imaging experiments were conducted on a custom-built Nikon TI microscope equipped with a 100x/NA 1 . 49 oil-immersion TIRF objective ( Nikon apochromat CFI Apo SR TIRF 100x Oil ) , EM-CCD camera ( Andor iXon Ultra 897 ) , a perfect focusing system ( Nikon ) and a motorized mirror to achieve HiLo-illumination ( Tokunaga et al . , 2008 ) . To image PA-JF646 dyes ( Grimm et al . , 2015; 2016 ) , a multi-band dichroic ( 405 nm/488 nm/561 nm/633 nm BrightLine quad-band bandpass filter , Semrock ) was used to reflect a 633 nm laser ( 1 W , Coherent , Genesis ) and 405 nm laser ( 140 mW , Coherent , Obis ) into the objective , and emission light was filtered using a bandpass emission filter ( FF01 676/37 Semrock ) . To image JF549 ( Grimm et al . , 2015 ) , the same multi-band dichroic ( 405 nm/488 nm/561 nm/633 nm quad-band bandpass filter , Semrock ) was used to reflect a 561 nm laser ( 1 W , Coherent , Genesis ) into the objective and emission light was filtered using a bandpass emission filter ( Semrock 593/40 nm ) . The laser intensity was controlled using an acousto-optic transmission filter . For JF549 experiment at 2 Hz , a low constant laser intensity was used to minimize photobleaching . For experiments at 225 Hz , stroboscopic pulses of 1 ms 633 nm laser per frame were used at maximal laser intensity to minimize motion-blurring . When imaging single molecules at long exposure times ( 500 ms , 2 Hz ) fast-moving molecules motion-blur into the background and mostly bound molecules appear as single diffraction limited spots ( Chen et al . , 2014 ) . Thus , bound Sox2 molecules were identified using SLIMfast ( Normanno et al . , 2015 ) ( Source code 1 ) , a custom-written MATLAB implementation of the MTT algorithm ( Sergé et al . , 2008 ) using the following algorithm settings: Localization error: 10−6; deflation loops: 1; Blinking ( frames ) ; 1; maximum number of competitors: 3; maximal expected diffusion constant ( µm2/s ) : 0 . 1 . The length of each bound trajectory , corresponding to the time before unbinding or photobleaching , was determined and used to generate a survival curve ( fraction still bound ) as a function of time . The survival curve was then fitted to a two-exponential function:P ( t ) =Fe−koff , emp , nst+ ( 1−F ) e−koff , emp , st as previously described for Sox2 ( Chen et al . , 2014 ) , where koff , emp , ns corresponds to the empirical off rate for non-specific binding and koff , s corresponds to the empirical off rate for specific binding . The measured empirical off rate is the sum of the photobleaching rate and the actual off rate for unbinding of Sox2 from chromatin:koff , emp , s=kphotobleach+koff , s The photobleaching rate was measured using an ES cell line stably expressing H2b-Halo . Since H2b-Halo displays minimal unbinding ( e . g . no FRAP recovery ) , any apparent unbinding was interpreted as photobleaching . Thus , to determine kphotobleach the SPT experiment was repeated using identical settings and the apparent off-rate of H2b-Halo determined using two-exponential fitting , where kphotobleach corresponds to the slow component . Finally , the Sox2 residence time is then simply given by the inverse of the photobleaching-corrected off rate: τs=1koff , s . As with the experiments at 2 Hz , single molecules were localized and tracked using SLIMfast , a custom-written MATLAB implementation of the MTT algorithm ( Sergé et al . , 2008 ) , using the following algorithm settings: Localization error: 10−6 . 25; deflation loops: 0; Blinking ( frames ) ; 1; maximum number of competitors: 3; maximal expected diffusion constant ( µm2/s ) : 20 . To determine the fraction bound from the single particle tracking measurements at 225 Hz , the displacements ( ‘jump lengths’ ) at several Δτ ( 4 . 5 ms , 9 . 0 ms , … , 31 . 5 ms ) where fit to a steady-state model consisting of a free Sox2 population ( with diffusion constant DFREE ) and a bound Sox2 population ( with diffusion constant DBOUND ) using a modeling approach similar to what has been previously described by Mazza et al . ( 2012 ) , but with some modifications . The combined displacement histograms at several Δτ ( 4 . 5 ms , 9 . 0 ms , … , 31 . 5 ms ) were fitted to:P ( r , Δτ ) =FBOUNDr2 ( DBOUNDΔτ+σ2 ) er24 ( DBOUNDΔτ+σ2 ) +ZCORR ( Δτ ) ( 1−FBOUND ) r2 ( DFREEΔτ+σ2 ) er24 ( DFREEΔτ+σ2 ) where:ZCORR ( Δτ ) =1Δz∫−Δz/2Δz/2{1−∑n=0∞ ( −1 ) n[erfc ( ( 2n+1 ) Δz2−z4DFREEΔτ ) +erfc ( ( 2n+1 ) Δz2+z4DFREEΔτ ) ]}dz and:Δz=0 . 700 μm+ 0 . 15716s−1/2D+0 . 20811 μm using custom-written least-squares fitting software ( MATLAB ) . The above model contains three fitted parameters ( DBOUND , DFREE and FBOUND ) . The localization error , σ , was roughly 35 nm . ZCORR ( Δτ ) corrects for free molecules moving out of the axial detection slice ( out-of-focus; axial detection slice experimentally measured to be ~0 . 7 µm ) . The ZCORR ( Δτ ) expression assumes absorbing boundaries , which overestimates the fraction of molecules moving out of focus ( Kues and Kubitscheck , 2002 ) . To correct of this , Monte Carlo simulations were performed to determine a corrected Δz , which is shown above . Full details on modeling displacements will be published elsewhere ( Hansen et al . ) and is available upon request . ATAC-seq was performed on 50 , 000 asynchronous and mitotic cells as previously described ( Buenrostro et al . , 2013 ) with the following modifications . For both asynchronous and mitotic samples , Nocodazole was added to cells at 100 ng/mL concentration and incubated at 37 C for 6 hr , and mitotic cells were collected by careful shake off . After PBS wash , cells were immediately resuspended in the transposase reaction mix ( 25 µL 2x TD buffer , 2 . 5 µL transposase , and 22 . 5 µL nuclease-free water ) . Transposition and DNA purification was performed as described ( Buenrostro et al . , 2013 ) . For PCR amplification , we determined the linear range to be between 10–16 cycles and performed amplification for library preparation using 12 cycles for all subsequent samples . We performed two replicates , and each sample and replicate was sequenced using one lane of Illumina Hi-Seq 2500 for 50 bp paired-end reads . Sequenced paired mates were mapped on mm10 genome build using Bowtie2 with the following parameters: --no-unal --local --very-sensitive-local --no-discordant --no-mixed --contain --overlap --dovetail --phred33 –I 10 –X 2000 . Raw mapped reads were visualized using Integrative Genomics Viewer ( IGV ) . From the length distribution of sequenced reads , two size classes were used: short reads ( under 100 bp ) and mono-nucleosome sized reads ( 180–247 bp ) . Fragment sizes corresponding to these size classes were mapped separately using the same parameters as above . Peak calling was performed using the Homer suite package for asynchronous and mitotic samples , and using either total reads , or each size class . Peaks for asynchronous and mitotic samples were merged for the corresponding size classes . Peak intensity scatter heatmap was plotted using the MATLAB package Heatscatter with the number of bins set at 100 . Sox2 binding sites were collated from published ChIP-seq data sets ( Chen et al . , 2008 ) with GEO Accession number GSE11431 . ATAC-seq read densities from the short reads size class were calculated in a 4 kb region centered at each peak in 25 bp bins were calculated using a Homer suite package ( Heinz et al . , 2010 ) , and visualized using Java TreeView ( Saldanha , 2004 ) . Sox2 motif and associated position weight matrix were obtained from JASPAR core 2014 database , and genomic locations containing Sox2 motif was generated using PWMTools with a cut-off p-value of 10−4 . The short read size class was used to calculate the read density at single base resolution for footprinting analysis . Footprinting was performed similarly for matched random genomic regions . Sequencing data are deposited into GEO under the accession number GSE85184 . Halo-Sox2 KI cells stably expressing H2B-GFP were grown on gelatin-coated chambered coverglass ( Lab-Tek #155411 ) , labeled with JF-549 at 100 nM , and washed as described above . Time lapse imaging was performed with Zeiss LSM 710 confocal microscope as described above , with images collected at 2 Hz for 2 . 5 min . PFA at varying concentrations were added 10 s after the start of image collection . Quantification of chromosome enrichment was performed using an in-house Matlab code ( Source code 2 ) . Briefly , for each frame , the H2B-GFP was thresholded to generate a mask for chromosomes . The mask was applied to the JF549 channel to calculate the mean intensity at the chromosomes , which was normalized to the total cell intensity . The normalized chromosome intensity at 60 s post PFA-addition was obtained for each experiment and plotted as mean ± SEM .

A kidney cell functions differently from a skin cell despite the fact that all the cells in one organism share the same DNA . This is because not all of the genes encoded within the DNA are active in the cells . Instead , cells can turn on just those genes that are specific to how that cell type works . One way that cells can regulate their genes is by using proteins called transcription factors that can bind to DNA to turn nearby genes on and off . When cells divide to form new cells , the DNA is condensed and gene activity is turned off . However , each dividing cell also has to ‘remember’ the program of genes that specifies its identity . After division , how do the cells know which genes to turn on and which ones to keep off ? It was thought that the transcription factors attached to the DNA were all detached from it during cell division . Through studies in mouse embryonic stem cells , Teves et al . now show that this finding is largely an artifact of the methods used to study the process . In fact , many transcription factors still bind to and interact with DNA during cell division . This provides an efficient way for the newly formed cells to quickly reset to the pattern of gene activity appropriate for their cell type . Having found that many key transcription factors are still bound to DNA during cell division , the next challenge is to find out what role this binding plays in allowing cells to ‘remember’ their identity .

2016

A dynamic mode of mitotic bookmarking by transcription factors

Lung disease causes significant morbidity and mortality , and is exacerbated by environmental injury , for example through lipopolysaccharide ( LPS ) or ozone ( O3 ) . Toll-like receptors ( TLRs ) orchestrate immune responses to injury by recognizing pathogen- or danger-associated molecular patterns . TLR4 , the prototypic receptor for LPS , also mediates inflammation after O3 , triggered by endogenous hyaluronan . Regulation of TLR4 signaling is incompletely understood . TLR5 , the flagellin receptor , is expressed in alveolar macrophages , and regulates immune responses to environmental injury . Using in vivo animal models of TLR4-mediated inflammations ( LPS , O3 , hyaluronan ) , we show that TLR5 impacts the in vivo response to LPS , hyaluronan and O3 . We demonstrate that immune cells of human carriers of a dominant negative TLR5 allele have decreased inflammatory response to O3 exposure ex vivo and LPS exposure in vitro . Using primary murine macrophages , we find that TLR5 physically associates with TLR4 and biases TLR4 signaling towards the MyD88 pathway . Our results suggest an updated paradigm for TLR4/TLR5 signaling . Lung disease is a major contributor to morbidity and mortality worldwide . In the US alone , over 15% of the population suffers from lung disease , at an annual cost of 120 , 000 deaths , and >$50 billion ( Redd , 2002; Kochanek et al . , 2011; Ford et al . , 2015 ) . Environmental lung injury , for example through inhaled lipopolysaccharide ( LPS ) or elevated ozone ( O3 ) levels , exacerbates lung disease ( Bell et al . , 2004; Katsouyanni et al . , 1995; Hubbell et al . , 2005; Thorne et al . , 2005 ) . For example , household LPS levels are directly associated with asthma symptoms and asthma-related health care utilization ( Thorne et al . , 2005 ) . Also , sound epidemiological studies suggest that for every 10 parts-per-billion ( ppb ) increase in O3 levels there is an associated mortality increase of 0 . 39–0 . 87% ( Katsouyanni et al . , 1995; Levy et al . , 2005 ) . It is therefore imperative to understand the mechanisms of environmentally induced lung injury . Innate immune activation is a major contributor to lung disease pathogenesis and environmentally-induced exacerbations ( Miller and Peden , 2014; Garantziotis and Schwartz , 2010 ) . Toll-like receptors ( TLRs ) orchestrate the innate immune response to lung injury by recognizing exogenous pathogen- or endogenous danger-associated molecular patterns . TLR4 is the prototypic receptor for LPS ( Kawai and Akira , 2010 ) , which is found in particulate-matter pollution and house dust ( Thorne et al . , 2005; Mueller-Anneling et al . , 2004 ) and is a major contributor to sepsis-induced lung injury ( Andonegui et al . , 2003; Baumgarten et al . , 2006 ) . TLR4 also mediates inflammation and airway hyperresponsiveness after O3 exposure ( Garantziotis et al . , 2010 ) , triggered by release of the endogenous sugar hyaluronan ( Garantziotis et al . , 2009 ) . Regulation of TLR4 signaling is still incompletely understood . TLR4 can heterodimerize with other TLR like TLR2 and TLR6 ( Stewart et al . , 2010; Wang et al . , 2014 ) ; in these cases , the partner TLRs serve to expand the TLR4 ligand spectrum . However , until now there has not been evidence of TLR4 interaction with other TLR , that modulates TLR4 signaling in response to its own ligands . TLR4-TLR5 interaction has been reported once ( Mizel et al . , 2003 ) , wherein TLR4 was shown to promote nitric oxide production after flagellin exposure . We therefore hypothesized that the reciprocal interaction may also be true , that is TLR5 participates in TLR4 signaling after environmental lung injury . TLR5 , the prototypic receptor for bacterial flagellin , is expressed in alveolar macrophages ( Shikhagaie et al . , 2014 ) , is induced after injury ( Menendez et al . , 2011 ) and regulates the immune response to injury ( Burdelya et al . , 2008; Uematsu et al . , 2008; Wilson et al . , 2012 ) . TLR5 plays an important role in immunity and metabolism and has been implicated in processes as varied as asthma ( Wilson et al . , 2012 ) , antiviral defense ( Zhang et al . , 2014 ) , ischemia-reperfusion injury ( Parapanov et al . , 2015; Fukuzawa et al . , 2011 ) , radiation–induced injury ( Burdelya et al . , 2008; Burdelya et al . , 2012 ) , and regulation of gut immunity ( Uematsu et al . , 2008 ) . Furthermore , known functional genetic polymorphisms in TLR5 are associated with susceptibility to infections ( Hawn et al . , 2003; Grube et al . , 2013; West et al . , 2013 ) and autoimmune disease ( Gewirtz et al . , 2006 ) . These findings suggest a clinically relevant role of TLR5 in human immune regulation in the response to injury . We show that TLR5 deficiency in mice significantly alters the in vivo response to TLR4 activators LPS , hyaluronan and O3 . Mechanistically , we show that after ultrapure LPS exposure , TLR5 co-immunoprecipitates with MyD88 , TLR4 and LPS . The presence of TLR5 promotes formation of the Myddosome , that is association of MyD88 and IRAK4 , and biases TLR4 signaling towards the MyD88 pathway . Finally , we demonstrate that human carriers of a dominant-negative TLR5 allele have decreased inflammatory response to O3 exposure in vivo and LPS exposure in vitro . Our results thus suggest that TLR5 participates in TLR4 signaling and modulates environmental lung injury in disease-relevant exposures that lead to TLR4 activation . We first investigated the effect of TLR5 on TLR4 signaling in vivo , by exposing Tlr5-deficient mice or wildtype controls to LPS via intraperitoneal administration . As expected , this led to substantial lung inflammation in wildtype mice , which was ameliorated in the absence of TLR5 ( Figure 1A ) . TLR5 deficiency also ameliorated cellular influx and lung injury as evidenced by lung lavage protein ( Figure 1B ) . Furthermore , expression of inflammatory cytokines in the lung was significantly ameliorated in Tlr5-deficient mice ( Figure 1C , D ) . This was largely mirrored in a reduction of inflammatory gene expression in the liver ( Figure 1—figure supplement 1A ) . To ensure that our results were not affected by obesity- , microbiome- or breeding-related immune perturbations , we performed experiments with mice that were either purchased from a commercial vendor ( using C57BL/6 as controls ) or bred in our NIEHS colony ( using wild-type littermates as controls ) and treated some mice with neomycin to reduce bacterial burden in the intestinal tract . Our results did not vary regardless of mouse provenance or antibiotic treatment ( Figure 1—figure supplement 1B ) . To determine if the TLR5 effect on TLR4 signaling has broader biological and clinical relevance in the lung , we explored TLR5-mediated effects on sterile lung injury after exposure to the ambient pollutant , O3 . O3 exposure is associated with increased morbidity and mortality in human patients with cardiopulmonary disease ( Katsouyanni et al . , 1995; Levy et al . , 2005 ) ; furthermore , it is now understood that TLR4 mediates the development of inflammation and airway hyperresponsiveness ( AHR ) after O3 exposure ( Garantziotis et al . , 2010 ) . We used an O3 dose that is equivalent to human exposure during a high-O3 day ( Slade et al . , 1997; Hatch et al . , 1994 ) . Tlr5-deficient mice had ameliorated airway cytokine expression and almost abolished AHR after O3 exposure ( Figure 2A , B ) . Because hyaluronan is the endogenous danger-associated molecular pattern that activates TLR4 and mediates the response after O3 exposure ( Garantziotis et al . , 2010; Garantziotis et al . , 2009 ) , we then investigated the effect of TLR5 on hyaluronan signaling . Tlr5-deficient mice had substantially reduced inflammatory gene induction and significantly diminished AHR after instilled ultrapure ( pharmaceutical grade ) hyaluronan exposure ( Figure 3A , B ) . We then compared the effects of Tlr4- , Tlr5- and Myd88 deficiency on the hyaluronan response in vitro . We incubated tracheal rings of these strains with short fragments of hyaluronan , which induces hyperresponsiveness to methacholine-induced constriction through a TLR4-MyD88 pathway ( Garantziotis et al . , 2010; Li et al . , 2011 ) We found that deficiency in either of the 3 molecules abolished the response to hyaluronan ( Figure 3C ) suggesting that TLR4 and TLR5 are non-redundant and necessary for the hyaluronan-MyD88 signaling response . We then performed a more global analysis of the TLR5 effects on TLR4 signaling . We analyzed gene expression patterns using the NanoString platform ( www . nanostring . com ) and utilizing the Mouse Innate Immunity Panel Codeset ( Ns_Mm_Myeloid_v2 . 0 ) and focused specifically on 242 genes that were more than 2-fold upregulated after LPS exposure ( Supplementary files 1a and 1b and Figure 1—figure supplement 2 ) and sorted them according to magnitude of TLR5 effect . We showed that the presence or absence of functional TLR5 is associated with differential regulation of immune genes in this panel . Interestingly , there was a linear correlation between the magnitude of the TLR5 effect and the proportion of genes that are either published or predicted to be downstream of the NFκB pathway ( Figure 1—figure supplement 2C ) : among the genes that were 70–90% upregulated in Tlr5-sufficient mice compared to Tlr5-deficient mice , almost 90% were in the NFκB pathway , while this proportion fell to 50% among the genes that were not different between genotypes . ( R2 = 0 . 89 , p=0 . 0013 ) . This may suggest that TLR5 preferentially impacts gene expression downstream of NFκB activation . Alternatively , because these data are from whole lung tissue , this could reflect decreased recruitment of inflammatory cells , although we are not aware that inflammatory cells preferentially express NFκB pathway genes . In aggregate , the presented results support that TLR5 promotes TLR4 signaling in several models of TLR4 activation through pathogen- or danger-associated molecular patterns ( PAMPs or DAMPs , that is LPS or hyaluronan respectively ) and promotes TLR4-mediated inflammation and airway hyperresponsiveness in vivo . We then investigated the effect of TLR5 in human TLR4 signaling . In humans , a dominant-negative TLR5 single nucleotide polymorphism ( SNP ) ( rs5744168 , TLR5392STOP ) ( Hawn et al . , 2003 ) is found with a prevalence of 8–10% in Caucasians and 3% in African Americans . We hypothesized that carriers of this SNP may have reduced TLR4-mediated inflammation . We used the Environmental Polymorphisms Registry ( Chulada et al . , 2011 ) , a NIEHS-supported cohort , to recruit carriers of this allele , as well as ‘wildtype’ controls . Others have reported that whole blood from rs5744168 minor-allele carriers does not differ in the response to LPS compared to ‘wildtype’ ( West et al . , 2013 ) , and we confirmed this finding ( Figure 4—figure supplement 1A ) . We believe this happens because whole blood consists of different cell types , which differentially express TLR5 , thereby confounding the effect on TLR4 signaling . We then investigated the effect of this TLR5 SNP on purified , primary monocyte-derived macrophages . Macrophages from rs5744168 minor-allele carriers had a decreased response to flagellin and ultrapure LPS , but not Pam3CSK4 ( Figure 4A and Figure 4—figure supplement 1B ) , thus confirming that human TLR5 genetic variation specifically determines the response to LPS . This was not due to altered expression of TLR4 or CD14 , which was not changed by the rs5744168 genotype ( Figure 4—figure supplement 1C ) . We then investigated the effect of TLR5 in O3-induced inflammation in healthy human volunteers . We exposed human volunteers to O3 , and isolated alveolar macrophages through bronchoscopy 24 hr after exposure , which represents the peak of O3-induced inflammation and symptoms in humans . TLR5 expression was modestly increased in alveolar macrophages of human volunteers after O3 exposure ( Figure 4B , p=0 . 05 by Wilcoxon pairwise signed rank test ) . There was no association between TLR5 expression and TLR4 expression after O3 exposure . We found , that TNFα expression by alveolar macrophages after O3 exposure was not increased in any of the TLR5-deficient individuals ( rs5744168 minor-allele carriers ) , while it was increased in wildtype-allele carriers ( Figure 4C and Figure 4—figure supplement 1D ) . To investigate the mechanistic role for TLR5 in the response to LPS-induced TLR4 activation , we then compared primary bone marrow derived macrophages ( BMDM ) from Tlr5-deficient and -sufficient mice . Tlr5-deficient BMDM had significantly decreased expression of TNFα and IL-6 after ultrapure LPS exposure in vitro , by an average of 30–50% ( Figure 5A ) . To ensure that the altered response was not due to LPS contaminants despite the ultrapure preparation , we also assayed Tlr4-deficient BMDM and saw no response to ultrapure LPS ( Figure 5A ) . The effect of Tlr5-deficiency was also observed when using other sources of Tlr4 activation , such as Monophosphoryl Lipid A ( Figure 5—figure supplement 1A ) . To ascertain that our observation was not due to an off-target effect of genetic Tlr5 ablation on macrophage biology , we used RAW264 . 7 cells , a murine macrophage cell line , which are naturally deficient in Tlr5 ( McDonald et al . , 2007 ) . We transfected these cells with a murine Tlr5 construct , or empty vector , and noted that the Tlr5-transfected cells had significantly higher TNFα expression in response to ultrapure LPS ( and the TLR5 ligand flagellin , as expected ) . The TLR5-dependent effect was specific to LPS exposure , as responses to poly ( I:C ) ( TLR3 ligand ) , Pam3CSK4 ( TLR2 ligand ) , and ODN ( TLR9 ligand ) were not affected by the presence of TLR5 ( Figure 5B ) . We then interrogated the effect of TLR5 on TLR4 signaling . TLR4 signals through both MyD88 and TRIF pathways , whereas TLR5 is so far only known to signal through MyD88 ( Kawai and Akira , 2010 ) . We found that , in primary murine and human BMDMs , MyD88 co-immunoprecipitated with TLR5 after ultrapure LPS exposure ( Figure 6A , Figure 6—figure supplement 1 ) . This association was directly dependent upon TLR4 activation , since it was not observed in Tlr4-deficient primary BMDM ( Figure 6B ) . TLR4 , as expected , also immunoprecipitated with MyD88 after LPS exposure ( Figure 6B , Figure 6—figure supplement 1 ) . Furthermore , the association of MyD88 with TLR4 and TLR5 was specific , since other TLR like TLR6 and TLR7 did not immunoprecipitate with MyD88 after LPS exposure ( Figure 6B and C ) . MyD88 signaling occurs through the complexing of Myd88 with IRAK-4 which stabilizes formation of the so-called Myddosome ( Gay et al . , 2011 ) . Tlr5-deficient BMDM had decreased immunoprecipitation of IRAK-4 with MyD88 ( Figure 6C ) . There was also significant reduction in the phosphorylation of IKKα/β , IκB , p65 , JNK1/2 and ERK1/2 in Tlr5-deficient BMDMs compared with wildtype cells during early activation ( Figure 7A , B , Figure 7—figure supplement 1A ) , which is MyD88- but not TRIF-dependent ( Kawai et al . , 2001; Yamamoto et al . , 2003 ) . Nuclear IRF3 ( a specific readout of TRIF-dependent signaling Honda and Taniguchi , 2006 ) was not affected in Tlr5-deficient cells ( Figure 7—figure supplement 1B ) . In aggregate , these results support that TLR5 directly interacts with MyD88 after LPS exposure and enhances MyD88-dependent TLR4 signaling by promoting efficient assembly of Myddosome . We then investigated whether TLR5 participates directly in the TLR4 signaling complex , or whether it affects TLR4 signaling indirectly . We first evaluated whether TLR5 affects TLR4 cell surface expression and trafficking using bone marrow-derived macrophages ( BMDM ) from genetically deficient or wild-type mice . There were no differences in basal levels of cell surface TLR4 or CD14 between Tlr5-deficient and –sufficient BMDM , nor in LPS-induced internalization of TLR4 ( Figure 8—figure supplement 1A–D ) . Because we were unable to find commercially available validated antibodies against TLR4 that could be used in co-immunoprecipitation experiments , we utilized an induced expression system using tagged TLR4 and TLR5 in HEK293 cells and found that TLR4 and TLR5 reciprocally co-immunoprecipitated in transfected HEK293 cells ( Figure 8A ) . We then overexpressed TLR4 and TLR5 in HeLa cells and confirmed their interaction through a Proximity Ligation Assay ( Figure 8—figure supplement 1E ) . Furthermore , we exposed TLR5-hemagglutinin tagged expressing RAW264 . 7 cells to biotin-tagged LPS , and ( after thoroughly washing the cells ) could co-precipitate TLR5 with LPS ( Figure 8B ) . These results suggest that TLR5 directly participates in the TLR4 signaling complex after LPS exposure , and , in aggregate , support a functional interaction of TLR5 with TLR4 in the response to environmental injury . The important novel finding from our work , is that TLR5 heteromerization with TLR4 modulates canonical TLR4 signaling and promotes activation of the MyD88 pathway . Recent evidence highlights the role of molecules of the TLR4 receptor complex in modulating TLR4 signaling . For example , elegant work has demonstrated that CD14 , which is necessary for LPS binding to TLR4 , also controls TLR4 endocytosis after LPS ligation , and thus is necessary for TRIF signaling , which is thought to occur in the endosomal compartment ( Zanoni et al . , 2011 ) . Our work further suggests that TLR4 signaling is modulated by the addition of TLR5 to the receptor complex . Unlike CD14 , TLR5 does not appear to regulate LPS-induced internalization of TLR4 ( Figure 8—figure supplement 1C ) . Our findings rather support a model in which TLR5 promotes TLR4/MyD88 signaling at the plasma membrane by enhancing the assembly of the Myddosome . Importantly , our work suggests that TLR5 regulation of TLR4 signaling is biologically significant . TLR5-deficient mice had approx . 30–50% decreased cytokine expression in local and systemic LPS models of lung inflammation , while airway hyperresponsiveness after ozone or hyaluronan exposure was significantly reduced , and inflammatory gene induction after hyaluronan exposure was abolished . It is possible that the ‘fine-tuning’ , MyD88-promoting effects of TLR5 are particularly evident in lower-grade inflammation such as ozone- or hyaluronan-induced , which explains the larger impact on TLR5 deficiency on human and murine inflammation after ozone exposure compared to LPS exposure . Using primary murine and human macrophages , we demonstrate that , in the physiological state , TLR5 co-immunoprecipitates with MyD88 after ultrapure LPS exposure , but only in the presence of TLR4 . This indicates that TLR5 is recruited into the Myddosome assembly , along with TLR4 , upon TLR4 activation . Indeed , it has been postulated that the ability of the Myddosome to form 7:4 and 8:4 MyD88:IRAK4 stoichiometries is a potential mechanism through which clusters of activated TLR receptors can be formed and different TLR receptors can be recruited into the same assembly ( Motshwene et al . , 2009 ) . Higher-order assembly of receptor complexes in lipid raft microdomains is likely to be crucial in the fine-regulation of immune responses ( Gay et al . , 2011 ) . Our work suggests that TLR5 may be part of the higher-order receptor assembly that regulates TLR4 signaling . TLR4 is a promiscuous receptor , having been found to heterodimerize with TLR2 and TLR6 ( Stewart et al . , 2010; Wang et al . , 2014 ) but the effect of heteromerization until now has always been to expand the TLR4 ligand spectrum . To our knowledge , this is the first work to demonstrate that TLR heteromerization may serve to modulate canonical TLR signaling . Further research will be necessary to uncover the precise mechanism of this effect . TLR5 heteromerization with TLR4 may help recruit additional MyD88 moieties to the signaling complex , and thus promote downstream signaling . Alternatively , or perhaps in addition , endogenous activators of both TLR4 and TLR5 , like hyaluronan or HMGB1 ( Das et al . , 2016 ) may engage these receptors and promote the signaling response . A recent paper supported the role or TLR5 in TLR4 signaling , showing similar effects of the rs5744168 minor allele polymorphism on TNFα and IL-8 expression by human monocytes after LPS exposure , and also demonstrating that TLR5 does not influence the gene expression of TLR4 ( 45 ) , as we also show . By contrast , that paper could not confirm that TLR5 is modifying the NFκB pathway . This apparent discrepancy may be due to different research methodologies: These authors used gene silencing which only resulted in approx . 50% reduction of TLR5 expression , as well as reporter cell lines with transfection-induced TLR5 expression , as opposed to our use of primary cells with genetically knocked-out gene function . Furthermore , the authors of that study did show an effect for TLR5 on NFκB when transfecting with lower ( more physiological ) doses of TLR5 DNA ( Dickey et al . , 2019 ) . In summary , our results suggest a new model of TLR4-TLR5 complex formation in response to the PAMP LPS or the DAMP hyaluronan ( Figure 9 ) . We propose that the hitherto accepted model of TLR4 signaling through pure TLR4 homodimers rather reflects the TLR5-deficient state . In TLR5-expressing cells , TLR5 participates in a heteromeric higher-order TLR4 receptor complex and potentiates MyD88 signaling by promoting efficient assembly of the Myddosome . This may suggest that exposures that induce TLR5 expression , such as DNA injury , p53 activation ( Menendez et al . , 2011 ) or flagellated bacterial infection may in parallel prime the MyD88-dependent pro-inflammatory response to LPS , due to the presence of more TLR5 receptors which may promote TLR4 signaling . Since TLR5 signals through MyD88 , in a finite TLR4 receptor pool the presence of TLR5/TLR4 higher-order complexes would bias TLR4 signaling towards MyD88 . Notably , cell surface expression of TLR4 on immune cells is low ( a few hundred or thousand molecules per cell ) compared to other TLRs ( Visintin et al . , 2001; Juarez et al . , 2010 ) , supporting that the presence of relatively few TLR5 receptor molecules may suffice to augment MyD88 signaling downstream of TLR4 . C57Bl/7J mice and B6 . 129S1-Tlr5tm1Flv/J ( TLR5-deficient ) mice were purchased from the Jackson Laboratory ( Bar Harbor , ME ) . The TLR5-deficient allele was generated in the 129S1 genome and subsequently backcrossed to C57BL/6 before being transferred to the Jackson Laboratory by the donating investigator . When possible , wildtype littermate control mice were used in our study in parallel to commercial wildtype C57BL/6J controls . To ensure that our results were not due to locality-influenced microbiome changes , we repeated experiments in mice that were bred at the NIEHS vivarium , as well as mice purchased from the Jackson Laboratory and studied within 1 week of arrival . In some experiments , mice received neomycin water to control gut microbiome . Results were comparable independent of provenance or antibiotic dosing . Mice were given access to water and chow ad libitum and were maintained at a 12 hr dark-light cycle . No differences in body weight were observed at the ages studied ( 6–12 weeks old ) . All experiments are approved by the NIEHS Institutional Animal Care and Use Committee . Mice received ultrapure E . coli O111:B4 LPS ( List Biological Labs , Campbell , CA ) ( 50 μl of 1 mg/ml in PBS ) , or control PBS vehicle only , by oropharyngeal aspiration and were phenotyped 24 hr later . For systemic LPS exposure , mice received 10 mg/kg LPS or PBS by intraperitoneal injection and were phenotyped 24 hr later . In other experiments , mice were exposed to 2 ppm ozone for 3 hr , in a chamber with 20 exchanges/hour , 50–65% relative humidity and a temperature of 20–25° C as previously described ( Garantziotis et al . , 2009 ) and were phenotyped 24 hr later . Control mice received filtered air in an identical setup . In other experiments , mice received 50 µl of a 3 mg/ml solution of sonicated , LPS-free , pharmaceutical-grade hyaluronan with an average molecular weight of 100–300 kDa ( derived from Healon , Abbott Laboratories , Abbott Park , IL ) or PBS vehicle by retropharyngeal aspiration , and were phenotyped 2 hr later . Airway responsiveness to 25–100 mg/ml methacholine ( MCh; Sigma ) was measured 24 hr following O3 or 2 hr following hyaluronan challenge . Briefly , mice were anesthetized with urethane ( 2 g/kg; Sigma ) , tracheotomized with a tracheal cannula ( Harvard Apparatus ) with Luer adapter , and mechanically ventilated on a 42°C water-heated pad at a rate of 150 breaths/min , a tidal volume of 10 ml/kg and a positive end-expiratory pressure ( PEEP ) of 3 cm H2O with a computer-controlled small animal ventilator ( FlexiVent , Scireq , Montreal , Canada ) . To block spontaneous breathing , mice were given pancuronium bromide i . p . ( 0 . 8 mg/kg; Sigma-Aldrich ) 5 min prior to assessment of airway responses . To measure airway responsiveness , a single-frequency forced oscillation waveform , followed by a broadband forced oscillation waveform ( matched to the animal breathing frequency ) were applied using the Flexiware 7 . 6 software default mouse inhaled dose-response script . The resulting pressure , volume , and flow signals were fit to either the Single Compartment or Constant Phase model of the lung to obtain total respiratory system resistance ( Rrs ) and elastance ( Ers ) or Newtonian resistance ( Rn , generally understood as proximal airway resistance ) , tissue damping ( G , generally understood as peripheral tissue resistance ) , and tissue elastance ( H ) , respectively ( Irvin and Bates , 2003 ) . The peak response at each dose was averaged and graphed along with the average baseline measurement for each group . TLR4-FLAG M2 or MyD88-V5 and TLR5-HA were overexpressed in HeLa cells grown on glass coverslips . 24 hr post-transfection the cells were fixed with 4% paraformaldehyde ( PFA ) for 10 min at room temperature ( RT ) and blocked with 10% normal goat serum for 1 hr at RT . The cells were next permeabilized with 0 . 1% TritonX-100 in goat serum for 15 min at RT and incubated with primary antibodies ( dilution 1:1000 ) against epitope-tags overnight: rabbit anti-FLAG M2 ( Cell Signaling ) , mouse anti-V5 ( Invitrogen ) and mouse anti-HA ( Sigma ) . Duolink , based on in situ proximity ligation assay ( PLA ) , was performed according to manufacturer instructions ( Sigma ) . Bone marrow was collected from tibias and femurs of wildtype and Tlr5-deficient mice and cell single preparations were made . Growth medium for maturation of BMDM consisted of DMEM-F12 containing 10 mM L-glutamine , 10% embryonic stem cell qualified fetal bovine serum , 1% antibiotic and antifungal mix and 30 ng/mL murine M-CSF . Cells were cultivated in an incubator at 37°C , 5% CO2 for up to 7 days with medium change every 48 hr after first medium change 72 hr after platting . Tlr5-deficient BMDM were evaluated for responsiveness to flagellin and were found to be unresponsive ( Figure 5—figure supplement 1A ) . We used HeLa cells and Raw264 . 7 cells . Neither cell line is among the list of commonly misidentified cell lines . All cell lines were procured by ATCC and were free of mycoplasma contamination . In order to access cell surface expression of TLR4 , Wildtype , Tlr4-deficient or Tlr5-deficient BMDM were harvested , washed with PBS and were exposed to ultrapure LPS for 0 , 15 , 30 , 60 or 90 min . Cells were washed with cold PBS and gently lifted from the culture dishes using a cell lifter . Cells numbers were estimated and cells were aliquoted in 1 × 106 cells per tube in the FACS buffer ( 0 . 5% BSA , 0 . 1% NaN3 , and 2 mM EDTA in PBS ) . Cells were blocked for 20 min on ice in a blocking solution ( FACS buffer , 10% species specific serum , and 1% FCR block ) . Cells were stained using APC anti-mouse CD284 ( TLR4 ) Antibody ( clone SA15-21 ) , anti-CD14 Antibody ( Biolegend ) or isotype controls for 30 min on ice . Cells were washed two times with 1 mL FACS buffer after staining , suspended in 500 uL FACS buffer containing 1 mg/mL propidium iodide ( to identify dead cells ) and analyzed on a BD FACSAria II equipment . Wildtype or Tlr5-deficient BMDM were harvested , washed once with cold PBS , and lysed for 30 min at 4°C in 1% TritonX-100 , 20 mM HEPES ( pH 7 . 4 ) , 150 mM NaCl , 1 . 5 mM MgCl2 , 2 mM EGTA , protease and phosphatase inhibitors ( Roche ) . Cellular debris was removed by centrifugation at 16 , 000rcf for 10 min . For immunoblotting , cell extracts were fractionated by SDS-PAGE and transferred to Immobilon-P transfer membranes ( Millipore ) , using either a wet transfer apparatus ( Bio-Rad ) or with a dry transfer system ( iBlot ) from Invitrogen . Immunoblot analysis was performed , and the bands were visualized with HRP-coupled goat anti-rabbit , goat anti mouse , or donkey anti-goat Ig as appropriate ( Rockland ) , using the ECL western blotting detection system ( GE Healthcare ) . Protein levels were equilibrated with the Protein Assay Reagent ( Bio-Rad ) . For coimmunoprecipitations , cells were harvested , washed once with cold PBS , and lysed in a TritonX-100-containing buffer ( 0 . 5% TritonX-100 , 20 mM HEPES ( pH 7 . 4 ) , 150 mM NaCl , 1 . 5 mM MgCl2 , 2 mM EGTA , protease and phosphatase inhibitors ( Roche ) . Cell extracts were incubated with 1 μg of Ab ( anti-HA , Sigma ) or normal IgG ( negative control ) for 2 hr , followed by incubation for 12 hr with 30 μl of protein G-Sepharose beads ( prewashed and resuspended in lysis buffer at a 1:1 ratio ) . After incubations , the beads were washed four times with lysis buffer , separated by SDS-PAGE , and analyzed by immunoblotting . For TLR4 , TLR5 , TLR6 and TLR7 antibodies , blocking agent was 5% BSA and antibody dilution was 1:10 , 000 . For Myd88 antibody , blocking agent was 5% milk and antibody dilution was 1:1000 . Antibodies used were TLR4: Cat#482300 ( Life Technology ) . TLR5: Cat#PA1-41139 ( Invitrogen ) , TLR6: Cat#AF1533 ( R and D Systems ) , TLR7: Cat# MAB7156 ( R and D Systems ) , Myd88: Cat#4283 ( cell signaling ) . ELISA assay was performed using either R and D Duoset assay kits or Luminex multiplex assays according to manufacturer recommendations . The complete details of the human exposure studies and the subject characteristics were previously published ( Frush et al . , 2016 ) . After obtaining informed consent through a Duke University Institutional Review Board approved protocol , healthy human subjects were exposed to filtered air and ozone ( 200 parts per billion ) in a crossover challenge designed study . Exposures were for 135 min , during which participants alternated between resting and walking on a treadmill at 2–3 mph to mimic an individual performing mildly strenuous activity under ambient conditions . Ozone was created from a 100% O2 source by cold plasma corona discharge ( Ozotech , Yreka CA ) , and mixed with filtered air before addition to chamber and was continuously monitored . The order of filtered air or ozone exposure was randomized for every participant , with at least a 21 day washout period . Approximately 20 hr after exposure , participants underwent a flexible bronchoscopy with bronchoalveolar lavage . Following bronchial alveolar lavage , human alveolar macrophages were isolated . After red cell lysis and counting , the macrophages were re-suspended in medium ( RPMI1640 with 10% heat-inactivated FBS , 100 units/ml penicillin , and 100 μg/ml streptomycin ) and plated in a 24 well plate at a density of 200 , 000 cells per well . The cells were maintained in a CO2 incubator at 37° C for 2 hr . After 2 hr , the medium was replaced to remove non-adherent cells and then 2 hr later , the supernatant was collected and the cells were harvested for RNA . RNA extraction was performed using the Fourth Edition Qiagen Protocol ( Qiagen , RNeasy Mini Kit , 4th edition , Valencia , CA ) , followed by DNase treatment ( DNase I , Ambion , Austin , TX ) and cDNA synthesis ( BioRad ) . RT-PCR was performed on an ABI SDS 7500 ( Applied Biosystems ) using SYBR Green Reagent ( Clontec Laboratories Inc , Mountain View , CA ) . TLR5 expression was determined in comparison to the 18 s RNA housekeeping gene and the data reported as fold change over the matched filter air sample for each individual subject . The following primers were used for RT-PCR: 18 s ( Fwd: GTAACCCGTTGAACCCCATT , Rev: CCATCCAATCGGTAGTAGCG ) ; TLR4 ( Fwd: GGCCATTGCTGCCAACAT , Rev: CAACAATCA CCTTTCGGCTTTT ) , TLR5 ( Fwd: TGTATGCACTGTCACTCTGACTCTGT , Rev: AGCCCCGGAACTTTGTGACT ) . Human TNF-α was measured from the cell supernatants via ELISA ( MAX Standard Set kit , BioLegend ) according to manufactures instructions . Readings were taken using BMG LABTECH Omega ( Software Version 1 . 20 ) . In a second study , participants were invited according to their rs5744168 genotype . Peripheral blood was drawn , and monocyte-derived macrophages isolated after 7 days in culture and exposed to ultrapure LPS or flagellin as indicated . Data are represented as mean ± s . e . m . and were analyzed depending on experimental design by either analysis of variance ( one-way or two-way ANOVA ) followed by Tukey’s post hoc test or by unpaired t test with Welch's or Holm-Sidak correction as appropriate . Tlr5 gene expression and TNF-α production data from ozone expose human volunteer macrophages is presented as individual values and analyzed by Wilcoxon pairwise signed rank test . All clinical studies described in this work were approved by the Institutional Review Boards of the NIEHS and Duke University respectively . Written informed consent was received from all participants prior to inclusion in the described studies .

Immune cells in the lung help guard against infections . On the surface of these cells are proteins called TLR receptors that recognize dangerous molecules or DNA from disease-causing microbes such as bacteria . When the immune cells detect these invaders , the TLR receptors spring into action and trigger an inflammatory response to destroy the microbes . This inflammation usually helps the lung clear infections . But it can also be harmful and damage the lung , for example when inflammation is caused by non-infectious substances such as pollutants in the atmosphere . There are several TLR receptors that each recognize a specific molecule . In 2010 , researchers showed that the receptor TLR4 is responsible for causing inflammation in the lung after exposure to pollution . Another receptor called TLR5 also helps activate the immune response in the lung . But it was unclear whether this receptor also plays a role in pollution-linked lung damage . Now , Hussain , Johnson , Sciurba et al . – including one of the researchers involved in the 2010 study – have investigated the role of TLR5 in immune cells from the lungs of humans and mice . The experiments showed that TLR5 works together with TLR4 and helps trigger an inflammatory response to both pollutants and bacteria . Hussain et al . found that people lacking a working TLR5 receptor ( which make up 3–10% of the population ) are less likely to experience lung inflammation when exposed to pollution or bacterial proteins that activate TLR4 . These findings suggest that people without TLR5 may be protected from pollution-induced lung injury . Further research into the role of TLR5 could help develop genetic tests for identifying people who are more sensitive to damage from pollution . This information could then be used to determine the likelihood of a patient experiencing certain lung diseases .

2020

TLR5 participates in the TLR4 receptor complex and promotes MyD88-dependent signaling in environmental lung injury

The evolutionary increase in size and complexity of the primate neocortex is thought to underlie the higher cognitive abilities of humans . ARHGAP11B is a human-specific gene that , based on its expression pattern in fetal human neocortex and progenitor effects in embryonic mouse neocortex , has been proposed to have a key function in the evolutionary expansion of the neocortex . Here , we study the effects of ARHGAP11B expression in the developing neocortex of the gyrencephalic ferret . In contrast to its effects in mouse , ARHGAP11B markedly increases proliferative basal radial glia , a progenitor cell type thought to be instrumental for neocortical expansion , and results in extension of the neurogenic period and an increase in upper-layer neurons . Consequently , the postnatal ferret neocortex exhibits increased neuron density in the upper cortical layers and expands in both the radial and tangential dimensions . Thus , human-specific ARHGAP11B can elicit hallmarks of neocortical expansion in the developing ferret neocortex . The expansion of the neocortex during primate evolution is thought to constitute one important basis for the unparalleled cognitive abilities of humans . The size of the neocortex is mainly regulated by the proliferative capacity of neural progenitor cells during cortical development and the length of the neurogenic period ( Azevedo et al . , 2009; Borrell and Götz , 2014; Dehay et al . , 2015; Kaas , 2013; Kalebic et al . , 2017; Krubitzer , 2007; Lui et al . , 2011; Molnár et al . , 2006; Rakic , 2009; Sousa et al . , 2017; Wilsch-Bräuninger et al . , 2016 ) . Two major classes of neural progenitors can be distinguished: apical progenitors ( APs ) , whose cell bodies reside in the ventricular zone ( VZ ) , and basal progenitors ( BPs ) , whose cell bodies reside in the subventricular zone ( SVZ ) . Whereas APs are highly proliferative in the neocortex of all mammalian species studied ( Götz and Huttner , 2005; Rakic , 2003a ) , BPs are highly proliferative only in species with an expanded neocortex ( Borrell and Götz , 2014; Florio and Huttner , 2014; Lui et al . , 2011; Reillo et al . , 2011 ) . Specifically , a subtype of BPs , called basal ( or outer ) radial glia ( bRG ) , are thought to play a key role in the evolutionary expansion of the neocortex ( Borrell and Götz , 2014; Florio and Huttner , 2014; Lui et al . , 2011 ) . Importantly , in species with an expanded neocortex , such as primates or the ferret , the SVZ has been shown to be divided into two distinct histological zones: the inner and outer SVZ ( ISVZ and OSVZ , respectively ) ( Dehay et al . , 2015; Reillo and Borrell , 2012; Smart et al . , 2002 ) . The OSVZ is uniquely important for the evolutionary expansion of the neocortex , as proliferative bRG are particularly abundant in this zone ( Betizeau et al . , 2013; Fietz et al . , 2010; Hansen et al . , 2010; Poluch and Juliano , 2015; Reillo and Borrell , 2012; Reillo et al . , 2011; Smart et al . , 2002 ) . Increased proliferative capacity of bRG results in an amplification of BP number and is accompanied by a prolonged phase of production of late-born neurons ( Geschwind and Rakic , 2013; Otani et al . , 2016; Rakic , 2009 ) . As the mammalian cerebral cortex is generated in an inside-out fashion , these late-born neurons occupy the upper-most layers of the cortex ( Lodato and Arlotta , 2015; Molnár et al . , 2006; Molyneaux et al . , 2007; Rakic , 1972; Rakic , 2009; Sidman and Rakic , 1973 ) . Thus , an increased generation of upper-layer neurons and increased thickness of the upper layers are also hallmarks of an expanded neocortex . The evolutionary expansion of the neocortex is characteristically accompanied by an increase in the abundance of proliferative bRG , in the length of the neurogenic period , and in the relative proportion of upper-layer neurons within the cortical plate ( Borrell and Götz , 2014; Dehay et al . , 2015; Florio and Huttner , 2014; Geschwind and Rakic , 2013; Lui et al . , 2011; Molnár et al . , 2006; Sousa et al . , 2017; Wilsch-Bräuninger et al . , 2016 ) . This is most obvious when comparing extant rodents , such as mouse , with primates , such as human . Carnivores , such as ferret , display intermediate features ( Borrell and Reillo , 2012; Hutsler et al . , 2005; Kawasaki , 2014; Reillo et al . , 2011 ) . Specifically , ferrets exhibit a gyrified neocortex and , during development , a pronounced OSVZ populated with proliferative bRG ( Barnette et al . , 2009; Borrell and Reillo , 2012; Fietz et al . , 2010; Kawasaki , 2014; Kawasaki et al . , 2013; Poluch and Juliano , 2015; Reillo et al . , 2011; Sawada and Watanabe , 2012; Smart and McSherry , 1986a;Smart and McSherry , 1986b ) . In this context , it should be noted that in evolution , the split between the lineages leading to mouse and to human occurred a few million years later than that leading to ferret and human ( Bininda-Emonds et al . , 2007 ) . In addition to the above-mentioned features associated with neocortex expansion in general , certain specific aspects of human neocortex expansion are thought to involve human-specific genomic changes . Recent transcriptomic studies established that certain previously identified human-specific genes ( Bailey et al . , 2002; Dennis and Eichler , 2016 ) are preferentially expressed in neural progenitor cells and have implicated these genes in human neocortex expansion ( Fiddes et al . , 2018; Florio et al . , 2015; Florio et al . , 2018; Florio et al . , 2016; Suzuki et al . , 2018 ) . Among these genes , the one that showed the most specific expression in human bRG compared to neurons was ARHGAP11B ( Florio et al . , 2015 ) . ARHGAP11B arose in evolution after the split of the human lineage from the chimpanzee lineage , as a product of a partial gene duplication of ARHGAP11A , a gene encoding a Rho GTPase activating protein ( Dennis et al . , 2017; Florio et al . , 2015; Florio et al . , 2016; Kagawa et al . , 2013 ) . Forced expression of ARHGAP11B in the embryonic mouse neocortex leads to an increase in BP proliferation and pool size ( Florio et al . , 2015 ) . However , as described above , the mouse exhibits only a minute amount of bRG , a cell type thought to be instrumental for neocortex expansion , and the role of ARHGAP11B on the pool size of bRG , therefore , remains elusive . Additionally , the role of ARHGAP11B on the production of upper-layer neurons , another hallmark of the evolutionary expansion of the neocortex , is also unknown . Here , we study the effects of forced expression of ARHGAP11B in the developing ferret neocortex , which already exhibits several features of an expanded neocortex , including an abundance of bRG and of upper-layer neurons , and as such is a suitable model organism to address the role of ARHGAP11B in the evolutionary expansion of the neocortex . We first examined the ability of ARHGAP11B to increase BP abundance in ferret . To this end , we immunostained E40/P0 ferret neocortex for PCNA , a marker of cycling cells , in order to identify progenitor cells ( Figure 1A and Figure 1—figure supplement 2A ) . We observed an increase in the proportion of PCNA+ FP+ cells in OSVZ of the ARHGAP11B-expressing embryos compared to control ( Figure 1B ) . The abundance of PCNA+ FP+ cells was increased in both ISVZ and OSVZ , but this increase was particularly strong in the OSVZ ( Figure 1—figure supplement 2B ) . Of note , we did not detect any increase in the abundance of FP– progenitor cells in the SVZ , suggesting that ARHGAP11B does not promote any non-cell-autonomous effects ( Figure 1—figure supplement 2C ) . We next immunostained the E40/P0 ferret neocortex for phospho-vimentin ( PhVim ) , a marker of mitotic cells ( Figure 1C ) . Our analysis revealed no effect of ARHGAP11B expression on apical mitoses ( Figure 1D , left-most ) and on basal mitoses in the abventricular VZ ( Figure 1D , second column from left ) compared to control . In contrast , a 3-fold increase in the abundance of basal mitotic cells in the SVZ was detected ( Figure 1D , sum of ISVZ and OSVZ ) . This increase was observed for the ISVZ ( 2-fold , Figure 1D , second column from right ) , but was especially prominent for the OSVZ ( 5-fold , Figure 1D , right-most column ) . A comparably large increase ( 5-fold ) was detected when examining mitotic bRG , that is , PhVim+ BPs exhibiting a PhVim+ process in the OSVZ ( Figure 1E , F ) . bRG accounted for ≈50% of all BPs upon ARHGAP11B expression , and their relative proportion was not significantly changed compared to control or non-electroporated regions ( Figure 1—figure supplement 2D ) . Of note , this strong increase in FP+ basal mitoses was not accompanied by any change in FP– mitotic cells ( Figure 1—figure supplement 2E ) nor by a change in thickness of the ferret germinal zones ( Figure 1—figure supplement 2F ) . Taken together , these data indicate that ARHGAP11B markedly increases the abundance of BP , in particular bRG , when expressed in the embryonic ferret neocortex . We next analyzed the ARHGAP11B-increased bRG in more detail . Proliferative neural progenitors , in particular apical radial glia ( aRG ) and bRG , characteristically express the transcription factor Sox2 ( Pollen et al . , 2015 ) . We therefore immunostained E40/P0 ferret neocortex for Sox2 ( Figure 2A ) and detected a 40% increase in the proportion of Sox2+ FP+ cells in the germinal zones ( GZs ) ( Figure 2B ) . This increase was exclusively due to an increase in BPs , as we observed a doubling of the proportion of Sox2+ FP+ cells in both the ISVZ and OSVZ , but no increase in the VZ ( Figure 2C ) , upon ARHGAP11B expression . These data in turn are consistent with the effects of ARHGAP11B described above ( Figure 1—figure supplement 2B ) . We then analyzed the FP+ cells for the expression of the transcription factor Tbr2 ( Figure 2A ) , a marker of certain BPs ( Englund et al . , 2005 ) . In embryonic mouse neocortex , Tbr2 is not only expressed in the predominant type of BP , the basal intermediate progenitors ( bIPs ) which are known to be neurogenic ( Haubensak et al . , 2004; Miyata et al . , 2004; Noctor et al . , 2004 ) , but also ( in contrast to an earlier report ( Wang et al . , 2011 ) ) in the vast majority of mitotic bRG ( Florio et al . , 2015 ) , which exhibit a low proliferative capacity ( Wang et al . , 2011 ) . In contrast , human bRG , which exhibit a high proliferative capacity ( Hansen et al . , 2010; LaMonica et al . , 2013 ) , largely lack Tbr2 expression ( Fietz et al . , 2010; Hansen et al . , 2010 ) . Extrapolating from these data on mouse and human BPs to the bRG in embryonic ferret neocortex , many of which express Tbr2 ( Reillo et al . , 2011 ) , it appears justified to assume that a portion of the latter bRG may be neurogenic and exhibit a reduced proliferative capacity . Consistent with this assumption and with the effects of ARHGAP11B expression on neural progenitors in the developing ferret neocortex described so far , we observed , upon expression of ARHGAP11B for 7 days , a decrease in the proportion of Tbr2+ FP+ progenitors in all GZs , which was statistically significant for the VZ ( Figure 2D , E ) . In order to potentially obtain further cues as to the proliferative capacity of the ARHGAP11B-increased bRG in the developing ferret neocortex , we focused our attention on progenitor cells that ( i ) exhibited a radial morphology , ( ii ) expressed Sox2 , but ( iii ) lacked Tbr2 expression ( Figure 2F ) . Upon ARHGAP11B expression , we observed , in the sum of the GZs , a 20% increase in the proportion of radial Sox2+ cells that were Tbr2– ( Figure 2G ) . This was largely due to an increase in the proportion of these cells in the OSVZ ( Figure 2H ) , where more than 90% were Tbr2– . Considering that the ISVZ is the GZ with the highest amount of Tbr2+ BPs ( Figure 2A ) , we examined the relative proportions of Sox2+ Trb2– and Sox2+ Tbr2+ BPs in the ISVZ and did not observe any significant difference in these proportions between control and ARHGAP11B expression ( Figure 2—figure supplement 1 ) . These findings are consistent with the notion that the ARHGAP11B-increased radial Sox2+ Tbr2– cells in the OSVZ are bRG . Studies in fetal human neocortex have established that such cells in the OSVZ are highly proliferative ( Hansen et al . , 2010 and LaMonica et al . , 2013; for reviews see Florio and Huttner , 2014 and Lui et al . , 2011 ) . Hence , our finding that expression of ARHGAP11B in developing ferret neocortex results in a marked increase in the proportion of these cells suggests that this human-specific gene is sufficient to promote , in a gyrencephalic carnivore , the generation of bRG with putatively increased proliferative capacity . We investigated the potential consequences of the ARHGAP11B-elicited increase in the abundance of BPs , notably of Sox2+ Tbr2– bRG , for neurogenesis in the developing ferret neocortex . To this end , we immunostained E40/P0 ferret neocortex for Tbr1 , a transcription factor which is a marker of deep-layer neurons ( Kolk et al . , 2006 ) ( Figure 3—figure supplement 1A ) , and for Satb2 , a transcriptional regulator which is expressed in neurons that establish callosal projections and that are highly enriched in the upper layers of the cortical plate ( CP ) ( Alcamo et al . , 2008; Britanova et al . , 2008 ) ( Figure 3A ) . The vast majority ( >90% ) of the FP+ neurons in the CP of both control and ARHGAP11B-expressing ferret neocortex were found to be Satb2+ ( Figure 3—figure supplement 1B ) . This high percentage is consistent with our experimental approach in which we targeted the embryonic ferret neural progenitors by in utero electroporation at E33 , that is the time when the generation of the upper-layer neurons starts ( Jackson et al . , 1989; Martínez-Martínez et al . , 2016 ) . Analysis of the distribution of Satb2+ FP+ neurons at E40/P0 between the CP on the one hand side , and the GZs plus the intermediate zone ( IZ ) on the other hand side , revealed that around 60% of the neurons had reached the CP in both control and ARHGAP11B-expressing ferret neocortex ( Figure 3B left ) . We next examined electroporated ferret neocortex at P10 ( Figure 3—figure supplement 2; analysis confined to gyri ) , which is the stage when neuron production is completed and neuron migration is terminating in the motor and somatosensory areas ( Jackson et al . , 1989; Smart and McSherry , 1986a; Smart and McSherry , 1986b ) . Consistent with this , our analysis of the control brains revealed that more than 90% of the Satb2+ FP+ neurons had reached the CP ( Figure 3B middle ) . In contrast , only 70% of the Satb2+ FP+ neurons were found in the CP of the ARHGAP11B-expressing neocortex ( Figure 3B middle ) . However , at P16 , nearly all Satb2+ FP+ neurons were found in the CP in both control and ARHGAP11B-expressing neocortex ( Figure 3B right; again , analysis confined to gyri ) . These data suggested that upon ARHGAP11B expression , either neurons migrate more slowly to the CP , or the neurogenic period is extended . To explore the latter scenario , we injected EdU into P5 ferret kits ( i . e . 12 days after electroporation ) , which is the stage when neuronal progenitors undergo their very last neuron-generating cell divisions in the motor and somatosensory areas of the neocortex ( Jackson et al . , 1989; Smart and McSherry , 1986a; Smart and McSherry , 1986b ) . Analysis 11 days after EdU injection , at P16 ( Figure 3C ) , revealed a 4-fold increase in the proportion of FP+ cells that were EdU+ , in neocortical gyri of ARHGAP11B-expressing kits compared to control ( Figure 3D ) . This was consistent with a prolonged , and hence increased , production of cells in ARHGAP11B-expressing kits , which in turn would be in line with the above described finding that ARHGAP11B increases the abundance of proliferative bRG . Importantly , among the EdU+ FP+ cells of the ARHGAP11B-expressing neocortex , 20% were Satb2+ neurons ( Figure 3C' and E ) . In contrast , we did not detect a single Satb2+ EdU+ FP+ neuron in any of the control neocortices ( Figure 3E ) . Collectively , these data indicate that neurogenesis in ARHGAP11B-expressing ferret neocortex continues longer than in control neocortex . In light of the extension of the neurogenic period upon ARHGAP11B expression , we examined a potential increase in the abundance of the last-born neurons , that is , the upper-layer neurons . To this end , we first performed Nissl staining of the P16 ferret neocortex to visualize all neurons and the various layers of the CP , and immunostaining for Satb2 , which is expressed in the majority of upper-layer neurons ( Alcamo et al . , 2008; Britanova et al . , 2008; Lodato and Arlotta , 2015 ) ( Figure 4A ) . These analyses , carried out on gyri , revealed ( i ) an increase in the abundance of FP+ cells in the CP ( Figure 4—figure supplement 1A ) , and ( ii ) an alteration in the distribution of the FP+ cells between layers II-VI of the CP , with a greater proportion of cells in layers II-IV ( Figure 4—figure supplement 1B ) , in the ARHGAP11B-expressing neocortex compared to control . A similar abundance increase ( Figure 4—figure supplement 1C ) and altered distribution ( Figure 4B ) were observed for Satb2+ FP+ neurons . Furthermore , the proportion of FP+ cells in the CP that were Satb2+ neurons was increased upon ARHGAP11B expression ( Figure 4C ) . We sought to corroborate these data by immunostaining the P16 ferret neocortex for Brn2 ( Pou3f2 , Figure 4—figure supplement 2A ) , a transcription factor considered to be a marker of layer II and III neurons ( Dominguez et al . , 2013; McEvilly et al . , 2002; Sugitani et al . , 2002 ) . Quantification of Brn2+ FP+ cells in layers II and III of gyri revealed a marked increase in the abundance of these cells ( Figure 4—figure supplement 2B ) and in the proportion of FP+ cells that were Brn2+ neurons ( Figure 4D ) , in ARHGAP11B-expressing kits compared to control . Of note , the effects of ARHGAP11B expression on the Brn2+ neurons were of a greater degree than those on the Satb2+ neurons ( compare Figure 4—figure supplement 2B with Figure 4—figure supplement 1C; Figure 4 , panel D with panel C ) , in line with the notion that Brn2 exhibits a greater specificity for layer II and III neurons than Satb2 ( Dominguez et al . , 2013; Sugitani et al . , 2002 ) . Taken together , our data indicate that the extension of the neurogenic period upon ARHGAP11B expression in the developing ferret neocortex is accompanied by an increase in the abundance of layer II and III neurons . Finally , we explored whether the ARHGAP11B-elicited increase in the abundance of BPs , notably of proliferative bRG , resulting in an extended neurogenic period and in an increased abundance of upper-layer neurons , had any consequences for the size and morphology of the ferret neocortex . To this end , we analyzed the neocortex at P16 ( Figure 4—figure supplement 3A–C ) , the developmental stage when cortical folds have already formed in the ferret ( Barnette et al . , 2009; Fernández et al . , 2016; Matsumoto et al . , 2017; Sawada and Watanabe , 2012; Shinmyo et al . , 2017 ) , and quantified a set of morphological parameters ( Figure 4—figure supplement 3D ) , each on two coronal sections located slightly rostrally ( Figure 4—figure supplement 3B and C , position 1 ) and slightly caudally ( Figure 4—figure supplement 3B and D , position 2 ) , respectively , to a middle position along the rostro-caudal axis . This revealed no significant differences between control and ARHGAP11B-expressing kits with regard to brain mass ( Figure 4—figure supplement 3E ) and gross neocortex morphology , including the gyrification index of the electroporated area ( referred to as local GI ) , the size of the posterior sigmoid gyrus , lateral gyrus and coronal gyrus , and the depth and thickness of the cruciate sulcus , suprasylvian sulcus and lateral sulcus ( Figure 4—figure supplement 3F–I; for the gyri and sulci positions , see Figure 4—figure supplement 3B–D ) . However , we did detect an increase , upon ARHGAP11B expression , in the thickness of the three gyri analyzed ( Figure 4E ) . We therefore investigated the underlying cause of this thickness increase and , considering the increased abundance of upper-layer neurons upon ARHGAP11B expression ( Figure 4—figure supplement 1C , Figure 4—figure supplement 2B ) , measured the thickness of CP layers II-IV in the gyri of control and ARHGAP11B-expressing ferret neocortex at P16 . This revealed a 100 µm increase in the thickness of these upper layers in ARHGAP11B-expressing neocortex compared to control ( Figure 4F , note also the thicker layer II in Figure 4A and another example in Figure 4—figure supplement 5G ) . Hence , ARHGAP11B is sufficient to promote neocortex expansion in ferret in the radial dimension . We then explored whether ARHGAP11B would also be able to promote neocortex expansion in ferret in the tangential dimension . Here , we exploited the previous findings that during the first week of the ferret postnatal neocortex development , migrating neurons change their migration mode from radial to tangential , which results in an increased lateral dispersion of the late-born neurons ( Gertz and Kriegstein , 2015; Reillo et al . , 2011; Smart and McSherry , 1986a; Smart and McSherry , 1986b ) . In line with this , the abundance of FP+ cells per 200 µm-wide field of cortical wall decreased about 3-fold from E40/P0 to P16 ( Figure 4—figure supplement 4A ) . Yet , the increase in FP+ cells per field of cortical wall due to ARHGAP11B expression was not only observed at P16 , confirming the above described analysis of FP+ cell numbers in the CP ( Figure 4—figure supplement 1A ) , but was already detected at E40/P0 ( Figure 4—figure supplement 4A ) . The decrease in FP+ cell abundance per 200 µm-wide field of cortical wall from E40/P0 to P16 ( Figure 4—figure supplement 4A ) was not accompanied by an increase in cell death ( Figure 4—figure supplement 4B–G ) . Rather , this decrease reflected an increase in the lateral spread of the progeny of the targeted cells , as evidenced by the ARHGAP11B-elicited increase in the lateral length of the area harbouring FP+ cells that was observed at P16 ( Figure 4G , Figure 4—figure supplement 5A top , B ) . Of note , the increase in the lateral spread of FP+ cells persisted along the rostro-caudal axis of the FP+ region of the neocortex ( Figure 4—figure supplement 5C ) and did not reflect a difference in the electroporation efficiency , as the lateral length of the FP+ area on both the basal and apical side was not different between control and ARHGAP11B-expressing E40/P0 ferret neocortex ( Figure 4—figure supplement 5D , E ) . Importantly , the total lateral length of the dorsal neocortex was also increased upon ARHGAP11B expression ( Figure 4H , Figure 4—figure supplement 5A bottom ) . This increase was observed at different positions along the rostro-caudal axis ( Figure 4—figure supplement 5F ) . These data further imply that the increase in the lateral spread ( Figure 4G , Figure 4—figure supplement 5A top , B ) did not occur at the expense of the other , FP– , areas . We conclude that ARHGAP11B is sufficient to promote neocortex expansion in ferret also in the tangential dimension . Given that the ARHGAP11B-elicited increase in the lateral length of the FP+ area ( ≈25% , Figure 4G ) was greater than the ARHGAP11B-elicited tangential expansion ( ≈10% , Figure 4H ) , we explored whether this discrepancy could be resolved by ARHGAP11B expression increasing cell density , specifically in the upper layers of the CP where most of the ARHGAP11B-increased neurons resided . Indeed , our analysis revealed an increase in total cell density ( revealed by DAPI staining ) ( Figure 4—figure supplement 5G , H ) and in neuronal density ( revealed by Satb2 immunostaining , Figure 4—figure supplement 5G , I ) in layers III and IV at P16 , but not in layer II , which in general exhibited the highest cell density . Taken together , these data indicate that the ARHGAP11B-induced increase in BP pool size led to an increase in upper-layer neurons that in turn resulted in ( i ) a thickening of the upper layers , ( ii ) a higher cell density in those upper layers that were able to accommodate additional neurons , ( iii ) a greater lateral spread of upper-layer neurons , and ( iv ) consequently , a greater lateral length of the dorsal neocortex . Neocortex expansion involves not only an increase in the number of neurons , but also of glial cells . Although we on purpose performed in utero electroporation of the ferret neocortex at E33 in order to target OSVZ progenitors generating upper-layer neurons ( Jackson et al . , 1989; Martínez-Martínez et al . , 2016 ) , and therefore would expect only a small proportion of the FP+ cells to be glial cells , we nonetheless examined the possible occurrence of GFAP+ and S100ß+ cells ( as indicators of an astrocytic lineage ( Reillo et al . , 2011; Voigt , 1989 ) ) and of Olig2+ cells ( as an indicator of an oligodendrocytic lineage ( Mo and Zecevic , 2009; Voigt , 1989 ) ) among the FP+ progeny of the targeted cells . An additional reason to examine astrocytes was that radial glial progenitor cells at later stages of human , monkey and ferret neocortex development are known to differentiate into astrocytic progenitors which then give rise to post-mitotic astrocytic cells ( Rakic , 1978; Rakic , 2003b; Reillo et al . , 2011; Voigt , 1989 ) . Upon immunostaining of the P16 ferret neocortex for GFAP ( Figure 4—figure supplement 5A , B ) and Olig2 ( Figure 4—figure supplement 5E ) , no FP+ cells in the CP of control neocortex that were GFAP+ or Olig2+ could be detected . In contrast , while we still did not detect any Olig2+ FP+ cells in the CP of ARHGAP11B-expressing neocortex , about 3% of the CP FP+ cells were GFAP+ ( Figure 4—figure supplement 5C ) . Given that ARHGAP11B expression increases the pool size of FP+ cells ( Figure 4—figure supplement 1A , Figure 4—figure supplement 4A ) , this translates into an ARHGAP11B-induced appearance of GFAP+ FP+ cells in the CP . All of these GFAP+ FP+ cells were Ki67– ( data not shown ) and hence post-mitotic astrocytes . To corroborate these data , we performed immunostaining of the P16 ferret neocortex for another astrocyte marker , S100ß ( Figure 4–figure supplement 5D ) ( López-Hidalgo et al . , 2016 ) , and readily identified FP+ S100ß+ astrocytes upon ARHGAP11B expression . These data are consistent with at least two scenarios related to the increased abundance of proliferative bRG upon ARHGAP11B expression . Either this increase results not only in an extension of the neurogenic period but also in the initiation of astrogenesis . Or the greater bRG pool size gives rise to a detectable level of astrocytic cells once the bRG differentiate along the astrocytic lineage ( Rakic , 2003b; Reillo et al . , 2011; Voigt , 1989 ) . The present effects of ARHGAP11B on BPs in developing ferret neocortex are substantially greater than , and significantly different from , those previously observed upon transient ARHGAP11B expression in embryonic mouse neocortex ( Florio et al . , 2015 ) , with regard to BP quantity and quality . As to quantity , in embryonic mouse neocortex , ARHGAP11B expression resulted in a doubling and tripling of the pool size of total BPs and BPs in mitosis , respectively ( Florio et al . , 2015 ) , whereas in embryonic ferret neocortex , ARHGAP11B expression led to nearly tripling and quintupling of the pool size of total BPs and BPs in mitosis ( notably mitotic bRG ) , respectively ( for a data summary , see Supplementary file 1 ) . The greater effect of ARHGAP11B expression on mitotic BP abundance than total BP abundance , in both embryonic mouse and ferret neocortex , indicates that ARHGAP11B expression increases the proportion of the total duration of the BP cell cycle that is used for M-phase . This in turn would be consistent with ARHGAP11B promoting , in embryonic ferret neocortex , the proliferative rather than neurogenic mode of bRG division . This notion would be in line with the previous findings that both , cortical progenitors in embryonic mouse neocortex ( Arai et al . , 2011 ) and radial glial progenitors in postnatal ferret neocortex ( Turrero García et al . , 2016 ) , which are not yet committed to neurogenesis spend a greater proportion of their total cell cycle in M-phase than those committed to neurogenesis . Regarding the regulation of cell cycle length , it appears worthwhile to consider the present increase in BP abundance upon ARHGAP11B expression in developing ferret neocortex in the context of previous findings showing that shortening the duration of the G1 phase in progenitors of both mouse ( Lange et al . , 2009; Pilaz et al . , 2009 ) and ferret ( Nonaka-Kinoshita et al . , 2013 ) developing neocortex results in increased BP proliferation and neocortex expansion . Furthermore , manipulation of cyclins or cdk/cyclin complexes was shown to lead to an increase in upper layer and callosal projection neurons ( Lange et al . , 2009; Nonaka-Kinoshita et al . , 2013; Pilaz et al . , 2009 ) , which again is consistent with the phenotypes observed in the present study upon expression of ARHGAP11B in the developing ferret neocortex . As to BP quality , mouse bRG have been shown to undergo mostly neurogenic cell divisions and to display a limited proliferative capacity ( Shitamukai et al . , 2011; Wang et al . , 2011; Wong et al . , 2015 ) . In contrast , human bRG are known to be highly proliferative , which is thought to be one of the key factors contributing to the evolutionary expansion of the human neocortex ( Fietz et al . , 2010; Florio and Huttner , 2014; Hansen et al . , 2010; Lui et al . , 2011 ) . The proliferative capacity of bRG in other mammalian species that have been studied appears to follow the general rule that highly proliferative bRG are more often found in species with an expanded neocortex ( Betizeau et al . , 2013; Reillo et al . , 2011 ) . Hence , the ARHGAP11B-promoted increase in the number of proliferative ( Sox2+ Tbr2– ) bRG in the developing ferret neocortex fulfills a key requirement for further neocortex expansion . In this context , it should be emphasized that those of the human-specific genes ( Dennis and Eichler , 2016 ) that have been shown to be preferentially expressed in cortical progenitor cells ( Florio et al . , 2018 ) and that have been ectopically expressed in embryonic mouse neocortex , that is ARHGAP11B ( Florio et al . , 2015 ) and NOTCH2NL ( Fiddes et al . , 2018; Florio et al . , 2018; Suzuki et al . , 2018 ) , show an increase in BP proliferation that mostly involves basal intermediate progenitors . To the best of our knowledge , our study provides first evidence of a human-specific gene increasing the number of proliferative bRG in a gyrencephalic neocortex . As bRG are considered to be instrumental for the evolutionary expansion of neocortex , this suggests that the ferret likely is a model system superior to mouse for studying the role of human-specific genes in neocortex development . An increase in the upper layers is a fundamental characteristic of neocortex expansion ( Fame et al . , 2011; Hutsler et al . , 2005; Molnár et al . , 2006; Tarabykin et al . , 2001 ) . Layers II-IV are evolutionarily the most novel and mammalian-specific layers ( Molnár et al . , 2006 ) . Layers II and III , in particular , underwent disproportional expansion during primate evolution . As the layers II-III neurons are the last-born neurons , an increase in the length of the neurogenic period might result in an increase in the number of late-born neurons . In fact , it has been proposed that the length of the neurogenic period may be a contributing factor in explaining the evolutionary increase in neocortex size and neuron number between humans and other great apes ( Lewitus et al . , 2014 ) . Consistent with this , we observed a prolonged generation of late-born neurons , increased thickness of layers II-IV , and a marked increase in Satb2+ ( 3 . 3-fold ) and , in particular , Brn2+ ( 5 . 6-fold ) neurons in the upper layers of the CP , upon ARHGAP11B expression ( for a data summary , see Supplementary file 1 ) . Thus , also with regard to the key product generated by cortical progenitor cells , that is the neurons , ARHGAP11B is sufficient to promote another feature in the developing ferret neocortex that would be consistent with further neocortex expansion . In this context , Satb2+ neurons are of special interest , as Satb2 is essential for establishing callosal projections to the contralateral hemisphere ( Alcamo et al . , 2008; Britanova et al . , 2008 ) . Callosal projection neurons are considered to play a key role in the high-level associative connectivity , thus contributing significantly to human cognitive abilities , with their impairment causing cognition-related pathologies ( Fame et al . , 2011 ) . Given that the evolutionary increase in neocortex size is accompanied by an increase in callosal projection neurons , our data show that ARHGAP11B elicits yet another , specific feature in the developing ferret neocortex associated with further neocortex expansion . A hallmark of neocortex expansion is neocortical folding ( Borrell , 2018; Borrell and Götz , 2014; Kroenke and Bayly , 2018 ) . Transient expression of ARHGAP11B in the embryonic mouse neocortex induced cortical folding in about half of the cases in this normally lissencephalic rodent ( Florio et al . , 2015 ) . However , expression of ARHGAP11B in the developing ferret neocortex did not significantly increase the GI , gyrus size , or sulcus depth and thickness in this gyrencephalic carnivore . The major morphological signs of neocortical expansion upon ARHGAP11B expression in developing ferret neocortex that we could detect were ( i ) an increased gyrus thickness , which reflected the already discussed increase in upper-layer thickness due to the increase in upper-layer neurons , and ( ii ) a larger lateral area of the CP where these neurons were found . The latter phenotype resulted in a ≈10% increase in the lateral length of the dorsal neocortex , which however did not appear to be sufficient to significantly increase the gyrus size or GI . The reason why a 3–5–fold increase in upper-layer neurons ( Supplementary file 1 ) resulted in more modest increases in upper-layer thickness and dorsal neocortex lateral length ( 24% and 10% , respectively ) could be that either ( i ) the increase in the number of FP+ neurons occurred at the expense of FP– neurons , or ( ii ) that the FP+ neurons added themselves to the FP– neurons already existing in the control condition , thereby increasing neuronal cell density in the upper layers . Our results show that the latter scenario is the case , as we detected a significant increase in both the total cell density and neuronal density in the upper cortical layers upon ARHGAP11B expression . Analysis of cortical neuron density as a function of cortical neuron number across carnivores has revealed that neuron density tends to decrease with increasing neuron number ( Herculano-Houzel et al . , 2015; Jardim-Messeder et al . , 2017; Lewitus et al . , 2012 ) . This trend is much less prominent in primates , which in general exhibit a greater cortical neuron density per cortical neuron number than carnivores . Thus , the increase in upper-layer neuron density upon increasing upper-layer neuron number by the human-specific ARHGAP11B can be regarded as a form of ‘primatization’ of the ferret neocortex . In this context , it is interesting to note that neocortical neuron density across carnivores shows a much greater range of variation than that across primates ( Herculano-Houzel et al . , 2015; Lewitus et al . , 2012 ) . We therefore hypothesize that the ferret neocortex is plastic enough to accommodate additional neurons , without having to increase neocortex size or folding . A corollary of this notion with regard to ARHGAP11B is that its primary function is to increase the proliferation and thus pool size of BPs , notably of bRG , which results in a lengthening of the neurogenic period and an increased generation of neurons , in particular of upper-layer neurons . In embryonic mouse neocortex , the ARHGAP11B-elicited increase in BPs could induce cortical folding as an indicator of neocortex expansion ( Florio et al . , 2015 ) . In contrast , in the developing ferret neocortex studied here , the ARHGAP11B-elicited BP increase did not further increase the GI , although it did increase the neurogenic period and upper-layer neurons . This in turn resulted in a greater thickness of the upper layers , increased neuron density , and a larger lateral length of the dorsal neocortex . The difference between the phenotypes observed in mouse and ferret developing neocortex therefore appears to reflect how species in different mammalian orders tend to deal with an increase in cortical neuron number , rather than the primary function of ARHGAP11B . All experimental procedures were conducted in agreement with the German Animal Welfare Legislation after approval by the Landesdirektion Sachsen ( licences TVV 2/2015 and TVV 21/2017 ) . Timed-pregnant ferrets ( Mustela putorius furo ) were obtained from Marshall BioResources ( NY , USA ) or Euroferret ( Copenhagen , Denmark ) and housed at the Biomedical Services Facility ( BMS ) of MPI-CBG . Observed mating date was set to E0 . Animals were kept in standardized hygienic conditions with free access to food and water and with an 16 hr/8 hr light/dark cycle . All experiments were performed in the dorsolateral telencephalon of ferret embryos , at a medial position along the rostro-caudal axis , in the prospective motor and somatosensory cortex . All plasmids used in this study were previously published ( see the Key Resources table ) . All plasmids were extracted and purified using the EndoFree Plasmid Maxi kit ( QIAGEN ) following the manufacturer’s instructions . In utero electroporation of ferrets was performed as originally established by Dr . Hiroshi Kawasaki with the modifications listed below ( Kawasaki et al . , 2012 ) . Pregnant jills ( with embryos at E33 ) were kept fasted for at least 3 hr before the surgery . They were first placed in the narcosis box with 4% isoflurane . When in deep anesthesia , the ferrets were placed on the operation table and attached to the narcosis mask with a constant 3% isoflurane flow . Subsequently , the ferrets were injected subcutaneously with analgesic ( 0 . 1 ml Metamizol , 50 mg/kg ) , antibiotic ( 0 . 13 ml Synulox , 20 mg/kg or 0 . 1 ml amoxicilin , 10 mg/kg ) and glucose ( 10 ml 5% glucose solution ) . A drop of Dexpanthenol Ointment solution was placed on their eyes to prevent eye dehydration during the surgical procedure . The ferret bellies were then shaved , sterilized with iodide and surgically opened . The uterus was exposed and embryos were injected intraventricularly with a solution containing 0 . 1% Fast Green ( Sigma ) in sterile PBS , 2 . 5 μg/μl of either pCAGGS vector ( control ) or pCAGGS-ARHGAP11B vector , in either case together with 1 μg/μl of pCAGGS vector encoding a fluorescent protein ( pCAGGS-GFP or pCAGGS-mCherry ) . For embryonic studies where the position of the electroporated embryos in the uterus was known , the same fluorescent protein-encoding vector was co-electroporated together with either control or ARHGAP11B-expressing vectors . For postnatal studies , pCAGGS and pCAGGS-ARHGAP11B were co-electroporated with different fluorescent protein-encoding vectors to enable distinction of kits . For different experiments , these vectors were alternated . Electroporations were performed with six 50-msec pulses of 100 V at 1 s intervals . Following electroporation , uterus was placed back in the peritoneal cavity and the muscle layer with the peritoneum were sutured using a 4–0 suture . The skin was sutured intracutaneously using the same thickness of the suture . Animals were carefully monitored until they woke up and then underwent postoperative care for the following 3 days ( 2 x daily 10 mg/kg amoxicilin , 3 x daily 25 mg/kg Metamizol ) . For the neurogenic length experiments , a single pulse of EdU was injected intraperitoneally at P5 , that is 13 days after in utero electroporations , and the animals were sacrificed 11 days later , at P16 . The protocol for EdU injection was the same as the one previously published for ferret kits ( Turrero García et al . , 2016 ) . Ferret embryos were isolated at E37 and E40 . Ferret kits were sacrificed at P0 , P10 and P16 . For the isolation of embryos , pregnant jills underwent the second surgery that followed the same pre-operative care , anesthesia and analgesia as the first surgery . The sutures from the first operation were carefully removed and the uterus was exposed . A caesarian section was made and embryos were removed from the uterus . Subsequently a complete hysterectomy was performed , after which the muscle layer with peritoneum and skin were sutured and the animal underwent the same post-operative care as after the first surgery . Embryonic brains were isolated and fixed at 4°C for 48 hr in 4% paraformaldehyde in 120 mM phosphate buffer pH 7 . 4 . For the postnatal time points , the kits were sacrificed by intraperitonal injection of 4 mg/kg Xylazin +40 mg/kg Ketamin . When in deep anaesthesia , the kits were perfused intracardially with PBS , followed by perfusion with 4% paraformaldehyde in 120 mM phosphate buffer pH 7 . 4 at room temperature . Kit brains were then isolated and fixed for 48 hr in 4% paraformaldehyde in 120 mM phosphate buffer pH 7 . 4 at 4°C . Ferret jills whose kits were used postnatally also underwent the second surgery with the hysterectomy . Upon this surgery , they underwent the same post-operative care as after the first surgery . All jills were kept at the BMS of the MPI-CBG for at least two weeks after the second surgery . Afterwards they were donated for adoption . No adult ferrets were sacrificed in this study . Upon fixation ferret brains were sectioned either on a vibratome or cryostat . Vibratome sections were 70 μm thick . Thickness of cryosections varied from 35 to 60 μm . Vibratome sections were either immunostained freshly prepared or conserved in cryoprotectant solution ( 30% sucrose , 30% ethylene glycol , 1% PVP40 , 1 . 3 mM NaH2PO4 , 3 . 9 mM Na2HPO4 , 15 mM NaCl ) and stored at –20°C for later use . For analysis of ARHGAP11B expression , total RNA was isolated from two to three cryosections per developmental stage , using the RNeasy FFPE RNA isolation kit ( Qiagen ) including DNase-treatment following the manufacturer’s instructions . An additional DNase-treatment was performed on the isolated RNA using the DNA-free DNA Removal Kit ( Life technologies ) . cDNA was synthesized using random hexamers and Superscript III Reverse Transcriptase ( Life Technologies ) . qPCR was performed using the Light Cycler SYBR green Master mix ( Roche ) on a Light Cycler 96 ( Roche ) . Gene expression data were normalized based on the housekeeping gene Hprt1 . Primer sequences are provided in the Key Resources Table . Immunofluorescence was performed as previously described ( Kalebic et al . , 2016 ) . Antigen retrieval ( 1 hr incubation with 10 mM citrate buffer pH 6 . 0 at 70°C in a water bath or oven ) was performed for the following samples: all E37 samples , all E40/P0 samples , and vibratome sections of P10 and P16 samples which were used for immunostainings of the nuclear markers ( Satb2 , Brn2 , Sox2 , Olig2 and Ki67 ) . When immunofluorescence for Tbr2 ( E40/P0 ) was performed , a modified antigen retrieval protocol was used ( 1 hr incubation with 10 mM citrate buffer pH 6 . 0 supplemented with 0 . 05% Tween-20 at 80°C in a water bath ) . Antigen retrieval was followed by three washes with PBS . All the immunostainings , except the one for ARHGAP11B , were done as follows . Samples were subjected to permeabilization for 30 min in 0 . 3% Triton X-100 in PBS at room temperature , followed by quenching for 30 min in 0 . 1 M glycine in PBS at room temperature . Blocking was performed in a blocking solution ( 0 . 2% gelatin , 300 mM NaCl , 0 . 3% Triton X-100 in PBS ) for 30 min . Primary antibodies were incubated in the blocking solution for 48 hr at 4°C . Subsequently , the sections were washed three times in the blocking solution , incubated with secondary antibodies ( 1:500 ) and DAPI ( Sigma ) in the blocking solution for 1 hr at room temperature , and washed again three times in the blocking solution before being either used for Nissl post-staining or directly mounted on microscopy slides with Mowiol . For the immunostaining for ARHGAP11B , permeabilization was performed for 1 hr in the 0 . 3% Triton X-100 solution at room temperature . Quenching was performed as for the other immunostainings . Blocking was done in 10% horse serum supplemented with 0 . 3% Triton X-100 ( HS blocking solution ) for 1 hr . Primary antibodies were incubated in the same solution for 24 hr , at 4°C . Subsequently , sections were washed five times in the HS blocking solution , incubated with the secondary antibodies ( 1:1000 ) and DAPI in the HS blocking solution for 1 hr at room temperature , washed five times in PBS and mounted on microscopy slides with Mowiol . ARHGAP11B antibody used in this study is a mouse monoclonal antibody raised against a recombinant full length ARHGAP11B . As the first 220 amino acids of ARHGAP11B are 89% identical to the first 220 amino acids of the ferret Arhgap11a , we showed by immunostaining that the antibody we used did not recognize the endogenous ferret Arhgap11a ( Figure 1—figure supplement 1E ) . Staining with the Nissl fluorescent stain , EdU detection , annexin V labeling , and TUNEL labeling were performed after the antibody stainings . NeuroTraceTM 640/660 deep-red fluorescent Nissl stain ( Molecular probes ) was used , following the manufacturer's instructions . EdU staining was performed using Click-iT EdU Alexa Fluor 647 Imaging kit ( Invitrogen ) , following the manufacturer's instructions . Annexin V labeling was performed using the Annexin V-Cy5 bright fluorescence reagent ( BioVision ) , following the manufacturer's instructions . TUNEL labeling was performed using the In situ cell death detection kit , TMR red ( Sigma-Aldrich ) , following the manufacturer's instructions . Images of whole ferret brains ( Figure 4—figure supplement 3A ) were obtained as follows . Images were acquired using an iPhone 7 and taking a photography of the brain through a UV-protection filter mounted on a SZX 16 Olympus stereomicroscope , equipped with a fluorescence lamp . Fluorescent images were acquired using a Zeiss LSM 880 upright single-photon point scanning confocal system . For the embryonic stages , images were taken as either 1 μm single optical sections with the 40x objective or 2 μm single optical sections with the 20x objective . For the postnatal stages , images were taken as either 2 μm single optical sections with the 20x objective or 7 . 2 μm single optical sections with the 10x objective . When the images were taken as tile scans , the stitching of the tiles was performed using the ZEN software . Subsequently , all images were analyzed and processed with ImageJ ( http://imagej . nih . gov/ij/ ) . All cell counts were performed in standardized microscopic fields using Fiji , processed using Excel ( Microsoft ) , and results were plotted using Prism ( GraphPad Software ) . For each condition , data ( typically at least three microscopic fields ) from one experiment ( see definition below ) were pooled , and the mean of the indicated number of experiments was calculated . Whenever possible the quantifications were done blindly . The definition of the morphological parameters is depicted graphically in the Figure 4—figure supplement 3D . All the morphological parameters except the lateral length of the dorsal neocortex were quantified at two positions ( referred to as positions 1 and 2 ) in the somatosensory cortex , as defined in Figure 4—figure supplement 3B , for the following gyri and sulci: posterior sigmoid gyrus , coronal gyrus , lateral gyrus , cruciate sulcus , suprasylvian sulcus and lateral sulcus ( Sawada and Watanabe , 2012 ) . All the morphological parameters are presented as a ratio of the value of the electroporated hemisphere and the value of the contralateral hemisphere , as established previously ( Matsumoto et al . , 2017 ) . Calculations were as follows . Local Gyrification Index ( Local GI ) : ThelengthoftheinnercontourThelengthoftheoutercontour Local gyrification index is calculated as a ratio of the local GI of the electroporated area and the local GI of the equivalent area on the contralateral side . Gyrus size is calculated as a ratio of the size of an electroporated gyrus ad the size of the equivalent gyrus of the contralateral side . Sulcus depth is calculated as a ratio of the length of the line connecting the outer contour and the bottom of a sulcus and the equivalent line on the contralateral side . Sulcus thickness is calculated as a ratio of the thickness measured from the ventricular surface to the bottom of an electroporated sulcus and the equivalent thickness on the contralateral side . Gyrus thickness is calculated as a ratio of the thickness measured from the ventricular surface to the top of an electroporated gyrus and the equivalent thickness on the contralateral side . The upper layers thickness was measured as distance between the top of layer II and bottom of layer IV . Lateral length of the FP+ area was measured as the distance between the medial-most and the lateral-most FP+ cell on a coronal cross-section , following the inner contour of the neocortex . The lateral length of the dorsal neocortex was measured at three positions along the rostro-caudal axis ( positions 1 and 2 , and position 1 . 5 located between positions 1 and 2 ( Figure 4—figure supplement 5C bottom ) ) , and was defined as a distance between the cingulate gyrus and ectosylvian gyrus along the inner contour , as depicted in Figure 4—figure supplement 5A bottom . Total cell density was measured as number of DAPI+ nuclei in 50 µm x 50 μm fields . Neuronal density was measured as number of Satb2+ nuclei in the 50 µm x 50 μm fields . All statistics analyses were conducted using Prism ( GraphPad Software ) . Sample sizes are reported in each figure legend , where the term ‘one experiment’ would refer to one embryo for analysis at E37 or E40 , and to one kit for postnatal analysis . Total number of litters analyzed was as follows: E37 , two litters; E40/P0 , four litters; P10 , two litters; P16 , three litters . Embryos or kits from all litters were included in the statistical analyses . Tests used were Two-way ANOVA with Bonferroni posttest and Student's t-test . For each quantification , the statistical test and significance are indicated in the figure legend .

The human brain owes its characteristic wrinkled appearance to its outer layer , the cerebral cortex . All mammals have a cerebral cortex , but its size varies greatly between species . As the brain evolved , the neocortex , the evolutionarily youngest part of the cerebral cortex , expanded dramatically and so had to fold into wrinkles to fit inside the skull . The human neocortex is roughly three times bigger than that of our closest relatives , the chimpanzees , and helps support advanced cognitive skills such as reasoning and language . But how did the human neocortex become so big ? The answer may lie in genes that are unique to humans , such as ARHGAP11B . Introducing ARHGAP11B into the neocortex of mouse embryos increases its size and can induce folding . It does this by increasing the number of neural progenitors , the cells that give rise to neurons . But there are two types of neural progenitors in mammalian neocortex: apical and basal . A subtype of the latter – basal radial glia – is thought to drive neocortex growth in human development . Unfortunately , mice have very few basal radial glia . This makes them unsuitable for testing whether ARHGAP11B acts via basal radial glia to enlarge the human neocortex . Kalebic et al . therefore introduced ARHGAP11B into ferret embryos in the womb . Ferrets have a larger neocortex than mice and possess more basal radial glia . Unlike in mice , introducing this gene into the ferret neocortex markedly increased the number of basal radial glia . It also extended the time window during which the basal radial glia produced neurons . These changes increased the number of neurons , particularly of a specific subtype found mainly in animals with large neocortex and thought to be involved in human cognition . Introducing human-specific ARHGAP11B into embryonic ferrets thus helped expand the ferret neocortex . This suggests that this gene may have a similar role in human brain development . Further experiments are needed to determine whether ferrets with the ARHGAP11B gene , and thus a larger neocortex , have enhanced cognitive abilities . If they do , testing these animals could provide insights into human cognition . The animals could also be used to model human brain diseases and to test potential treatments .

2018

Human-specific ARHGAP11B induces hallmarks of neocortical expansion in developing ferret neocortex

The mechanistic target of rapamycin complex 1 ( mTORC1 ) stimulates a coordinated anabolic program in response to growth-promoting signals . Paradoxically , recent studies indicate that mTORC1 can activate the transcription factor ATF4 through mechanisms distinct from its canonical induction by the integrated stress response ( ISR ) . However , its broader roles as a downstream target of mTORC1 are unknown . Therefore , we directly compared ATF4-dependent transcriptional changes induced upon insulin-stimulated mTORC1 signaling to those activated by the ISR . In multiple mouse embryo fibroblast and human cancer cell lines , the mTORC1-ATF4 pathway stimulated expression of only a subset of the ATF4 target genes induced by the ISR , including genes involved in amino acid uptake , synthesis , and tRNA charging . We demonstrate that ATF4 is a metabolic effector of mTORC1 involved in both its established role in promoting protein synthesis and in a previously unappreciated function for mTORC1 in stimulating cellular cystine uptake and glutathione synthesis . Pro-growth signals in the form of growth factors , hormones , and nutrients impinge on cellular metabolic programs in a coordinated fashion involving both acute , post-translational regulation and transcriptional control of nutrient transporters and metabolic enzymes . The mechanistic target of rapamycin complex 1 ( mTORC1 ) acts as a central point of integration for these signals and propagates a downstream metabolic response that increases anabolic processes while decreasing specific catabolic processes . Through a variety of downstream effectors , mTORC1 stimulates the synthesis of the major macromolecules comprising cellular biomass , including protein , lipid , and nucleic acids , along with metabolic and adaptive pathways that support this anabolic program ( Valvezan and Manning , 2019 ) . A particularly interesting feature of this coordinated metabolic program downstream of mTORC1 is the co-opting of key nutrient-sensing transcription factors that are established to be activated , independent of mTORC1 , in response to depletion of specific nutrients ( Torrence and Manning , 2018 ) . In their canonical roles , these transcription factors serve to mount an adaptive response by upregulating genes that allow cells to overcome the specific nutrient deficiency . Perhaps the best characterized of these transcription factors with dual regulation is the hypoxia-inducible factor 1 ( HIF1 ) comprising the labile HIF1α protein heterodimerized with the aryl hydrocarbon receptor nuclear translocator ( ARNT or HIF1β ) . Oxygen depletion ( i . e . , hypoxia ) results in the rapid stabilization of HIF1α and allows the HIF1 heterodimer to induce genes involved in glucose uptake , glycolysis , and angiogenesis to adapt to hypoxia and decrease mitochondrial respiration ( Nakazawa et al . , 2016 ) . On the other hand , in response to upstream growth factor signaling pathways , activation of mTORC1 stimulates an increase in HIF1α protein synthesis , leading to elevated expression of HIF1 gene targets ( Brugarolas et al . , 2003; Düvel et al . , 2010; Hudson et al . , 2002; Laughner et al . , 2001; Thomas et al . , 2006; Zhong et al . , 2000 ) . The result is an mTORC1-mediated increase in glucose uptake and glycolysis even when oxygen is not limiting ( e . g . , normoxia ) , a process referred to as aerobic glycolysis , which can support the production of biosynthetic precursors in the form of glycolytic intermediates . Similarly , the sterol regulatory element ( SRE ) -binding protein ( SREBP ) family of transcription factors are independently regulated by both adaptive nutrient signals and growth signals controlling mTORC1 . The SREBPs are canonically activated upon sterol depletion and induce expression of the enzymes required for de novo synthesis of fatty acid and sterol lipids ( Horton et al . , 2002 ) . However , insulin and growth factor signaling can also induce lipid synthesis via mTORC1-stimulated activation of SREBP and its lipogenic gene targets ( Düvel et al . , 2010; Owen et al . , 2012; Peterson et al . , 2011; Porstmann et al . , 2008 ) . Recent studies have suggested that the regulation of nutrient-sensing transcription factors by mTORC1 signaling extends to the activating transcription factor 4 ( ATF4 ) ( Adams , 2007; Ben-Sahra et al . , 2016; Torrence and Manning , 2018 ) . ATF4 is a basic leucine zipper ( bZIP ) transcription factor that is selectively translated in response to specific forms of cellular stress to induce the expression of genes involved in adaptation to stress ( Walter and Ron , 2011 ) . This adaptive program is referred to as the integrated stress response ( ISR ) and is initiated by stress-activated protein kinases , including general control nonderepressible 2 ( GCN2 ) activated upon amino acid deprivation and protein kinase RNA-like endoplasmic reticulum kinase ( PERK ) activated by ER stress , among others , which phosphorylate eIF2α on Ser51 ( Harding et al . , 2000 ) . Phosphorylation of eIF2α serves to globally attenuate mRNA translation to conserve amino acids and energy and decrease the cellular protein load as one adaptive measure to overcome these stresses ( Clemens , 1996 ) . Importantly , a small number of mRNAs , including that encoding ATF4 , exhibit increased translation upon eIF2α-Ser51 phosphorylation ( Vattem and Wek , 2004 ) . The stress-induced increase in ATF4 leads to the expression of a canonical set of ATF4 target genes , including those involved in nonessential amino acid ( NEAA ) biosynthesis and amino acid transport , as part of the adaptive cellular response specific to stresses such as amino acid depletion ( Harding et al . , 2003 ) . ATF4 functions in heterodimers with other bZIP transcription factors and also co-regulates many of its target genes with additional transcription factors as part of the cellular stress response ( Newman and Keating , 2003; Wortel et al . , 2017 ) . However , whether and how the many distinct upstream stresses that activate ATF4 influence its heterodimerization partners and the induction of specific sets of genes is not well understood . While ATF4 is a major downstream effector of the ISR , evidence has emerged that ATF4 can also be activated by pro-growth signals that stimulate mTORC1 signaling ( Adams , 2007; Ben-Sahra et al . , 2016 ) , and cis-regulatory elements for ATF4 binding are enriched in the promoters of mTORC1-induced genes ( Düvel et al . , 2010 ) . Importantly , the mTORC1-mediated activation of ATF4 involves its increased translation in a manner that is independent of the ISR and phosphorylation of eIF2α ( Ben-Sahra et al . , 2016; Park et al . , 2017 ) . These findings suggest that , similar to HIF1 and SREBP , ATF4 induction may be mobilized as part of the broader anabolic program downstream of mTORC1 . Indeed , our previous findings indicate that mTORC1 promotes de novo purine synthesis , in part , through induction of mitochondrial one-carbon metabolism via ATF4 activation and expression of its gene target MTHFD2 ( Ben-Sahra et al . , 2016 ) . How the ATF4-dependent gene program compares between its adaptive role in the ISR and its activation as a downstream effector of mTORC1 signaling and whether ATF4 contributes to established or new functions of mTORC1 are unknown ( Figure 1A ) . Here , we find that the mTORC1-ATF4 program represents a small subset of ATF4-dependent genes induced by ER stress and includes genes encoding the enzymes required for tRNA charging , NEAA synthesis , and amino acid uptake . Consistent with regulation of these enzymes by mTORC1 through ATF4 , ATF4 contributes to the induction of protein synthesis downstream of mTORC1 . We also find that mTORC1 signaling promotes glutathione synthesis through ATF4 and its specific regulation of the cystine transporter SLC7A11 . Thus , ATF4 is an anabolic effector of mTORC1 signaling , necessary for both its canonical regulation of protein synthesis and its induced synthesis of glutathione , the most abundant antioxidant in cells . To identify the ATF4-dependent gene targets downstream of mTORC1 , we compared the insulin-induced , rapamycin-sensitive transcripts between wild-type MEFs and those with biallelic loss of ATF4 via CRISPR/Cas9 gene deletion ( see Materials and methods ) . Consistent with our previous studies ( Ben-Sahra et al . , 2016 ) , insulin stimulated an increase in ATF4 protein in MEFs , which was decreased with rapamycin ( Figure 1B , Figure 1—figure supplement 1 ) . In parallel , these cells were treated with a time course of tunicamycin , an inhibitor of N-glycosylation that potently induces ER stress and an increase in ATF4 , to identify ATF4 gene targets downstream of the ISR . RNA-seq analysis revealed that 20% of transcripts ( 253 total ) significantly upregulated upon insulin stimulation were significantly blocked in their induction with rapamycin treatment . Approximately 30% of these mTORC1-regulated genes ( 77 total ) lost their insulin responsiveness with ATF4 deletion . In comparison , 36% of transcripts significantly induced with tunicamycin treatment at 4 hr were dependent on ATF4 ( 774 total ) . Importantly , the expression of just 61 genes was found to overlap between these two modes of ATF4 regulation , being ATF4 dependent in response to both mTORC1 activation and the ISR ( Figure 1C , Figure 1—source data 1 ) . The RNA-seq analysis demonstrated that only 8% of ATF4 gene targets induced by ER stress were also significantly stimulated by mTORC1 signaling ( e . g . , insulin induced and rapamycin sensitive ) . Interestingly , these 61 shared genes showed significant KEGG pathway enrichment for aminoacyl tRNA biosynthesis , amino acid metabolism , and one-carbon metabolism ( Figure 1D ) . The genes shared between the ISR and mTORC1 signaling were greatly enriched among those exhibiting the most significant increase upon tunicamycin treatment , with 75% ( 46 genes ) lying within the top 100 of the 774 tunicamycin-induced genes ( Figure 1E ) . It is worth noting that many of the top ATF4-dependent genes that scored as being induced by the ISR alone showed some degree of rapamycin-sensitive induction with insulin but did not reach statistical significance in the RNA-seq analyses . These data indicate that the subset of ATF4-dependent genes induced by mTORC1 signaling largely comprised those that are also most sensitive to ATF4 induction by the ISR . Among the 61 shared ATF4-dependent transcripts , those involved in amino acid synthesis and transport , one-carbon metabolism , and aminoacyl tRNA charging often displayed comparable fold changes between insulin stimulation and tunicamycin treatment , while canonical genes of the ER stress response , such as Herpud1 and Ddit3/Chop , showed much greater induction with tunicamycin ( Figure 1F ) . In addition to aminoacyl tRNA synthetase genes , expression of the Xpot gene encoding Exportin-T , which is the major Ran GTPase family member for nuclear to cytosolic export of mature tRNAs ( Arts et al . , 1998; Kutay et al . , 1998 ) , was found to be similarly regulated by ATF4 in response to mTORC1 and ISR activation . Transcripts encoding known negative regulators of mTORC1 signaling are also among these 61 shared ATF4-induced genes , including Ddit4/Redd1 and Sesn2 ( Brugarolas et al . , 2004; Condon et al . , 2021; Lee et al . , 2010; Reiling and Hafen , 2004; Wolfson et al . , 2016 ) . These targets likely contribute to the ATF4-dependent inhibition of mTORC1 signaling ( S6K1 phosphorylation ) observed upon tunicamycin treatment ( Figure 1B ) , while in the context of mTORC1 signaling , these ATF4 targets might play a role in negative feedback regulation of mTORC1 . These data suggest that a specific subset of ATF4-dependent ISR-induced genes are likewise regulated by growth factor signaling through mTORC1 and are enriched for specific processes including aminoacyl tRNA synthesis , amino acid synthesis and uptake , and one-carbon metabolism . ATF4 is known to form heterodimers with other bZIP transcription factors to engage its gene targets , while also co-regulating genes with transcription factors that bind additional promoter elements ( Kilberg et al . , 2009; Newman and Keating , 2003; Wortel et al . , 2017 ) . Thus , we used bioinformatic tools to determine whether the promoters of ATF4 gene targets shared between the ISR and mTORC1 signaling might be distinct from those induced by the ISR alone . Indeed , CiiiDER analysis ( Gearing et al . , 2019 ) revealed that there are predicted promoter-binding sequences that distinguish the 61 shared target genes from the top 200 ATF4-dependent genes induced by the ISR alone ( Figure 2A , Figure 2—source data 1 ) . Regulatory elements for the C/EBP family of transcription factors , which are well established to heterodimerize with ATF4 to induce its canonical downstream targets ( Cohen et al . , 2015; Ebert et al . , 2020; Huggins et al . , 2015 ) , were the most enriched in the promoters of ATF4 gene targets with shared regulation . On the other hand , binding elements for the TEAD family of transcription factors , which function with YAP/TAZ in the Hippo signaling pathway , were enriched in the promoters of ATF4-dependent gene targets significantly induced only by tunicamycin , consistent with published work indicating a functional connection between the unfolded protein response and YAP-TEAD activation ( Wu et al . , 2015; Xiao et al . , 2019 ) . We next analyzed the 61 ATF4-dependent genes with shared regulation for physical evidence of promoter binding of specific transcription factors using the Cistrome Data Browser , a portal for mining existing chromatin immunoprecipitation-DNA sequencing ( ChIP-seq ) data ( Mei et al . , 2017; Zheng et al . , 2019 ) . Importantly , this second unbiased analysis also revealed C/EBP isoforms as most commonly binding to the promoters of these genes ( Figure 2B ) . As all members of the C/EBP family have the potential to heterodimerize with ATF4 and contribute to the induction of these gene targets ( Newman and Keating , 2003 ) , we first determined the effects of siRNA-mediated knockdown of individual isoforms , relative to ATF4 knockdown , on expression of three representative genes in Tsc2-/- MEFs , which exhibit growth factor-independent activation of mTORC1 signaling . This analysis revealed that knockdown of ATF4 , C/EBPβ , C/EBPδ , or C/EBPγ each led to decreased transcript levels of the shared mTORC1 and ISR gene targets Slc7a5 , Mthfd2 , and Aars ( Figure 2C ) . However , this analysis was complicated by the finding of substantial co-dependence for expression among these bZIP transcription factors , with knockdown of any one of the C/EBP family members or ATF4 significantly changing expression of at least one other family member . C/EBPδ knockdown , for instance , decreased expression of all genes measured , including ATF4 , which was also reflected in loss of ATF4 protein ( Figure 2C , D ) . It is worth noting that we were unable to identify reliable antibodies to specific C/EBP family members for use in MEFs . Among C/EBP family members , C/EBPγ has been found in other settings to regulate many of the genes revealed in our analysis to be induced through shared regulation of ATF4 ( Huggins et al . , 2015 ) , and its knockdown significantly decreased expression of the three ATF4 target genes tested without effects on ATF4 protein levels ( Figure 2C , D ) . Based on this finding , we knocked down ATF4 or C/EBPγ in wild-type MEFs and stimulated the cells with insulin in the presence or absence of rapamycin to determine whether C/EBPγ impacted the mTORC1 and ATF4-dependent regulation of these genes . Indeed , knockdown of C/EBPγ attenuated the insulin-induced expression of these genes , albeit to a lesser extent than ATF4 knockdown ( Figure 2E ) . C/EBPγ knockdown also blocked the ability of insulin to increase ATF4 transcript levels , suggesting that following the induction of ATF4 mRNA translation downstream of mTORC1 ( Ben-Sahra et al . , 2016; Park et al . , 2017 ) , it stimulates its own expression via ATF4-C/EBPγ heterodimers . Thus , C/EBPγ and likely other ATF4-binding partners of the C/EBP family contribute to the induction of ATF4 gene targets following mTORC1-mediated activation of ATF4 . To validate and expand the findings from the RNA-seq analysis , a NanoString codeset was designed to simultaneously quantify transcripts of genes involved in the enriched processes above ( see Materials and methods ) . As positive and negative controls , respectively , we included the glycolytic targets of HIF1 , established previously to be regulated downstream of mTORC1 ( Düvel et al . , 2010 ) , and the mitochondrial tRNA synthetases , not believed to be regulated by ATF4 . Using this codeset , we analyzed gene expression in three settings of mTORC1 activation: ( 1 ) wild-type MEFs stimulated with insulin in the presence or absence of rapamycin , ( 2 ) growth factor-independent activation of mTORC1 via genetic loss of the TSC protein complex in Tsc2-/- MEFs , and ( 3 ) Tsc2-/- MEFs with siRNA-mediated knockdown of ATF4 ( Figure 3A , Figure 3—source data 1 ) . ATF4 protein levels were robustly upregulated with either genetic or insulin-stimulated mTORC1 activation in these settings , with both rapamycin and ATF4-targeting siRNAs blocking this induction ( Figure 3—figure supplement 1A ) . Interestingly , the majority of transcripts analyzed in the functional categories that encode enzymes of NEAA synthesis , one-carbon metabolism , amino acid transporters , and cytosolic aminoacyl tRNA synthetases ( and Xpot ) were increased with mTORC1 activation in a manner sensitive to both rapamycin and siRNA-mediated knockdown of ATF4 . However , the HIF1 targets of glycolysis were mTORC1-regulated but independent of ATF4 , and transcripts encoding the mitochondrial tRNA synthetases were reproducibly regulated by neither mTORC1- nor ATF4 . We further confirmed the mTORC1- and ATF4-mediated regulation of a representative subset of these transcripts via qPCR ( Figure 3B , C , Figure 3—figure supplement 1B ) . Consistent with previous studies ( Ben-Sahra et al . , 2016; Park et al . , 2017 ) , both ATF4 transcript and protein levels were induced by mTORC1 signaling in these settings ( Figure 3B–E ) . Transcriptional changes in ATF4 gene targets were reflected in corresponding changes in the abundance of representative protein products , with varying degrees of rapamycin sensitivity , likely reflecting inherent differences in the turnover rates of these proteins ( Figure 3D , E , Figure 3—figure supplement 1C , D ) . We next wanted to confirm that these mTORC1- and ATF4-induced changes were independent of the ISR . While we have shown previously that mTORC1 regulates ATF4 in a manner that is independent of eIF2α-S51 phosphorylation ( Ben-Sahra et al . , 2016 ) , chronic activation of mTORC1 upon loss of TSC2 is known to cause a basal increase in ER stress and activation of the ISR ( Ozcan et al . , 2008 ) . Therefore , we utilized MEFs with endogenous , homozygous knock-in of the Eif2a-S51A mutation ( Eif2aA/A ) , which fail to induce ATF4 downstream of cellular stress ( Figure 3—figure supplement 1E; Scheuner et al . , 2001 ) . Consistent with mTORC1-dependent , ISR-independent regulation , insulin increased ATF4 protein levels and expression of its gene targets involved in amino acid synthesis and one-carbon metabolism , amino acid transporters , and aminoacyl tRNA synthetases in a rapamycin-sensitive manner in these cells , as shown by NanoString analysis and confirmed for a subset of genes by qPCR ( Figure 3F , G , Figure 3—figure supplement 1F ) . Notably , like insulin stimulation , genetic activation of mTORC1 via siRNA-mediated knockdown of Tsc2 in the eIF2αA/A MEFs also increased ATF4 protein levels , further confirming that this regulation can occur independent of the ISR ( Figure 3H ) . Growth factor-independent activation of mTORC1 also occurs upon loss of the PTEN tumor suppressor , and rapamycin was found to decrease ATF4 protein levels and ATF4-dependent expression of representative gene targets in established PTEN-deficient prostate cancer cells , LNCaP and PC3 ( Figure 3I–L , Figure 3—figure supplement 1G–J ) . The transcription factor c-MYC can be activated downstream of mTORC1 in some settings and has previously been shown to regulate many of the same ATF4 target genes encoding the enzymes of amino acid synthesis and transport , as well as aminoacyl tRNA synthetases ( Csibi et al . , 2014; Stine et al . , 2015; Wall et al . , 2008; Zirin et al . , 2019 ) . A survey of expression for 11 of such genes in Tsc2-/- cells finds that the majority are not significantly affected by siRNA-mediated knockdown of c-MYC , whereas a few others are modestly but significantly decreased ( e . g . , Gars , Slc1a5 , Psat1 ) , albeit to a lesser degree than with siRNA-mediated knockdown of ATF4 ( Figure 3—figure supplement 1K , L ) . To determine whether ATF4 activation is both necessary and sufficient for mTORC1-mediated regulation of these gene targets related to amino acid acquisition and utilization , we knocked out Atf4 using CRISPR/Cas9 in Tsc2-/- MEFs and confirmed biallelic disruption ( Figure 4A ) . Protein levels of identified ATF4 targets were decreased in Tsc2-/- Atf4-/- MEFs and fully rescued with expression of wild-type ATF4 but not a DNA-binding domain mutant ( ATF4DBD ) ( Figure 4B ) . As mTORC1 regulates ATF4 translation through a mechanism requiring its 5′-UTR ( Ben-Sahra et al . , 2016; Park et al . , 2017 ) , the stably rescued cell lines , which express an ATF4 cDNA lacking the 5′-UTR , exhibit protein expression of ATF4 and its encoded gene targets that are resistant to rapamycin ( Figure 4C , Figure 4—figure supplement 1 ) . The expression of select ATF4 target transcripts was markedly decreased in Tsc2-/- Atf4-/- MEFs to a similar extent to that measured in Atf4 wild-type cells treated with rapamycin ( Figure 4D ) . These transcript levels were rescued with the expression of wild-type ATF4 , but not ATF4DBD , in a manner that was completely or partially rapamycin resistant ( Figure 4D ) . These collective data show that mTORC1 signaling drives the expression of genes involved in tRNA export and charging , amino acid uptake , and NEAA synthesis through its downstream regulation of the ATF4 transcription factor . mTORC1 induces protein synthesis through multiple downstream targets ( Valvezan and Manning , 2019 ) . Given that the major mTORC1-regulated ATF4 target genes identified above are involved in amino acid uptake , synthesis , and tRNA charging , we hypothesized that ATF4 induction through mTORC1 signaling might contribute to the canonical increase in protein synthesis upon mTORC1 activation . Relative rates of protein synthesis were measured via [35S]-methionine incorporation into newly synthesized proteins in Tsc2+/+ and Tsc2-/- cells treated with control siRNAs or Tsc2-/- cells treated with siRNAs targeting ATF4 or Rheb , the small GTPase target of TSC2 that is an essential upstream activator of mTORC1 . siRNA-mediated knockdown of either ATF4 or Rheb substantially decreased ATF4 protein levels in Tsc2-/- cells ( Figure 5A ) . Importantly , the elevated rate of protein synthesis in Tsc2-/- MEFs was decreased with ATF4 knockdown to a similar extent to that observed with Rheb knockdown ( Figure 5A , B ) . No change in mTORC1 signaling or phosphorylation of eIF2α was observed with ATF4 knockdown in this setting ( Figure 5—figure supplement 1A ) . Protein synthesis was also measured in the Tsc2-/- Atf4-/- cell lines described above . Notably , the cells lacking ATF4 exhibited increased uptake of [35S]-methionine relative to parental cells and those reconstituted with ATF4 ( Figure 5—figure supplement 1B ) . Despite this unexplained difference in methionine uptake , Atf4 knockout cells exhibited a reduced rate of protein synthesis that was similar to the parental lines treated with rapamycin ( Figure 5C , D ) . However , rapamycin treatment further reduced protein synthesis in the Atf4 knockout cells . Importantly , cells reconstituted with the rapamycin-resistant ATF4 cDNA exhibited rescued protein synthesis , surpassing that observed in cells with endogenous ATF4 , but this enhanced protein synthesis was still significantly reduced with rapamycin . Like Tsc2-/- cells , wild-type cells cultured in the presence of growth factors exhibited reduced protein synthesis upon deletion of Atf4 , again with a reduction similar to that from rapamycin treatment ( Figure 5—figure supplement 1C , D ) . Together , these data suggest that ATF4 induction downstream of mTORC1 is necessary but not sufficient for mTORC1-regulated protein synthesis , consistent with the multiple mechanisms through which mTORC1 controls this key anabolic process . Among the 61 shared mTORC1- and ISR-induced ATF4 gene targets identified , the cystine-glutamate antiporter Slc7a11 was the gene with the highest fold induction by RNA-seq analysis upon insulin treatment ( Figure 1F ) . SLC7A11 ( also known as xCT ) associates with SLC3A2 ( CD98 ) at the plasma membrane and serves as the primary transporter of cystine , the oxidized form of cysteine and predominant cysteine species in both plasma and cell culture media , whereas reduced cysteine is transported through neutral amino acid systems ( Figure 6A; Bannai and Kitamura , 1980; Conrad and Sato , 2012 ) . Transcript levels of Slc7a11 were sensitive to rapamycin in Tsc2-/- MEFs and greatly decreased with ATF4 knockout ( Figure 6B ) . Upon reconstitution with rapamycin-resistant ATF4 , expression of Slc7a11 was rescued and no longer sensitive to rapamycin , while the ATF4DBD mutant was unable to restore Slc7a11 transcript levels . A similar pattern of mTORC1- and ATF4-regulated expression was measured for Slc3a2 ( Figure 6C ) . SLC7A11 protein , detected using an antibody validated with siRNA knockdown ( Figure 6—figure supplement 1A ) , decreased in Tsc2-/- MEFs treated with mTOR inhibitors and were increased in wild-type MEFs stimulated with insulin in an mTOR-dependent manner ( Figure 6D , E , Figure 6—figure supplement 1B , C ) . SLC7A11 transcript levels were also decreased with both ATF4 knockdown and rapamycin in the PTEN-deficient cancer cell lines LNCaP and PC3 , although SLC7A11 expression was relatively more resistant to rapamycin in PC3 cells ( Figure 6—figure supplement 1D , E ) . SLC7A11 protein levels likewise decreased in LNCaP and PC3 cells treated with mTOR inhibitors , without significant changes to the SLC3A2 gene product CD98 ( Figure 6F , Figure 6—figure supplement 1F–H ) . These data confirm and extend the findings from RNA-seq and NanoString analyses ( Figure 1F , Figure 3A , F ) and demonstrate that ATF4 is both necessary and sufficient for the mTORC1-mediated induction of Slc7a11 expression . ATF4 is known to be important for the uptake and synthesis of NEAAs ( Harding et al . , 2003 ) . In agreement with this , we observed that Tsc2-/- Atf4-/- cells fail to proliferate in Dulbecco’s Modified Eagle’s Medium ( DMEM ) , which only contains a subset of NEAAs , while addback of wild-type ATF4 , but not ATF4DBD , restored proliferation ( Figure 6G ) . Supplementation of DMEM with a mixture of all NEAAs , including cysteine , allowed the Tsc2-/- Atf4-/- MEFs to proliferate at the same rate as the ATF4-reconstituted cells , while NEAAs lacking cysteine completely failed to support proliferation of these cells ( Figure 6H ) . Furthermore , supplementation with excess reduced cysteine alone , but not equimolar concentrations of oxidized cysteine in the form of cystine , was able to significantly increase proliferation of the Tsc2-/- Atf4-/- MEFs , albeit to a lesser extent than NEAAs plus cysteine . The majority of these cells die after 72 hr in DMEM , and exogenous expression of either ATF4 or SLC7A11 restores their survival ( Figure 6I ) . Taken together , these data indicate that a defect in the acquisition of cysteine , which normally occurs through SLC7A11-mediated uptake of cystine , underlies the inability of Tsc2-/- Atf4-/- cells to proliferate or survive in DMEM and suggest a key role for mTORC1 signaling in controlling cystine uptake through ATF4 . To directly test whether mTORC1 influences cystine uptake , we employed both genetic ( Tsc2 loss ) and physiological ( insulin stimulation ) activation of mTORC1 , measuring [14C]-cystine uptake in the presence or absence of rapamycin or the xCT inhibitor erastin ( Figure 6A; Yang and Stockwell , 2008 ) . Both rapamycin and erastin significantly decreased [14C]-cystine uptake into Tsc2-/- MEFs ( Figure 6J ) . In wild-type MEFs , insulin stimulated an increase in cystine uptake that was inhibited with rapamycin , and this mTORC1-regulated cystine transport was completely lost with ATF4 knockout , reduced to levels of erastin-treated cells ( Figure 6K ) . Additionally , Tsc2-/- Atf4-/- MEFs showed a decrease in cystine uptake when compared to parental Tsc2-/- MEFs , which could be rescued with re-expression of ATF4 ( Figure 6L ) . However , cystine consumption in cells reconstituted with rapamycin-resistant ATF4 was still significantly sensitive to rapamycin treatment , suggesting the existence of additional , ATF4-independent mechanisms influencing the transport or cellular incorporation of cystine downstream of mTORC1 . As mTORC2 has been previously suggested to directly regulate xCT ( Gu et al . , 2017 ) , we utilized Rictor-/- MEFs , which lack mTORC2 activity , to determine whether mTORC2 was contributing to the decreased cystine uptake observed upon treatment with mTOR inhibitors . While Rictor-/- MEFs displayed increased uptake of cystine relative to their wild-type counterparts , cystine uptake , as well as ATF4 protein levels , was sensitive to rapamycin and Torin1 in both cell lines ( Figure 6—figure supplement 1I , J ) . Thus , mTORC1 promotes cellular cystine uptake , at least in part , through the activation of ATF4 and induction of Slc7a11 expression , which supports cell proliferation and survival . Cysteine , generally acquired through cystine uptake and reduction , is an essential component of the tripeptide glutathione ( Figure 6A ) , the most abundant antioxidant in cells ( Meister , 1983 ) . We hypothesized that the regulation of cystine uptake through mTORC1 and ATF4 might influence cellular glutathione content . Indeed , mTORC1 inhibition with rapamycin or Torin1 significantly decreased total glutathione levels in Tsc2-/- MEFs , albeit less than buthionine-sulfoximine ( BSO ) , a direct inhibitor of glutathione synthesis ( Figure 7A; Griffith and Meister , 1979 ) . Similar to BSO treatment , mTOR inhibitors decreased both reduced ( GSH ) and oxidized ( GSSG ) forms of glutathione to the same degree , indicating effects on total glutathione abundance rather than its redox state ( Figure 7—figure supplement 1A ) . Stable reconstitution of Tsc2-/- MEFs with TSC2 also decreased total glutathione levels ( Figure 7B ) . To examine this response in vivo , we employed a mouse model of tuberous sclerosis complex involving xenograft tumors derived from the rat TSC2-/- tumor cell line ELT3 ( Hodges et al . , 2002 ) . To avoid major differences in tumor size from the treatments , we treated tumor-bearing mice for just 5 days with either vehicle or rapamycin , prior to harvesting tumors for immunoblot analysis and metabolite profiling . Importantly , we found that rapamycin treatment strongly decreased ATF4 protein levels in these tumors with a concomitant decrease in glutathione levels , measured by LC-MS in tumor metabolite extracts ( Figure 7C , D ) . An analysis of published metabolomics data ( Tang et al . , 2019 ) also revealed that rapamycin treatment significantly decreased glutathione levels in human TSC2-deficient angiomyolipoma cells ( Figure 7—figure supplement 1B ) . Likewise , inhibition of mTORC1 signaling with rapamycin or Torin1 in LNCaP and PC3 cells resulted in a significant decrease in total glutathione levels , although the degree of decrease varied between the two cell lines ( Figure 7E ) , perhaps reflecting the above finding that SLC7A11 expression is more resistant to mTOR inhibitors in PC3 cells ( Figure 6—figure supplement 1D–H ) . To determine the role of ATF4 and SLC7A11-dependent cystine uptake in glutathione synthesis downstream of mTORC1 signaling , we compared Tsc2-/- MEFs with or without Atf4 knockout . Total glutathione levels were greatly decreased in Tsc2-/- Atf4-/- MEFs , and exogenous expression of ATF4 or SLC7A11 , but not ATF4DBD , was able to restore glutathione levels to these cells ( Figure 7F ) . Supplementation with all NEAAs or just cysteine , transported through neutral amino acid systems , but not equimolar concentrations of cystine , transported through SLC7A11 , also rescued total glutathione levels , as measured by either enzymatic assay or LC-MS ( Figure 7F , Figure 7—figure supplement 1C ) . Furthermore , insulin stimulated an increase in glutathione levels in wild-type MEFs in a manner completely sensitive to rapamycin , an effect ablated in Atf4 knockout cells , which had a very low abundance of glutathione ( Figure 7G ) . Likewise , the rapamycin-sensitive nature of glutathione in Tsc2-/- MEFs was completely lost in Tsc2-/- Atf4-/- MEFs ( Figure 7H ) . Glutathione levels were restored to these cells upon exogenous expression of ATF4 , but not the ATF4DBD mutant . However , glutathione was still significantly reduced with rapamycin treatment in cells expressing the rapamycin-resistant ATF4 , suggesting possible ATF4-independent mechanisms also contributing to this regulation ( Lam et al . , 2017 ) . As one possible contributing factor , we found that the transcript encoding both the catalytic ( GCLC ) and regulatory ( GCLM ) subunits of glutamate-cysteine ligase , the first enzyme of glutathione synthesis , was sensitive rapamycin , in a manner unaffected by ATF4 knockdown ( Figure 7—figure supplement 1D , E ) . We also found that GCLC and GCLM protein levels could be modestly induced by insulin through mTORC1 signaling in both wild-type and ATF4 knockout cells , but their protein abundance was unaffected by rapamycin in Tsc2-/- MEFs ( Figure 7—figure supplement 1F , G ) . Thus , the mechanism underlying the apparent ATF4-independent effects of mTORC1 signaling on glutathione levels remains unknown . These collective data show that mTORC1 signaling induces glutathione synthesis , at least in part , through the activation of ATF4 and SLC7A11-dependent cystine uptake . Our findings expand the functional repertoire of mTORC1 signaling as it relates to the control of anabolic processes and cellular metabolism through its noncanonical activation of ATF4 . Importantly , less than 10% of stress-responsive , ATF4-dependent targets were found to be significantly stimulated through the mTORC1-mediated activation of ATF4 in response to insulin . Among others , we found that genes involved in amino acid biosynthesis , transport , and tRNA charging were induced by mTORC1-ATF4 signaling , many to a comparable level to that of ER-stress induction with tunicamycin . While the molecular nature of this selective induction remains unknown , our data suggest that the 61 ATF4-dependent genes shared in their induction between mTORC1 signaling and the ISR represent targets most highly responsive to increases in ATF4 levels . Since mTORC1 signaling leads to a more modest increase in ATF4 protein levels than does the ISR , the selective induction of these genes might be reminiscent of the dose-dependent activation of MYC target genes documented in other studies ( Sabò et al . , 2014; Schuhmacher and Eick , 2013; Walz et al . , 2014 ) . Our bioinformatic analyses and functional data also indicate involvement of the C/EBP family of transcription factors as heterodimerization partners of ATF4 for the regulation of these gene targets shared between mTORC1 signaling and the ISR . Consistent with the specific ATF4 target genes induced by mTORC1 signaling , including those involved in amino acid acquisition and tRNA charging , we found that ATF4 activation contributes to both canonical ( e . g . , protein synthesis ) and new ( e . g . , glutathione synthesis ) functions of mTORC1 . As mTORC1 stimulates protein synthesis through multiple downstream targets ( Holz et al . , 2005; Jefferies et al . , 1994; Ma and Blenis , 2009; Raught et al . , 2004; Thoreen et al . , 2012 ) , it was not surprising to find that ATF4 was necessary but not sufficient for the increased rate of protein synthesis accompanying mTORC1 activation . We also demonstrate that mTORC1 signaling regulates the abundance of total cellular glutathione , both reduced and oxidized , at least in part through the ATF4-dependent induction of the cystine transporter SLC7A11 , a major source of the cysteine that is limiting for glutathione synthesis . Importantly , this mTORC1-ATF4-mediated transcriptional upregulation of SLC7A11 leading to increased cystine uptake would temporally follow the inhibition of SLC7A11 recently found to be mediated through mTORC2 and Akt-dependent transient phosphorylation of the transporter ( Gu et al . , 2017; Lien et al . , 2017 ) . Our findings are consistent with a recent study indicating rapamycin-sensitive expression of xCT in TSC models , which the authors attribute to the OCT1 transcription factor ( Li et al . , 2019 ) . However , our study indicates that the mTORC1-mediated activation of ATF4 is both necessary and sufficient for this regulation . The transcription factor NRF2 ( also known as NFE2L2 ) is activated by oxidative stress and is a master regulator of the enzymes required for glutathione synthesis , as well as SLC7A11 to increase cystine uptake ( Habib et al . , 2015; Sasaki et al . , 2002; Ye et al . , 2014 ) . While NRF2 depletion has been described to decrease the viability of cells with TSC gene loss ( Zarei et al . , 2019 ) , we have no evidence from this or previous studies that mTORC1 signaling influences the levels or activity of NRF2 ( Zhang et al . , 2014 ) . mTORC1 serves to couple growth signals to the coordinated control of anabolic processes , including the biosynthesis of protein , lipids , and nucleotides , as well as metabolic pathways that support this anabolic state ( Valvezan and Manning , 2019 ) . This metabolic program is orchestrated to provide biosynthetic precursors and directly promote the synthesis of macromolecules while also maintaining cellular homeostasis and preventing nutrient or metabolic stress . For example , mTORC1 signaling promotes metabolic flux through the NADPH-producing oxidative branch of the pentose phosphate pathway , thereby providing the reducing power essential to support an mTORC1-stimulated increase in de novo lipid synthesis ( Düvel et al . , 2010 ) . Importantly , NADPH is also essential to reduce cystine , taken up through SLC7A11 , into two molecules of cysteine for use in glutathione synthesis , in addition to being required to regenerate reduced glutathione following its oxidation . Supporting this logic of a coordinated metabolic program downstream of mTORC1 , pro-growth signaling through mTORC1 likely promotes glutathione synthesis to help buffer against the oxidative stress that accompanies anabolic metabolism and increased rates of protein synthesis ( Han et al . , 2013; Harding et al . , 2003; Kong and Chandel , 2018 ) . Our findings further support the addition of ATF4 to SREBP and HIF1 , as nutrient- and stress-sensing transcription factors that are independently co-opted by mTORC1 signaling to drive the expression of metabolic enzymes and nutrient transporters . Unlike adaptive signals stemming from the depletion of individual nutrients , such as amino acids , sterols , or oxygen , which generally attenuate mTORC1 signaling as part of the adaptive response , pro-growth signals that activate mTORC1 can stimulate these transcription factors in concert to support a broader anabolic program . It will be important in future studies to understand the dual regulation of these transcription factors by both pro-growth and adaptive mechanisms as it relates to settings of physiological ( fasting and feeding ) and pathological ( tumor development ) nutrient fluctuations . MEFs and PC3 cells were maintained in DMEM ( Corning/Cellgro , 10-017-CV ) with 10% fetal bovine serum ( FBS , Corning/Gibco ) . LNCaP cells were maintained in RPMI-1640 ( Corning/Cellgro 10-040-CV ) with 10% FBS . Tsc2-/- ( p53-/- ) MEFs and littermate-derived wild-type counterparts were provided by David Kwiatkowski ( Brigham and Women’s Hospital , Boston , MA ) . eIF2αS/S ( WT ) and eIF2αA/A ( S51A knock-in mutant ) MEFs were provided by Randal Kaufman ( Sanford-Burnham-Prebys Medical Discovery Institute , La Jolla , CA ) and were not used above passage 3 ( after received ) . Rictor+/+ and Rictor-/- MEFs were provided by D . A . Guertin and D . M . Sabatini ( Whitehead Institute , Massachusetts Institute of Technology , Cambridge , MA ) . Cancer cell lines were obtained from ATCC . Atf4-/- MEF lines generated in this study were maintained in DMEM with 10% FBS , supplemented with 55 µM 2-mercaptoethanol ( Thermo , 21985023 ) , and 1X MEM NEAA mix ( NEAA , final concentrations: 100 µM each of alanine , aspartate , asparagine , glutamate , glycine , proline , and serine; Thermo 11140050 ) . In experiments with supplementation of excess cysteine , cells were plated in DMEM with 10% FBS , 1 mM cysteine , and , where indicated , 1X MEM NEAA mix . Cells were transfected with 20 nM of the indicated siRNAs using Opti-MEM ( Thermo , 31985062 ) and RNAimax ( Thermo , 13778150 ) according to the manufacturer’s protocol . siRNAs were from GE Life Sciences/Dharmacon: non-targeting pool ( D-001810-10-50 ) , Myc ( L-040813-00-0010 ) , Atf4 ( mouse , L-042737-01-0020 ) , Rheb ( L-057044-00-0020 ) , RhebL1 ( L-056074-01-0020 ) , Tsc2 ( L-047050-00-0020 ) , C/ebpα ( L-040561-00-0005 ) , C/ebpβ ( L-043110-00-0005 ) , C/ebpδ ( L-060294-01-0005 ) , C/ebpγ ( L-065627-00-0005 ) , and ATF4 ( human , L-005125-00-0020 ) . Forty-eight hours post transfection , cells were treated as indicated prior to lysis for immunoblotting , RNA extraction , or protein synthesis assays . For C/EBP isoform knockdown experiments , transfection of siRNAs was performed a second time , 24 hr after the first transfection . For protein synthesis assays involving siRNAs , transfection was also performed a second time , 24 hr after the first transfection , which was necessary to achieve sufficient knockdown of Rheb to reduce mTORC1 signaling . Tsc2+/+ ( WT ) and Tsc2-/- MEFs lacking Atf4 were generated by CRISPR-Cas9-mediated deletion using pSpCas9n ( BB ) −2A-GFP ( PX461 ) vector ( Addgene , 48140 ) according to the previously described protocol ( Ran et al . , 2013 ) . The paired nickase guides were designed using E-CRISP ( Heigwer et al . , 2014 ) and targeted the sequences AGCATAGCCCCTCCACCTCC and GACAATCTGCCTTCTCCAGG in exon 2 of ATF4 . Forty-eight hours post transfection , single GFP-positive cells were sorted into 96-well plates . Cells were cultured in DMEM with 10% FBS supplemented with 1X MEM NEAA and 55 µM 2-mercaptoethanol . Single cell clones were grown for immunoblot analysis , and those showing loss of ATF4 protein were selected for sequence analysis involving the isolation of genomic DNA ( Qiagen , 69504 ) , PCR amplification using KOD Xtreme Hot Start DNA Polymerase ( Millipore , 71975 ) , and the primers TCGATGCTCTGTTTCGAATG and CTTCTTCCCCCTTGCCTTAC flanking the targeted deletion site , with sequencing on an ABI3730xl DNA analyzer at the DNA Resource Core of Dana-Farber/Harvard Cancer Center ( funded in part by NCI Cancer Center support grant 2P30CA006516-48 ) . The mutations were identified using CRISP-ID software ( Dehairs et al . , 2016 ) . For the final clones selected , the mutations generated in the WT MEFs include an out-of-frame 17-bp deletion starting at the codon encoding T237 and a large out-of-frame 73-bp deletion starting after the codon encoding G219 , both resulting in premature STOP codons . The mutations generated in the Tsc2-/- MEFs include an out-of-frame 245-bp deletion starting at the codon encoding G190 , resulting in a premature STOP codon after the D191 codon , and an in-frame deletion removing the sequences encoded between E210 and E284 . For generation of Atf4 expression vectors , the murine Atf4 cDNA was amplified from the plasmid 21845 from Addgene ( Harding et al . , 2000 ) . Restriction enzyme cloning with AgeI and ClaI was used to insert the Atf4-coding sequence ( lacking the 5′ and 3′-UTR ) into the pTRIPZ plasmid for doxycycline-inducible expression . The ATF4DBD mutant , in which amino acids 292–298 of the DNA-binding domain are changed from RYRQKKR to GYLEAAA ( He et al . , 2001; Lange et al . , 2008 ) , was generated by DpnI-mediated site-directed mutagenesis using KOD Xtreme Hot Start DNA Polymerase . The dox-inducible pTRIPZ and SLC7A11 plasmids were a gift from Alex Toker ( Beth Israel Deaconess Medical Center , Boston , MA ) . cDNA expression was induced with 1 μg/mL of doxycycline ( Sigma-Aldrich , D3447 ) for 12–24 hr before assays were conducted . GFP was inserted into pTRIPZ to produce the control vector . Lentivirus was generated in HEK293T cells transfected with pMD2 . G and psPAX2 ( Addgene , 12259 and 12260 ) and the given pTRIPZ constructs . Forty-eight hours post transfection , the virus-containing medium was used to infect the Atf4 knockout cells , which were selected with 8 μg/mL puromycin . TSC2 addback cell lines were generated by retroviral infection following transfection of PT67 cells with pBabe hygro IRES-EV or pBabe hygro IRES-TSC2 . Cells were selected with 400 μg/mL hygromycin B ( Thermo , 10687010 ) . Wild-type and Atf4-/- MEFs were grown to 70% confluence in 6 cm plates and were serum-starved in the presence of 2-mercaptoethanol and 1X MEM NEAA mixture and treated with vehicle ( DMSO ) or 20 nM rapamycin ( LC Laboratories , R5000 ) for 30 min prior to stimulation with vehicle ( water ) or 500 nM insulin ( Alpha Diagnostic , INSL 16 N-5 ) for 16 hr or treated with vehicle ( DMSO ) or 2 μg/mL tunicamycin ( Sigma-Aldrich , T7765 ) for 4 , 8 , or 16 hr . RNA was harvested with TRIzol according to the manufacturer’s protocol ( Thermo , 15596018 ) . All samples passed RNA quality control measured by NanoDrop 1000 Spectrophotometer ( NanoDrop Technologies ) and 2100 Bioanalyzer ( Agilent Technologies ) . cDNA libraries were generated to produce 150-bp paired-end reads on an Illumina NovaSeq with read depth of 20 million paired reads per sample . Reads were aligned and annotated to the Ensembl Mus musculus GRCm38 . p6 genome assembly using the align and featureCounts functions from the Rsubread package ( 2 . 0 . 1 ) in R ( 3 . 6 . 3 ) ( Liao et al . , 2019 ) . Differential gene expression analysis was performed using the voom and eBayes functions from the EgdeR ( 3 . 28 . 1 ) and Limma ( 3 . 42 . 2 ) packages , respectively ( Ritchie et al . , 2015; Robinson et al . , 2010 ) . Transcripts found to be significantly induced by tunicamycin were further limited to those with a greater than 1 . 2-fold increase . The enrichKEGG function from the clusterProfiler package ( 3 . 14 . 3 ) was used to perform KEGG pathway over-representation tests ( Yu et al . , 2012 ) . Gene set enrichment analysis was evaluated using GSEA software from the Broad Institute ( Subramanian et al . , 2005 ) . Computations were run on the FASRC Cannon cluster , supported by the Faculty of Arts and Sciences Division of Science Research Computing Group at Harvard University . Pseudogenes and unannotated genes were excluded from Figure 1C and Figure 1—source data 1 . The complete RNA-seq data can be found at GEO under the accession number GSE158605 . CiiiDER software was downloaded from CiiiDER . org with the M . musculus GRCm38 . 94 genome files . Searches were run against the JASPAR transcription factor-binding profile database . Searches were run on promoter regions spanning +1500 to −500 bp from the predicted transcriptional start site using a site identification deficit threshold of 0 . 1 . The background gene list ( ISR only ) comprised the 200 ATF4-dependent genes most significantly increased in expression upon tunicamycin treatment that were not in the list of 61 genes shared in their regulation by mTORC1 signaling and the ISR . Results of this analysis are included in Figure 2—source data 1 . Each of the 61 shared mTORC1 and ISR genes was analyzed using the CistromeDB Toolkit ( http://dbtoolkit . cistrome . org/ ) of existing genome-wide ChIP-seq data . A half-decay distance of 1 kb to the transcription start site was used . The top 20 transcription factors or chromatin regulators found in ChIP-seq experiments to bind to each gene were compiled , and the number of genes each factor bound to within the list of 61 was determined , with the top 9 regulators graphed , excluding the general factors EP300 and POL2RA . Cells were lysed in ice-cold Triton lysis buffer ( 40 mM HEPES pH 7 . 4 , 120 mM NaCl , 1 mM EDTA , 1% Triton X-100 , 10 mM sodium pyrophosphate , 10 mM glycerol 2-phosphate , 50 mM NaF , 0 . 5 mM sodium orthovanadate , 1 μM Microcystin-LR , and Sigma protease inhibitor cocktail P8340 ) . For immunoblots on SLC7A11 , cells were lysed in 1% SDS lysis buffer ( 10 mM Tris pH 7 . 4 , 1 mM EDTA , 10 mM sodium pyrophosphate , 10 mM glycerol 2-phosphate , 50 mM NaF , 0 . 5 mM sodium orthovanadate , and Sigma protease inhibitor cocktail P8340 ) . Samples were centrifuged at 20 , 000 × g for 10 min at 4°C , and protein concentration in the supernatant was determined by Bradford assay ( Bio-Rad , 5000202 ) and normalized across samples . Proteins were separated by SDS-PAGE , transferred to nitrocellulose membranes , and immunoblotted with indicated antibody . Primary antibodies used were MTHFD2 ( Protein Tech , 12270-1-AP ) , PSAT1 ( Protein Tech 20180-1-AP ) , phospho ( P ) -S6K1 T389 ( Cell Signaling Technologies [CST] , 9234 ) , ATF4 ( CST , 11815 ) , eIF2α ( CST , 9722 ) , P-eIF2α S51 ( CST , 9721 ) , S6K1 ( CST , 2708 ) , CD98 ( CST , 13180 ) , AARS ( Bethyl Antibodies , A303-475A-M ) , GARS ( Bethyl Antibodies , A304-746A-M ) , LARS ( Bethyl Antibodies , A304-316A-M ) , XPOT ( Aviva Biotechnologies , ARP40711_P050 ) , TSC2 ( CST , 4308 ) , SLC7A11 ( mouse , Abcam , ab175186 ) , SLC7A11 ( human , CST , 12691 ) , GCLC ( Abcam , ab190685 ) , GCLM ( Abcam , ab126704 ) , RICTOR ( CST , 9476 ) , RHEB ( CST , 13879 ) , 4EBP1 ( CST 9644 ) , c-MYC ( CST , 9402 ) , and β-actin ( Sigma-Aldrich , A5316 ) ; secondary antibodies used were IRDye 800CW Donkey anti-Mouse IgG ( H + L ) ( LI-COR , 926-32212 ) and Donkey anti-Rabbit IgG ( H + L ) ( LI-COR , 925–32213 ) , and HRP-conjugated anti-mouse and anti-rabbit secondary antibodies from ( CST , 7074 and 7076 ) . Immunoblots of MTHFD2 , PSAT1 , AARS , GARS , LARS , and XPOT were imaged using Odyssey CLx Imaging System ( LI-COR Biosciences ) . β-Actin was developed with Odyssey CLx Imaging System or enhanced chemiluminescence assay ( ECL ) . All remaining immunoblots were developed using ECL . Immunoblots were quantified using the Odyssey CLx Imaging System and were normalized to β-actin . The ATF4 immunoblot corresponding to Figure 1B ( Figure 1—figure supplement 1A ) and the SLC7A11 immunoblots in Figure 6D , E ( Figure 6—figure supplement 1B , C ) was quantified using ImageJ and normalized to β-actin . RNA was harvested using TRIzol from cells at 70% confluence . All samples passed RNA quality control measured by NanoDrop 1000 Spectrophotometer . Parallel plates were lysed for immunoblots . The isolated RNA was analyzed using a custom NanoString probe library according to the manufacturer’s instructions ( NanoString Technologies ) . Briefly , sample RNA was hybridized to RNA-probes at 65°C for 16 hr , excess probe was washed away , and an nCounter SPRINT was used to quantify specific mRNA molecules present in each sample . Direct mRNA counts were normalized to internal control genes , and mRNA expression was analyzed using nSOLVER software . Heatmaps were generated with Morpheus software from the Broad Institute ( https://software . broadinstitute . org/morpheus ) . Transcripts with 100 counts or fewer , a value based off of negative control samples , were not included in our analyses . Results of this analysis are included in Figure 3—source data 1 . For gene expression analysis , RNA was isolated with TRIzol according to the manufacturer’s protocol . All samples passed RNA quality control measured by NanoDrop 1000 Spectrophotometer . cDNA was generated with Superscript III First Strand Synthesis System for RT-PCR ( Thermo , 18080051 ) . Quantitative RT-PCR was performed with a CFX Connect Realtime PCR Detection System ( Bio-Rad ) using iTaq Universal SYBR Green Supermix ( Bio-Rad , 1725125 ) . Samples were quantified by the ∆∆CT method , normalized to β-actin ( mouse samples ) or RPLP0 ( human samples ) to quantify relative mRNA expression levels . qPCR primers: Atf4GATGGGTTCTCCAGCGACAAGCCGGAAAAGGCATCCTCCTTCPsat1GCTGTCGCCTTAGCACCATGGATCTCCAACAATACCGAGTGMthfd2TCCTTGTTGTCTGCGTTGGCCTTCATTTCGCACTGCCGCCSlc7a5GGACAAGGTGATGCGTCCAAGCCAACACAATGTTCCCCACAarsTTGCTATTCCCTCGGAGCACCTCCTCGGGAACCTTAGCTCGarsGGCAGAGGTCTCTGAGCTGGCACGATGGTCATAAGCTGCCarsGAGCAGGCTGCCGACTACATATAGCTACGCGTGCTGAGGNarsGAGCCGGCCTGTGTAAAGATGACCCAGCCAAACACCTTCAIarsTATTGCATCACCTCCAGACGCTGAACCATTCTGTTGCTGGGAGclcTGTACTCCACCCTCGTCACCCCTGCTGTCCCAAGGCTCGGclmTGGGCACAGGTAAAACCCAACTGGGCTTCAATGTCAGGGASlc7a11ATCTCCCCCAAGGGCATACTGCATAGGACAGGGCTCCAAASlc3a2TGATGAATGCACCCTTGTACTTGGCTCCCCAGTGAAAGTGGASlc1a5GTAAAATACCGCAATCCTGTATCCCGATAGCGAAGACCACCAGG XpotGCTTCAGGCTCAGATGCAGAAAAGCAAGGCGAACACTTGGc-MycAGAGCTCCTCGAGCTGTTTGTTCTCTTCCTCGTCGCAGATC/ebpαCAAGAACAGCAACGAGTACCGGTCACTGGTCAACTCCAGCACC/ebpβCGCCTTATAAACCTCCCGCTTGGCCACTTCCATGGGTCTAC/ebpδCGACTTCAGCGCCTACATTGACTAGCGACAGACCCCACACC/ebpγTCGGATCACATTGCTCTGATTTCTGTGCCTGAGTATGAATGACACTActinCACTGTCGAGTCGCGTCCTCATCCATGGCGAACTGGTGPSAT1AAAAACAATGGAGGTGCCGCGGCTCCACTGGACAAACGTAASNSTGGCTGCCTTTTATCAGGGGTCTGCCACCTTTCTAGCAGCMTHFD2GGCAGTTCGAAATGAAGCTGTTGCCAACCAGGATCACACTCASLC7A5GACTACGCCTACATGCTGGAAGCAGCAGCACGCAGAGAARSCCATTCAGAAGGGCACAGGTTATCCACGCCCTGTGTTGTCGARSGCCAGCAGGGAGATCTTGTGCCAGCTCCTTTGCTTCCAGAXPOTGACGCAGAGCGACTAGAGGTAAACATCTTCCCTATCACTCCATCSLC7A11AAGGTGCCACTGTTCATCCCATGTTCTGGTTATTTTCTCCGACARPLP0CCTCGTGGAAGTGACATCGTATCTGCTTGGAGCCCACATT Cells were cultured as indicated , washed twice with PBS , and changed to methionine/cystine/glutamine-free DMEM ( Thermo , 21013024 ) supplemented with 0 . 5 mM L-cystine ( Sigma , 57579 ) and L-glutamine ( 4 mM ) with 50 μCi/mL 35S-methionine ( PerkinElmer , NEG009L005MC ) for 20 min . Cells were washed twice in ice-cold PBS and lysed in ice-cold Triton lysis buffer . Total protein concentrations were normalized following a Bradford assay , and normalized samples were separated by SDS-PAGE and transferred to nitrocellulose . 35S-methionine incorporation into protein was analyzed by autoradiography , and relative rates of protein synthesis were quantified using ImageJ Software ( NIH ) to quantify radiolabeled protein per lane for each sample . For isogenic Atf4-/- cell lines , exogenous cDNA expression was induced for 16 hr with 150 ng/mL doxycycline , and cells were treated as indicated in the presence of doxycycline ( 150 ng/mL ) and 1X MEM NEAA mixture plus 1 mM cysteine ( Sigma , C7477 ) . Protein synthesis was assayed with 20 min labeling in the absence of NEAA and cysteine to avoid competition of 35S-methionine uptake with the supplemented amino acids . For methionine uptake assays , cells were cultured and labeled as described above for the protein synthesis assay , were washed three times in cold PBS , and lysed in Triton lysis buffer . For cystine uptake assays , cells were treated the same but labeled for the final 10 min with medium containing 0 . 1 µCi L-[1 , 2 , 1' , 2'-14C]-Cystine ( PerkinElmer , NEC854010UC ) and washed three times in ice-cold PBS containing cold cystine ( 1 mM ) , prior to lysis in Triton lysis buffer . Whole-cell radiolabel incorporation was quantified with a Beckman LS6500 scintillation counter . Cells from identically treated parallel plates were counted using a Beckman Z1-Coulter Counter to normalize uptake measurements to cell number . To quantify cell proliferation , cells were plated in DMEM in 6-well plates in triplicate in the presence of 2-mercaptoethanol , 1X MEM NEAA mixture , 10% FBS , and doxycycline ( 1 μg/mL ) . Twenty-four hours after plating , cells were washed twice with PBS and media was changed to DMEM with 10% dialyzed FBS , doxycycline ( 1 μg/mL ) , and the amino acid supplements indicated for each experiment , with media refreshed daily . Starting on day 0 , viable cells from triplicate wells corresponding to each condition or cell line were counted using a hemocytometer , excluding dead cells detected by trypan blue stain ( Sigma-Aldrich , T8154 ) . Cells were plated in DMEM in 6-well plates in triplicate in the presence of 2-mercaptoethanol , 1X MEM NEAA mixture , 10% FBS , and doxycycline ( 1 μg/mL ) . Twenty-four hours after plating , cells were washed twice with PBS and media was changed to DMEM with 10% dialyzed FBS and doxycycline ( 1 μg/mL ) . Seventy-two hours later , cells were detached with Accumax ( Sigma-Aldrich , A7089 ) and washed twice with cold PBS on ice . Cells were stained with annexin V and propidium iodide ( PI ) according to the manufacturer’s instruction ( BD , 556547 ) . Samples were analyzed using an LSRFortessa ( BD ) flow cytometer , and the fractions of stained cells were quantified using FloJo 10 . 6 , with Staurosporine ( 4 hr , 5 μM ) ( Tocris , 1285 ) used as a positive control for cell death and to help establish gating of the sorted cells . Cells were plated in 96-well plates at 5000 cells/well . Total glutathione , GSH , and GSSG levels were measured using the GSH/GSSG-Glo Assay ( Promega , V6611 ) according to the manufacturer’s protocol . Total glutathione levels were normalized to cell number determined from parallel plates . BSO ( Sigma , B2515 ) was used as a positive control to inhibit glutathione synthesis . For measurements via LC-MS/MS , metabolites were extracted from cells on dry ice using 80% methanol , and extracts were dried under nitrogen gas for metabolite profiling via selected reaction monitoring with polarity switching using a 5500 QTRAP mass spectrometer . Data were analyzed using MultiQuant 2 . 1 . 1 software ( AB/SCIEX ) to calculate the Q3 peak area . Normalized peak area of glutathione from human TSC2-/- angiomyolipoma ( 621-101 ) cells was determined from previously published data ( Tang et al . , 2019 ) . For xenograft tumor studies , experimental details were provided previously ( Valvezan et al . , 2017 ) . Briefly , mice bearing TSC2-/- ELT3 xenograft tumors were treated every other day for 5 days with vehicle or rapamycin ( 1 mg/kg on days 1 , 3 , and 5 ) and tumors were harvested for metabolite extraction , as above , 3 hr after the final treatment . For RNA-sequencing analysis , Benjamini–Hochberg false discovery rate ( FDR ) -adjusted p values were determined from empirical Bayes moderated t-statistics using the voom and eBayes functions from the limma package . Comparisons with FDR-adjusted p<0 . 05 were considered significant for the gene groups denoted compared to vehicle-treated controls . For KEGG enrichment , p values were FDR corrected . For CiiiDER transcription factor over-representation analysis , Fisher's exact test p values were used . Transcription factor binding elements with p<0 . 01 and test statistic >0 were considered over-represented in genes of interest . Unpaired two-tailed t-tests were used for NanoString analyses to calculate p values for rank ordering . All remaining statistical analyses were performed with Prism 8 software ( GraphPad Software , La Jolla , CA ) . Statistical analyses for qPCR data with two treatment groups were determined by unpaired two-tailed t-test , while those with greater than two treatment groups were determined by one-way analysis of variance ( ANOVA ) with Holm–Sidak method for multiple comparisons . Statistical analyses for protein synthesis assays were determined by one-way ANOVA with Holm–Sidak method for multiple comparisons from values quantified with ImageJ software ( US National Institutes of Health , Bethesda , MD ) . Statistical analyses for immunoblot quantification data with two treatment groups were determined by unpaired two-tailed t-test , while those with greater than two treatment groups were determined by one-way ANOVA with Holm–Sidak method for multiple comparisons . For proliferation assays , unpaired two-tailed t-test was used for comparisons to GFP-expressing cells . For cell death analysis , one-way ANOVA with Holm–Sidak method for multiple comparisons , summing the annexin V+/PI- , annexin V-/PI+ , and annexin V+/PI+ populations for each conditions . For glutathione quantification of experiments with two conditions , an unpaired two-tailed t-test was performed . For remaining glutathione assays and all cystine and methionine uptake experiments , one-way ANOVA with Holm–Sidak method for multiple comparisons was used . The source data for the RNA-sequencing experiment can be found at GEO under the accession number GSE158605 . The source data for Figure 1C can be found in Figure 1—source data 1 . The source data for Figure 2A can be found in Figure 2—source data 1 . The source data for NanoString heatmaps shown in Figure 3 can be found in Figure 3—source data 1 .

When building healthy tissue , the human body must carefully control the growth of new cells to prevent them from becoming cancerous . A core component of this regulation is the protein mTORC1 , which responds to various growth-stimulating factors and nutrients , and activates the chemical reactions cells need to grow . Part of this process involves controlling ‘nutrient-sensing transcription factors’ – proteins that regulate the activity of specific genes based on the availability of different nutrients . One of these nutrient-sensing transcription factors , ATF4 , has recently been shown to be involved in some of the processes triggered by mTORC1 . The role this factor plays in how cells respond to stress – such as when specific nutrients are depleted , protein folding is disrupted or toxins are present – is well-studied . But how it reacts to the activation of mTORC1 is less clear . To bridge this gap , Torrence et al . studied mouse embryonic cells and human prostate cancer cells grown in the laboratory , to see whether mTORC1 influenced the behavior of ATF4 differently than cellular stress . Cells were treated either with insulin , which activates mTORC1 , or an antibiotic that sparks the stress response . The cells were then analyzed using a molecular tool to see which genes were switched on by ATF4 following treatment . This revealed that less than 10% of the genes activated by ATF4 during cellular stress are also activated in response to mTORC1-driven growth . Many of the genes activated in both scenarios were involved in synthesizing and preparing the building blocks that make up proteins . This was consistent with the discovery that ATF4 helps mTORC1 stimulate growth by promoting protein synthesis . Torrence et al . also found that mTORC1’s regulation of ATF4 stimulated the synthesis of glutathione , the most abundant antioxidant in cells . The central role mTORC1 plays in controlling cell growth means it is important to understand how it works and how it can lead to uncontrolled growth in human diseases . mTORC1 is activated in many overgrowth syndromes and the majority of human cancers . These new findings could provide insight into how tumors coordinate their drive for growth while adapting to cellular stress , and reveal new drug targets for cancer treatment .

2021

The mTORC1-mediated activation of ATF4 promotes protein and glutathione synthesis downstream of growth signals

Hedgehog signaling plays very important roles in development and cancers . Vertebrates have three transcriptional factors , Gli1 , Gli2 and Gli3 . Among them , Gli3 is a very special transcriptional factor which closely resembles Cubitus interruptus ( Ci , in Drosophila ) structurally and functionally as a ‘double agent’ for Shh target gene expression . Here we show that Gli3 full-length , but not the truncated form , can be methylated at K436 and K595 . This methylation is specifically catalyzed by Set7 , a lysine methyltransferase ( KMT ) . Methylation at K436 and K595 respectively increases the stability and DNA binding ability of Gli3 , resulting in an enhancement of Shh signaling activation . Furthermore , functional experiments indicate that the Gli3 methylation contributes to the tumor growth and metastasis in non-small cell lung cancer in vitro and in vivo . Therefore , we propose that Set7 mediated methylation is a novel PTM of Gli3 , which positively regulates the transactivity of Gli3 and the activation of Shh signaling . Hedgehog ( Hh ) signaling plays critical roles in embryonic development and tumor growth ( Chen et al . , 2007; Hooper and Scott , 2005; Ingham and McMahon , 2001; Nieuwenhuis and Hui , 2005; Yao and Chuang , 2015 ) . Its misregulation leads to many types of cancers ( Gialmanidis et al . , 2009; Hui and Angers , 2011; Jiang and Hui , 2008 ) . Hh pathway is activated when Hh ligands ( Shh , Ihh or Dhh ) bind to their repressive receptor Patched ( Ptc ) , a twelve-transmembrane protein , leading to alleviation of its repression on the signaling transducer , Smoothened ( Smo ) , a seven-transmembrane protein . The unleashed Smo can then activate the Gli transcription factors , resulting in the expression of downstream target genes , like Gli1 and Ptch1 . In vertebrate , there are three Gli proteins , Gli1 , Gli2 and Gli3 . While Gli1 acts as a transcriptional activator and amplifies the Shh signal in a positive feedback loop , Gli2 and Gli3 mainly function as the Shh-regulated transcriptional activator and repressor . Among them , Gli3 draws considerable attentions due to its structural and functional similarity to Ci . Like Ci , Gli3 plays critical roles in modulating the switch-on and -off of Shh signaling . In the absence of Shh signaling , Gli3 is partially proteolyzed into a truncated transcriptional repressor ( Gli3R ) to block the expression of downstream target genes . In the presence of Shh signaling , Gli3 is in a full-length transactivation form and activates the expression of downstream genes like Gli1 and Ptch1 ( Dai et al . , 1999; 2002 ) . There is substantial evidence demonstrating that post-translational modifications ( PTMs ) play important roles in regulating protein functions , especially in regulating the transactivity of transcriptional factors ( Filtz et al . , 2014; Huang et al . , 2013; Inuzuka et al . , 2011; 2012; Liu et al . , 2014; Munro et al . , 2010 ) . Like many other transcriptional factors , Gli3 is also subjected to various PTMs . Among them , phosphorylation and ubiquitination modifications are well characterized in regulating the transactivity of Gli3 . For example , in the absence of Shh signal , Gli3 is phosphorylated by PKA , GSK3 and CKI , and subsequently ubiquitinated by SCFSlimb/β-TrcP for partial proteolyzation to confer it trans-repressive activity ( Chen et al . , 2009; Hsia et al . , 2015; Tempé et al . , 2006; Wang et al . , 2000; Wang and Li , 2006; Zhang et al . , 2009 ) . Whether other PTMs are involved in the regulation of Gli3 transactivity remains elusive . Protein methylation is one of the most common PTMs and plays an important role in regulating the transduction of signaling pathways , like MAPK , BMP , WNT , Hippo and JAK-STAT ( Bikkavilli and Malbon , 2012; Kim et al . , 2013; Mazur et al . , 2014; Oudhoff et al . , 2013; Viña et al . , 2013 ) . Protein methylation typically happens on arginine or lysine residues catalyzed by peptidylarginine methyltransferases ( PRMTs ) or lysine methyltransferases ( KMTs ) respectively . So far , near 50 KMTs and 9 PRMTs had been identified in human genome ( Biggar and Li , 2015 ) . Among them , Set7 is one of the most studied KMTs , regarding its pivotal role in methylation of non-histone proteins . Although Set7 was first identified as a histone lysine methyltransferase specifically for Histone 3 lysine 4 monomethylation , an epigenetic marker associated with transcriptional activation ( Nishioka et al . , 2002; Wang et al . , 2001 ) , accumulating evidence indicates that methylation of non-histone proteins including P53 , P65 , TAF10 and so on is the major biological function of this enzyme ( Biggar and Li , 2015; Chuikov et al . , 2004; Ea and Baltimore , 2009; Yang et al . , 2009 ) . Set7 mediated methylation of Lys372 in P53 increases its stability , resulting in the induction of P53 target genes ( Chuikov et al . , 2004 ) . P65 can be methylated by Set7 at Lys37 which enhances the DNA binding and improves the expression of NF-κb target genes ( Ea and Baltimore , 2009 ) . Previous sequence alignments of the methylated sites on the initial substrates of Set7 revealed a predicted consensus sequence motif for Set7: ( K/R ) - ( S/T/A ) -K-X ( Couture et al . , 2006 ) . Besides , a recent peptide-array based analysis redefined this recognition motif to: ( G/R/H/K/P/S/T ) - ( K>R ) - ( S>K/Y/A/R/T/P/N ) -K- ( Q/N ) - ( A/Q/G/M/S/P/T/Y/V ) ( Dhayalan et al . , 2011 ) , which dramatically expands the putative targets of Set7 . Here , we report that Gli3 full-length , but not the Gli3 repression form , can be methylated at the K436 and K595 sites in vivo and in vitro . This methylation is specifically catalyzed by Set7 . Moreover , the methylation modifications on K436 and K595 respectively increases the stability and the DNA binding capacity of Gli3 , resulting in enhanced activation of Shh signaling pathway . Furthermore , we demonstrate that this Set7 mediated Gli3 methylations contribute to the tumor growth and metastasis in non-small cell lung cancer in vitro and in vivo . These findings expanded our understanding of PTM-directed Gli3 transactivity regulation , and implied a therapeutic potential of Set7 in treating tumors dependent on Shh signaling . Given that the transcriptional activity of Gli3 is orchestratedly regulated by multiple PTMs , such as phosphorylation and ubiquitination , and that protein methylation plays an important role in regulating several key signaling pathways , we sought to examine whether Gli3 can be post-translationally modified by methylation . We performed a mass spectrometry analysis of flag-tagged Gli3 from the cell lysate of HEK293T . This mass spectrometry analysis showed two methylation modifications on Gli3 K436 and K595 residues ( Figure 1—figure supplement 1 ) . By comparing the flanking sequence of K436 and K595 with reported Set7 substrates , such as ERα ( Subramanian et al . , 2008 ) , P53 ( Chuikov et al . , 2004 ) , PCAF ( Masatsugu and Yamamoto , 2009 ) and Histone 3 ( Wang et al . , 2001 ) , we found strong similarities among them ( Figure 1A , upper panel ) , suggesting the possible involvement of Set7 in methylation of these two residues . Interestingly , these methylation signals were exclusively present in the Gli3 full-length but not the truncated repression form according to the mass spectrometry result ( Figure 1—figure supplement 1 ) . Through sequence alignments , we found that these two sites in Gli3 are evolutionally conserved in many species ( Figure 1—figure supplement 2 ) . To further test if the methylations on K436 and K595 are catalyzed by Set7 , in vitro methylation assays were performed by incubating MBP-tagged peptides of Gli3 , MBP-K436 ( amino acids from 344 to 463 of Gli3 ) or MBP-K595 ( amino acids from 464 to 666 of Gli3 ) with GST-Set7 and 3H-SAM ( Figure 1A , lower panel ) . As shown in Figure 1B , Set7 can specifically methylate the peptides , MBP-K436 and MBP-K595 , whereas the substitution of K436 or K595 with arginine ( K436R or K595R ) abolished Set7 mediated Gli3 methylation . In addition , GST pull-down assays further indicated the interaction between Set7 and Gli3 ( Figure 1C ) . 10 . 7554/eLife . 15690 . 003Figure 1 . Set7 methylates Gli3 full-length in vivo and in vitro . ( A ) Sequence alignment of the reported Set7 substrates with Gli3K436 and Gli3K595 ( upper ) . Schematic representation of Gli3 protein and the truncated peptide used in in vitro methylation assay ( B ) and GST pull-down assay ( C ) ( lower ) . ( B ) In vitro methylation assay with 3H-S-adenosine-methionine ( 3H-SAM ) , bacteria purified Set7 and MBP fusion protein . * and ** represent the MBP-K436 and MBP-K595 respectively . ( C ) GST pull-down assay using GST-Set7 and MBP tagged Gli3 truncated fragments described in ( A ) . * and ** represent the MBP-K436 and MBP-K595 respectively . ( D–F ) Western blot of immunoprecipitates ( top three panels ) and lysates ( bottom ) from HEK293T cells expressing indicated siRNAs or proteins . * and ** represent the full-length and repressor forms of Gli3 respectively . ( G–I ) Western blot of immunoprecipitates ( top three panels ) and lysates ( bottom ) from NIH-3T3 cells stably expressing indicated shRNAs or proteins . ( I’ ) Schematic representation of 6 PKA targeted serines which were mutated to nonphosphorylatable alanines in Gli3PA . ( J ) GST pull-down assay using GST-Set7 and flag-Gli3 in NIH-3T3 cells in the presence of Shh . Ctrl , Control . Me-K436 , antibody anti methylated Gli3-K436; Me-K595 , antibody anti methylated Gli3-K595 . WCL , whole cell lysis . The protein level of Gli3 in ( D–I ) are normalized to the same . DOI: http://dx . doi . org/10 . 7554/eLife . 15690 . 00310 . 7554/eLife . 15690 . 004Figure 1—figure supplement 1 . The mass spectrometry results show the methylation modification in the full-length Gli3 on K436 and K595 . ( A ) Sliver staining of the mass spectrometry sample . * represent the Gli3 full-length , ** represent the Gli3 repressor . ( B and C ) The mass spectrometry results of the methylation modification on K436 ( B ) and K595 ( C ) in the Gli3 full-length sample . DOI: http://dx . doi . org/10 . 7554/eLife . 15690 . 00410 . 7554/eLife . 15690 . 005Figure 1—figure supplement 2 . Sequence alignment of K436 and K595 sites of Gli3 in different species Sequence alignment of the methylation sites K436 . ( A ) and K595 ( B ) in Gli3 from different species suggested that these two sites are evolution conserved . DOI: http://dx . doi . org/10 . 7554/eLife . 15690 . 00510 . 7554/eLife . 15690 . 006Figure 1—figure supplement 3 . The methylation antibodies can specifically recognize the mono methylated Gli3 peptides . ( A ) The sequences of peptides used for preparing antibodies . ( B ) Dot blots of peptides described in ( A ) . Me-K436 , antibody against methylated Gli3-K436; Me-K595 , antibody against methylated Gli3-K595 . DOI: http://dx . doi . org/10 . 7554/eLife . 15690 . 00610 . 7554/eLife . 15690 . 007Figure 1—figure supplement 4 . The methylation antibodies can specifically recognize the mono methylated Gli3 full length in embroynic lung Indicated tissues from mouse embryos ( 14 . 5 dpc ) were isolated and lysed . Methylation signals on Gli3 full length were detected by western blots . DOI: http://dx . doi . org/10 . 7554/eLife . 15690 . 007 To facilitate studies on the Set7-mediated methylation of Gli3 in vivo , we generated polyclonal antibodies that specifically recognize monomethylated Gli3-K436 or Gli3-K595 , respectively ( Figure 1—figure supplement 3A ) . Dot plot assays confirmed the specificity of those antibodies ( Figure 1—figure supplement 3 ) . Consistent with the results of mass spectrometry , immunoprecipitation ( IP ) of endogenous Gli3 in HEK293T cells followed by immunoblot showed that only the full-length Gli3 contains the methylation signals on Gli3-K436 and Gli3-K595 ( Figure 1D ) . The methylation signals were decreased upon Set7 knockdown , indicating that methylation on these two sites of Gli3 is mediated by Set7 ( Figure 1D ) . Moreover , overexpression of Set7 drastically improved the methylation levels whereas the catalytically inactive mutant of Set7 , Set7H297A ( Wang et al . , 2001 ) , had no effect on those methylations of Gli3 ( Figure 1E ) . Furthermore , Set7 failed to methylate Gli3K436R or Gli3K595R , the arginine substitution mutants of Gli3 , in vivo ( Figure 1F ) . Since the methylation happens exclusively in the full-length Gli3 , which acts as a principal transcription activator of Shh pathway in response to Shh signal ( Dai et al . , 1999 ) , we then adopted , NIH-3T3 , a cell line fully responsive to the Shh signaling , to perform our further investigation . Cells with stable expression of individual unrelated Set7 shRNAs showed significantly reduced methylation signals on K436 and K595 of Gli3 compared to cells expressing luciferase control shRNA ( Figure 1G ) . In addition , the Gli3 arginine substitution mutants , Gli3K436R and Gli3K595R , showed abolished methylation signals when blotted with the matched antibodies ( Figure 1H ) . Given that only the full-length Gli3 can be methylated , we began to use a Gli3PA mutant , in which the six phosphorylatable serines in the PKA target clusters were mutated to nonphosphorylatable alanines , resulting in the resistance of Gli3PA to be processed into Gli3R ( Niewiadomski et al . , 2014 ) ( Figure 1I' ) . Similar to the result in Figure 1H , the methylation signals can be detected in Gli3PA but not in its arginine substitution mutants , Gli3PAK436R and Gli3PAK595R when blotted with respective antibodies ( Figure 1I ) . Given the Shh signaling plays an important role in the development of multiple embryonic tissues of mice ( Ingham and McMahon , 2001 ) , we started to examine whether these methylation signals exist in the Gli3 in several responsive mouse embryonic tissues , such as brain , lung , skin , skeleton and gut . Indicated tissues were collected from embryos 14 . 5 days post coitum ( dpc ) and Gli3 methylation signals were detected by western blot using anti-Gli3 antibody , Me-K436 antibody and Me-K595 antibody , respectively . As shown in Figure 1—figure supplement 4 , the methylation signals on Gli3 full length were only detected in embryonic lung tissues . Since Gli3 full-length can be methylated by Set7 in vivo and in vitro , we then examined whether Set7 interacts with Gli3 . GST pull-down assays indicated that GST-tagged Set7 can successfully pull down the flag-tagged Gli3 full-length of NIH-3T3 cells in the presence of Shh ( Figure 1J ) . Therefore , we demonstrate that Set7 can methylate the full-length , but not the repression form of Gli3 , at K436 and K595 residues , indicating that Gli3 is a novel substrate of Set7 . Set7 can modulate the activation of multiple signaling pathways , such as NF-κb , JAK-STAT , we sought to examine whether Set7 also regulates Shh signaling . We performed Shh luciferase reporter assays ( Kinzler and Vogelstein , 1990 ) in the condition of Set7 knockdown . In the absence of Shh , downregulation of Set7 had no effect on the luciferase activity . In contrast , Set7 knockdown dramatically decreased the luciferase activity in the presence of Shh , suggesting that Set7 plays a positive role on Shh pathway activation but has no effect on the basal level of Shh signaling ( Figure 2A and A’ ) . In line with this finding , overexpression of Set7 but not its inactive mutant Set7H297A can significantly enhance the luciferase activity in the presence of Shh ( Figure 2B ) , suggesting that its methyltransferase activity is required for Set7 mediated Shh activation . Furthermore , Set7 mediated increase of luciferase activity was in a dose-dependent manner ( Figure 2C ) . We then examined whether Set7 knockdown affects the expression of endogenous Shh target genes , like Gli1 and Ptch1 . In the presence of Shh , NIH-3T3 cells expressing Set7 siRNAs showed significant reduction in Set7 protein levels and a parallel decrease in Gli1 and Ptch1 mRNA levels compared to cells expressing control siRNAs ( Figure 2D–F ) . In addition , Gli1 protein level was also decrease in cells with Set7 knockdown ( Figure 2F ) . It’s of interest that either Shh or SAG , the agonist of Shh pathway , directed activation of Shh signaling can upregulate Set7 levels , whereas Gli1 knockdown can lead to a decrease of Set7 mRNA levels , suggesting the existence of a potential positive feedback loop between Set7 and Shh signaling ( Figure 2A' and Figure 2—figure supplement 1 ) . In addition , NIH-3T3 cells treated with Shh showed greatly increased methylation signals on Gli3 which may be probably due to increased Gli3 full-length or/and Set7 levels ( Figure 2A’ and Figure 2—figure supplement 2 ) , suggesting that the methylation signals of Gli3 is regulated by Shh input . Taken together , these findings indicate that Set7 can positively regulate the activation of Shh signaling . 10 . 7554/eLife . 15690 . 008Figure 2 . Methylation by Set7 promotes the Sonic Hedgehog signaling activity . ( A–C ) Shh luciferase reporter assays in NIH-3T3 cells expressing indicated siRNAs or proteins . ( A’ ) Set7 level changes in NIH-3T3 cells expressing indicated Set7 siRNAs . n . s . p>0 . 05 , ***p<0 . 005 versus siCtrl respectively in ( A ) . ( D and E ) qPCR to detect Shh target gene , Gli1 and Ptch1 , in Shh treated NIH-3T3 cells after Set7 knockdown . ***p<0 . 005 versus siCtrl . ( F ) Western blot to detect protein level changes of Set7 and Gli1 in ( D and E ) . ( G ) Shh luciferase reporter assay in Shh treated NIH-3T3 cells transfected with indicated siRNAs . ***p<0 . 001 , n . s . p>0 . 05 versus siCtrl siCtrl respectively . ( H–J ) mRNA level changes of Gli1 and Ptch1 ( H and I ) and Shh luciferase activity changes ( J ) in Shh treated NIH-3T3 cells transfected with unmethylatable mutants , Gli3K436R and Gli3K595R . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 versus Gli3WT ( K and L ) Shh luciferase assay in Shh treated NIH-3T3 cells expressing indicated siRNAs or/and proteins . , **p<0 . 01 , ***p<0 . 001 versus siCtrl Gli3WT or PCDNA Gli3WT respectively . . Results in ( A–E , G–L ) are shown as mean ± SEM ( n=3 ) . All the qPCR results are normalized to GAPDH . Ctrl , Control . DOI: http://dx . doi . org/10 . 7554/eLife . 15690 . 00810 . 7554/eLife . 15690 . 009Figure 2—figure supplement 1 . Sonic hedgehog signaling increases the mRNA level of Set7 . ( A and B ) qPCR of Set7 in NIH-3T3 cells treated with Shh ( A ) and SAG ( B ) . **p<0 . 01 ***p<0 . 001 versus Ctrl . ( C and D ) qPCR of Gli1 ( C ) and Set7 ( D ) in NIH-3T3 cells transfected with Gli1 siRNA . **p<0 . 01 ***p<0 . 001 versus siCtrl . Results are shown as mean ± SEM ( n=3 ) . All the qPCR results are normalized to GAPDH . DOI: http://dx . doi . org/10 . 7554/eLife . 15690 . 00910 . 7554/eLife . 15690 . 010Figure 2—figure supplement 2 . The methtylation of Gli3 is regulated by the Shh signaling Western blot of endogenous Gli3 immunoprecipitates from NIH-3T3 cells with or without Shh treatment . * and ** represent the full-length and repressor forms of Gli3 respectively . The protein density of Gli3 full length and methylation signals is analyzed using ImageJ . DOI: http://dx . doi . org/10 . 7554/eLife . 15690 . 01010 . 7554/eLife . 15690 . 011Figure 2—figure supplement 3 . In the Gli1-/-; Gli2-/- MEF cells , Set7 still regulates the Shh signaling . ( A ) qPCR to detect Ptch1 in Shh treated Gli1-/-;Gli2-/- MEF cells with Set7 knockdown . *p<0 . 05 versus Ctrl . ( B ) qPCR to detect Set7 in Shh treated Gli1-/-;Gli2-/- MEF cells with Set7 knockdown . ***p<0 . 001 versus Ctrl . Results are shown as mean ± SEM ( n=3 ) . All the qPCR results are normalized to GAPDH . DOI: http://dx . doi . org/10 . 7554/eLife . 15690 . 01110 . 7554/eLife . 15690 . 012Figure 2—figure supplement 4 . Non-methylation mutants show reduced Gli1 and Ptch1 mRNA level , and decreased Shh luciferase activity . ( A and B ) qPCR of Gli1 and Ptch1 ( A ) and Shh luciferase activity ( B ) in NIH-3T3 cells expressing indicated proteins . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . Results are shown as mean ± SEM ( n=3 ) . All the qPCR results are normalized to GAPDH . DOI: http://dx . doi . org/10 . 7554/eLife . 15690 . 012 Next , we started to investigate whether the positive regulation of Set7 on Shh signaling relies on its methylation on Gli3 . As shown in Figure 2G , although Set7 knockdown can suppress the activation of Shh signaling , this inhibition was abolished when the Gli3 level was depleted by RNAi . In contrast , depletion of both Gli1 and Gli2 in MEF cells showed no effect on releasing such inhibition due to Set7 knockdown ( Figure 2—figure supplement 3 ) . These findings apparently indicate the requirement of Gli3 on Set7 mediated Shh activation . Furthermore , compared to Gli3WT , the mRNA levels of Gli1 and Ptch1 and the Shh luciferase activity were decreased in cells expressing Gli3K436R and Gli3K595R , suggesting that Set7-directed Gli3 methylation plays an important role in maintaining Shh activation ( Figure 2H–J ) . In consistent with this finding , cells expressing the Gli3PA point mutants , Gli3PAK436R and Gli3PAK595R , showed decreased Shh luciferase activity and levels of Gli1 and Ptch1 transcripts compared to those with Gli3PAWT expression ( Figure 2—figure supplement 4 ) . To further confirm this notion , Shh luciferase reporter assays were performed . Either Set7 downregulation or upregulation can lead to the reduction or the induction of Shh luciferase activity respectively in cells with Gli3WT expression but not in cells expressing Gli3K436R and Gli3K595R ( Figure 2K and L ) . Thus , our findings strongly suggested that Gli3 methylation at K436 and K595 sites are required for Set7-mediated enhancement of Shh signaling activation . Gli3 can be regulated at multiple levels to affect the transduction of Shh signaling , such as trafficking to the primary cilium , changes of protein stability or binding ability to the promoters of target genes . In mammal , the primary cilium ( cilia ) is an essential organelle for Shh signaling transduction . Gli3 traffics in and out of the cilia all the time ( Figure 3—figure supplement 1A and B ) . In the presence of Shh , Gli3 initially accumulates in the cilium tip ( Figure 3—figure supplement 1B ) and then functions as Gli3A to bind to the promoter of Gli1 to further transduce the signaling ( Chen et al . , 2009; Dai et al . , 1999; 2002; Wen et al . , 2010; Zhang et al . , 2009 ) . In order to verify whether Set7 mediated Gli3 methylation affects its cilial translocation , we performed cilial staining in Shh treated NIH-3T3 cells with stable expression of Set7 shRNA , Set7 or Set7H297A . As shown in Figure 3—figure supplement 1C and D , the Gli3 accumulation on the tips of cilia was comparable among those three groups , suggesting that the Set7-related changes of Shh activation are not due to the cilial translocation of Gli3 . It has been reported that Set7 can influence the stability of its methylated substrates like P53 ( Chuikov et al . , 2004 ) . Given the importance of Gli3 stability on Shh signaling activation ( Tempé et al . , 2006; Zhang et al . , 2009 ) , we then tested whether Set7 can also affect the stability of Gli3 . Western blot showed that knockdown of Set7 caused a significant reduction of endogenous Gli3 at protein levels ( Figure 3A , left panel ) , whereas overexpression of Set7 improved the Gli3 protein level ( Figure 3A , right panel ) . This Set7 mediated induction of Gli3 seems due to PTMs , because the mRNA level of Gli3 had no change regardless whether Set7 was knocked down or overexpressed ( Figure 3B ) . To further confirm this point , we then tested whether Set7 can affect the protein levels of the unmethylatable Gli3 mutants , Gli3K436R and Gli3K595R . As expected , exogenous Gli3WT had the same change as the endogenous Gli3 . Interestingly , like the Gli3WT , the protein levels of Gli3K595R were still subjected to the modulation of Set7 , whereas the protein levels of Gli3K436R were not responding to either knockdown or overexpression of Set7 ( Figure 3C and D ) . These findings suggested that the methylation at K436 but not K595 causes stability change of Gli3 ( Figure 3C and D ) . Finally , the reduced half-life of Gli3K436R , but not Gli3WT or Gli3K595R was defined by western blot in cells treated with cycloheximide ( CHX ) ( Figure 3E and F ) . Therefore , these data indicated that Set7 mediated methylation on K436 increases the protein stability of Gli3 . 10 . 7554/eLife . 15690 . 013Figure 3 . Set7 improved the stability and DNA binding ability respectably through K436 and K595 methylation . ( A and B ) Western blot ( A ) and qPCR ( B ) of Gli3 in the condition of Set7 knockdown or overexpression in NIH-3T3 cells . n . s . p>0 . 05 versus siCtrl ( left ) or PCDNA ( right ) respectively . ( C and D ) Western blot of Gli3WT , Gli3K436R or Glli3K595R in the condition of Set7 knockdown or overexpression in NIH-3T3 cells . ( E and F ) Western blot ( E ) and statistic ( F ) of Gli3WT , Gli3K436R or Gli3K595R in NIH-3T3 cells treated with cycloheximide ( CHX ) for the indicated times . ( G ) Schematic representation of three Gli3 binding sites on the promoter regions of Gli1 . ( H ) qPCR of Set7 in Gli3WT-flag stable cell lines after Set7 knockdown in ( I ) . **p<0 . 01 , ***p<0 . 00 versus siCtrl . ( I ) ChIP-qPCR analysis using Ctrl IgG ( grey ) or anti-flag antibody ( black ) in NIH-3T3 cells transfected with Ctrl or Set7 siRNAs . ChIP signal levels are represented as fold change of input chromatin . ( J ) Western blot of flag tagged Gli3WT and mutants in the stable cell lines used in ( K ) . ( K ) ChIP-qPCR analysis using anti-flag antibody in NIH-3T3 cells expressing Gli3WT-flag , Gli3K436R-flag or Gli3K595R-flag . ChIP signal levels are represented as fold change of input chromatin . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , n . s . p>0 . 05 versus anti flag siCtrl ( I ) or Gli3WT-flag ( K ) . Data are represented as mean ± SEM ( n=3 ) . All the qPCR results are normalized to GAPDH . In all the experiments , cells are treated with Shh . Ctrl , Control . DOI: http://dx . doi . org/10 . 7554/eLife . 15690 . 01310 . 7554/eLife . 15690 . 014Figure 3—figure supplement 1 . Set7 does not influence the cilium translocation of Gli3 . ( A and B ) Immunofluorescent staining of Ac-tubulin ( blue ) and Gli3 ( red ) in the absence ( A-A’’ ) or presence ( B-B’’ ) of Shh . ( C and D ) Statistic results of Gli3 accumulation at cilia tips in NIH-3T3 cells with Set7 knockdown ( C ) or Set7/Set7H297A over-expression ( D ) . In all experiments of ( C and D ) , cells were treated with Shh . Data are represented as mean ± SEM ( n=3 ) . Ctrl , Control . DOI: http://dx . doi . org/10 . 7554/eLife . 15690 . 01410 . 7554/eLife . 15690 . 015Figure 3—figure supplement 2 . Set7 improves the DNA binding ability of Gli3 on Gli1 promoter . ( A ) ChIP-qPCR analysis at Gli1 loci ( Figure 3G ) using control IgG or anti-Gli3 antibody in NIH-3T3 cells transfected with control or Set7 siRNAs . ChIP signal levels are represented as fold change of input chromatin . **p<0 . 01 , ***p<0 . 001 , versus anti Gli3 siCtrl . ( B ) qPCR of Set7 in NIH-3T3 cells with Set7 knockdown . RNAi efficiency was shown . Data are represented as mean ± SEM ( n=3 ) . **p<0 . 01 versus siCtrl . Ctrl , Control . All the qPCR results are normalized to GAPDH . DOI: http://dx . doi . org/10 . 7554/eLife . 15690 . 01510 . 7554/eLife . 15690 . 016Figure 3—figure supplement 3 . Set7 improves the DNA binding ability of Gli3 on the promoter regions of Hhip and Ptch1 ChIP-qPCR assays at Hhip . ( A ) and Ptch1 ( B ) loci using control IgG or anti-flag antibody in NIH-3T3 Gli3-flag stable cells transfected with control or Set7 siRNA . ChIP signal levels are represented as fold change of input chromatin . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , versus anti flag siCtrl . DOI: http://dx . doi . org/10 . 7554/eLife . 15690 . 016 Gli3 full-length is a transcription activator and can bind to the promoter of Gli1 to further enhance the activation of Shh signaling ( Hui and Angers , 2011; Jiang and Hui , 2008 ) . To examine whether Set7 also affects the DNA binding ability of Gli3 , chromatin immunoprecipitation ( ChIP ) assays were performed in NIH-3T3 cells in the presence of Shh . As shown in Figure 3—figure supplement 2 , knockdown of Set7 greatly decreased the binding of Gli3 on Gli1’s promoter regions ( Figure 3G and Figure 3—figure supplement 2 ) . Although Shh treatment can almost completely block the processing of Gli3 full-length to Gli3R , trace Gli3R background may still disturb the result of ChIP assays . In order to address this concern , we established NIH-3T3 cell lines with stable expression of C-terminal flag-tagged Gli3 ( Gli3-flag ) . Thus , only the Gli3 full-length can be recognized by anti-flag antibody . ChIP assays with anti-flag antibody demonstrated that Set7 knockdown markedly affected the binding of Gli3 on Gli1’s promoter regions ( Figure 3G–I ) , suggesting that Set7 augments the DNA binding of Gli3 full-length . Given that K595 is on the zinc finger which is important for DNA binding of Gli3 ( Pavletich and Pabo , 1993 ) , it’s conceivable that the Set7 mediated methylation on K595 may contribute to increased DNA binding of Gli3 . Stable cell lines expressing Gli3WT-flag , Gli3 mutants ( Gli3K436R-flag and Gli3K595R-flag ) were established . Comparable expression of flag-tagged Gli3 was detected in those three cell lines ( Figure 3J ) . ChIP assays with anti-flag antibody showed dramatically impaired the DNA binding ability in Gli3K595R-flag , whereas the Gli3K436R-flag presented the similar DNA binding ability to Gli3WT-flag ( Figure 3K ) , suggesting that Set7 mediated methylation at K595 enhances the DNA binding ability of Gli3 . In addition , the binding ability of Gli3 was also examined in the promoter regions of Ptch1 and Hhip by ChIP assays . As shown in Figure 3—figure supplement 3 , the binding of Gli3 to these regions was markedly impaired after Set7 knockdown . In conclusion , Set7 improved the protein stability and DNA binding ability of Gli3 through methylation of K436 and K595 sites respectively . Misregulation of Shh pathway has been reported in many types of cancers , including basal cell carcinoma , medulloblastoma and lung cancer ( Gialmanidis et al . , 2009; Hui and Angers , 2011; Jiang and Hui , 2008; Yuan et al . , 2007 ) . Since Set7 can augment Shh activation by increasing Gli3 stability and its DNA binding ability on the promoter regions of Gli1 , we then determined whether this Set7-Gli3-Gli1 axis plays a role in tumor development . By comparison on the mRNA levels of Set7 and Gli1 between Beas2B , a normal bronchial epithelial lung cell line , and A549 , a non-small cell lung cancer cell line , we found that both Set7 and Gli1 were extensively expressed in A549 cells ( Figure 4—figure supplement 1A ) . Furthermore , the GEO database search ( GSE10245 ) showed a positive association between Set7 and Gli1 in NSCLC tumor samples , suggesting the involvement of this Set7-Gli3-Gli1 axis in the development of NSCLCs ( Figure 4—figure supplement 1B ) . To examine the existence of this regulation axis in A549 cells , we performed Shh luciferase reporter assays and qPCR to evaluate the activity of Shh signaling in A549 cells with Set7 knockdown . Depletion of Set7 displayed a decreased luciferase activity and reduced expression of Gli1 and Ptch1 ( Figure 4A and B ) . Our above findings indicate that unmethylatable mutations at either K436 or K595 cause a significant reduction of Shh signaling activation ( Figure 2H–J ) . To examine whether unmethylatable mutations at both sites will lead to a synergistic suppression on Shh signaling activation , we constructed a double arginine substitution mutant of Gli3 , Gli3KRKR , in which both K436 and K595 were mutated to arginine . To further examine whether Gli3 methylations at K436 and K595 are important for Set7-mediated Shh activation in A549 cells , we performed luciferase assays and qPCR . Compared to A549 cells with stable Gli3WT expression , reduced luciferase activity and transcripts of Gli1 and Ptch1 were detected in cells stably expressing Gli3K436R , Gli3K595R or Gli3KRKR ( Figure 4C–E ) . In addition , more severe suppression on luciferase activities and Ptch1 mRNA , but not Gli1 mRNA , was observed in cells expressing Gli3KRKR , suggesting that the methylation at these two sites plays a synergistic role in regulating Shh signaling activation ( Figure 4D and E ) . In line with the reduced Shh activation , MTT assays showed attenuated growth in cells stably expressing Gli3K436R , Gli3K595R or Gli3KRKR compared to cells with Gli3WT expression ( Figure 4F ) . Since anchorage-independent growth is tightly associated with tumor development , we then determined whether Set7-mediated Gli3 methylation affects this feature . The ability to form colonies in soft agar was dramatically impaired in stable transfectants with the expression of Gli3K436R , Gli3K595R or Gli3KRKR compared to Gli3WT ( Figure 4G and G’ ) . In agreement with these in vitro growth assays , A549 cells with stable expression of Gli3K436R , Gli3K595R or Gli3KRKR showed retarded tumor growth in a xenograft tumor model ( Figure 4H–J ) . 10 . 7554/eLife . 15690 . 017Figure 4 . Methylation of Gli3 positively regulated the A549 proliferation and migration ability . ( A and B ) Shh luciferase reporter assay ( A ) and qPCR ( B ) of Gli1 , Ptch1 and Set7 in A549 cells in the condition of Set7 knockdown . ( C ) Western of flagged Gli3WT , Gli3K436R , Gli3K595R or Gli3KRKR in A549 stable cells used in ( D-M ) . ( D and E ) qPCR ( D ) of Gli1 and Ptch1 and Shh luciferase reporter assay ( E ) in A549 cells expressing Gli3WT , Gli3K436R , Gli3K595R or Gli3KRKR . ( F ) MTT assays in A549 cells expressing indicated proteins . ( G and G’ ) Photography ( G ) and statistic results ( G’ ) of anchorage-independent growth assay in A549 cells expressing indicated proteins . ( H and I ) Statistic ( H ) and photography ( I ) results of tumor size in A549 cells expressing indicated proteins . ( J ) Tumor growth curves of A549 cells express indicated protein in a xenograft mouse model . ( K–M ) Migration ability evaluated by wound healing assay ( K ) and transwell migration assay ( L ) . Invasiveness evaluated by matrigel invasion assay ( M ) in A549 cells expressing indicated proteins . All the qPCR results are normalized to GAPDH . Data in ( D–H , J ) were represented as mean ± SEM ( n=3 ) . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 versus Gli3WT . Ctrl , Control . DOI: http://dx . doi . org/10 . 7554/eLife . 15690 . 01710 . 7554/eLife . 15690 . 018Figure 4—figure supplement 1 . The involvement of this Set7-Gli3-Gli1 axis in the development of NSCLCs ( A ) Comparison of mRNA levels of Gli1 and Ptch1 in Beas2B and A549 cells by qPCR . The qPCR results are normalized to GAPDH . Data wad represented as mean ± SEM ( n=3 ) . ( B ) Correlation of Set7 and Gli1 mRNA levels in NSCLC samples . DOI: http://dx . doi . org/10 . 7554/eLife . 15690 . 01810 . 7554/eLife . 15690 . 019Figure 4—figure supplement 2 . Methylation on the Gli3PA promoted the A549 tumorigenesis . ( A ) Western blot of flagged Gli3PAWT , Gli3PAK436R , Gli3PAK595R or Gli3PAKRKR in A549 stable cells used in ( B-H ) . ( B and C ) Hh luciferase reporter assay ( B ) and qPCR ( C ) of Gli1 and Ptch1 in A549 cells expressing flagged Gli3PAWT , Gli3PAK436R , Gli3PAK595R or Gli3PAKRKR . ( D ) MTT assays in A549 cells expressing indicated proteins . ( E and E’ ) Photography ( E ) and statistic results ( E’ ) of anchorage-independent growth assay in A549 cells expressing indicated proteins . ( F-H ) Migration ability evaluated by wound healing assay ( F ) and transwell migration assay ( G ) . Invasiveness evaluated by matrigel invasion assay ( H ) in A549 cells expressing indicated proteins . ( G’ and H’ ) Statistic results of migration ( G’ ) and invasion ( H’ ) ability of A549 cells expressing indicated proteins . Data in ( B-D , G’ , H’ ) were represented as mean ± SEM ( n=3 ) . Ctrl , Control . All the qPCR results are normalized to GAPDH . DOI: http://dx . doi . org/10 . 7554/eLife . 15690 . 019 A recent report indicates that the activation of Shh pathway is required for epithelial to mesenchymal transition ( EMT ) in NSCLCs ( Della Corte et al . , 2015 ) . Combined with our results above , we started to examine whether this Set7-Gli3-Gli1 axis also plays a role in cell migration or invasion . To evaluate cell migration changes , transwell migration and wound healing assays were performed in A549 cells . As shown in Figure 4K and L , A549 cells with the expression of Gli3K436R , Gli3K595R or Gli3KRKR had limited migration ability compared to Gli3WT . Furthermore , matrigel chamber assays were performed and the invasion ability of A549 cells stably expressing Gli3K436R , Gli3K595R or Gli3KRKR was substantially reduced ( Figure 4M ) . Moreover , all the in vitro functional experiments in Figure 4 were repeated in A549 cells stably transfected with Gli3PAWT , Gli3PAK436R , Gli3PAK595R or Gli3PAKRKR , and showed similar results ( Figure 4—figure supplement 2 ) . It further confirms that the methylation on Gli3 full-length regulates tumorigenesis in NSCLCs . In present study , we demonstrated that Set7 can exclusively methylate Gli3 full-length at K436 and K595 sites in vitro and in vivo . These methylations on Gli3 augment the activation of Shh signaling . To our knowledge , we for the first time identified Gli3 as a novel substrate of Set7 and demonstrated the involvement of Set7 in Shh activation . Furthermore , we showed that Set7 mediated methylation on K436 increases the protein stability of Gli3 whereas the methylation on K595 boosts the DNA binding of Gli3 . Finally , functional experiments indicated that this Set7-mediated Gli3 methylation contributes to the tumor growth and metastasis in NSCLCs in vitro and in vivo . Hh signaling is one of the most conserved signaling pathways . The silence or activation of Hh signaling is tightly regulated by various post-translational modifications ( PTMs ) at multiple levels . Gli3 is a key transcriptional factor determining the switch-off and -on of Shh signaling . The PTMs of Gli3 like phosphorylation ( Tempé et al . , 2006; Wang et al . , 2000; Wang and Li , 2006 ) and ubiquitination ( Chen et al . , 2009; Hsia et al . , 2015; Zhang et al . , 2009 ) are largely involved in this process . For example , in the absence of Shh , Gli3 full-length is phosphorylated by PKA , GSK3 and CK1 and is subjected to SCFSlimb/β-TrcP mediated ubiquitination followed with partial proteasome degradation to generate a truncated repression form , Gli3R ( Tempé et al . , 2006; Wang et al . , 2000; Wang and Li , 2006 ) . Protein methylation , as one of the most common PTMs , participates in the regulation of signaling transduction of multiple signaling pathways , like MAPK , BMP , WNT , Hippo and JAK-STAT ( Bikkavilli and Malbon , 2012; Kim et al . , 2013; Mazur et al . , 2014; Oudhoff et al . , 2013; Xu et al . , 2013 ) . Protein methylation is catalyzed by methyltransferase . Among them , Set7 draws tremendous attention to its predominant role in methylating non-histone proteins . In this study , we identify that Gli3 is a novel substrate of Set7 . By methylating Gli3 , Set7 positively regulates the activation of Shh signaling pathway . It’s of particular interest that , although Gli3R also has these two residues , K436 and K595 , the Set7 mediated methylation solely occurs in Gli3 full-length . One possible explanation comes from the differential cellular localization of Gli3R and Gli3 full-length in the absence of Shh . A recent study demonstrates that the stability of Dnmt1 can be dynamically regulated by Set7 and demethylase , Lsd1 ( Wang et al . , 2009 ) . Since Lsd1 is a nuclear protein , it is conceivable that Lsd1 can also remove the methylation modifications of Gli3R , resulting in these signals undetectable either by mass spectrometry or western blot . In addition , we found increased Set7 levels in cells treated with Shh or SAG , suggesting a positive feedback loop between Set7 and Shh signaling . Thus , in the presence of Shh signaling , although Gli3 full-length also translocates into the nucleus subjected to Lsd1 mediated demethylation , increased Set7 could possibly compensate this demethylation effect , resulting in augment in Shh signaling . Therefore , it will be particularly interesting to examine whether Lsd1 and Set7 can dynamically regulate the activation of Shh signaling by demethylation and methylation of Gli3 respectively . We demonstrated that Set7 methylates K436 and K595 of Gli3 . The methylation on K436 increases the protein stability and the methylation on K595 enhances the DNA binding of Glli3 . These two methylation modifications of Gli3 are not redundant , because arginine substitution in either point severely impairs the activation of Shh signaling , suggesting the role of Set7 in fine-tuning the transactivity of Gli3 . It has been recently reported that the activation of Shh signaling represses the canonical WNT signaling in colon epithelial cell differentiation , and that WNT pathway activation negatively regulates Shh pathway ( van den Brink et al . , 2004; Yanai et al . , 2008 ) , suggesting a mutual repelling crosstalk between these two pathways . Given that Set7 can inhibit WNT signaling by methylating β-catenin , and activate Shh signaling by methylating Gli3 , it’s very possible that Set7 is a key regulator involved in this WNT-Shh crosstalk . Abnormal activation of Shh signaling contributes to tumor development by promoting cell proliferation and EMT in many cancer types . Our present study indicates that Set7 mediated Gli3 methylation is critical for this oncogenic role of Shh pathway because cancer cells expressing Gli3K436R , Gli3K595R or Gli3KRKR showed severely reduced growth and metastasis in xenograft tumor models . This finding strongly suggests a therapeutic value of Ser7 in targeting Shh-dependent tumors , such as BCC , MB and Lung cancer . The Gli3 and Set7 expression plasmids were kindly provided by Dr . Chi-chung Hui and Dr . Min Wu respectively . The Gli3PA plasmid was obtained from Addgene ( Plasmid# 226672 ) . The unmethylatable mutants of Gli3 , Gli3K436R , Gli3K595R , Gli3KRKR , Gli3PAK436R , Gli3PAK595R , Gli3PAKRKR , and the methylase mutant of Set7 , Set7H297A , were generated by site-directed mutagenesis kit ( TOYOBO , Osaka , Japan ) . The DNA sequences of the mutants were verified by Sanger DNA sequencing . The GST-tagged expression plasmid , GST-Set7 and the MBP-tagged expression plasmids , MBP-K436 , MBP-K595 , MBP-K436R and MBP-K595R were cloned into PGEX3 vector ( Promega , Madison , WI ) and expressed in E coil . HEK293T , A549 and NIH-3T3 were purchased from Cell Bank , The Culture Collection , Chinese Academy of Sciences ( CBTCCCAS ) . CBTCCCAS has provided certifications for the source , identity and mycoplasma contamination of these cell lines . HEK293T cells and A549 cells were cultured in DMEM medium ( HyClone , Logan , UT ) with 10% FBS ( Gibco , Wellesley Hills , MA ) and transfected with Lipofectamine2000 ( Invitrogen , Waltham , MA ) according to the manufacturer’s protocol . NIH-3T3 cells were cultured in DMEM medium ( HyClone , UT ) with 10% NCS ( Gibco ) and transfected with Polyfect ( QIAGEN , Hilden , Germany ) according to the manufacturer’s protocol . Recombinant Shh protein ( R&D , Minneapolis , MN ) or SAG ( CALBIOCHEM , Darmstadt , Germany ) was prepared in DMEM medium with 0 . 5% NCS and added to cells 24 hr after transfection . Gli1-/-;Gli2-/- MEF cells ( Lipinski et al . , 2008 ) were kindly provided by Dr . Steven Y . Cheng . HEK293T cells were lysed using NP-40 buffer ( 50 mM Tris-Cl pH8 . 0 , 0 . 1 M NaCl , 10 mM Sodium fluoride , 1 mM Sodium vanadate , 1% NP-40 , 10% Glycerol , 1 . 5 mM EDTA , Protease Inhibitor Cocktail ) . Cell lysates were rotated for 30 min at 4°C and cleared by centrifuge at full speed . Lysates were then incubated with M2 beads ( Sigma-Aldrich , St Louis , MO ) for 24 hr at 4°C . After 4-time wash ( 15 min/time ) with NP-40 buffer , the protein samples were collected by boiling in 1×SDS loading buffer and subjected to standard SDS-PAGE and Western Blot . Primary antibodies used in this study: Mouse anti-flag ( Sigma-Aldrich ) , Mouse anti-Set7 ( Cell signaling , Danvers , MA ) , Goat anti-Gli3 ( R&D ) , Rabbit anti-GAPDH ( Sigma-Aldrich ) , Rabbit anti-monomethylated K436 ( Me-K436 , Abcolony , China ) , Rabbit anti-monomethylated K595 ( Me-K595 , Abcolony , China ) . Please also refer the figure legend of Figure 1—figure supplement 3 for more detailed information . Secondary antibodies used in this study were purchased from Millipore Company . Flag tagged Gli3 were transfected into the HEK293T cells . Cells were collected 48 hr after transfection . Then IP flag using M2 beads as described above . The protein samples were digested by Filter Aided Sample Preparation ( FASP ) Method ( Wisniewski et al . , 2009 ) . The tryptic peptides were separated by nanoflow liquid chromatography and analyzed by tandem mass spectrometry ( Thermo Electron Finnigan ) . The LTQ-Orbitrap equipped with an NSI nanospray source ( 1 . 5 kV ) was operated in data-dependent mode , in which the normalized collision energy was 35% . Full scan was detected in the Orbitrap analyzer ( R=60 , 000 at m/z 300 ) followed by MS/MS acquisition of the ten most-intense ions in LTQ . Mass calibration used an internal lock mass ( m/z 445 . 120025 ) , the dynamic exclusion repeat count was 1 , the repeat duration was 30 s , and the exclusion duration window was 120 s . Raw Orbitrap full-scan MS and ion trap MS2 spectra were processed by MaxQuant 1 . 3 . 0 . 5 ( Cox and Mann , 2008 ) . A composite target-decoy database was created with the program Sequence Reverser from the MaxQuant package . All identified MS/MS spectra were searched against this target/decoy database ( Human UniProtKB/Swiss-Prot database , 2014-10-29 download ) . Spectra were initially searched with a mass tolerance of 7 ppm in MS and 0 . 5 Da in MS/MS and strict trypsin specificity . Cysteine carbamidomethylation was searched as a fixed modification , whereas N-acetyl protein , oxidized methionine , and mono-methylated lysine ( putative methylation site ) were searched as variable modification . The estimated false discovery rate ( FDR ) of all peptide and protein identifications was fixed at maximum 1% ( Olsen et al . , 2006 ) . Prepare a 100 µl reaction mixture containing 75 µl MAB buffer ( 50 mM Tris pH8 . 5 , 20 mM KCl , 10 mM MgCl2 , 10 mM BME , 250 mM Sucrose ) , 4 µl of 3H labelled S-adenosyl methionine ( 3H-SAM ) ( Perkin elmer ) , 4 µl of GST-Set7 protein ( at 1 mg/ml ) , 10 µl 10×BSA and 4 µl of 10×protease inhibitor cocktail . Add 20 µl Mix contents in five new tubes . Transfer 5 µl ( 5 mg/ml ) purified MBP , MBP-K436 , MBP-K436R , MBP-K595 and MBP-K595R into the tubes , respectively . Pipette up and down gently and incubate the tubes at 37°C for overnight . To stop the reaction , add 8 µl of 4×SDS-PAGE loading buffer to each tube and heat the samples at 95°C for 5 min . Then run the SDS-PAGE gel and dry the gel . Put the dried gel into the film cassette without the wrap and a piece of Kodak BioMax MS film ( Sigma ) . Put the cassette at -80°C freezer for appropriate time and develop the film . Sequences of RNAi oligonucleotides are as follows: Nonspecific small interfering RNA ( siRNA ) : UUCUCCGAACGUGUCACGU Set7 siRNA-1 sense strand: CCUUUGAUCUGUAUCUCUUTT Set7 siRNA-2 sense strand: GGACCUAAUACUGUUAUGUTT Gli3 siRNA-1 sense strand: CCCGUGGGUAUGUCUAUAUTT Gli3 siRNA-2 sense strand: GCUCUAAGUAGGUAUUUAATT All RNAi oligonucleotides were purchased from Shanghai GenePharma Company , China . These RNAi oligonucleotides were transfected into cells by using the Lipofectamine RNAi mix transfection kit ( Invitrogen ) according to the manufacturer’s instructions . The siRNA in paper represent siRNA-1 . Set7 shRNA-1 sense strand: CCGGCCGTGTTCAGAGATACCAAATCTCGAGTATCTCTGAACACGGTTTTTG; Set7 shRNA-1 antisense strand: AATTCAAAAAACCGTGTTCAGAGATACCAAATCTCGAGTCTCT GAACACGG; Set7 shRNA-2 sense strand: CCGGCCTAATACTGTTATGTCGTTTCTCGAGAAACGACATAACAGTATTAGGTTTTTG; Set7 shRNA-2 antisense strand: AATTCAAAAACCTAATACTGTTATGTCGTTTCTCGAGAAACGACATAACAGTATTAG . This shRNA was cloned into the PLKO vector and transfected into HEK293T cells to make pseudoviruses . Then , the pseudoviruses were collected to infect the indicated cells . The lysis tubes ( Precellys ) were prepared and added 200 µl denature buffer ( 50 mM Tris-HCl , 0 . 5 mM EDTA , 1% SDS and proteinase inhibitor cocktail ) before use . The mouse embryos at 14 . 5 dpc were dissected and carefully put the indicated tissues into these tubes until all the embryos were dissected . Using the Precellys Homogenizers to homogenate tissues . Add 4×SDS loading buffer before western blot . NIH-3T3 cells cultured in 24-well plates were co-transfected with 1 . 15 µg 8×GliBS luciferase reporter , 115 ng pRL-TK Renilla , 230 ng indicated constructs per well . 48 hr after transfection , the luciferase activity was measured by Dual-GloTM luciferase assay system ( Promega ) in triplicate . Statistical significance was determined using student t-test . Cells were lysed in TRIzole ( Invitrogen ) and then performed the RNA isolation following the standard protocol . An optimized amount of RNA was used for Reverse transcription using ReverTra Ace qPCR RT Master Mix with gDNA Remover ( TOYOBO , Japan ) . Real-time PCR was performed on the CFX96 TouchTM Real-Time PCR Detection System ( BIO-RAD ) with SYBR Green Real time PCR Master Mix ( TOYOBO , Japan ) . 2-Ct method was adopted to quantify the results . Statistical significance was determined using student t-test . The primer pairs used for real-time PCR were listed below . Set7_Mus_RT_F: CACTCCTTCACTCCGAACTG Set7_Mus_RT_R: TTCAGCTCCACTTGATACCAC Gli3_Mus_RT_F: AGAAGCCCATGACATCTCAG Gli3_Mus_RT_R: GGTCTGCTACACTACCTCCA Gli1_Mus_RT_F: CTGAGACGCCATGTTCAATCC Gli1_Mus_RT_R ACCAGAAAGTCCTTCTGTTCCC Ptch1_Mus_RT_F: ACTACCCGAATATCCAGCACC Ptch1_Mus_RT_R: ATCCTGAAGTCCTTGAAGCCA GAPDH_Mus_RT_F: GAGAAACCTGCCAAGTATGATGAC GAPDH_Mus_RT_R: TGGAAGAGTGGGAGTTGCTG Set7_homo_RT_F: CATTTCTACCAATGCTCTTCTTCC Set7_homo_RT_R: ACATAACAGTATTAGGTCCCACAG Gli1_homo_RT_F: GGGATGATCCCACATCCTCAGTC Gli1_homo_RT_R: CTGGAGCAGCCCCCCCAGT Ptch1_homo_RT_F: CCACAGAAGCGCTCCTACA Ptch1_homo_RT_R: CTGTAATTTCGCCCCTTCC GAPDH_homo_RT_F: GAGTCAACGGATTTGGTCGT GAPDH_homo_RT_R: GACAAGCTTCCCGTTCTCAG NIH-3T3 cells were plated in 24-well plates , and fixed in 4% formaldehyde for 20 min . Cells were permeabilized with PBS/0 . 5% Triton X-100 for 3 min and nonspecific binding sites were blocked with 2% BSA in PBST ( PBS with 0 . 5% Tween-20 ) . Cells were stained with primary antibodies diluted in 2% BSA/PBST for 1 hr at room temperature . After washing four times with PBST , cells were incubated for 1 hr with appropriate secondary antibodies together with 1 g/ml of 4 , 6-diamidino-2-phenylindole ( DAPI ) in 2% BSA/PBST . Leica LAS SP5 confocal microscope was employed to take images . Primary antibodies used in this study: Mouse anti-ac-tubulin ( Sigma ) . Secondary antibodies used in this study were bought from Millipore Company . Cells were cross-linked for 10 min at 37°C by adding formaldehyde to a final concentration of 1% . The cross-linking was stopped by adding glycine to a final concentration of 0 . 125 M . Fixed cells were then washed with PBS , and sonicated in sonication buffer ( 50 mM Hepes-KOH , pH7 . 5 , 140 mM NaCl , 1 mM EDTA , pH8 . 0 , 1% Triton X-100 , 1% sodium deoxycholate , 0 . 1% SDS , and proteinase inhibitor cocktail ) with a Bioruptor Sonicator to yield genomic DNA fragments with size of about 250bp . Lysates were centrifuged , collected and incubated with M2 beads overnight on a rotator at 4°C . Beads were washed 4 times with ChIP wash buffer ( 0 . 1% SDS , 1% Triton X-100 , 2 mM EDTA , pH8 . 0 , 150 mM NaCl , and 20 mM Tris-Cl , pH8 . 0 ) and finally washed with ChIP final wash buffer ( 0 . 1% SDS , 1% Triton X-100 , 2 mM EDTA , pH8 . 0 , 500mM NaCl , and 20 mM Tris-Cl , pH8 . 0 ) . Genomic DNA was eluted with elution buffer ( 1% SDS and 100 mM NaHCO3 ) at 65°C for 30 min . 5 M NaCl was add to a final concentration of 200 mM for a further incubation at 65°C for 4 hr or overnight . Proteins were then removed by 0 . 25 mg/ml proteinase K in a 5 mM EDTA solution after 2 hr incubation at 55°C . Genomic DNA was purified using DNA purification kit ( QIAGEN ) for Real-time PCR . Statistical significance was determined using student t-test . Primer pairs used in this study are listed below . a-Forward: GGAGAGCAATTAGGAAGTTTGG a-Reverse: GGAGAGCAATTAGGAAGTTTGG b-Forward: TCTCTAGCTTCTATCCACCCA b-Reverse: TCTCTAGCTTCTATCCACCCA c-Forward: GATTGGACTCCTGACCTGTG c-Reverse: CATGTTAGGAAACCCACCCA Hhip-Forward: CTATATTAAGCCCAGACTTTCCAG Hhip-Reverse: CATCCCAGTGCAATGTTCAG Ptch1-Forward: TGGAGGGCAGAAATTACTCAG Ptch1-Reverse: TAATGGAAGTATTGCATGCGAG Cells were seeded in 96-well plates at 5000 cells/well . Relative cell growth rate was determined by MTT assay daily . 20 μl of MTT working solution ( 5 mg/ml ) was added into each well and incubated for 4 hr . Then discard the supernatant and dissolve the MTT formazan in 100 μl DMSO . The absorbance was measured at the wavelengths of 595 nm and 630 nm . Statistical significance was determined using student t-test . Cells ( 5×105 ) were seeded in 3 . 5 cm dishes and incubated overnight . The cell monolayers were scratched horizontally using a sterile 10 μl pipette tip . The floating cells were removed with PBS and cultured again in DMEM without serum . Photographic images along the scrape line were obtained at indicated time points by an inverted microscope . Briefly , 1ml culture medium with 1% agar was first plated into 3 . 5 cm dishes . After the bottom agar became solidified , each well received another 1 ml 0 . 3% agar in culture medium carrying 5000 cells . Cells were covered with 1 ml of complete medium and medium was changed every four days . Colonies were stained with 0 . 005% crystal violet and counted after 3-weeks of incubation . The transwell migration assay was carried out using a 24-well transwell cell culture chamber ( Corning , New York ) with an 8 μm pore size . Briefly , 1×104 cells were added into an upper chamber with 200 μl DMEM without serum . 700 μl DMEM containing 20% FBS was added into the lower chamber as a chemo attractant . After 48 hr , the non-migration cells were manually removed with a rubber swab . Cells that migrated to the lower side of the membrane were stained with crystal violet and photographed using an inverted microscope . The transwell invasion assay was carried out using a 24-well transwell cell culture chamber ( Corning , New York ) . The 8 μm pore size upper chambers were coated with growth factor-reduced Matrigel ( BD ) . Briefly , A549 cells were incubate in serum free medium for 24 hr . Matrigel were diluted three times with serum free DMEM . Add 50 μl above mixture to the upper chamber and incubate for at least 4 hr . Then 2×104 cells were added into an upper chamber with 200 μl DMEM without serum . 700 μl DMEM containing 20% FBS was added into the lower chamber as a chemoattractant . After 48 hr , the non-invasion cells were manually removed with a rubber swab . Cells that invade to the lower side of the membrane were stained with crystal violet and photographed using an inverted microscope . Female nude mice were obtained from Shanghai Experimental Animal Center and maintained in pathogen-free conditions . Exponential phase A549 cells were trypsinized and washed with PBS ( HyClone ) and suspended in fresh PBS in 4°C . 5X106 cells were injected subcutaneously into four-week-old female nude mice on both left and right flank . Tumor size was measured every 7 days , and tumor volume was estimated using the formula: tumor volume=0 . 5×length×width2 . At the end of this experiment , tumors were harvested , weighed and photographed . All procedures for animal experimentats were performed in accordance with the Institutional Animal Care and Use Committee guidelines of the Animal Core Facility of the Institutes of Biochemistry and Cell Biology ( SIBCB ) . The approval ID for using the animals was 087 by the Animal Core Facility of SIBCB . All the experiments were repeated more than twice .

Cells in mammals need to be able to communicate with each other to enable them to work together in tissues and organs . A signaling pathway called Hedgehog signaling plays a crucial role in carrying information between cells in developing embryos , but if it is active at other times it can also promote the development of cancers . The Hedgehog signaling pathway regulates the activity of several proteins , including one called Gli3 . When the Hedgehog signaling pathway is not active , Gli3 is able to switch off certain genes in the cells . On the other hand , when the signaling pathway is active , Gli3 changes shape so that it is able to activate its target genes instead . It is thought that this shape change is triggered by the addition ( or removal ) of chemical tags to Gli3 . So far , researchers have reported that several different types of chemical tags can modify the activity of Gli3 . However , it is not clear whether another type of chemical tag – known as a methyl tag – is involved in regulating Gli3 . Fu et al . studied Hedgehog signaling in mice . The experiments show that an enzyme called Set7 can modify Gli3 by adding methyl tags to certain sites in the protein . This modification makes the protein’s structure more stable and helps it to bind to the target genes . Further experiments show that these methyl groups contribute to the progression of lung cancer . Fu et al . ’s findings expand our understanding of how chemical tags can alter the cells’ response to Hedgehog signaling activity . Future challenges are to understand exactly how Set7 and Gli3 interact and to develop drugs that can block this interaction , which may have the potential to treat cancer .

2016

Set7 mediated Gli3 methylation plays a positive role in the activation of Sonic Hedgehog pathway in mammals

Oesophageal adenocarcinoma ( OAC ) is one of the most common causes of cancer deaths . Barrett’s oesophagus ( BO ) is the only known precancerous precursor to OAC , but our understanding about the molecular events leading to OAC development is limited . Here , we have integrated gene expression and chromatin accessibility profiles of human biopsies and identified a strong cell cycle gene expression signature in OAC compared to BO . Through analysing associated chromatin accessibility changes , we have implicated the transcription factor KLF5 in the transition from BO to OAC . Importantly , we show that KLF5 expression is unchanged during this transition , but instead , KLF5 is redistributed across chromatin to directly regulate cell cycle genes specifically in OAC cells . This new KLF5 target gene programme has potential prognostic significance as high levels correlate with poorer patient survival . Thus , the repurposing of KLF5 for novel regulatory activity in OAC provides new insights into the mechanisms behind disease progression . Oesophageal cancer is the eighth most common cancer worldwide , and its 5-year survival rate of 15% makes it the sixth most-common cause of cancer-related death ( Ferlay et al . , 2015; Pennathur et al . , 2013 ) . A subtype of oesophageal cancer , oesophageal adenocarcinoma ( OAC ) , is the predominant subtype in many Western countries and its incidence is rising rapidly ( Coleman et al . , 2018 ) . Patients with OAC often present at a late stage with advanced disease ( Smyth et al . , 2017 ) . The lack of molecular knowledge of OAC , combined with lack of tailored therapies , contribute to the low survival of OAC patients . The accepted model of OAC development is the progression from an intestinal metaplastic condition of the lower oesophagus , known as Barrett’s oesophagus ( BO ) , to OAC through increasing stages of dysplasia ( Burke and Tosh , 2012; Spechler and Souza , 2014 ) . Many mutations found in OAC are also present in BO , especially TP53 , which suggests a stepwise transition to OAC ( Ross-Innes et al . , 2015; Stachler et al . , 2015 ) . Focal amplifications differ as they largely occur in OAC compared to BO ( Lin et al . , 2012; Stachler et al . , 2015; Yamamoto et al . , 2016 ) . The amplified genes can be grouped into functional biological pathways with the RAS-ERK signalling pathway ( e . g . ERBB2; EGFR; KRAS ) and GATA transcription factors ( GATA4; GATA6 ) being the most common ( Frankell et al . , 2019; Lin et al . , 2012; Cancer Genome Atlas Research Network et al . , 2017 ) . The morphology of BO differs from the oesophageal epithelia by the presence of a columnar epithelium and secretory goblet cells , rather than squamous epithelium ( reviewed in Spechler and Souza , 2014 ) . Genomic and transcription events have been observed to differ between BO and OAC . Mutations in TP53 are more frequent in BO from patients that had progressed to OAC ( Stachler et al . , 2018 ) and SMAD4 mutations appear to occur exclusively in OAC , although at a low frequency ( Weaver et al . , 2014 ) . Increased TGFβ signalling through other SMAD family members , SMAD2/3 , promotes growth in OAC cells ( Blum et al . , 2019 ) . Additionally , increased expression and increased activity of AP-1 transcription factors occurs in the transition from BO to OAC ( Blum et al . , 2019; Britton et al . , 2017; Maag et al . , 2017 ) . Despite these studies , the definitive molecular mechanisms of progression to OAC are poorly understood and biomarkers to identify patients at risk of progression are lacking . Changes to the chromatin landscape have been implicated in many cancers and chromatin accessibility changes during tumourigenesis are a major factor in altering regulatory element activity ( Britton et al . , 2017; Corces et al . , 2018; Davie et al . , 2015; Denny et al . , 2016; Kelso et al . , 2017; Rendeiro et al . , 2016; Tome-Garcia et al . , 2018; Zhou and Guo , 2018 ) . We recently used Assay for Transposase-Accessible Chromatin using sequencing ( ATAC-seq ) to ascertain the molecular basis of BO and identified a set of transcription factors that define the BO chromatin landscape and are retained in OAC ( Rogerson et al . , 2019 ) . Here , we took a similar approach to discover important transcriptional regulators ( Figure 1A ) that are specifically operational in OAC and hence contribute to the molecular basis of disease progression from BO to OAC . We compared the open chromatin landscape in BO and OAC patient biopsies and uncovered KLF5 as an important transcriptional regulator that is repurposed to directly drive a cell cycle gene expression signature during the progression of BO to OAC . To begin to understand the molecular events that distinguish OAC form the BO precursor state we first established the differential gene expression profiles between BO and OAC . We analysed public human BO and OAC RNA-seq data ( Maag et al . , 2017 ) . These samples separate well after principal component analysis ( PCA ) , therefore we retained all samples for further analysis ( Figure 1—figure supplement 1A ) . Performing differential gene expression analysis , we identified 905 differentially expressed genes between BO and OAC ( ±1 . 5 x; Q-value <0 . 05; Figure 1B; Supplementary file 1 ) . Of these 905 genes , 465 are upregulated in OAC and 440 are downregulated in OAC compared to BO . To validate these findings , we analysed RNA-seq data from our own sample collection ( 3 BO and 3 OAC ) . Genes that were upregulated in OAC from the discovery dataset were significantly upregulated in the validation dataset and likewise for downregulated in OAC genes ( Figure 1C ) . To gain insights into biological pathways behind these differentially expressed genes , we used two approaches . Firstly , Gene Set Enrichment Analysis ( GSEA ) uncovered two cell cycle associated terms , ‘G2M checkpoint’ and ‘E2F1 targets’ , as the most significant upregulated gene sets in OAC ( Figure 1D ) . Conversely , ‘Fatty acid metabolism’ and ‘p53 pathway’ are the most significant downregulated gene sets ( Figure 1—figure supplement 1B ) . Secondly , biological pathway gene ontology analysis of upregulated genes revealed many cell cycle associated terms , such as ‘Nuclear division’ , ‘Regulation of mitotic cell cycle’ and ‘DNA replication’ ( Figure 1E ) . Example genes such as CDC25B , CENPI and E2F1 all showed significant upregulation in OAC compared to BO in both datasets ( Figure 1—figure supplement 1D ) . Downregulated genes uncovered metabolic associated terms , such as ‘alcohol metabolic process’ , ‘monocarboxylic acid metabolic process’ and ‘Lipid catabolic process’ ( Figure 1—figure supplement 1C ) . Representative example genes from these pathways such as IDI1 , ADH4 and CIDEC all show significant downregulation in both datasets ( Figure 1—figure supplement 1E ) . These initial results indicate a strong upregulation of genes associated with cell cycle processes during the progression from BO to OAC accompanied with the inactivation of genes controlled by the p53 pathway and genes associated with metabolism . To identify putative transcriptional regulators that may drive the transition to OAC and impact on this enhanced cell cycle profile , we analysed the accessible chromatin landscape using ATAC-seq from patient biopsies . To supplement our previous ATAC-seq datasets from BO and OAC patients ( Britton et al . , 2017; Rogerson et al . , 2019 ) , we performed ATAC-seq on two additional OAC biopsies , which were quality-checked and reproducible ( Figure 2—figure supplement 1A and B ) . We wanted to focus on the differentially expressed genes in OAC compared to BO , therefore we generated a set of accessible regions representing potential regulatory regions that are associated with this set of genes . We took all ATAC-seq peaks from all samples within +/- 250 kb of a TSS of a differentially expressed gene ( Figure 2A ) . After merging so that only unique peaks remained , 35 , 220 regions were used for further analyses ( Supplementary file 2 ) . We first performed principal component analysis on normalised ATAC-seq signal of all BO and OAC samples to identify differences between samples ( Figure 2B ) . This led to clustering of all BO samples and clustering of most OAC samples . OAC samples T_003 and T_005 did not cluster with the other OAC samples and were therefore removed from the subsequent differential accessibility analysis . We then carried out differential accessibility analysis between BO and OAC on this peak set ( Figure 2C; Supplementary file 2 ) . A total of 1495 regions were significantly differentially accessible ( ±2 x; Q-value <0 . 1 ) , the majority of which increased in accessibility ( 1327/1495 ) . An example gene locus which shows differential accessibility in OAC is centred on KRT19 ( Figure 2D ) . Within this locus , both gene promoters ( 14%; 1/7 ) and distal regulatory regions ( 86%; 6/7 ) gain accessibility in OAC . To assess whether the observed changes of accessibility near differentially expressed genes are common to other OACs , we compared our ATAC-seq data to independent , previously published ATAC-seq datasets from TCGA-ESCA oesophageal adenocarcinoma samples ( Figure 2D , bottom; Figure 2E; Corces et al . , 2018 ) . TCGA-ESCA samples showed similar open chromatin peak profiles and clustered with our OAC samples with the exception of one sample , which clusters with our BO samples ( Figure 2E ) . The chromatin accessibility profiles nearby genes differentially expressed in OAC are therefore reproducible across patients . Next , we harnessed the differential accessibility data to uncover the identities of transcription factors bound in these regions . De novo motif discovery of regions that become more accessible in OAC contain significantly enriched motifs for AP-1 , KLF , TBX , NFκB and p53 transcription factor families ( Figure 3A; Supplementary file 3 ) . AP-1 and KLF were clearly the most frequent motifs in the differential regions and showed the strongest match score for the consensus motif . Regions that showed decreased chromatin accessibility in OAC are enriched in EWSR1-FLI1 , ASCL2 , GLI2 , E2F and ZBTB18 motifs , albeit with relatively low match scores ( Figure 2—figure supplement 1C; Supplementary file 3 ) . To further assess which transcription factors might be involved in gene expression control , we carried out footprinting analysis on differential accessible regions from our ATAC-seq datasets ( Figure 3B; Bentsen et al . , 2020 ) . In differential accessible regions , motifs for KLF ( e . g . KLF4 , KLF5 and KLF1 ) and AP-1 ( e . g . FOS , JUNB , JUND , JUN and FOSL1/2 ) transcription factors showed the highest footprinting score in OAC , whereas motifs for homeobox transcription factors ( e . g . HNF1A , HOXA5 and NKX2-5 ) , ARID3A and MEF transcription factors ( e . g . MEF2A and MEF2C ) showed more footprinting in BO . To provide more evidence for transcription factor occupancy , we then plotted ATAC-seq signal across their motifs . Both FOS ( AP-1 ) and KLF4 ( KLF ) motifs show a clear increase in footprint depth in OAC , indicative of more transcription factor binding ( Figure 3C ) . We have previously identified AP-1 as an important regulator in OAC ( Britton et al . , 2017 ) , but the role of KLF transcription factors in OAC is poorly understood . We therefore focussed on the potential role of KLF transcription factors in the progression of BO to OAC . To identify a specific KLF transcription factor that may be bound to these accessible regions , we analysed the expression of individual KLF transcription factors in OAC samples ( Figure 3D ) . KLF5 was clearly the highest expressed among the KLF family in OAC . KLF5 has been previously implicated in oesophageal squamous cell carcinoma as a tumour suppressor ( Tarapore et al . , 2013 ) and has been identified as pro-tumorigenic in gastric cancer via amplifications ( Chia et al . , 2015 ) . To determine the gene regulatory functions of KLF5 , we carried out siRNA-mediated knockdowns of KLF5 in OE19 cells , a cell line we identified as having a similar chromatin landscape to OAC biopsies ( Rogerson et al . , 2019 ) and exhibits strong tumourigenic properties ( Hassan et al . , 2017 ) . Knockdown of KLF5 was evident after 3 days siRNA transfection ( Figure 3—figure supplement 1A ) and RNA-seq replicates were highly correlative ( Figure 3—figure supplement 1B ) . Carrying out differential expression analysis identified 4934 genes ( 2637 upregulated and 2297 downregulated ) with significant changes in gene expression ( ±1 . 3 x; Q-value <0 . 05; Figure 3—figure supplement 1C; Supplementary file 4 ) . Biological pathway GO term analysis revealed several enriched terms including ‘DNA replication’ and ‘Regulation of mitotic cell cycle’ for downregulated genes , and terms involving ‘Oxidative phosphorylation’ and ‘mitochondrial gene expression’ for upregulated genes ( Figure 3—figure supplement 1D and E ) . The terms associated with downregulated genes are reminiscent of the terms enriched in genes upregulated in OAC ( see Figure 1 ) . Moreover , GSEA also found similar gene sets: ‘mitotic spindle’; ‘G2M checkpoint’ and ‘E2F targets’ for downregulated genes and ‘oxidative phosphorylation’; ‘xenobiotic metabolism’ and ‘fatty acid metabolism’ for upregulated genes ( Figure 3—figure supplement 1F and G ) . Since the genes regulated by KLF5 are involved in similar processes as the genes aberrantly expressed in OAC , we asked whether any of the same genes are in each dataset . 21% ( 97/465 ) of the genes upregulated in OAC significantly overlap with those downregulated with siKLF5 ( Figure 3E ) and many of these are associated with cell cycle related functions , including genes encoding core cell cycle proteins like CCNE1 , E2F1 and various MCM proteins ( Figure 3—figure supplement 1H ) . Further analysis of the biological pathways enriched within these 97 genes identified very similar GO terms to those enriched in genes upregulated in OAC compared to BO ( Figure 3F ) . GSEA also identified the same gene set terms: ‘G2M checkpoint’ and ‘E2F1 targets’ ( Figure 3G ) . Next , we asked whether these genes are directly regulated by KLF5 , and carried out replicate ChIP-seq for KLF5 in OE19 cells which were highly correlated ( Figure 3—figure supplement 2A and B ) . We therefore took the overlap of peaks between biological replicates forward for downstream analyses , resulting in 13 , 542 peaks ( Figure 3—figure supplement 2C; Supplementary file 5 ) . These peaks are highly enriched in the KLF5 motif , demonstrating the validity of the dataset , and also in AP1 ( FRA1 ) and GATA ( GATA6 ) motifs , which we have previously revealed in genome wide studies as implicated in OAC ( Britton et al . , 2017; Rogerson et al . , 2019; Figure 3—figure supplement 2D ) . Focussing on the 97 genes that are upregulated in OAC and also downregulated after KLF5 depletion , 97% have a KLF5 ChIP-seq peak within 0 . 5 Mb of the TSS and the median distance between a KLF5 ChIP-seq peak and the TSS of all significantly downregulated genes was 11 , 975 bp ( Figure 3—figure supplement 2E ) . In contrast , KLF5-binding regions are further away ( >20 kb ) from the TSS of genes that were either unaffected by KLF5 depletion or whose expression was increased . This is indicative of direct activation by KLF5 . An example gene is CDC25B which harbours multiple KLF5 ChIP-seq peaks surrounding its locus ( Figure 3H ) . Collectively , these results suggest a direct activator role of KLF5 in controlling cell cycle genes in OAC . Having determined a role for KLF5 in controlling cell-cycle-associated gene expression in OAC cells , we sought to determine the mechanism through which KLF5 acquires these functions . We first asked whether KLF5 expression changes in the transition from BO to OAC , however no increase in expression was found ( Figure 4A ) . An alternative mechanism might be through redistributing the binding of KLF5 to different regulatory elements in OAC . We therefore hypothesised that KLF5 is active in both BO and OAC but may regulate specific genes in OAC by binding at different loci . CP-A cells are derived from non-dysplastic BO and do not exhibit strong tumourigenic properties ( Lin et al . , 2012 ) so we compared KLF5 expression in BO-derived CP-A and OAC-derived OE19 cells and found that KLF5 is expressed at similar levels in ( Figure 4—figure supplement 1A ) . We therefore we used these cell-lines to model KLF5 activity in BO and OAC . To gain a more comprehensive view of KLF5 function , we performed ChIP-seq for KLF5 in CP-A cells and used spike-in normalisation to better assess differential binding relative to OE19 cells . Anti-KLF5 antibodies precipitated KLF5 in CP-A cells ( Figure 4—figure supplement 1B ) and biological replicates were highly reproducible ( Figure 4—figure supplement 1C ) . We took the overlap of peaks between biological replicates forward for downstream analyses , resulting in 13 , 526 peaks ( Figure 4—figure supplement 1D; Supplementary file 5 ) . Motif analysis showed high enrichment of the KLF5 motif further demonstrating the quality of the data ( Figure 4—figure supplement 1E ) . KLF5 peaks from CP-A and OE19 cells were merged , generating a combined peak set of 21 , 353 peaks . Differential binding analysis revealed an altered KLF5 binding profile between CP-A and OE19 cells ( Figure 4B , C; Supplementary file 5 ) : 8608 peaks show more binding in OE19 cells ( 40% ) , 9230 peaks are shared between the two cell-lines ( 44% ) and 3515 show more binding in CP-A cells ( 16% ) . An example locus is CCNE1 , which demonstrates increased KLF5 binding in OE19 cells at the promoter and putative enhancers associated with open chromatin regions ( Figure 4D ) . Reciprocally , specific binding of KLF5 in CP-A cells is evident at the NKX3-1 locus and common binding of KLF5 in both CP-A and OE19 cells is evident at the JAG1 locus ( Figure 4—figure supplement 1F ) . We next asked whether the regions that exhibit differential binding are enriched for specific transcription factor motifs . Regions that are bound by KLF5 in OE19 cells are enriched in motifs for KLF , GATA , Forkhead , AP-1 and TCF transcription factors , whereas regions bound by KLF5 in CP-A cells are enriched for a different set of motifs with , TEAD , RUNX and p53 transcription factor families in addition to KLF and AP-1 motifs detected ( Figure 4E; Supplementary file 6 ) . These results are in-keeping with our previous work showing AP-1 and GATA6 functionality in OAC ( Britton et al . , 2017; Rogerson et al . , 2019 ) . Regions specifically bound by KLF5 in OAC cells also exhibited increased accessibility in OE19 cells and importantly , accessibility is also elevated around these binding sites in OAC tissue ( Figure 4—figure supplement 2A ) . These findings are therefore consistent with a broad role of KLF5 in OAC . To further probe the potential biological significance of the differentially bound KLF5 regions , we associated these with the nearest gene and determined the enriched GO terms for genes associated with cell-type-specific KLF5 peaks that also show preferential expression in BO or OAC . OE19-specific KLF5-binding events are associated with genes involved in ‘cell division’ control , whereas CP-A-specific KLF5 binding is associated with ‘epithelial cell differentiation’ ( Figure 4—figure supplement 2B ) . The latter observation is consistent with the potential loss of cell identity in OAC . However , since oncogenic events during the progression from BO to OAC are poorly understood , we decided to focus on regions that acquire KLF5 binding in OE19 cells . To relate specific KLF5-binding events to gene expression changes , we took the set of 97 genes that are upregulated in OAC and downregulated by KLF5 depletion ( i . e . activated by KLF5; Figure 3E ) and found that there are 371 OE19-specific KLF5-binding peaks within a 0 . 5 Mb locus centred on the TSS . To further explore how KLF5 activates these genes in OAC , we assessed the transcription factor binding motif distribution ( identified in Figure 4E ) within this set of OE19-specific KLF5 peaks . We detected KLF5 binding motifs in 257/371 of these regions , and strikingly , 56% ( 145/257 ) of the peaks also house a mixture of FOXA , AP-1 , GATA and TCF motifs , in addition to the KLF motif ( Figure 4F; Figure 4—figure supplement 2C; Supplementary file 7 ) . This suggests that KLF5 functions in a combinatorial manner with these other transcription factors to activate gene transcription during progression from BO to OAC . However , a large portion of these peaks ( 44% ) contain only a KLF motif suggesting a more independent role for KLF5 in these regions ( Figure 4F ) . To test whether these motif enrichments reflect transcription factor binding , we integrated our ChIP-seq data of transcription factors active in OAC ( GATA6 and HNF4A; Rogerson et al . , 2019 ) , with KLF5-binding data . Since only GATA motifs are enriched in these regions we would expect co-binding with GATA6 and not HNF4A . We therefore compared ChIP-seq profiles for these transcription factors , and see extensive co-binding of GATA6 at these sites but no evidence of co-binding with HNF4A ( Figure 4—figure supplement 2D ) . Finally , the predicted target gene co-regulation by KLF5 and GATA6 was validated by depletion of each factor in OE19 cells , which leads to a large significant overlap in downregulated genes ( Figure 4G ) . However , this co-regulated gene set contains only two of the cell cycle associated genes regulated by KLF5 , suggesting that this combination of transcription factors is not directly involved in controlling this process . Turning back to the cell cycle genes directly activated in OAC through OAC-specific binding of KLF5 , we tested whether KLF5 is relevant for their expression in CP-A cells . As expected from the lower KLF5-binding levels in these cells , depletion of KLF5 had little effect on these genes ( Figure 4—figure supplement 2E ) , consistent with a newly acquired function in OAC cells . To establish whether the 371 KLF5-bound regions that are associated with KLF sensitive genes are relevant to OAC , we turned back to our ATAC-seq data and clustered the data to reveal two clusters . One set of regions is already partially open in CP-A cells that increase in accessibility in OE19 cells ( cluster 1 ) and another set are closed in CP-A cells and become more accessible ( cluster 2 ) ( Figure 4H , Figure 4—figure supplement 2F , left ) . Importantly , the same pattern of accessibility is evident using ATAC-seq signal from BO and OAC tissue ( Figure 4H , Figure 4—figure supplement 2F , right ) . To identify any potential differences between these clusters , we performed motif analysis ( Figure 4—figure supplement 2G; Supplementary file 8 ) . The most common motif in both clusters were KLF motifs and the most striking difference is the large proportion of AP1 motifs specifically associated with cluster one suggesting a potential role for AP1 in priming binding of KLF5 to these regions . Together , these results indicate an altered DNA-binding profile for KLF5 in BO and OAC , and this altered binding is associated with chromatin opening . This altered binding profile for KLF5 in OAC reflects a direct role in controlling genes involved in cell cycle . Our results indicate a role of KLF5 in controlling increased cell cycle gene expression in OAC; however , it is unclear how this relates to genetic events that potentially impact on the same process . Genomic amplifications in signalling receptors are common in OAC , such as ERBB2 ( 32% OAC have an ERBB2 amplification; Cancer Genome Atlas Research Network et al . , 2017 ) and occur during the transition from BO to OAC ( Stachler et al . , 2015 ) . As the ERK pathway is implicated in promoting cell proliferation and is controlled by ERBB2 , we investigated whether ERBB2 signalling impacts on KLF5-mediated gene regulatory events . First , we sought evidence for a link with transcription factor activity , and performed ATAC-seq on OE19 cells to investigate whether depletion of ERBB2 could alter chromatin accessibility . OE19 cells contain an amplification of the ERBB2 locus ( Dahlberg et al . , 2004 ) and are dependent on ERBB2 for their proliferation ( Hong et al . , 2012 ) . ERBB2 levels were efficiently reduced after 72 hr of siRNA treatment and phosphorylation of downstream targets ( ERK and AKT ) was reduced ( Figure 5—figure supplement 1A ) . ATAC-seq data were reproducible and good quality ( Figure 5—figure supplement 1B and C ) . We performed differential accessibility analysis , which identified 717 regions with decreased chromatin accessibility and 733 regions with increased accessibility ( Figure 5A; Supplementary file 9 ) . De novo motif analysis of the regions that exhibit reduced chromatin accessibility following ERBB2 depletion , revealed that the majority contain AP-1-binding motifs as expected from the established connections between ERK pathway signalling and AP1 transcription factors . However , the binding motif for KLF transcription factors was also detected , albeit in a subset of the regions ( Figure 5B; Supplementary file 10 ) . We then used our KLF5 ChIP-seq dataset from OE19 cells to validate KLF5 binding at regions with reduced chromatin accessibility following ERBB2 depletion ( Figure 5—figure supplement 1D ) . These regions are relevant in the context of OAC as they also show increased chromatin accessibility in OAC tissue compared to BO ( Figure 5C ) . The convergence of ERBB2 signalling on KLF5 transcription factor activity suggested that they might also converge on the same genes . We therefore carried out RNA-seq in OE19 cells treated with siRNA against ERBB2 . The RNA-seq data were highly reproducible ( Figure 5—figure supplement 1E ) and resulted in 778 genes down- and 664 genes up-regulated ( two-fold change; FDR < 0 . 05 , FPKM > 1 ) ( Figure 5—figure supplement 1F ) . There is a large , statistically significant overlap between directly activated KLF5 target genes and genes downregulated by ERBB2 depletion . Moreover , a closer comparison reveals that the expression of the majority of the directly activated KLF5 target genes was reduced upon ERBB2 knockdown ( Figure 5D ) . Most of these common target genes are cell cycle related . These results therefore indicate that ERBB2 and KLF5 converge on a similar set of regulatory regions to drive the expression of cell cycle regulatory genes . To establish whether ERBB2 can redistribute KLF5 binding and activate its target genes , we created BO-derived CP-A cell lines that stably over express ERBB2 to mimic the effect of amplification seen in OAC . These cells exhibit high levels of ERBB2 expression , maintain ERK and AKT activation in serum starved conditions ( Figure 5—figure supplement 2A ) , and exhibit growth factor-independent proliferation ( Figure 5—figure supplement 2B ) . Several cell cycle related genes that are activated by KLF5 in OAC cells are also activated by ERBB2 overexpression in BO cells ( Figure 5—figure supplement 2C ) . However , we were unable to detect any increases in KLF5 occupancy at a panel of KLF5-binding regions associated with cell cycle genes ( Figure 5—figure supplement 2D ) . These findings therefore reaffirm the convergence of ERBB2 signalling and KLF5 on the activation of a cell cycle gene signature but ERBB2 is not sufficient to trigger KLF5 redistribution . Finally , we assessed whether defective KLF5-driven cell cycle gene regulation led to proliferative defects in OAC cells . We first depleted KLF5 in OE19 cells using siRNA which resulted in the reduction of KLF5 protein ( Figure 5—figure supplement 3A ) , and the growth of cells was significantly impeded after siKLF5 treatment ( Figure 5E ) . Second , we validated this growth defect by using CRISPR interference technology . Stable transfection of dCas9-KRAB and subsequent transfection of sgRNAs targeting the promoter of KLF5 ( sgKLF5 ) into OE19 cells resulted in the reduction of KLF5 protein levels ( Figure 5—figure supplement 3B ) . CRISPRi knockdown of KLF5 also significantly reduced the growth of OE19 cells ( Figure 5F ) , mirroring the result with siKLF5 . We further explored the role of KLF5 in cell growth and cell cycle progression by performing similar assays while perturbing KLF5 target genes ( CCNE1 , CDC25B , KIF14 , CLSPN and NR4A1 ) . All these genes showed significant reductions in expression upon siRNA treatment ( Figure 5—figure supplement 3C ) . The growth of OE19 cells was significantly reduced with the treatment of siRNA against CCNE1 , KIF14 and CLSPN ( Figure 5G ) . Knockdown of these genes also significantly altered cell cycle patterns , particularly knockdown of CLSPN which induced a prominent S-phase block ( Figure 5—figure supplement 3D ) . These results provide more evidence for the role of KLF5 in the growth of cells and highlight the role of KLF5 target genes in this phenotype . To assess whether the expression of KLF5 and its target genes has any clinical relevance , we sourced OAC expression and survival data ( Cancer Genome Atlas Research Network et al . , 2017 ) and plotted a survival of patients with high and low expression ( ±median ) of KLF5 itself and KLF5 target genes up to 24 months ( Figure 5H ) . Those with a higher expression of KLF5 showed no difference in patient survival , whereas patients with high target gene expression exhibited a significantly lower survival rate compared to those with low expression . This result is in keeping with the hypothesis that it is the activation of KLF5 target genes by its redistribution across chromatin , rather than its expression level that is important . It is noteworthy that CLSPN expression alone is predictive of increased patient survival ( Figure 5H ) and its enhanced expression in OAC compared to BO makes this a useful potential biomarker ( Figure 5I ) . Collectively , these results confirm the functional role of KLF5 in cell cycle control in OAC and convergence of action with the ERBB2 signalling pathway . This is clinically important as patients with highly expressed KLF5 target genes have a worse prognosis that those without . Genome sequencing efforts of patients with BO and OAC have provided insights into the molecular causes of BO and OAC and show the mutational relationships between these disease states ( Ross-Innes et al . , 2015; Stachler et al . , 2015 ) . This has provided evidence for a model of OAC developing from BO . The molecular mechanisms involved in progression to OAC are poorly understood; however , BO offers a therapeutic window of opportunity to identify those more at risk of OAC development . In addition to genetic events , epigenetic changes and alterations to the chromatin landscape are also likely to play an important role in disease progression . Here , we demonstrate that there are marked changed in chromatin accessibility and associated gene expression , indicating active changes at the chromatin level during carcinogenesis . One of the major contributing factors to this change is the transcription factor KLF5 . KLF5 is re-purposed in OAC cells and its chromatin-binding profile is massively rewired to drive increased expression of cell cycle associated genes ( Figure 5J ) . Conversely , this rewiring results in the loss of KLF5 binding to many regulatory regions occupied in Barrett’s cells . This loss is potentially associated with the loss of cell identity , and may also contribute to the development of the cancer phenotype . Cell cycle deregulation is one of the key hallmarks of cancer ( Hanahan and Weinberg , 2011 ) and here we uncovered a cell cycle gene expression signature , comprised of genes that are overexpressed in OAC . Recent research identified the cell-cycle as a perturbed pathway in OAC and suggested the possibility of CDK4/6 inhibitors as a therapeutic treatment ( Frankell et al . , 2019; Mourikis et al . , 2019 ) . We have previously uncovered a deregulated FOXM1 regulatory network active in OAC , a key regulator of late cell cycle gene expression ( Wiseman et al . , 2015 ) . By integrating ATAC-seq data to identify upstream regulators of this signature , we also uncovered AP-1 and KLF5 as putative transcription factors in this process . We have previously identified AP-1 as an important factor in OAC ( Britton et al . , 2017 ) and others have shown an increase in AP-1 family transcription factors between non-dysplastic BO and low-grade dysplastic BO ( Maag et al . , 2017 ) . What is less clear is the role of KLF5 in the progression of BO to OAC . KLF5 has been shown to have a tumour promoting function in pancreatic ( He et al . , 2018 ) and basal-like breast cancer ( Qin et al . , 2015 ) . KLF5 is also frequently amplified in gastric cancer ( Chia et al . , 2015; Zhang et al . , 2018 ) and has recently been shown to regulate gene expression in OAC in combination with other transcription factors , GATA6 , ELF3 and EHF ( Chen et al . , 2020 ) . This was reinforced by a recent study that identified KLF5 as a master transcription factor on which OAC cell-lines were dependent ( Reddy et al . , 2019 ) . Paradoxically , KLF5 has been shown to have a tumour suppressor role in oesophageal squamous cell carcinoma ( Tarapore et al . , 2013 ) and breast cancer ( Chen et al . , 2002 ) . The expression of the related protein , KLF4 , together with three other genes , was able to stratify OAC from BO , albeit KLF4 expression is reduced in progression from BO to OAC ( Maag et al . , 2017 ) . Previous studies have begun to suggest a role for KLF5 in cell cycle control . For example , KLF5 binds to a CCNE1 promoter proximal element in bladder cancer cells ( Pattison et al . , 2016 ) and KLF5 increases the expression of Ccnb1 and Mcm2 downstream of oncogenic Ras in fibroblasts ( Nandan et al . , 2005 ) . Here , we provide evidence that KLF5 exhibits a widespread role; directly controlling cell proliferation through activation of cell cycle associated genes . We also show that reduction of KLF5 levels , or several of its target genes , in OAC cells impairs growth . Indeed , this is exemplified by CLSPN which may have therapeutic potential as its gene product , Claspin , has recently been shown to have a broader role in cancer cell viability by protecting cancer cells from replication stress ( Bianco et al . , 2019 ) . KLF5 directly binds and regulates core cell cycle genes for example CDC25B , CCNE1 and MCM2 , some of which are cell cycle transcription factors for example E2F1 , MYBL2 , thus providing a mechanism for propagating its effects on cell cycle control . We also show KLF5 expression is almost unchanged between BO and OAC . By profiling KLF5 chromatin binding in BO and OAC cells , we have demonstrated an altered KLF5 binding profile . The regions bound by KLF5 specifically in OAC cells are enriched in motifs for several transcription factors , including the GATA family which suggests a combinatorial regulatory code . This is in keeping with our finding that there is extensive overlap between the binding of KLF5 and GATA6 which is reinforced by recent studies that show that KLF5 binds with GATA6 in OAC ( Chen et al . , 2020 ) and gastric cancer ( Chia et al . , 2015 ) . The overlap in regulatory potential with GATA6 provides a plausible link to one of the major genetic events that drive the BO to OAC transition . Our work also suggests a link to another major pathway that is activated through gene amplification in OAC , the ERBB2-driven RAS-ERK pathway . Knockdown of ERBB2 reduced the expression of many KLF5 target genes and KLF5 motifs were found at regions with reduced chromatin accessibility upon ERBB2 knockdown . However , ERBB2 overexpression in BO cells is insufficient to trigger KLF5 redistribution , indicating that other pathways contribute to KLF5 redistribution in OAC , but this needs further investigation . Nevertheless , it is clear that ERBB2 signalling and KLF5 activity converge on the same cell cycle genes and both are required for their activation , indicating functional synergy . The signalling pathways are more unclear in the context of BO , the precancerous precursor . We see enrichment of the TEAD motif only in CP-A cells and not OE19 cells , suggesting that KLF5 may be operating through the Hippo signalling pathway in BO . In other contexts , KLF5 has been shown to cooperate with TEAD transcription factors , downstream of YAP/TAZ ( Wang et al . , 2015 ) and KLF5 is stabilised by YAP in breast cancer cells ( Zhi et al . , 2012 ) . Further work is needed to substantiate these links in BO . In summary , we have used integrative analysis of RNA-seq and ATAC-seq from BO and OAC patient samples to uncover a cell cycle signature regulated by KLF5 . Using a multi-omics approach , we found an oncogenic role of KLF5 in OAC , a transcription factor that has not been shown to be mutated , amplified and/or over-expressed in OAC . This study highlights the power of supplementing expression data with genome-wide chromatin profiling methods such as ATAC-seq . This provides molecular insights into the mechanisms by which BO progresses to OAC and identifies a signature of transcription factor gene targets that have potential prognostic significance and could be used as biomarkers in the clinic . OE19 and CP-A cells were purchased from ATCC and tested negative for mycoplasma . OE19 cells were maintained in Gibco RPMI 1640 ( ThermoFisher , 52400 ) supplemented with 10% Gibco fetal bovine serum ( ThermoFisher , 10270 ) and 1% Gibco penicillin/streptomycin ( ThermoFisher , 15140122 ) . CP-A cells were cultured in keratinocyte serum free media ( ThermoFisher , 17005042 ) supplemented with 5 μg/L EGF ( ThermoFisher , 10450–013 ) , 50 mg/L bovine pituitary extract ( ThermoFisher , 13028014 ) and 10% Gibco fetal bovine serum ( ThermoFisher , 10270 ) and 1% Gibco penicillin/streptomycin ( ThermoFisher , 15140122 ) . Cell-lines were authenticated by STR profiling and routinely tested for mycoplasma . OE19-dCas9-KRAB stable cells were generated by transfecting 1 × 106 OE19 cells with 7 . 5 μg Cas9 plasmid with guides targeting the AAVS1 locus ( Addgene #42230; 5’- GGGGCCACTAGGGACAGGAT-3’ ) and 7 . 5 μg donor plasmid ( pAAVS1-Puro-DNR; Origene GE100024 ) containing doxycycline inducible dCas9-KRAB with Fugene HD ( Promega , E2311 ) , as per manufacturer’s instructions . After 7 hr , media was replaced and supplemented with 7 . 5 μM RS-1 ( Sigma-Aldrich , R9782 ) and 1 μM SCR7 pyrazine ( Sigma-Aldrich , SML1546 ) , to promote homologous recombination and to inhibit non-homologous end joining respectively . Media was changed the next day and cells were selected with puromycin ( 0 . 75 μg/ml ) for 14 days . Selected colonies were re-plated to grow single clones and clones screened for dCas9-KRAB protein expression by immunoblotting . OE19-dCas9-KRAB cells were cultured with 100 ng/mL doxycycline ( Sigma-Aldrich , D3447 ) to induce dCas9-KRAB . To create CP-A-ERBB2 ( overexpressing ERBB2 ) and CP-A-empty ( control ) cells we first created the pHAGE-empty plasmid ( pAS4940 ) by excising the ERBB2 coding sequence from pHAGE-ERBB2 ( addgene #116734 ) using Xho1 ( NEB , R0146S ) followed by re-ligation of the vector . HEK293T cells were transfected with either pHAGE-ERBB2 or pHAGE-empty target plasmids , plus pMD2 . G ( Addgene , #12259 ) , psPAX2 ( Addgene , #12260 ) using Polyfect ( Qiagen , 301107 ) . Viral particles were precipitated from media using PEG-it ( System Biosciences , LV810A-1 ) . CP-A-empty and CP-A-ERBB2 stable cells were generated by transfecting 1 × 106 CP-A cells with lentiviral particles containing either pHAGE-ERBB2 or pHAGE-empty using polybrene ( EMD Millipore , TR-1003 ) at MOI of 1 for 24 hr . Transfected cells were grown for 2 days in full media before selection using puromycin ( 0 . 75 μg/ml ) for 14 days . Fresh frozen OAC 2 mm biopsies were obtained by consenting patients undergoing endoscopy . Tissue collection was granted by the ethics committee of Salford Royal NHS Foundation Trust ( 04/Q1410/57 ) . Patient consent was obtained in written form and signed by the patient and doctor . Patient biological replicates are defined as separate patients , and cell-line biological replicates are defined as separate cell-lines cultures , processed at the same time . Cells were lysed directly in RIPA buffer and incubated on ice for 5 min . The lysate was then sonicated in a water bath sonicator ( Diagenode Bioruptor ) for 5 min , 30 s on/off and protein quantified using Pierce BCA Assay Kit ( ThermoFisher , 23227 ) . Lysates were supplemented with SDS-PAGE loading dye to a final concentration of 1x and boiled for 10 min . Equal amounts of protein were separated on a 10% polyacrylamide gel and transferred to a nitrocellulose membrane ( GE life sciences , 1060002 ) using a Pierce Power Station ( ThermoFisher ) . Membranes were blocked using Odyssey blocking buffer ( Licor , 927–40000 ) and then incubated with antibodies against KLF5 ( abcam , ab137676 ) , Tubulin ( Sigma-Aldrich , T9026 ) , ERBB2 ( ThermoFisher , MA5-14057 ) , phospho-ERBB2 ( Cell Signalling Technologies , 6942S ) , AKT ( Cell Signalling Technologies , 2920S ) , phospho-AKT ( Cell Signalling Technologies , 9106S ) , ERK1/2 ( Cell Signalling Technologies , 4695S ) or phospho-ERK1/2 ( Cell Signalling Technologies , 9106S ) overnight at 4°C . Membranes were incubated with IRDye secondary antibodies ( Licor , 925–32212 , 925–32213 ) and imaged using a Li-Cor Odyssey scanner . RT-qPCR was carried out using QuantiTect SYBR Green RT-PCR Kit ( Qiagen , 204243 ) using the primer pairs detailed in Supplementary file 11 . Relative gene expression was calculated using the ΔΔCT method relative to levels of GAPDH mRNA . 200 , 000 cells were plated on a six-well plate and incubated for 24 hr . 100 pmol either control non-targeting ( siNT; Dharmacon , D-001810-10-0020 ) , siKLF5 ( Dharmacon , L-013571-00-0005 ) , or siERBB2 ( Dharmacon , L-003126-00-0005 ) SMARTpool siRNA was transfected per well using Lipofectamine RNAiMAX ( Thermofisher , 13778150 ) as per the manufacturer’s instructions and incubated for 72 hr . Modified full length sgRNAs were designed using E-CRISP ( Heigwer et al . , 2014; available at http://www . e-crisp . org/E-CRISP ) using the KLF5 TSS ( ±200 bp ) as input and obtained from Synthego . 9 pmol of either control non-targeting sgRNA ( 5’- GUAAGGCUAUGAAGAGAUAC-3’ ) or sgKLF5 pool ( 5’-GUGCGCUCGCGGUUCUCUCG-3’; 5’- AGGACGUUGGCGUUUACGUG-3’; 5’- GCGUCAAGUGUCAGUAGUCG-3’ ) was transfected per well using Lipofectamine RNAiMAX ( Thermofisher , 13778150 ) . Media was changed after 72 hr for longer treatments . RNA was extracted from cells using a RNeasy RNA extraction Kit ( Qiagen , 74136 ) and quality checked using Nanodrop 1000 ( ThermoFisher ) . Paired-end RNA-seq libraries were generated using TruSeq stranded mRNA library kit ( Illumina ) and sequenced on a HiSeq 4000 platform ( Illumina ) by the University of Manchester Genomic Technologies Core Facility . Reads were trimmed using Trimmomatic v0 . 32 ( Bolger et al . , 2014 ) quality checked using FastQC ( available at: http://www . bioinformatics . bbsrc . ac . uk/projects/fastqc ) and aligned to RefSeq transcript annotation of GRCh37 ( hg19 ) using STAR ( Dobin et al . , 2013 ) . Reads aligned to chromosomes 1–22 and chromosome X were retained . The Cufflinks package v2 . 2 . 1 ( Trapnell et al . , 2012 ) was used to calculate gene expression levels using Cuffnorm , and to analyse differential gene expression using Cuffdiff . Default parameters were used in both instances . Significant gene expression changes were defined by a fold change of ±1 . 3 and a Q-value of <0 . 05 . For ERBB2 knockdown experiments , counts for genes were determined using featureCounts ( Liao et al . , 2014 ) . Log2 transformed counts were obtained using DESeq2 variance stabilising transformation ( VST ) function . 200 , 000 cells were plated on a six-well plate and siRNA/sgRNA treatment started after 24 hr incubation . At specific time-points after treatment plates were washed with PBS and fixed with 4% paraformaldehyde for 10 min . Plates were then washed twice with PBS and kept at 4°C . Cells were then stained by first incubating plates at room temperature for 10 min in 0 . 1% Triton X-100 with gentle shaking and then incubated at room temperature for 30 min in 0 . 1% crystal violet ( Sigma-Aldrich , HT90132 ) with gentle shaking . Plates were extensively washed with water multiple times and left to dry . The dye was solubilised with 10% acetic acid for 10 min with gentle shaking and absorbance was read at 590 nm . Values for siNT at each time-point were used as 100% growth . Cells were trypsinised and collected as a single-cell suspension , washed with cold PBS , then fixed in 70% ethanol and stored at −20°C for at least 2 hr . Cells were then resuspended in staining solution ( 50 μg/mL propidium iodide ( Sigma , P4170 ) , 100 μg/mL RNase ( Sigma , R4642 ) ) and incubated at room temperature for 30 min . Cells were analysed by the University of Manchester Flow Cytometry Core Facility on a LSRFortessa . Percentages of cells in each cell cycle phase were calculated using ModFit LT ( http://www . vsh . com/products/mflt/ ) . Patient samples were processed as previously described ( Britton et al . , 2017 ) and omni-ATAC-seq was performed as previously described ( Corces et al . , 2017 ) . ATAC-seq libraries ( ~8 per lane ) were sequenced on a HiSeq 4000 platform ( Illumina ) by the University of Manchester Genomic Technologies Core Facility . Reads were quality checked using FastQC ( available at: http://www . bioinformatics . bbsrc . ac . uk/projects/fastqc ) and aligned to GRCh37 ( hg19 ) using Bowtie2 v2 . 3 . 0 ( Langmead and Salzberg , 2012 ) with the following options: -X 2000 –dovetail . Unique reads ( >q30 ) aligned to chromosomes 1–22 and chromosome X were retained . Peaks were called using MACS2 v2 . 1 . 1 ( Zhang et al . , 2008 ) with the following parameters: -q 0 . 01 –nomodel --shift −75 --extsize 150 -B –SPMR . Peaks called from individual samples were merged using mergePeaks . pl ( using d = 250 parameter ) from the HOMER package v4 . 9 ( Heinz et al . , 2010 ) and resized to peak summit ±250 bp to generate a peak set on which to perform differential accessibility analyses . Amplifications in patient biopsies were removed as described previously ( Denny et al . , 2016 ) using a fold change of 16 . bedGraph files were converted into BigWig files using bedGraphtoBigWig and visualised in the UCSC Genome Browser ( Kent et al . , 2002 ) . For comparing BO and OAC ATAC-seq , the Cufflinks package v2 . 2 . 1 ( Trapnell et al . , 2012 ) was used to calculate chromatin accessibility levels using Cuffnorm , and differential chromatin accessibility was analysed using Cuffdiff . Default parameters were used in both instances . Significant chromatin accessibility changes were defined as a fold change of ±2 and a Q-value of <0 . 1 . For ERBB2 knockdown experiments differential accessibility was calculated using DESeq2 ( Love et al . , 2014 ) . Alignment files of biological repeats were combined , peaks recalled and peaks from both conditions were then merged using using mergePeaks . pl ( using d = 250 parameter ) from the HOMER package v4 . 9 ( Heinz et al . , 2010 ) and resized to peak summit ±250 bp . featureCounts from the SUBread package ( Liao et al . , 2014 ) was used to count reads within peaks from ATAC-seq samples and these were used an input for DESeq2 to calculate differential binding using default settings . A linear fold change of ±2 and a Q-value of <0 . 05 were used as a cut-off for further analyses . ATAC-seq fragment size was visualised using a custom python script . Correlation plots between technical replicates were visualised using multiBamSummary and plotCorrelation from the deepTools package ( Ramírez et al . , 2016 ) . Tag density plots and heatmaps were also generated using computeMatrix and plotProfile or plotHeatmap tools from the deepTools package . ATAC-seq counts were also visualised using Morpheus ( https://software . broadinstitute . org/morpheus/ ) and hierarchical clustering was performed with this software using 1-Pearson’s correlation unless otherwise stated . Correlation plots of samples were visualised using the similarity matrix tool from Morpheus . To analyse ATAC-seq or ChIP-seq peaks for enriched transcription factor motifs , genomic coordinates were analysed using findMotifsGenome . pl with –cpg –mask –size 200 -bg parameters from the Homer package ( v4 . 7; Heinz et al . , 2010 ) . Background sequences were total accessible regions from all samples for ATAC-seq analysis and whole genome for ChIP-seq analysis . Pre-ranked genes ( ranked by log2 ( fold change ) ) were subject to gene set enrichment analysis from hallmark gene sets ( h . all . v6 . 2 ) using GSEAPreranked from GSEA v3 . 0 ( Subramanian et al . , 2005 ) . Gene ontology analysis was carried out using Metascape ( Zhou et al . , 2019; metascape . org ) . To analyse footprinting signatures in ATAC-seq data the TOBIAS package was used ( v0 . 5 . 1; Bentsen et al . , 2020; available at https://github . molgen . mpg . de/loosolab/TOBIAS ) . Merged BAM files from each condition were processed using ATACorrect , footprint scores calculated using FootprintScores and differential footprinting analysis using BINDetect . Footprinting plots across identified footprints at TF motifs were plotted using plotProfile from the deepTools package ( v2 . 5 . 0; Ramírez et al . , 2016 ) . ChIP-qPCR and ChIP-seq analysis was carried out as described previously ( Wiseman et al . , 2015 ) . For ChIP-qPCR , 2 . 5 × 106 cells and 1 μg antibody were used , and analysed using a Rotor-Gene SYBR Green PCR Kit ( Qiagen , 204074 ) . For ChIP-seq , 1 × 107 cells , 5 µg target protein antibody , 1 µg Spike-in antibody ( Active Motif , 61686 ) and 50 μl Protein A Dynabeads were used . 20 ng Spike-in Drosophila chromatin ( Active Motif , 53083 ) was supplemented to chromatin preps for Spike-in normalisation , as described previously ( Egan et al . , 2016 ) . Normal rabbit IgG ( Millipore , 12–370 ) antibody was used in parallel as a control . DNA libraries were prepared using TruSeq ChIP sample prep kit ( Illumina ) and sequenced on a HiSeq 4000 ( Illumina ) platform . Sequencing reads were aligned to GRCh37 ( hg19 ) and dm6 using Bowtie2 v2 . 2 . 3 ( Langmead et al . , 2009 ) . Reads aligning to the Drosophila genome were counted and used to generate scale factors . BAM files were then scaled to the sample with the lowest number of Drosophila reads . Only reads with a mapping quality >q30 were retained . Peak calling was performed on individual replicates using MACS2 v2 . 1 . 1 ( Zhang et al . , 2008 ) using default parameters with additional –SPMR parameter . bedGraph files were converted to bigwig using BedGraphtoBigWig script and visualised in the UCSC Genome Browser . The overlap of peaks between two biological replicates was calculated using BEDtools v2 . 26 . 0 ( Quinlan and Hall , 2010 ) using bedtools intersect with default settings with -f 0 . 3 parameter . Peaks present in both datasets were taken forward for further analysis . Differential binding analysis was performed using DESeq2 ( Love et al . , 2014 ) . Overlaps from biological repeats were merged using bedtools merge to generate a final set of peaks . featureCounts from the SUBread package ( Liao et al . , 2014 ) was used to count reads within peaks from ChIP-seq samples and these were used an input for DESeq2 to calculate differential binding using default settings . A linear fold change of ±2 and a Q-value of <0 . 05 were used as a cut-off for further analyses . Heatmaps of ChIP-seq signal were generated using computeMatrix and plotHeatmap from the deepTools package ( Ramírez et al . , 2016 ) . Tag density plots were generated using computeMatrix and plotProfile tools from the deepTools package . Correlation of biological replicates was visualised using multiBigwigSummary and plotCorrelation . Euler diagrams were generated using the Euler R package ( available at eulerr . co ) . Principal component analysis was performed using the prcomp function in R ( v3 . 5 . 1 , R Core Team , 2018 ) using log2 transformed RNA-seq or ATAC-seq normalised counts . Principal component scores were then plotted in Excel . Average expression of KLF5 or KLF5 target genes ( OAC upregulated and siKLF5 downregulated ) was calculated per patient and patients were ranked by average target gene expression . The median was calculated and patients were classified as either above or under median expression . Survival ( months ) was plotted for each group and a Log-rank test was carried out using GraphPad Prism v8 . To determine statistical significance between two groups , a Student’s unpaired two-tail T- test was carried out using GraphPad Prism v7 . To assess the changes in expression of a group of genes , a one-way ANOVA test was carried out in GraphPad Prism v7 . To assess the significance of gene/region overlaps derived from sequencing data , a hypergeometric distribution test was carried out using the phyper function in R . p-values<0 . 05 were considered as significant . All data were obtained from ArrayExpress , unless stated otherwise . ATAC-seq data from human BO , OAC tissue and OE19 cells were obtained from E-MTAB-5169 ( Britton et al . , 2017 ) and E-MTAB-6751 ( Rogerson et al . , 2019 ) . BO and OAC RNA-seq data were obtained from E-MTAB-4054 ( Maag et al . , 2017 ) and European Genome-phenome Archive ( EGA ) ( EGAD00001005915 ) . GATA6 and HNF4A ChIP-seq were obtained from E-MTAB-6858 and siGATA6 RNA-seq from E-MTAB-6756 . The Cancer Genome Atlas OAC ATAC-seq data were obtained from the GDC data portal ( portal . gdc . cancer . gov; Corces et al . , 2018 ) . All sequencing data are deposited in ArrayExpress . Additional OAC ATAC-seq data are available at E-MTAB-8447 and additional BO and OAC RNA-seq data are available at E-MTAB-8584 . siKLF5 RNA-seq data are available at E-MTAB-8446 . KLF5 ChIP-seq data are available at E-MTAB-8568 . siERBB2 ATAC-seq and RNA-seq data are available at E-MTAB-8576 and E-MTAB-8579 respectively . CP-A ATAC-seq data are available at E-MTAB-8994 .

Acid fluids present in the gut can sometimes ‘go up’ and damage the oesophagus , the pipe that connects the mouth and the stomach . As a result , a small number of individuals can develop Barrett’s oesophagus , a condition where cells in the lining of the lower oesophagus show abnormal shapes . In certain patients , these cells then become cancerous , but exactly how this happens is unknown . This lack of understanding contributes to late diagnoses , limited treatment and low survival rates . Many cancers feature ‘signature’ mutations in a set of genes that controls how a cell can multiply . Yet , in the case of cancers of the lower oesophagus , known genetic changes have had a limited impact on our understanding of the emergence of the disease . Here , Rogerson et al . focused instead on non-genetic changes and studied transcription factors , the proteins that bind to regulatory regions of the DNA to switch genes on and off . A close inspection of cancer cells in the lower oesophagus revealed that , in that state , a transcription factor called KLF5 controls the abnormal activation of genes involved in cell growth . This is linked to the transcription factor adopting a different pattern of binding onto regulatory regions in diseased cells . Crucially , when the cell growth genes regulated by KLF5 are activated , patients have lower survival rates . Further work is now required to examine whether this finding could help to identify patients who are most at risk from developing cancer . More broadly , the results from the work by Rogerson et al . demonstrate how transcription factors can be repurposed in a disease context .

2020

Repurposing of KLF5 activates a cell cycle signature during the progression from a precursor state to oesophageal adenocarcinoma

Plasmodium falciparum parasites , the causative agents of malaria , modify their host erythrocyte to render them permeable to supplementary nutrient uptake from the plasma and for removal of toxic waste . Here we investigate the contribution of the rhoptry protein RhopH2 , in the formation of new permeability pathways ( NPPs ) in Plasmodium-infected erythrocytes . We show RhopH2 interacts with RhopH1 , RhopH3 , the erythrocyte cytoskeleton and exported proteins involved in host cell remodeling . Knockdown of RhopH2 expression in cycle one leads to a depletion of essential vitamins and cofactors and decreased de novo synthesis of pyrimidines in cycle two . There is also a significant impact on parasite growth , replication and transition into cycle three . The uptake of solutes that use NPPs to enter erythrocytes is also reduced upon RhopH2 knockdown . These findings provide direct genetic support for the contribution of the RhopH complex in NPP activity and highlight the importance of NPPs to parasite survival . Malaria is caused by infection of the blood with Apicomplexan parasites of the genus Plasmodium . Critical for the proliferation and survival of Plasmodium in the blood is their ability to quickly penetrate host erythrocytes and acquire nutrients required for rapid growth . To facilitate this , the invasive merozoite forms of Plasmodium spp . sequentially secrete proteins from their apical organelles , the micronemes , rhoptries and dense granules . Proteins localizing to the micronemes and rhoptry neck are implicated in the irreversible attachment of the parasite to the host cell and are critical for invasion ( reviewed in [Harvey et al . , 2012; Weiss et al . , 2016] ) . Dense granule proteins are secreted once Plasmodium parasites have invaded their host cell ( Riglar et al . , 2011 ) , contributing to remodeling of the host cell ( de Koning-Ward et al . , 2016 ) . However , the role of proteins that localize to the rhoptry bulb is less clear and although they have been implicated in roles ranging from rhoptry biogenesis , erythrocyte invasion , formation of the parasitophorous vacuole ( PV ) in which the parasite is encased , as well as modification of the host cell ( Kats et al . , 2006; Counihan et al . , 2013 ) , functional data supporting these roles is very limited . RhopH2 is one of ~15 known proteins that localize to the rhoptry bulb in Plasmodium merozoites ( Counihan et al . , 2013; Ling et al . , 2003 ) . It is found in a high molecular weight complex with RhopH1 and RhopH3 ( Cooper et al . , 1988 ) that is discharged from merozoites , associating with the erythrocyte surface upon merozoite contact ( Sam-Yellowe et al . , 1988; Sam-Yellowe and Perkins , 1991 ) . The localization of RhopH proteins in the newly-infected erythrocyte is less clear as multiple localizations , including the PV membrane ( PVM ) , Maurer’s clefts and the cytosolic face of the erythrocyte membrane have been described for its constituents using different experimental approaches ( Perkins and Ziefer , 1994; Ndengele et al . , 1995; Sam-Yellowe et al . , 2001; Hiller et al . , 2003; Vincensini et al . , 2005 , 2008 ) . RhopH2 and RhopH3 are each encoded by a single gene . In contrast , RhopH1 in P . falciparum , the most pathogenic of the species infecting humans , is encoded by a multi-gene family comprising five variant genes termed clag 2 , 3 . 1 , 3 . 2 , 8 and 9 ( with clag3 . 1 and 3 . 2 mutually exclusively transcribed ) ( Gupta et al . , 2015; Kaneko et al . , 2001 , 2005; Ling et al . , 2004 ) . Of all the RhopH proteins , putative functions have only been assigned for RhopH1/Clag3 and Clag9 ( Gupta et al . , 2015 ) , although there is conflicting evidence for the involvement of Clag9 in cytoadherence ( Trenholme et al . , 2000; Goel et al . , 2010; Nacer et al . , 2011 ) . Via a high throughput drug-screening approach Clag3 has been linked to plasmodial surface anion channel ( PSAC ) activity ( Nguitragool et al . , 2011 ) . PSAC is a type of new permeability pathway ( NPP ) induced in the erythrocyte membrane by Plasmodium spp . that increases the cell’s porosity to organic and inorganic solutes . P . falciparum Clag3 null-mutants exhibit delayed in vitro growth , although NPP activity has not been investigated ( Comeaux et al . , 2011 ) . Intriguingly , Clag3 exhibits no homology to known ion channel proteins and lacks conventional membrane spanning regions to form a pore through the erythrocyte membrane , although it exists as both an integral and peripheral membrane protein in the infected erythrocyte ( Nguitragool et al . , 2011; Zainabadi , 2016 ) . Thus whether Clag3 forms ion channels directly and exclusively or if other parasite proteins or host cell membrane components contribute to a functional NPP is unknown . Alternatively , Clag3 may participate indirectly , for example , by regulating NPP activity . Both the rhopH2 gene and rhopH3 gene are refractory to deletion ( Cowman et al . , 2000; Janse et al . , 2011 ) . As RhopH1 is encoded by a multi-gene family , it is difficult to establish without genetically disrupting all but one clag variant within a parasite , whether the clag genes serve complementary functions or play distinct roles , including in NPP activity . To address these questions , we characterized RhopH2 in P . falciparum and conditionally depleted its expression in P . falciparum and the rodent malaria parasite P . berghei to investigate its contribution to erythrocyte invasion , parasite growth and erythrocyte permeability . Depletion of RhopH2 in cycle one did not affect transition into cycle two , suggesting RhopH2 plays no direct role in invasion . However , NPP activity was greatly reduced and parasite growth slowed as parasites progressed into trophozoite stage in cycle two , possibly due to nutrient depravation and/or accumulation of waste products . Transition into cycle three was curtailed by interesting phenomena including reduced schizont rupture and merozoite malformation that may be linked to reduced de novo pyrimidine synthesis . Taken together , RhopH2 appears to be important for NPP activity and for the exchange of nutrients and wastes with the blood plasma to facilitate parasite growth and proliferation . Conditional gene knockdown approaches were utilized herein to gain insight into the functional role of RhopH2 in Plasmodium parasites . This involved transfecting pRhopH2-HAglmS into P . falciparum that when correctly integrated into the rhopH2 locus , would lead to incorporation of a triple hemagglutinin ( HA ) and single strep II tag at the C-terminus of RhopH2 and the glucosamine ( GlcN ) -inducible glmS ribozyme ( Prommana et al . , 2013 ) within its 3' untranslated region ( UTR ) ( Figure 1a ) . Diagnostic PCR of transfectants resistant to WR99210 selection after three rounds of drug cycling confirmed that transgenic parasites , termed PfRhopH2-HAglmS , harbored the expected integration event ( Figure 1b ) . This was further validated by western blotting of parasite lysates from clonal PfRhopH2-HAglmS parasites using an anti-HA antibody; RhopH2 typically runs at 140 kDa by SDS-PAGE ( Cooper et al . , 1988; Ling et al . , 2003 ) and the observed 150 kDa band of RhopH2-HA is consistent with its anticipated size ( Figure 1c ) . Immunofluorescence analysis ( IFA ) confirmed RhopH2-HA localized to the rhoptry and co-localized with other rhoptry bulb proteins , RhopH1 , RhopH3 and RAMA but not with the rhoptry neck protein , RON4 , the micronemal marker , AMA-1 or the plasma membrane protein MSP1 ( Figure 1d ) . Comparison of the wildtype 3D7 and RhopH2-HAglmS parasite lines revealed that the addition of the epitope tags and ribozyme sequence did not impact on RhopH2-HAglmS to grow normally ( Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 23217 . 003Figure 1 . Generation of transgenic parasites in which RhopH2 is epitope-tagged . ( a ) The P . falciparum RhopH2 targeting construct was designed to integrate into the endogenous locus by a single crossover recombination event . The predicted structure of the endogenous locus before and after integration is shown . Haemagglutinin ( HA ) and strep II ( Str ) epitope tags , selectable marker ( sm ) , glmS ribozyme and untranslated regions ( UTR ) are shown . Arrows indicate oligonucleotides used in diagnostic PCR analysis and indicative product size . ( b ) Diagnostic PCR showing the PfRhoph2 gene contains the integrated sequence . Oligonucleotide pairs shown in ( a ) were used on genomic DNA prepared from drug-resistant parasites after transfection with the targeting construct ( integrant ) or 3D7 ( WT ) . DO354 and DO228 oligonucleotides , which recognize the rhoph2 locus , serve as a positive control for the PCR . ( c ) Western blot analysis showing the integrant line expresses the HA epitope tags . The predicted molecular mass of epitope-tagged RhopH2 is 164 kDa . PfHSP101-HA ( 101-HA ) serves as a positive control . ( d ) Immunofluorescence analysis ( IFA ) on schizonts fixed with acetone/methanol and labelled with anti-HA antibody to detect RhopH2 and other antibodies , as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 23217 . 00310 . 7554/eLife . 23217 . 004Figure 1—figure supplement 1 . Comparison of growth between P . falciparum wildtype ( 3D7 ) and RhopH2-HAglmS parasite lines . Parasite lines were grown in the absence of GlcN for three cycles and growth was measured by calculating the percentage parasitemia ( left panel ) or by lactate dehydrogenase assay ( LDH ) ( right panel ) . Shown is the mean ± SD fold-increase in parasitemia or LDH activity ( n = 6 independent biological replicates ) . An unpaired t-test was used to calculate statistical significance . DOI: http://dx . doi . org/10 . 7554/eLife . 23217 . 004 As RhopH2 has been described to reside at several different locations post-invasion , we took advantage of our RhopH2-HA line to characterize the expression and localization of RhopH2 at different times post-invasion using anti-HA antibodies . Western blot analysis revealed weak expression of RhopH2 during the ring and trophozoite stages , with a peak of expression at schizont stage ( Figure 2a ) , in keeping with when RhopH2 is maximally transcribed ( Ling et al . , 2003 ) . IFA confirmed RhopH2 synthesized during schizogony is carried in during invasion and localizes to the interface between the parasite and host cell ( Figure 2b ) . Weak labeling could also be observed at the erythrocyte membrane . As the parasite matured , RhopH2 could be detected in the erythrocyte cytoplasm , often exhibiting distinct punctate labeling , and the intensity of labeling at the erythrocyte membrane became more pronounced . RhopH2 did not co-localize with SBP1 , a Maurer’s cleft resident protein , indicating RhopH2 is not trafficked to the erythrocyte membrane via these membranous structures as previously suggested ( Sam-Yellowe et al . , 2001; Vincensini et al . , 2005 ) ( Figure 2c ) . 10 . 7554/eLife . 23217 . 005Figure 2 . Expression , localisation and solubility profile of P . falciparum RhopH2 . ( a ) Western blot analysis of RhopH2-HA expression across the erythrocytic cycle . Immunoblots were probed with the antibodies as indicated . ( b ) Immunofluorescence analysis ( IFA ) on erythrocytes infected with PfRhopH2-HAglmS and fixed with acetone/methanol . RhopH2 is labeled with the anti-HA antibody . The bars represent 5 µm . ( c ) IFA on erythrocytes infected with PfRhopH2-HAglmS , fixed with acetone/methanol and probed with anti-HA ( for RhopH2 ) and antibodies to the Maurer’s cleft protein SBP1 show that RhopH2 and SBP1 do not co-localise . ( d ) Solubility of RhopH2-HAglmS . Upper panel: Infected erythrocytes were synchronized and saponin-lysed when parasites reached ring ( R ) or schizont ( S ) stage and the pelleted material was sequentially dissolved in the buffers as indicated in the order of left to right ( upper panel ) . Supernatant fractions were analysed by western blotting with the indicated antibodies . Insoluble material represents protein remaining in the pellet fraction after 1% Triton X-100 treatment . Lower panel: Alternatively , infected erythrocytes were saponin-lysed when parasites were at ring stages , split into equal portions and pelleted before dissolving in one of the indicated buffers . Both supernatant ( Sn ) and pellet ( P ) fractions were analysed by western blotting . DOI: http://dx . doi . org/10 . 7554/eLife . 23217 . 005 Although RhopH2 has been shown to be present in detergent resistant membranes at schizont stages ( Sanders et al . , 2007; Hiller et al . , 2003 ) , localizes to the erythrocyte cytosolic face of the PVM ( Hiller et al . , 2003 ) and is present at the erythrocyte membrane , it is unclear how RhopH2 associates with these membranes . The hydrophobic region at I739-H759 is not universally predicted as a conventional transmembrane domain bioinformatically ( eg . TMHMM , SOSUI , TMPred ) . Therefore , we examined the solubility profile of RhopH2 at both schizont and ring-stages . We found that in contrast to EXP2 , which is a component of the Plasmodium translocon of exported proteins ( PTEX ) that resides at the PVM and requires Triton X-100 to be extracted from the membranes ( de Koning-Ward et al . , 2009 ) , the majority of RhopH2 could already be extracted with carbonate when sequential solubility assays were conducted ( Figure 2d , top panel ) . This indicates that RhopH2 is peripherally associated with membranes and is not an integral membrane protein . However , when erythrocytes infected with ring-stage parasites were saponin-lysed , pelleted by centrifugation and resuspended directly ( rather than sequentially ) in various detergents/buffers , RhopH2 could be extracted with carbonate and mostly with urea ( which also extracts peripheral membrane proteins ) , whereas it remained in the Triton X-100 pellet fraction ( Figure 2d , bottom panel ) . Combined , this data indicates that while RhopH2 predominantly has a peripheral association with the membranes at its respective locations , RhopH2 may be interacting with erythrocyte cytoskeletal proteins or is present in lipid rafts during the ring-stages , leading to its insolubility in Triton X-100 when resuspended in this buffer directly . To gain insight into proteins that interact with RhopH2 post-invasion , we next investigated the interactome of RhopH2 in ring and trophozoite stages by immunoprecipitating RhopH2 from PfRhopH2-HAglmS lysates using anti-HA antibodies and identifying proteins that had been affinity purified by mass-spectrometry ( Figure 3a ) . Bead-only and irrelevant protein controls ( Elsworth et al . , 2016 ) were used to identify non-specific interactions including ribosomal , nuclear and cytosolic proteins , which were subtracted to attain a list of likely specific interactions . 10 . 7554/eLife . 23217 . 006Figure 3 . The RhopH2 interactome . ( a ) Coomassie-stained SDS-PAGE gel of elution fractions from immune-precipitations performed with HA antibodies on lysates made from erythrocytes infected with RhopH2-HAglmS parasites at ring ( R ) or trophozoite stage ( T ) . ( b ) Bar graph showing the total number of peptides of particular subclasses of proteins that were affinity purified with PfRhopH2-HA . ( c ) Pie charts showing the number of peptides from the respective RhopH proteins that affinity purified with RhopH2-HA . The numbers of peptides identified are indicated in brackets . Note RhopH1 includes all CLAG peptides . ( d ) Western blot of blue-native PAGE performed on erythrocytes infected with trophozoite stage RhopH2-HAglmS parasites that had been solubilized in either 0 . 25% Triton X-100 or 1% ASB detergent reveal RhopH2 is present in ~670 and ~410 kDa species . ( e ) Pie chart showing the numbers of the most abundant peptides from PEXEL proteins that affinity purified with PfRhopH2-HA from trophozoite stage parasites . ( f ) Pie chart showing the numbers of the most abundant peptides from host erythrocyte proteins that affinity purified with PfRhopH2-HA in ring stage parasites . DOI: http://dx . doi . org/10 . 7554/eLife . 23217 . 006 In both parasite stages , RhopH2 was pulled down as well as other members of the RhopH complex , relatively few PV proteins such the PTEX complex , many exported PEXEL proteins , especially in trophozoites , and a large number of erythrocyte cytoskeletal proteins , particularly in the ring stages ( Figure 3b ) . That RhopH2 was interacting with the other members of the RhopH complex ( Figure 3c ) is consistent with an earlier report demonstrating the RhopH complex persists intact for at least 18 hr post-invasion ( Lustigman et al . , 1988 ) . There were more peptides recovered for RhopH1 ( particularly Clags3 . 1 , 3 . 2 and 9 ) and RhopH3 than there were for RhopH2 at the ring and trophozoite stages ( Figure 3c ) . Given that predicted molecular weights for the Clags ( 160–171 kDa ) and RhopH2 ( 163 kDa ) are similar but that of RhopH3 is somewhat smaller ( 104 kDa ) , this indicates that each component of the RhopH complex may not be in a 1:1:1 stoichiometry . Blue-Native PAGE gel analysis revealed that RhopH2 is present in a ~670 kDa complex that has a molecular mass larger than the predicted ~425 kDa ( Figure 3d ) . Apart from not being in an equimolar ratio , other non-RhopH proteins may also be present in the ~670 kDa complex . We also observed a smaller ~410 kDa complex when using the zwitterionic detergent 3- ( tetradecanoylamidopropyl dimethylammonio ) propane 1-sulfonate ( ASB-14 ) , which could contain a subset of the RhopH proteins and/or other proteins ( Figure 3d ) . Almost as abundant as RhopH peptides identified from the trophozoite-stage immunoprecipitations were exported proteins ( Figure 3b and e ) . The most predominant peptides from known exported proteins included those of mature parasite-infected erythrocyte surface antigen ( MESA; a protein that interacts with host protein 4 . 1 ) ( Waller et al . , 2003 ) , small exported membrane protein 1 ( SEMP1; a non-essential protein that localizes to the Maurer’s clefts and is partially translocated to the erythrocyte membrane ) ( Dietz et al . , 2014 ) , glycophorin-binding protein 130 ( GBP130; an exported soluble protein ) ( Maier et al . , 2008 ) , a variety of Plasmodium helical interspersed subtelomeric proteins ( PHIST; some of which have been shown to interact with PfEMP1 ) ( Proellocks et al . , 2014; Oberli et al . , 2014 , 2016 ) and HSP70-x ( localizes to J-dots ) ( Külzer et al . , 2012 ) . Peptides from exported proteins were more abundant in the pull-down performed on lysates from trophozoites compared to ring-stages , in keeping with protein export peaking during the trophozoite stage . The exception was ring erythrocyte surface antigen ( RESA ) in which more peptides were observed in the ring-stage pull-down . RESA is one of the first proteins exported into the erythrocyte that ultimately localizes to the ankyrin-band 3 complex at the erythrocyte cytoskeleton . A large number of peptides to erythrocyte cytoskeletal proteins were also identified , including tubulin , spectrin , ankyrin , protein 4 . 1 , band 3 and actin ( Figure 3f ) . Whether RhopH2 is indirectly interacting with these cytoskeletal components via exported proteins , or specifically interacting with all or a subset of these cytoskeletal proteins is unknown , especially since many cytoskeletal elements are bound together in the cell . Taken together , these results indicate that after invasion , the RhopH complex traffics from the PVM to the erythrocyte membrane and either en route or when it reaches its final destination , RhopH2 interacts with a number of exported parasite proteins that also bind to components of the host cytoskeleton . As the epitope-tagged RhopH2 line harbors a glmS riboswitch sequence , the ability to regulate RhopH2 expression in parasites via the addition of GlcN was investigated to gain functional insight into this protein . Erythrocytes infected with synchronized ring-stage parasites were treated for up to two cycles with 2 . 5 mM GlcN and assessed for protein knockdown via western blot ( Figure 4a ) and parasite growth via Giemsa-stained smears relative to parasites grown in the absence of GlcN ( Figure 4b–c ) . RhopH2 is normally transcribed around the onset of schizogony ( Ling et al . , 2003; Bozdech et al . , 2003; Le Roch et al . , 2004 ) and the addition of 2 . 5 mM GlcN resulted in knockdown of RhopH2 expression in schizonts by 84% within the first cycle . By late in the second cycle , RhopH2 protein levels were reduced by 92% in those parasites that made it to schizont stage ( Figure 4a ) . RhopH2-HAglmS parasites ( +GlcN ) appeared morphologically normal by the end of the first cycle ( Figure 4b ) . In separate experiments whereby RhopH2-HAglmS parasites expressing GFP at the end of cycle one were incubated with donor erythrocytes , the conversion of schizonts to ring stage parasites and therefore invasion efficiency was not specifically affected by the knockdown of RhopH2 ( Figure 4c ) . In contrast , a striking growth defect in RhopH2-HAglmS ( +GlcN ) parasites was observed in the second cell cycle around the ring to trophozoite transition stage , with late-ring stage parasites appearing irregular in shape and trophozoites exhibiting an abnormal stunted phenotype rather than progressing to mature trophozoites ( Figure 4b ) . In addition , RhopH2-HAglmS ( +GlcN ) parasites that transitioned to schizonts at the end of the second cycle harboured significantly lower numbers of merozoites per schizont ( mean of 19 merozoites cf 12 merozoites for –GlcN and +GlcN cultures , respectively p<0 . 0001 ) ( Figure 4d ) . Moreover , the time required to complete the second cycle and commence the third cycle was delayed ( ~92 hr cf ~108 hr for -GlcN and +GlcN cultures , respectively ) ( Figure 4b , e ) . This all translated to ~4 fold reduction in the number of ring-stage parasites observed at the beginning of the third cycle when compared to parasites not exposed to GlcN ( Figure 4e ) . 10 . 7554/eLife . 23217 . 007Figure 4 . Reduction in PfRhopH2 expression leads to altered growth phenotypes in vitro . ( a ) Effect of glucosamine on PfRhopH2 protein expression . Upper panel: overview of experiment . Synchronised cultures of PfRhopH2-glmS were treated with glucosamine ( GlcN ) at the indicated time and material harvested , as indicated . Lower panels: infected erythrocytes were harvested by saponin lysis and subject to SDS-PAGE and western blotting . PfRhopH2-HA was detected using an anti-HA antibody and EXP2 ( used as a loading control ) detected with a specific polyclonal EXP2 antibody . Right panel: Densitometry performed on bands observed in western blot using ImageJ was performed to calculate the ratio of EXP2 or RhopH2 protein levels in parasite lines grown in the presence ( + ) or absence ( - ) of GlcN ( n = 3 independent experiments ) . Shown is the mean ± SEM ( n = 3 ) . ( b ) Representative Giemsa-stained smears parasites depleted of RhopH2 progress to schizont stage in cycle one but parasite growth is slowed around the trophozoite stage ( n = 3 independent experiments ) . ( c ) Analysis of the number of schizonts in cultures of wildtype ( 3D7 ) and RhopH2-HAglmS parasites grown in the absence ( − ) or presence ( + ) of 2 . 5 mM GlcN that invaded donor erythrocytes within 3 or 5 hr post-incubation ( hpi ) , as measured by FACS ( n = 3 ) . Shown is the mean ± SEM . ( d ) Box plot indicating the number of merozoites formed per schizont in cultures of RhopH2-HAglmS grown in 0 mM ( 35 schizonts examined ) or 2 . 5 mM ( 51 schizonts examined ) GlcN . The central bar in the box plot denotes the median whilst the whiskers delineate the 10th and 90th percentiles . p<0 . 0001 by unpaired t-test . ( e ) Parasitemias of cultured PfRhopH2-HAglmS parasites grown in 0 mM or 2 . 5 mM GlcN , determined by counting a minimum of 1000 erythrocytes . Depletion of PfRhopH2 expression increases the length of the cell cycle and has a marked effect on the numbers of parasites progressing to cycle 3 . Shown is the mean ± SEM ( n = 3 ) . ( f ) Growth of 3D7 and PfRhopH2-HAglmS parasites when cultured in various concentrations of GlcN , as measured by lactate dehydrogenase assay ( LDH ) . The LDH activities of 3D7 and RhopH2-HAglmS cultured in the absence of GlcN at cycle three were normalized to 100% , and activity of all lines ( ± GlcN ) across the three cycles was measured relative to this . Shown is the mean ± SD ( n = 3 ) . An unpaired t-test revealed RhopH2-HAglmS parasites grew significantly slower than 3D7 in all concentrations of GlcN by 36 hpi ( p<0 . 01 ) ( g ) Measurement of nanoluciferase ( Nluc ) released into the culture media and in pelleted erythrocytes infected with 3D7 or RhopH2-HAglmS parasites expressing Hyp1-Nluc . Measurements commenced around the time 3D7 parasites were starting to egress and invade new erythrocytes . The data represents the mean ± SD of one biological replicate completed in triplicate , with results expressed as percentage Nluc activity in the media relative to the pellet fraction . DOI: http://dx . doi . org/10 . 7554/eLife . 23217 . 007 In separate experiments , parasite lactate dehydrogenase ( pLDH ) activity was also measured on trophozoite stages of RhopH2-HAglmS- or 3D7-parasitized erythrocytes grown in the presence of increasing concentrations of GlcN as a surrogate for parasite proliferation ( Figure 4f ) . The pLDH activities of 3D7 ( -GlcN ) and RhopH2-HAglmS ( -GlcN ) at cycle three were normalized to 100% , with activity of all parasite lines ( ± GlcN ) across the three cycles measured relative to this . While 3D7 parasite growth only begun to be affected by the addition of 2 mM GlcN by cycle 3 , in strong contrast , growth of RhopH2-HAglmS parasites was majorly reduced at all GlcN concentrations and also relative to 3D7 cultured in the same GlcN concentrations . The results also concur with the experiments above in that the effect of GlcN on pLDH activity could already be seen at cycle two and was drastically amplified when parasites transitioned into cycle three ( Figure 4f ) . To further validate the effects of knocking down RhopH2 upon parasite maturation and transition into cycle three , erythrocytes infected with RhopH2-HAglmS and the 3D7 parental line at trophozoite stage were transfected with an exported nanoluciferase fusion protein ( Hyp1-Nluc ) . This enabled schizont rupture and merozoite egress at the end cycle two to be monitored via measuring the amount of nanoluciferase released into the culture media compared to the cell pellet . Infected erythrocytes were supplemented with GlcN in cycle one when the parasites were at trophozoite stage and when the 3D7 line had grown to late-schizont stage in cycle two and new ring-stage parasites were beginning to be visible in Giemsa stain ( indicative of the start of merozoite egress and commencement of transition into cycle 3 ) , the media and cell pellets were harvested and every two hours thereafter for eight hours . The percentage ratio of nanoluciferase activity of media/pellet was then determined . This revealed that egress of RhopH2-HAglmS ( +GlcN ) line was markedly delayed compared to 3D7 ( +GlcN ) ( p<0 . 001 for 2 . 5 mM GlcN ) ( Figure 4g ) . These results validate that the slower growth of RhopH2-HAglmS ( +GlcN ) observed in cycle 2 is due to the specific effect of depletion of RhopH2 expression . Since the growth experiments revealed a defect in parasite transition from cycle two to cycle three , video microscopy of live schizont-stage parasites at the end of the second cycle on GlcN was performed to visualize whether knockdown of RhopH2 was impacting on erythrocyte egress and invasion . No obvious differences were observed in general schizont morphology or in the ability of the erythrocyte to burst , indicating egress per se was not actually affected . Rather , instead of the merozoites dispersing rapidly after egress , GlcN treatment caused the merozoites to remain clumped together ( Figure 5a , Videos 1–2 ) , a phenotype not observed in 3D7 ( +GlcN ) parasites ( not shown ) . Occasionally remnants of membranes could be observed around the merozoites , but even when these broke down , the merozoites remained clumped ( Figure 5a , see 3 mM 5s versus 29 . 1s ) . Nine and eleven schizont ruptures were observed ± GlcN treatment respectively , and as a consequence of merozoite clumping following GlcN treatment , less than two merozoites per rupture were released and able to contact new erythrocytes compared with six merozoites without GlcN ( Figure 5b ) . The net effect was fewer average invasions per rupture , with only 0 . 25 after GlcN treatment compared to 2 . 6 without treatment ( Figure 5c ) . Whilst 28 out of the 63 merozoites that made erythrocyte contact without GlcN went on to invade erythrocytes ( Figure 5d ) , only two out of the 14 merozoites treated with GlcN invaded erythrocytes , indicating a success rate of 0 . 44 and 0 . 14 invasions per contact , respectively ( Figure 5d ) . From these results , it was inferred that the lower fold-increase in parasitemia from cycle two to the next after RhopH2 knockdown stemmed from a combined effect of reduction in the number of parasites reaching schizogony in cycle two and a reduced invasion rate . The latter most likely stems from an indirect effect of RhopH2 knockdown that results in a clumping of merozoites incapable of breaking free to invade a new host cell and a reduced competency of merozoites forming at the end of cycle two to successfully invade an erythrocyte . 10 . 7554/eLife . 23217 . 008Figure 5 . Merozoites depleted of PfRhopH2 show defect in parasite invasion the following cycle . ( a ) Panel of images from videos of PfRhopH2-HAglmS schizonts observed rupturing and releasing merozoites at the end of cycle 2 , post-addition of 0 or 3 mM GlcN . The number of seconds post-rupture is indicated . ( b ) The number of merozoites contacting nearby erythrocytes per schizont rupture following GlcN treatment is shown . ( c ) The number of erythrocyte invasions per schizont rupture is shown . ( d ) The proportion of merozoite-erythrocyte contacts that successfully result in invasion are indicated . For ( b ) and ( c ) , the central bar denotes median , the box denotes 25–75th percentile and the whiskers the data range . ****p<0 . 0001 by unpaired t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 23217 . 00810 . 7554/eLife . 23217 . 009Video 1 . Plasmodium falciparum RhopH2-HAglmS schizont rupturing and releasing merozoites which invade nearby human erythrocytes . Successful invasions are indicated with white arrows . Time in seconds from egress is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 23217 . 00910 . 7554/eLife . 23217 . 010Video 2 . A rupturing Plasmodium falciparum RhopH2-HAglmS schizont that had been treated with 3 mM glucosamine for 2 cell cycles to knockdown RhopH2-HA expression . At 0 s the erythrocyte membrane surrounding the schizont begins to break down but the merozoites do not disperse until about 68 s later . None of the merozoites appeared to invade neighbouring erythrocytes . DOI: http://dx . doi . org/10 . 7554/eLife . 23217 . 010 To unequivocally show that the growth defects in P . falciparum were a consequence of RhopH2 knockdown , conditional regulation of RhopH2 in P . berghei was also performed . This also provided insight into the consequences of depleting RhopH2 expression on parasite growth in vivo . In this case , the P . berghei rhoph2 locus was modified to insert an anhydrotetracycline ( ATc ) -regulated transactivator element ( TRAD ) downstream of the endogenous rhoph2 promoter and a minimal promoter with TRAD binding sites upstream of the rhoph2 coding sequence . P . berghei ANKA parasites transfected with linearized pTRAD4-RhopH2ss and surviving pyrimethamine drug pressure were analyzed by diagnostic PCR and Southern blot , confirming that the targeting construct had integrated correctly into the rhoph2 locus and the line was clonal ( Figure 6—figure supplement 1a–c ) . Transcription of rhoph2 in this line , termed PbRhopH2-iKD , was highly responsive to ATc , with ~11-fold reduction of rhoph2 mRNA in schizont stages as determined by qRT-PCR and RT-PCR ( Figure 6—figure supplement 1d ) . The growth of the PbRhopH2-iKD line was specifically sensitive to ATc treatment . PbRhopH2-iKD parasites grew poorly in mice that had been pre-exposed to ATc 24 hr prior to infection ( Figure 6a ) . Conversely , growth of parental P . berghei parasites was unaffected by the presence of ATc ( Figure 6a ) as has been shown previously ( Pino et al . , 2012; Elsworth et al . , 2014 ) . The slower growth of PbRhopH2-iKD exposed to sucrose compared to parental P . berghei parasites exposed to ATc is most likely due to the transactivator not being able to induce transcription of rhopH2 to the same level as the native promoter . 10 . 7554/eLife . 23217 . 011Figure 6 . Depletion of RhopH2 in P . berghei leads to altered growth phenotypes in vivo and in vitro . ( a ) Representative growth curve ( n = 2 ) of P . berghei iRhopH2 and wildtype ( WT ) parasites . Groups of 5 mice were pre-treated for 24 hr with either 0 . 2 mg/ml ATc or sucrose ( vehicle control ) , then infected with the PbiRhopH2 iKD line or WT PbANKA . Parasitaemia was calculated at the indicated timepoints . Error bars represent standard error of the mean . An unpaired t-test revealed growth of RhopH iKD +ATc was significantly impaired at all time points ( p<0 . 0001 ) and that of RhopH2 iKD + sucrose was slower that PbAWT +ATc by day five post infection ( p=0 . 026 ) ( b ) Representative Giemsa-stained smears showing effect of RhopH2 knockdown with ATc on parasite growth and schizont formation . Schematic shows experimental outline . ( c ) Depletion of RhopH2 protein levels also impacts on the number of merozoites formed per schizont ( n = 59 and 55 schizonts examined for parasites grown in the absence and presence of ATc , respectively , and taken from three individual experiments ) . The central bar in the box plot denotes the median whilst the whiskers delineate the 10th and 90th percentiles . p<0 . 0001 by unpaired t-test . ( d ) Representative invasion assay ( n = 2 ) performed with merozoites from mechanically ruptured schizonts cultured in vitro ± ATc showing percentage of parasites from n = 50–100 that were at ring ( R ) , early trophozoite ( ET ) , late trophozoite ( LT ) or schizont ( S ) stage of development . DOI: http://dx . doi . org/10 . 7554/eLife . 23217 . 01110 . 7554/eLife . 23217 . 012Figure 6—figure supplement 1 . Characterization of inducible P . berghei RhopH2 parasites . ( a ) The P . berghei inducible RhopH2 targeting construct was designed to integrate into the endogenous locus by double-crossover recombination . The predicted structure of the endogenous locus before and after integration is shown . Green and pink bars indicate regions used to generate probes for Southern blot , arrows indicate oligonucleotides used in diagnostic PCR analysis and indicative product size . ( b ) Diagnostic PCR showing PbRhopH2 iKD parasites have integrated the targeting sequence . Oligonucleotide pairs shown in A were used on genomic DNA prepared from drug-resistant parasites after transfection with the targeting construct ( integrant ) or P . berghei ANKA ( WT ) . A product should only be observed for WT with DO291 and DO67 oligonucleotides . ( c ) Southern blot showing homologous integration into Pbrhoph2 gene as predicted . Plasmid DNA from the targeting construct and genomic DNA from integrant and WT were digested with restriction enzymes ( RE ) and probed with the 3' targeting sequence or TRAD sequence . In both cases , the endogenous locus ( E ) has been modified and integration ( In ) bands of the predicted size are seen . ( d ) RT-PCR showing reduced expression of PbRhopH2 in the presence of ATc . Upper panel: experimental outline . Mice infected with inducible RhopH2 line ( PbiRhopH2 ) were treated with ATc or vehicle control ( -ATc ) for 24 hr prior to harvest and overnight culture in vitro ± ATc , upon which RNA was extracted from the schizont stages . Lower panel: Diagnostic PCR using oligonucleotides specific for rhoph2 or exp2 . RhopH2 cDNA is only detected in the absence of ATc . Amplification products using gDNA as a template are shown in the last two lanes as controls . DOI: http://dx . doi . org/10 . 7554/eLife . 23217 . 012 Since the more mature stages of P . berghei sequester in vivo , erythrocytes infected with PbRhopH2-iKD parasites ( 1 × 107 ) were inoculated into ATc-pretreated mice and harvested the following cycle when the parasites were at ring-stage . They were then cultured ex vivo in the presence of ATc to examine the development of parasites across the entire cell cycle . Parasites in which RhopH2 had been depleted , exhibited delayed progression to trophozoite stage and the schizont stages displayed aberrant morphology , often appearing vacuolated and containing fewer merozoites ( Figure 6b–c ) . A synchronous in vitro invasion and growth assay using merozoites that had been generated from cultured schizonts confirmed these findings , showing that parasites depleted of RhopH2 could invade erythrocytes but exhibited a delay in the transition from the early to more mature trophozoite forms ( Figure 6d ) , consistent with our findings in P . falciparum . RhopH2 is localized on the host cytosolic side of the PVM immediately after invasion and was found to affinity purify some components of the PTEX and a variety of exported proteins . This raised the question of whether the RhopH complex helps traffic proteins that exit PTEX through the erythrocyte cytoplasm , particularly as protein export is required to support parasite growth ( Elsworth et al . , 2014; Dietz et al . , 2014; Beck et al . , 2014 ) . However , no defect in the export of PfEMP1 , or trafficking of either RESA to the erythrocyte membrane or SBP1 to the Maurer’s clefts was evident after knocking down RhopH2 expression with GlcN ( Figure 7a ) . In contrast , the localization of RhopH3 and to a lesser extent RhopH1/clag3 was affected when RhopH2 expression was knocked down ( Figure 7—figure supplement 1 ) . Moreover , RhopH2-HAglmS parasites supplemented with a reduced concentration of 0 . 5 mM GlcN that still gave efficient RhopH2 knockdown ( Figure 7b ) and which were harvested at mid-trophozite stage before parasites growth was impaired ( Figure 7c ) , could similarly export a nanoluciferase reporter ( Hyp1-NLuc ) as RhopH2-HAglmS ( -GlcN ) or 3D7 parasites ( ± ) GlcN ( Figure 7d ) . 10 . 7554/eLife . 23217 . 013Figure 7 . RhopH2 is not involved in the trafficking of exported proteins in the erythrocyte cytoplasm . ( a ) Representative IFAs of erythrocytes infected with RhopH2-HAglmS parasites grown in 0 mM or 2 . 5 mM GlcN using the indicated antibodies show trafficking of RESA , SBP1 and PfEMP1 is unaffected upon RhopH2 knockdown . Scale bar = 5 µm ( b ) Western blots of the parasites probed with an anti-HA antibody indicate that PfRhopH2 has been substantially knocked down with 0 . 5 mM GlcN relative to an EXP2 loading control . ( c ) Giemsa stained images of the trophozoites that were analysed . ( d ) Proportion of luciferase activity exported into the erythrocyte cytosol , secreted into the parasitophorous vacuole or present in the parasite cytoplasm of RhopH2-HAglmS and 3D7 wildtype parasites transfected with Hyp1-Nluc and grown in ± GlcN . Bars denote mean ± SD ( n = 3 ) . An unpaired t-test revealed there was no significance different in the exported NLuc fractions ± GlcN for 3D7 ( p=0 . 8579 ) and RhopH2-HAglmS ( p=0 . 1801 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23217 . 01310 . 7554/eLife . 23217 . 014Figure 7—figure supplement 1 . Localization of RhopH1/clag3 and RhopH3 in infected erythrocytes when RhopH2 expression is knocked down . Representative immunofluorescence analysis of erythrocytes infected with RhopH2-HAglmS parasites grown in 0 mM or 2 . 5 mM GlcN . Cells fixed with acetone/methanol and labelled with anti-HA antibody to detect RhopH2 IFAs and other antibodies , as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 23217 . 014 Since RhopH2 forms a complex with RhopH1 , a protein implicated in NPP activity , and depletion of RhopH2 leads to growth defects around the time that NPPs are active in the infected erythrocyte , we next assessed whether RhopH2 contributes to NPP function . Sorbitol transport into infected erythrocytes requires NPP activity , resulting in hypotonicity-induced cell lysis ( Wagner et al . , 2003; Nguitragool et al . , 2011 ) . Thus RhopH2-HAglmS-parasitized erythrocytes transfected with the Hyp1-Nluc reporter were treated with sorbitol buffer containing NanoGlo . The degree of lysis and hence channel activity could be quantified by measuring the amount of NanoGlo hydrolysed by Hyp1-Nluc which is released during cell lysis ( Azevedo et al . , 2014 ) . We established that GlcN-mediated knockdown of RhopH2 dramatically reduced the capacity of infected erythrocytes to be lysed by sorbitol , suggesting RhopH2 contributes to NPP activity ( Figure 8a ) . In contrast , 3D7-parasitized erythrocytes treated with GlcN were not affected in their ability to be lysed by sorbitol indicating that depletion of RhopH2 and not treatment with GlcN was responsible for inhibition of NPP function ( Figure 8a ) . As the ability of parasitized erythrocytes to be lysed only commences >24 hpi , Giemsa smears of parasites used in the sorbitol assays were examined but this revealed the parasites were all similarly aged ( Figure 8b ) . When an iso-osmotic solution of alanine was used instead of sorbitol , similar results were obtained , with increasing concentrations of GlcN reducing the capacity of RhopH2-HAglmS parasitized erythrocytes to be lysed ( Figure 8c ) . More lysis inhibition was also observed in 32 hr trophozoites compared to 24 hr trophozoites , consistent with the NPPs being more developed in older parasites . 10 . 7554/eLife . 23217 . 015Figure 8 . Knockdown of RhopH2 impairs sorbitol and alanine uptake . ( a ) GlcN-mediated knockdown of RhopH2 in PfRhopH2-HAglmS parasites expressing an exported Hyp1-Nluc reporter leads to a dramatic reduction in the capacity of infected erythrocytes to be lysed by the addition of sorbitol . In contrast erythrocytes infected with 3D7 parasites expressing Hyp1-Nluc are sensitive to sorbitol-mediated lysis . The % lysis was determined by the amount of NanoGlo substrate hydrolysed by Hyp1-Nluc , with 100% lysis defined as the Nluc activity ( RLU/min ) in parasites incubated in 280 mM sorbitol buffer with no GlcN . Data represents mean ± SD of three biological replicates completed in triplicate . ( b ) Giemsa stained images of the trophozoites analysed in the sorbitol uptake assays . ( c ) Analysis of sorbitol and alanine-mediated lysis of erythrocytes infected with PfRhopH2-HAglmS parasites at 24 and 32 hr post infection ( hpi ) at various concentrations of GlcN . The % lysis was determined by the amount of NanoGlo substrate hydrolysed by Hyp1-Nluc . Data represents mean ± SD of one biological experiment completed in triplicate . DOI: http://dx . doi . org/10 . 7554/eLife . 23217 . 015 Given that RhopH2 depletion appeared to affect NPP activity , we next examined the effect of depleting RhopH2 on the metabolism of P . falciparum-infected erythrocytes . This was undertaken by conducting comparative untargeted metabolomics on 3D7 and RhopH2-HAglmS parasites incubated in the presence and absence of 2 . 5 mM GlcN . Overall , ~1000 metabolites from diverse pathways were detected and assigned putative identities based on accurate mass , and confirmed using retention time where standards were available ( Creek et al . , 2012 ) . A Principal Component Analysis ( PCA ) of all metabolite features across the four sample groups , 3D7 ±GlcN ) and RhopH2-HAglmS ( ±GlcN ) showed that replicates from groups 3D7 ( ±GlcN ) and RhopH2-HAglmS ( -GlcN ) clustered together , and that these were metabolically different to the induced RhopH2 knockdown , RhopH2-HAglmS ( +GlcN ) ( Figure 9a ) . This indicates that knockdown of RhopH2 causes a reproducible metabolic shift in the parasites . The heat map demonstrates a substantial impact of GlcN on global metabolite levels , even in wild-type 3D7 parasites ( Figure 9b ) . Nevertheless , as indicated in the PCA analysis , the inclusion of the 3D7 ( +GlcN ) control allowed detection of several metabolites that were specifically perturbed in response to RhopH2 knockdown , including selected vitamins/cofactors , nucleotides , amino acids and glycolytic metabolites ( Figure 9b ) . A detailed scrutiny of individual metabolites showed that while glucosamine treatment appeared to elevate metabolite levels in general , the RhopH2 knockdown resulted in decreased levels of folate and thiamin phosphates , which are essential vitamins and cofactors for cellular growth ( Figure 9c ) . The other class of metabolites to significantly decrease upon RhopH2 depletion were intermediates in the de novo pyrimidine synthesis pathway , N-carbamoyl L-aspartate , dihydroorotate and orotate . These metabolites are essential nucleotide precursors in P . falciparum , however , levels of downstream nucleotides were not affected at this time-point ( Supplementary file 1 ) . Few other metabolites were significantly and specifically depleted in the RhopH2 knockdown , with the exception of the glycolytic intermediates 3-phosphoglycerate and phosphoenolpyruvate ( Supplementary file 1 ) . The only putatively identified metabolite to extensively accumulate ( >5-fold higher than all controls ) in the RhopH2 knockdown was the urea cycle intermediate argininosuccinate , however , the other urea cycle intermediates were not significantly perturbed . Interestingly , a general increase in amino acid levels was also observed in the RhopH2 knockdown ( Figure 9d ) . 10 . 7554/eLife . 23217 . 016Figure 9 . Metabolomics analysis of 3D7 and RhopH2-HAglmS parasites +/- GlcN treatment . ( a ) Principal Component Analysis scores plot of the first two principal components based on all metabolite features across the four sample groups . ( b ) Heat map of relative abundance of all the putative metabolites detected in this study grouped according to metabolite classes . ( c ) Fold change of metabolites showing a decrease in abundance , involved in vitamin and co-factor metabolism , de novo pyrimidine synthesis and glycolysis in the RhopH2-HAglmS ( +GlcN ) and 3D7 ( +Furosemide ) parasites compared to 3D7 ( untreated ) represented by the dotted vertical line . Error bars indicate relative standard deviation from n = 3 independent biological replicates . Thiamine monophosphate and orotate were not detected in the furosemide treatment experiment . ( d ) Fold change of metabolites ( amino acids and a urea cycle intermediate ) showing an increase in abundance in the RhopH2-HAglmS ( +GlcN ) and 3D7 ( +Furosemide ) parasites compared to 3D7 ( untreated ) represented by the dotted vertical line . Error bars indicate relative standard deviation from n = 3 independent biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 23217 . 016 In order to compare these metabolic perturbations to the effect of pharmacological NPP inhibition , erythrocytes infected with 3D7 were treated with furosemide and metabolite levels compared to untreated controls . Consistent with the RhopH2 knockdown , levels of folate and phosphoenolpyruvate were significantly lower in furosemide-treated parasites , and threonine , histidine , asparagine , serine and argininosuccinate levels all increased ( Figure 9d ) . Interestingly , the general depletion of de novo pyrimidine synthesis intermediates was not observed with furosemide , with N-carbamoyl L-aspartate levels found to be significantly higher following furosemide treatment . A manuscript utilizing the same strategy to deplete RhopH2 in Plasmodium falciparum was also published by Ito and colleagues ( Ito et al . , 2017 ) . These authors derive very similar conclusions about the role of RhopH2 in nutrient uptake . 10 . 7554/eLife . 23217 . 017Figure 10 . Scheme illustrating how RhopH2 knockdown effects blood stage development . ( a ) Knockdown of RhopH2 expression in cycle one appears to impair uptake of plasma nutrients in cycle two which delays development and replication in cycles 2 and 3 . ( b ) ( 1 ) The RhopH complex is probably introduced onto the surface of the parasitophorous vacuole membrane ( PVM ) during merozoite invasion . ( 2 ) Shortly after invasion the PTEX complex begins exporting parasite-synthesised proteins secreted into the parasitophorous vacuole ( PV ) , out into the erythrocyte cytoplasm . Some of the exported proteins such as PHISTs , MESA , LyMP , GBP130 and SEMP1 travel and bind to the erythrocyte cytoskeleton . The RhopH complex either ( 3 ) binds to these exported proteins at the erythrocyte surface or ( 4 , 5 ) assembles with these proteins en route to the surface . ( 6 ) Once at the surface , the RhopH/exported protein complex forms NPPs either by forming their own membrane-spanning pore or by ‘opening up’ an erythrocyte pore . The NPPs function to permit the entry of nutrients and to dispose of parasite waste products . DOI: http://dx . doi . org/10 . 7554/eLife . 23217 . 017 Experiments involving the use of animals were performed in accordance with the recommendations of the Australian Government and the National Health and Medical Research Council Australian code of practice for the care and use of animals for scientific purposes . The protocols were approved by the Deakin University Animal Welfare Committee ( approval number G37/2013 ) . To create a transgenic P . falciparum line in which RhopH2 expression could be knocked down , the transfection construct pRhopH2-HAglmS was created . This construct contains 1035 bp of sequence immediately upstream of the stop codon of RhopH2 ( Pf3D7_0929400 ) that had been PCR amplified from P . falciparum 3D7 genomic DNA ( gDNA ) with the primers DO227 and DO228 ( see Supplementary file 2 for oligonucleotide sequences ) and cloned into the BglII and PstI sites of pPfTEX88-HAglmS ( Chisholm et al . , 2016 ) . To engineer the PbRhopH2 inducible knockdown ( iKD ) line , the first 1477 bp of the PbRhopH2 coding sequence ( PbANKA_0830200 ) that had been PCR amplified with the primers DO291F and DO67R was cloned into the PstI and NheI sites of the modified pPRF-TRAD4-Tet07-HAPRF-hDHFR ( Pino et al . , 2012 ) described in Elsworth et al . ( 2014 ) . Also cloned into the NheI and BssHII sites of this vector were 1279 bp of the rhoph2 5' UTR sequence immediately upstream of the RhopH2 start codon , which had been PCR amplified using the primers DO62F and DO63R . Before transfection into P . berghei ANKA parasites , pTRAD4-iRhopH2ss was linearized with NheI . Blood-stage P . falciparum strain 3D7 was cultured continuously ( Trager and Jensen , 1976 ) and transfected as previously described ( Fidock and Wellems , 1997 ) . Transgenic parasites were selected with 2 . 5 nM WR99210 ( Jacobus ) or 5 μg/mL blasticidin S ( Sigma-Aldrich , Australia ) . P . berghei transgenic parasites were generated using the reference clone 15cy1 from the P . berghei ANKA strain . Transfection of parasites and selection of the transgenic parasites intravenously injected into 6- to 8-week-old female BALB/c mice was performed as previously described ( Janse et al . , 2006 ) . Erythrocytes infected with PfRhopH2-HAglmS parasites were treated at ring stage with 2 . 5 mM glucosamine ( GlcN ) or 0 mM GlcN as a control ( day 1 ) . Parasites were harvested at schizont stage , or in the following cycle at mid ring stage or schizont stage and treated with 0 . 05% saponin to remove haemoglobin . Western blots of parasite proteins fractionated on 8% Bis-Tris gels ( Life Technologies , Carlsbad , CA , USA ) were blocked in 5% skim milk in PBS and then incubated with mouse anti-HA ( 1:1000; Roche , Indianapolis , IN , USA ) for detection of RhopH2 and rabbit anti-EXP2 ( 1:1000 ) as a loading control . After washing , the membranes were probed with horseradish peroxidase-conjugated secondary antibodies ( 1:5000; Thermo Scientific , Waltham , MA , USA ) and detection was performed using the Clarity ECL Western blotting substrate ( Biorad , Hercules , CA , USA ) . The membrane was imaged using a Fujifilm LAS-4000 Luminescent Image Analyzer and ImageJ software ( NIH , version 1 . 46r ) was used to measure intensity of bands . Mice infected with erythrocytes infected with the PbRhopH2 iKD line were administered drinking water containing 0 . 2 mg/ml ATc ( Sigma ) made in 5% sucrose or 5% sucrose only as vehicle control when the parasitemia reached ~1% . After 24 hr when the parasites were predominantly at ring stage , mouse blood was harvested by cardiac bleed and cultured in vitro until parasites reached schizont stage ( ~16 hr ) in RPMI 1640 medium containing L-glutamine ( Life Technologies ) supplemented with 25 mM HEPES , 0 . 2% bicarbonate , 20% fetal bovine serum and 1 µg/ml ATc ( or vehicle as a control ) at 36 . 5°C . Experiments were performed on two independent occasions . The infected erythrocytes were lysed with 0 . 05% saponin prior to RNA extraction . To detect transcripts in P . berghei parasites by qRT-PCR , RNA was extracted from blood stage parasites using TRIsure reagent ( Bioline , UK ) . cDNA was then made using the iScript reverse transcription supermix ( Biorad ) according to the manufacturer’s instructions . cDNA ( or gDNA as a control ) was used in PCR reactions using oligonucleotides to rhopH2 ( O614F/O615R and O605F/O616R ) or gapdh ( O567F/O568R ) . The expression levels of rhopH2 were normalized against the gapdh house-keeping gene , with gene expression values calculated based on the 2ΔΔCt method . For analysis of P . falciparum growth , erythrocytes infected with PfRhopH2-HAglmS parasites at ring stage were sorbitol synchronized twice and the following cycle ( Cycle 1 ) , 2 . 5 mM GlcN was added to ring stage parasites , with 0 mM GlcN serving as the negative control . Parasitemias in Giemsa-stained smears were determined by counting a minimum of 1000 erythrocytes and comparative growth analysis was performed using a student’s t-test . Parasite growth of triplicate samples was also assessed using a modified Malstat assay protocol ( Makler and Hinrichs , 1993 ) . For this , GlcN ( Sigma ) was added to blood cultures of synchronized PfRhopH2-HAglmS ring stage parasites ( ~5% parasitemia and 2% hematocrit ) in cycle 1 . In cycle 1 and cycle 2 , when parasites were at trophozoite stage , three aliquots were removed for subsequent proliferation assays and the cultures then diluted 1/5 to 1/10 and seeded into new plates with fresh erythrocytes and GlcN . A final three aliquots were removed in cycle three when parasites were at trophozoite stage . To quantitate parasite biomass , 30 µL of culture was mixed with 75 µL Malstat reagent ( 0 . 1 M Tris pH 8 . 5 , 0 . 2 g/mL lactic acid , 0 . 2% v/v Triton X-100 and 1 mg/mL acetylpyridine adenine dinucleotide ( Sigma ) , 0 . 01 mg/ml phenozine ethosulfate ( Sigma ) and 0 . 2 mg/mL nitro blue tetrazolium ( Sigma ) . Once the no drug control wells had developed a purple color the absorbance was measured at 650 nm in a spectrophotometer . The cumulative absorbance values were calculated by subtracting the absorbance of uninfected erythrocytes from infected erythrocytes and multiplying by the combined dilution factor . The pLDH activities of 3D7 and RhopH2-HAglmS cultured in the absence of GlcN at cycle three were normalized to 100% , and activity of all lines ( ± GlcN ) at each day was measured relative to this . Data was analysed using a student’s t-test . Female Balb/c mice at 6 weeks of age were randomized into groups of five mice per experiment and then given drinking water containing either 0 . 2 mg/mL ATc ( Sigma ) made in 5% ( w/v ) sucrose or 5% sucrose only as a vehicle control . After 24 hr pre-treatment , mice were infected intraperitoneally ( i . p ) with 1 × 106 PbRhopH2 iKD parasitized erythrocytes . From 3 days post infection , parasitemias were monitored daily by Giemsa-stained tail blood smears , with mice humanely culled once the parasitemias reached >20% . Parasitemias in Giemsa-stained smears were determined by counting a minimum of 1000 erythrocytes . Comparative growth experiments were analyzed using a students t-test , with p<0 . 05 considered significant . To establish synchronous P . berghei infections , blood was harvested from donor mice infected with PbRhopH2 iKD when the parasitemia was ~3% . The blood was then cultured overnight in vitro in RPMI/20% FCS in the presence or absence of 1 µg/mL ATc until parasites reached schizont stage . The schizonts were purified on Nycodenz ( ELITech Group , Australia ) and isolated merozoites were incubated with uninfected erythrocytes in vitro as previously described and invasion allowed to proceed for 30 min ( Matthews et al . , 2013 ) . Following merozoite invasion , parasites were maintained in culture for a further 36 hr , with smears made at intervals and stained with Giemsa to monitor parasite growth . Erythrocytes infected with PfRhopH2-HAglmS at either ring or schizont stage were lysed with 0 . 05% ( w/v ) saponin in PBS . For sequential solubility assays , the pelleted parasite material was resuspended in a hypotonic lysis buffer ( 1 mM HEPES , pH 7 . 4 ) and after a 30 min incubation on ice , the material was centrifuged at 100 , 000 g for 30 min at 4°C . The supernatant , which contains soluble proteins , was removed and kept for analysis . The pellet was then resuspended in 0 . 1 M Na2CO3 ( pH 11 . 5 ) to extract proteins peripherally-associated with membranes . After another 30 min incubation on ice and centrifugation step , the pellet was resuspended in 1% ( w/v ) Triton X-100 in PBS and incubated at room temperature for 30 min to extract integral membrane proteins and re-centrifuged . The starting material , soluble fractions and the Triton X-100 insoluble fraction were electrophoresed by SDS-PAGE and transferred to nitrocellulose membrane for Western blotting . In an alternative approach , parasitized erythrocytes that had been hypotonically lysed with 1 mM HEPES , pH7 . 4 to remove soluble proteins were split into five equal fractions and resuspended in either 10 mM Tris-HCl , 0 . 1 M Na2CO3 ( pH 11 . 5 ) , 2% Triton X-100 , 6 M urea ( extracts peripheral and soluble proteins ) or 2% SDS ( solubilizes membrane proteins ) . Samples were incubated on ice for 1 hr and then centrifuged at 100 , 000 g for 30 min at 4°C . Pellet fractions were washed in 10 mM Tris-HCl . Both the soluble and insoluble fractions were analysed by Western blotting using mouse anti-HA ( 1:1000 ) , rabbit anti-EXP2 ( 1:1000 ) , rabbit anti-HSP101 ( 1:1000 ) and rabbit anti-SERA5 ( 1:1000 ) . Immunoprecipitations were performed on synchronised ring stage and trophozoite P . falciparum RhopH2-HAglmS-infected erythrocytes harvested with 0 . 05% ( w/v ) saponin in PBS . Parasite pellets were solubilized in 1% ( w/v ) Triton X-100 containing Complete protease inhibitors ( Roche ) . After a 30 min incubation on ice , the material was centrifuged at 17 , 000 g for 10 min at 4°C and supernatants were added to 100 μl PBS-washed anti-HA-agarose beads ( mAb clone HA-7 ) ( Sigma ) and mixed overnight at 4°C . The beads were washed in 0 . 5% Triton X-100 in PBS plus protease inhibitors . Bound proteins were eluted with 100 μL 1x non-reducing sample buffer ( 50 mM Tris-HCl pH 6 . 8 , 10% glycerol , 2 mM EDTA , 2% SDS , 0 . 05% bromophenol blue ) , then reduced and electrophoresed by SDS-PAGE . After staining the gel with Imperial Protein Stain ( Thermo Scientific ) , protein bands were manually excised and subjected to manual in-gel reduction , alkylation , and tryptic digestion , and extracted peptides were analysed by LC-MS/MS using an Orbitrap Lumos mass spectrometer ( Thermo Scientific ) fitted with nanoflow reversed-phase-HPLC ( Ultimate 3000 RSLC , Dionex , Australia ) . The nano-LC system was equipped with an Acclaim Pepmap nano-trap column and an Acclaim Pepmap RSLC analytical column . 1 μL of the peptide mix was loaded onto the enrichment ( trap ) column at an isocratic flow of 5 μL/min of 3% CH3CN containing 0 . 1% formic acid for 6 min before the enrichment column was switched in-line with the analytical column . The eluents used for the LC were 0 . 1% v/v formic acid ( solvent B ) and 100% CH3CN/0 . 1% formic acid v/v . The gradient used was 3% B to 20% B for 95 min , 20% B to 40% B in 10 min , 40% B to 80% B in 5 min and maintained at 80% B for the final 5 min before equilibration for 10 min at 3% B prior to the next sample . The mass spectrometer was equipped with a NanoEsi nano-electrospray ion source ( Thermo Fisher ) for automated MS/MS . The resolution was set to 120000 at MS1 with lock mass of 445 . 12003 with HCD Fragmentation and MS2 scan in ion trap . The top 3 s method was used to select species for fragmentation . Singly charged species were ignored and an ion threshold triggering at 1e4 was employed . CE voltage was set to 1 . 9 kv . Late trophozoite-stage ( 24–36 hr post invasion [hpi] ) P . falciparum-infected erythrocytes were lysed in 0 . 09% saponin in 5 mM Tris pH 7 . 5 and washed three times in PBS to remove haemoglobin . Following centrifugation , the parasite pellet was solubilized by sonication in 0 . 25% ( v/v ) Triton X-100 or 1% ( v/v ) ASB-14 ( 3- ( tetradecanoylamidopropyl dimethylammonio ) propane 1-sulfonate ) , the latter because it is often used for solubisation of proteins for 2D electrophoresis ) , then incubated with mixing at 4°C for 30 min . Insoluble material was pelleted ( 14 000 g for 30 min at 4°C ) . The supernatants were electrophoresed on NativePAGE Novex 3–12% Bis-Tris protein gels as per manufacturer’s instructions ( Invitrogen ) and transferred to PVDF for Western blotting . Bound antibody probes were detected with LiCor Odyssey Fc infrared imager followed by analysis with ODYSSEY v1 . 2 software . IFA was performed on thin smears of infected erythrocytes fixed with ice cold 90% acetone/10% methanol for 2 min . Cells were blocked in 1% ( w/v ) BSA/PBS for 1 hr . All antibody incubations were performed in 0 . 5% ( w/v ) BSA/PBS . Primary antibodies for P . falciparum were used at the following concentrations: rat anti-HA ( 1:100 , Life Technologies ) , mouse anti-HA ( 1:250 , Life Technologies ) , chicken anti-HA ( 1/200 , Abcam ) , rabbit anti-RhopH1/clag3 ( 1:200 ) ( Kaneko et al . , 2005 ) , rabbit anti-RhopH3 ( 1:250 ) , rabbit anti-RAMA-D ( 1:1000 ) , rabbit anti-AMA1 ( 1:300 ) , rabbit anti-RON4 ( 1:300 ) , rabbit anti-SBP1 ( 1:200 ) , mouse anti-RESA ( 1:1000 ) and mouse anti-MSP1-19 mAb 17B6 ( 20 µg/mL ) . After a one-hour incubation in primary antibody , cells were washed three times in PBS and incubated with the appropriate AlexaFluor 488/568-conjugated secondary antibodies ( 1:2000 ) for 1 hr . Cells were washed three times in PBS , and mounted with Prolong Gold Antifade reagent ( Life Technologies ) containing 4' , 6-diamidino-2-phenylindole ( DAPI ) ( VectorLabs , Australia ) . Images were taken on an Olympus IX71 microscope and processed using ImageJ v1 . 46r . Ring stage PfRhopH2-HAglmS infected erythrocytes ( cycle 1 ) were cultured at 4% hematocrit in the presence of 3 mM GlcN ( or 0 mM GlcN as a control ) until parasites reached the late schizont stage of cycle 2 . The culture was then diluted to 0 . 16% in RPMI media and 2 mL of this was allowed to settle to produce a monolayer onto a 35 mm Fluorodish ( World Precision Instruments , Sarasota , FL , USA ) . Live parasite imaging was performed at 37°C on a Zeiss AxioObserver Z1 fluorescence microscope equipped with humidified gas chamber ( 90% N2 , 1% O2 , and 5% CO2 ) . Late stage schizonts were observed until they looked ready to rupture and time-lapse videos were recorded with an AxioCam MRm camera at four frames per second . ImageJ and Prism ( Graphpad , La Jolla , CA , USA ) were used to perform image and statistical analyses . Quantitation of invasion was performed using an unpaired student’s t-test . PfRhopH2-HAglmS and 3D7 parasites transfected with pHGBHRB ( a plasmid encoding a GFP reporter under the expression of under the HSP86 5' UTR ) ( Wilson et al . , 2010 ) , were used for invasion assays . Tightly synchronized parasitized erythrocytes purified using a VarioMACS magnetic cell separator were mixed with erythrocytes ( 1:50 ratio ) that had been stained with 10 µM amine-reactive fluorescent dye 7-hydroxy-9H- ( 1 , 3-dichloro-9 , 9-dimethylacridin-2-one ) succinimidyl ester ( Cell Trace Far Red DDAO-SE ) ( Invitrogen , Carlsbad , CA , USA ) in RPMI-1640 for 1 hr at 37°C according to manufacturer’s protocol . At designated time points , erythrocytes were harvested and stained for 20 min at room temperature in the dark with the DNA dye Hoechst 34580 ( 2 μM ) ( Invitrogen ) made in RPMI-1640 . Following a washing step , stained samples were examined using a BD FACS Canto II flow cytometer ( BD Biosciences , Australia ) with 100 , 000 events recorded . Experiments were carried out in triplicate . The collected data was analysed with FlowJo software ( Tree Star , Ashland , Oregon ) . Data was analysed for statistical significance using an unpaired student’s t-test . Wildtype P . falciparum 3D7 and RhopH2-HAglmS-infected erythrocytes were transfected with a Nanoluciferase ( Nluc ) ( Hall et al . , 2012 ) protein N-terminally appended with the N-terminus of the PEXEL protein Hyp-1 as described in ( Azevedo et al . , 2014 ) but containing the blasticidin deaminase gene instead of the hDHFR gene . The infected erythrocytes were sorbitol synchronized and when the parasites reached late trophozoite stage , the cultures were treated with either 0 . 5 mM GlcN or no GlcN for 48 hr . Infected erythrocytes were subsequently transferred to a 96 well plate at 1% hematocrit , 1% parasitemia and the GlcN concentration was maintained prior to measurement of Nluc signal . A series of wells containing infected erythrocytes lacking exported Nluc were used to control for background luminescence . When performing the assay , the control well lacking Hyp1-Nluc was spiked with recombinant Nluc ( 1 ng/µL ) to control for Nluc quenching by haemoglobin . Subsequently 5 µL of resuspended culture was added to Greiner Lumitrac 96 well microplate in duplicate before adding 90 µL of either Background buffer ( 10 mM Tris phosphoric acid pH7 . 4 , 127 mM NaCL , 5 mM Na2EDTA , 5 mM DTT ) , Equinotoxin ( EQT ) buffer ( Background buffer with EQT ( 5 µg/mL ) prepared in house as per [Jackson et al . , 2007] ) , EQT/saponin buffer ( EQT buffer with 0 . 03% ( w/v ) saponin ) or hypotonic buffer ( 10 mM Tris phosphoric acid pH 7 . 4 , 5 mM Na2EDTA , 0 . 2% NP40 , 5 mM DTT ) , which allow differential fractionation of the infected erythrocyte . The cells were incubated for 10 min at RT to allow for lysis to occur . Following this , 5 µl of diluted NanoGlo ( Promega , Australia , diluted 1: 500 in background buffer ) was injected to each well , the plate shaken ( 700 rpm/30 s ) and relative light units were then measured with CLARIOstar plate reader ( BMG Labtech , Australia ) . Experiments were repeated on three independent occasions and two technical replicates were completed per biological replicate . The export of Nluc was calculated as follows: the mean ( X¯ ) was calculated before adjusting for the spike in control ( His-Nluc ) and subtracting the background ( buffer one ) . Error was estimated with standard deviation ( SD ) and coefficient of variation ( CV ) . Subsequently , percentage of Nluc was calculated for each compartment using the new mean:% Exported fraction=X¯EQTX¯ Hypox100 Standard deviation for exported fraction was calculated as follows:CVexported fraction=CVEQT 2+CVHypo2SDexported fraction=CV x % Exported fraction% Secreted fraction= ( X¯EQT+X¯SAP ) −X¯EQTX¯Hypox100 Standard deviation for secreted fraction was calculated as follows:CVsecreted fraction=SDEQT2+SDEQT+SAP2X¯EQT+SAP−X¯EQTSDsecreted fraction=CVsecreted fraction2+CVHypo2 x %secreted fraction% Parasite fraction=X¯Hypo− ( X¯EQT+X¯SAP ) X¯Hypox100 Standard deviation for parasite cytosol fraction was calculated as follows:CVparasite fraction=SDEQT+SAP2+SDHypo2X¯Hypo−X¯EQT+SAPSDparasite fraction=CVparasite fraction2+CVHypo2 x %parasite fraction Experiments were then combined depending on the weight of data reliability with error weighted mean and error weighted standard deviation . Weight of data depending reliability . Error weighted mean was calculated as follows:X¯weigthed=Value1Abs ( SD1 ) +Value2Abs ( SD2 ) +Value3Abs ( SD3 ) Abs ( 1SD1 ) +Abs ( 1SD2 ) +Abs ( 1SD3 ) Error weighted standard deviation was calculated as follows:SDweighted = ( Value1−X¯Weighted ) 2Abs ( SD1 ) + ( Value2−X¯Weighted ) 2Abs ( SD2 ) + ( Value3−X¯Weighted ) 2Abs ( SD3 ) 1SD1+1SD2+1SD3 Value : Percentage in relative compartment , where numbers refer to three biological replicates . SD: Standard deviation calculated for each compartment , where numbers refer to three biological replicates . Data were analysed for statistical significance using a two-tailed , unpaired student’s t test with equal variances . P . falciparum RhopH2-HAglmS and 3D7 parasites expressing Hyp1-Nluc were treated with 0–3 mM GlcN when parasites were at trophozoite ( 28–36 hr post invasion ) stage and parasites were then grown for a further 48 hr until trophozoites stage . After washing the parasitized erythrocytes twice in PBS , 10 µL at 1% hematacrit and 1% parasitemia ( or PBS as a control ) was dispensed in triplicate into a Thermo Scientific 96 well U bottom microplate and loaded into a Clariostar luminometer ( BMG labtech ) . To each well , 40 µL of sorbitol or alanine lysis buffer containing the NanoGlo substrate ( 280 mM sorbitol or 280 mM L-alanine , 20 mM Na-HEPES , 0 . 1 mg/ml BSA , pH 7 . 4 , Nano-Glo [1:1000 dilution] ) was added and the relative light units ( RLU ) measured every 3 min with gain set to 2500 . The percent lysis was determined by non-linear regression , exponential growth equation as analysed by GraphPad Prism software . The PBS control was subtracted and the value multiplied by 100 to get a percentage lysis . A value of 100% lysis is defined as the Nluc activity in relative light units ( RLU/min ) of parasites in 280 mM sorbitol or alanine buffer with no GlcN . A value of 0% lysis is defined as the Nluc activity of parasites in PBS containing nanoglo substrate . The rate of lysis was derived from a kinetic assay measuring the increase in RLU per minute ( Dickerman et al . , 2016 ) . Data was analysed for statistical significance using an unpaired student’s t-test . Erythrocytes infected with RhopH2-HAglmS and 3D7 parasites expressing NLuc ( as described above ) were sorbitol synchronized and subsequently GlcN-treated at trophozoite stage ( cycle 1 ) . Heparin was added ( 100 μg/ml ) to prevent any early invasion events and was subsequently removed when schizonts were observed ( GlcN concentrations were maintained ) . Parasites were allowed to invade over a six-hour window ( cycle 2 ) and were subsequently sorbitol synchronized prior to seeding into 96 well plates ( 100 µl/1% Hematocrit/1% parasitemia ) . Giemsa smears were taken at late schizont stage ( end of cycle 2 ) and when rings were observed , the cultures were pelleted ( 500g/3 min ) and 50 µl of supernatant containing released Nluc was removed for analysis ( media fraction ) . Infected erythrocyte cell pellets were also collected . Fractions were collected every two hours for a total of 8 hr . Prior to analysis of total Nluc content , 50 μl of media containing 1% hemocrit was added to each media fraction and 50 μl of media was added to each pellet sample to maintain equivalent volumes . Each fraction was fully re-suspended and 10 µl was added to 90 µl lysis buffer ( 10 mM tris phosphoric acid , 5 mM Ka2EDTA , 0 . 2% NP40 , 5 mM DTT , Nano-Glo ( 1:1000 dilution ) ) in a Greiner Lumitrac 96 well microplate prior to shaking ( 700 pm/30 s ) . Relative light units were measured with a ClARIOstar multimode plate reader ( BMG Labtech ) and data was subsequently analysed using GraphPad PRISM software . Tightly synchronized cultures of P . falciparum 3D7 or RhopH2-HAglmS ring stage parasites were exposed to either 0 mM or 2 . 5 mM GlcN in cycle one and they were harvested in the second cycle when they had sufficient haemazoin pigment ( ~24 hr post-invasion ) to facilitate magnetic purification using a VarioMACS magnetic cell separator . For furosemide treatment , 500 µM of furosemide was added to the cultures shortly after invasion when parasites were in the early ring stages of cycle two and the cultures were harvested at ~24 hr post-invasion . Morphology of parasites was monitored by light microscopy to obtain developmentally similar stages of parasites under GlcN and furosemide treatment and untreated control cultures . Metabolism was quenched by rapidly cooling down the cultures to 4°C , culture medium was removed following centrifugation at 3000 g for five minutes ) and metabolites were extracted from 4 . 5 × 107 cells using 150 µl of extraction buffer consisting of chloroform/methanol/water ( 1:3:1 v/v ) ( spiked with 1 µM PIPES , CHAPS and Tris as internal standards ) followed by vortex mixing for 1 hr at 4°C . After mixing , cellular debris was removed by centrifugation at 4°C ( >15000 g for 10 min ) and the supernatant was kept at −80°C prior to analysis . Three biological replicates were prepared for each cell line and treatment . Samples were analysed by hydrophilic interaction liquid chromatography coupled to high resolution-mass spectrometry ( LC-MS ) according to a previously published method ( Stoessel et al . , 2016 ) . All samples were analyzed as a single batch , in randomized order and pooled quality control samples were analyzed regularly throughout the batch to confirm reproducibility . Approximately 250 metabolite standards were analyzed immediately preceding the batch run to determine accurate retention times to facilitate metabolite identification . Additional retention times for metabolites lacking authentic standards were predicted computationally as previously described ( Creek et al . , 2011 ) . Data was analysed using the IDEOM workflow ( Creek et al . , 2011 , 2012 ) . Peak areas for significant metabolites were confirmed by manual integration with Tracefinder software ( Thermo Scientific ) . Multivariate statistical analysis utilized principal component analysis ( PCA ) on log-transformed and auto-scaled metabolite peak intensity data using the web-based analytical tool , MetaboAnalyst ( Xia et al . , 2015 ) . The IDEOM files containing all metabolomics data are uploaded on Figshare and can be accessed at https://figshare . com/s/c38c0a98fb01634677f6 .

Malaria is a life-threatening disease that affects millions of people around the world . The parasites that cause malaria have a complex life cycle that involves infecting both mosquitoes and mammals , including humans . In humans , the parasites spend part of their life cycle inside red blood cells , which causes the symptoms of the disease . In order to thrive , malaria parasites need to make the red blood cell more permeable so that they can absorb nutrients from the blood stream and get rid of toxic waste products they generate . Previous research has shown that the parasites can produce a protein that makes red blood cells more permeable to a range of nutrients . Understanding how the parasites can do this , and if they could change the permeability of their host red blood cell to prevent anti-malaria drugs from entering may help researchers to identify new approaches to starve the parasite . Counihan et al . investigated whether the parasites also use other proteins to modify red blood cells and demonstrated that a protein called RhopH2 can make the blood cells more permeable . The experiments used a genetically modified version of the parasite that lacked RhopH2 . Counihan et al . show that essential nutrients and vitamins were depleted in these parasites and that the parasites were much slower to grow and reproduce . The next important step would be to identify all proteins that are involved in making red blood cells more permeable and how they achieve this , and use this knowledge to help generate anti-malarial drugs .

2017

Plasmodium falciparum parasites deploy RhopH2 into the host erythrocyte to obtain nutrients, grow and replicate

Successful fertilization in angiosperms depends on the proper trajectory of pollen tubes through the pistil tissues to reach the ovules . Pollen tubes first grow within the cell wall of the papilla cells , applying pressure to the cell . Mechanical forces are known to play a major role in plant cell shape by controlling the orientation of cortical microtubules ( CMTs ) , which in turn mediate deposition of cellulose microfibrils ( CMFs ) . Here , by combining imaging , genetic and chemical approaches , we show that isotropic reorientation of CMTs and CMFs in aged Col-0 and katanin1-5 ( ktn1-5 ) papilla cells is accompanied by a tendency of pollen tubes to coil around the papillae . We show that this coiled phenotype is associated with specific mechanical properties of the cell walls that provide less resistance to pollen tube growth . Our results reveal an unexpected role for KTN1 in pollen tube guidance on the stigma by ensuring mechanical anisotropy of the papilla cell wall . Following deposition of dehydrated pollen grains on the receptive surface of the female organ , the stigma , pollen rehydrates , germinates and produces a pollen tube that carries the male gametes toward the ovules where the double fertilization takes place . This long itinerary through the different tissues of the pistil is finely controlled , avoiding misrouting of the pollen tube and hence assuring proper delivery of the sperm cells to the female gametes . How the pollen germinates a tube and how pollen tube growth is regulated have been the object of many investigations ( Palanivelu and Tsukamoto , 2012; Mizuta and Higashiyama , 2018 ) . The use of in vitro pollen germination as well as semi-in vivo fertilization assays , together with the analysis of mutants defective in pollen or ovule functions , have rapidly expanded our knowledge of the mechanisms that sustain pollen tube growth , its guidance toward the ovule and the final delivery of male gametes within the embryo sac ( Dresselhaus and Franklin-Tong , 2013; Higashiyama and Yang , 2017; Cameron and Geitmann , 2018 ) . At the cellular level , cytoskeleton has been extensively studied during pollen tube elongation ( Fu , 2015 ) , highlighting the critical role played by the actin microfilaments in pollen-tube tip growth through delivery of materials for the biosynthesis of the plasma membrane and cell wall . By contrast , CMTs seem to have a lower importance in the pollen tube elongation process , drugs affecting CMT polymerisation having no significant effects on pollen-tube growth rate , yet altering the capacity of pollen tubes to change their growth direction ( Gossot and Geitmann , 2007 ) . In Arabidopsis thaliana , pollen tubes grow within the cell wall of papillae of the stigmatic epidermis , and then through the transmitting tissue of the style and ovary ( Lennon and Lord , 2000 ) . The transmitting tissue has an essential function in pollen tube guidance , providing chemical attractants and nutrients ( Crawford and Yanofsky , 2008; Higashiyama and Hamamura , 2008 ) . In contrast to these accumulating data showing the existence of factors mediating pollen tube growth in the pistil , whether guidance cues exist at the very early stage of pollen tube emergence and growth in the papilla cell wall remains largely unknown . The cell wall constitutes a stiff substrate and hence a mechanical barrier to pollen tube progression . There are numerous examples in animal cells demonstrating that mechanical properties of the cellular environment , and in particular rigidity , mediate cell signalling , proliferation , differentiation and migration ( Discher et al . , 2005; Fu et al . , 2010; Ermis et al . , 2018 ) . In plant cells , cell wall rigidity depends mainly on its major component , cellulose , which is synthesized by plasma membrane-localized cellulose synthase complexes ( CSCs ) moving along cortical microtubule ( CMT ) tracks ( Paredez et al . , 2006 ) . While penetrating the cell wall , the pollen tube exerts a pressure onto the stigmatic cell ( Sanati Nezhad and Geitmann , 2013 ) . Physical forces are known to reorganise the cortical microtubules ( CMTs ) , which by directing CSCs to the plasma membrane , reinforce wall stiffness by novel cellulose microfibril ( CMF ) synthesis ( Paredez et al . , 2006; Sampathkumar et al . , 2014 ) . Hence , there is an intricate interconnection between CMT organisation , CMF deposition and cell wall rigidity ( Xiao and Anderson , 2016 ) . A major regulatory element of CMT dynamics is the KATANIN ( KTN1 ) microtubule-severing enzyme , which allows CMT reorientation following mechanical stimulation ( Uyttewaal et al . , 2012; Sampathkumar et al . , 2014; Louveaux et al . , 2016 ) . Here we investigated whether the CMT network of papilla cells might contribute to pollen tube growth and guidance in stigmatic cells by combining cell imaging techniques , genetic tools , chemical analysis of the cell wall and atomic force microscopy . We show that isotropic reorientation of CMTs occurs in aged Col-0 and ktn1-5 papilla cells , which is accompanied by a change in the growth direction of pollen tubes that tend to make coils on papillae . We show that CMT reorganisation is associated with isotropic rearrangement of CMFs , high content of crystalline cellulose and particular cell wall mechanics of ktn1-5 stigmas . Altogether , our results indicate that cytoskeleton dynamics and mechanical properties of the cell wall , which both depend on KTN1 activity , have a major role in guiding early pollen tube growth in stigma papillae . To assess the functional role of stigmatic CMTs in pollen-papilla cell interaction , we first analysed their organisation in papillae at stages 12 to 15 of stigma development as described ( Smyth et al . , 1990; Figure 1A , B ) . We generated a transgenic line expressing the CMT marker MAP65 . 1-citrine under the control of the stigma-specific promoter SLR1 ( Fobis-Loisy et al . , 2007 ) . Before ( stage 12 ) and at anthesis ( stage 13 ) , where flowers are fully fertile ( Kandasamy et al . , 1994 ) , the CMTs were aligned perpendicularly to the longitudinal axis of papilla cells and were highly anisotropic ( median value of 0 . 40 ) ( Figure 1C , D ) . At stage 14 , when anthers extend above the stigma , the CMT pattern became less organised , with a higher variability in anisotropy values . Finally , at stage 15 when the stigma extends above anthers , CMT anisotropy had a median value of 0 . 09 indicative of an isotropic orientation of CMTs ( Figure 1C , D ) . These findings reveal that the papilla CMT cytoskeleton is dynamic during development , with a change of CMT array orientation from anisotropy to isotropy . We then wondered whether this change in CMT organisation could be correlated with pollen tube growth . To this end , we self-pollinated Col-0 papillae from stages 12 to 15 and examined pollen tube growth one hour after pollination by scanning electron microscopy ( SEM ) ( Figure 2A ) . At stage 12 and 13 , we found most ( ~60% ) pollen tubes grew straight in the papillae , whereas about 30% and 10% of tubes made half-turn or one turn around stigmatic cells , respectively ( Figure 2B ) . At later stages of development , the tendency to coil around the papillae increased , with 3 . 5-fold more pollen tubes at stage 14 and 5 . 5-fold more at stage 15 making one or more than one turn around papillae . These results suggest that CMT organisation in the papilla impacts the direction of growth of pollen tubes and that loss of CMT anisotropy is associated with coiled growth . We next asked whether the pressure exerted by the growing pollen tube in the cell wall may induce changes in the CMT organisation of the papilla cell . To this end , we set up a live imaging system where pollinated stigmas were maintained in air so as to prevent immediate hydration and burst of pollen grains . During the time course of the experiment , starting at the beginning of pollen tube emergence on the stigma papilla at stage 13 ( t0 ) up to 13 min after germination , we found no major alteration of CMT arrays from early to late penetration stages ( Figure 3A ) . This result was in contradiction with previous immunostaining analysis that showed fragmentation of CMTs following compatible pollination in Brassica napus ( Samuel et al . , 2011 ) . To confirm our result and gain better insight into CMT organisation , we then observed pollinated stigmas at a single time point , that is 30 min after pollination , using a mounting medium to get higher resolution images . Again , we did not detect any reorganisation or degradation of CMTs ( Figure 3B ) . However , we found the stability of CMTs to be very sensitive to strong pressure , as when the coverslip was applied not cautiously on the stigma , a break-down of CMT arrays occurred ( Figure 3—figure supplement 1 ) . These results indicate that the pressure exerted by the pollen tube while growing in the papilla cell wall does not alter the organisation of stigmatic CMTs . To test the direct impact of stigma CMTs on pollen tube growth direction , we examined whether the destabilization of CMTs in Col-0 papillae could affect pollen tube growth . To this end , we treated stigmas by local application of the depolymerizing microtubule drug oryzalin in lanolin pasted around the style . After 4 hr of drug treatment , no more CMT labelling was detected in papillae , while CMTs were clearly visible in mock-treated ( DMSO ) stigmas ( Figure 4A ) . Stigmas were then pollinated with Col-0 pollen and one hour later observed by SEM . Pollen tubes were found to make one turn more frequently on drug-treated than on control papillae , with more than a 3-fold increase of tubes making one turn ( Figure 4B , C ) . To confirm the relation between stigma CMTs and pollen behaviour , we examined pollen tube growth on stigmas of the katanin1-5 mutant , which is known to exhibit reduced CMT array anisotropy in root cells ( Bichet et al . , 2001; Burk et al . , 2001; Burk and Ye , 2002 ) . Because the CMT organisation in ktn1-5 papillae is unknown , we crossed ktn1-5 with the MAP65-1-citrine marker line and found that CMT arrays were more isotropic in ktn1-5 papillae when compared with those of the WT ( Figure 5A , C ) . We then analysed Col-0 pollen behaviour on ktn1-5 stigmatic cells at stage 13 ( Figure 5B ) . We observed that Col-0 pollen tubes acquired a strong tendency to coil around ktn1-5 papillae , with more than 60% of tubes making one or more than one turn around papillae , sometimes making up to six turns , before reaching the base of the cell ( Figure 5D; Figure 5—figure supplement 1A–C ) . In some rare cases , pollen tubes even grew upward in the ktn1-5 mutant and appeared blocked at the tip of the papilla ( Figure 5—figure supplement 1B , D ) . The number of coils was significantly higher than on oryzalin-treated papillae . Indeed , 25% of the pollen tubes made at least two coils in ktn1-5 papillae whereas this percentage represented only 4% on the oryzalin treated stigmas ( Figure 4C; Figure 5D ) . To check whether papilla receptivity to pollen might be impaired in ktn1-5 stigmas , we examined hydration and tube growth of Col-0 pollen deposited on stage-13 Col-0 or ktn1-5 stigmas . We found no significant differences in pollen hydration related to the stigma type ( Figure 5—figure supplement 2A ) and pollen tubes grew in the transmitting tract and reached the ovules in ktn1-5 pistils like in Col-0 ( Figure 5—figure supplement 2B ) . We then wondered whether alteration of the KATANIN function in pollen might affect pollen tube growth directionality on Col-0 or ktn1-5 stigmas . Cross-pollination revealed that ktn1-5 pollen tubes grew like Col-0 pollen tubes on stage-13 Col-0 papillae , while they coiled on ktn1-5 papillae ( Figure 5—figure supplement 2C ) . Seed counts showed that fertility slightly decreased when ktn1-5 pollen was used to pollinate Col-0 stigmas , and was strongly affected when ktn1-5 was used as female , irrespective of whether the pollen was Col-0 or ktn1-5 ( Figure 5—figure supplement 2D ) . This indicates that the fertilisation process is impaired in ktn1-5 ovules . Altogether , these results show that the coiled phenotype observed on ktn1-5 stigmas is only dependent on defects occurring at the papilla level , that the stigma receptivity of ktn1-5 mutant is not altered and confirm that stigmatic CMTs , somehow , contribute to the directional growth of pollen tubes in papilla cells . The katanin1/fra2 mutant was initially described as a mutant impaired in cell wall biosynthesis and CMT array organisation ( Burk et al . , 2001 ) . This prompted us to investigate whether other mutants affected in cell wall biogenesis might exhibit the coiled pollen tube phenotype . We selected mutants impaired in the cellulose synthase complex ( kor1 . 1 , prc1 and any1 ) , hemicellulose biosynthesis ( xxt1 xxt2 , xyl1 . 4 ) and pectin content ( qua2 . 1 ) , for which expression of the corresponding genes in stigma was confirmed ( Figure 5—source data 2 ) . Strikingly , none of the 6 cell wall mutants displayed the coiled pollen tube phenotype ( Figure 5—figure supplement 3A–F ) . To ascertain that the cell wall composition was altered in the stigma papillae of ktn1-5 and cell wall mutants , we undertook sugar analysis of ktn1-5 and xxt1 xxt2 stigmas and compared them with that of Col-0 . We chose the xxt1 xxt2 mutant because its cell wall composition had been already well described ( Cavalier et al . , 2008; Xiao et al . , 2016 ) . The relative abundance of monosaccharides in pistil cell walls varied between the three genotypes with major differences detected in the ktn1-5 cell wall composition that had an increased abundance of glucose , this latter deriving either from hemicellulose and amorphous cellulose or from crystalline cellulose ( Figure 5—figure supplement 4A ) . ktn1-5 cell wall was characterized by the highest relative abundance of crystalline cellulose , which was 25% higher than in Col-0 and xxt1 xxt2 . To explore whether side chains of hemicellulose might also be affected in the katanin mutant , we performed an enzymatic fingerprinting of xyloglucans from dissected stigmas of Col-0 and ktn1-5; xxt1xxt2 was not analysed as it lacks xyloglucan side chains . We found in ktn1-5 a decrease in relative abundance of short side chains with terminal xylose ( XXG and XXXG ) and an increase in the longer ones with terminal fucose ( e . g XXFG ) ( Figure 5—figure supplement 4B ) . The chemical analysis confirms that the cell wall composition is altered in ktn1-5 stigmas and presents some specific features compared with Col-0 and xxt1 xxt2 stigmas . However , combined with the results of pollination on the six cell wall mutants , we cannot establish a causal effect of chemical alterations on the pollen tube coiled phenotype . This suggests that the relation between CMT organisation , cell wall composition and pollen tube trajectory is more complex than anticipated . Because CMTs guide cellulose deposition , they also control the directional elongation of plant cells ( Baskin , 2005 ) . Indeed , katanin1/fra2 cells are wider and shorter than wild type ( Bichet et al . , 2001; Burk et al . , 2001 ) , and we may predict that the shape of papilla cells in the mutant might be similarly altered . The contribution of stigmatic CMTs to pollen tube growth may thus be mediated by papilla cell shape only; for instance , wider papilla cells in ktn1-5 would promote pollen tube coiling . To check that possibility , we measured the length and width of papilla cells in Col-0 ( at stage 13 and 15 ) , ktn1-5 , xxt1 xxt2 and any1 ( Figure 5—figure supplement 5 ) . We did not find any correlation between papilla length and the coiled phenotype . Indeed , coiled phenotype was observed in stage-15 WT papillae that were longer than those at stage-13 , as well as in ktn1-5 papillae that had a length similar to stage-13 WT papillae . The correlation between papilla width and coiled phenotype was also not clear-cut . As anticipated , ktn1-5 mutant exhibited wider papillae than Col-0 . However , xxt1 xxt2 and any1 papillae were also wider than the WT but did not display the coiled phenotype . Altogether , this suggests that papilla morphology is not sufficient to explain the coiled phenotype . Because the main role of CMTs in plant cells is to guide the trajectory of CSCs , thereby impacting the mechanical anisotropy of the cell wall , we analysed the cell walls of Col-0 and ktn1-5 papillae following pollination . First , using Transmission Electron Microscopy ( TEM ) , we found Col-0 pollen tubes penetrated the cuticle and grew between the two layers of the papilla cell wall , as previously described ( Kandasamy et al . , 1994 ) , in both Col-0 and ktn1-5 papilla cells ( Figure 6A ) . We did not detect any significant difference in the ultrastructure of cell walls ( Figure 6—figure supplement 1 ) . Interestingly , as the pollen tube progresses through the papilla cell wall , it generates a bump ( external deformation ) and an invagination ( internal deformation ) in the cell wall ( Figure 6A ) . As such deformations could reflect differences in wall properties , we quantified the external ( extD ) and internal ( intD ) deformation following pollination of Col-0 and ktn1-5 stigmas . To visualize more clearly this deformation , we pollinated stigmas expressing the plasma membrane protein LTI6B fused to GFP ( LTI6B-GFP ) with pollen whose tube was labelled with the red fluorescent protein RFP driven by the ACT11 promoter ( Rotman et al . , 2005; Figure 6B ) . We found that Col-0 pollen tubes grew with almost equal extD and intD values in Col-0 cell wall . However , the ratio between extD and intD was about three when Col-0 pollen tubes grew in ktn1-5 papilla cells ( Figure 6B–D ) . These quantitative data were consistent with the observation that pollen tubes appeared more prominent on ktn1-5 stigmatic cells using SEM ( Figure 5B; Figure 5—figure supplement 1B ) . Assuming that stigmatic cells are pressurized by their turgor pressure , this may reflect the presence of softer walls in ktn1-5 papillae . To test this hypothesis , we assessed the stiffness of Col-0 and ktn1-5 papilla cell walls using Atomic Force Microscopy ( AFM ) with a 200 nm indentation . We found that cell wall stiffness in ktn1-5 papilla cells was about 30% lower than that in stage 13 WT cells ( Figure 6E , F ) . To confirm this result , we investigated the stiffness of the papilla cell wall on WT stigmas at stage 15 , where increased coiled pollen tubes were detected ( Figure 2A , B ) . We found the cell wall to be softer than that of papillae at stage 13 but stiffer than that of the ktn1-5 ( Figure 6E , F ) . We then reasoned that the presence of softer walls should also affect the pollen tube growth rate , stiffer walls reducing growth . Thus , measuring the growth rate would provide an indication of the resistance force encountered by the tube while growing in the papilla cell wall . We thus monitored the growth rate of Col-0 pollen tubes in Col-0 and ktn1-5 papillae . We found that pollen tubes grew faster ( ~x 1 . 8 ) within ktn1-5 papillae ( Figure 6G ) . Similarly , we found that pollen tube growth on stage 15 papillae was faster ( ~x 1 . 6 ) than on stage 13 papillae ( Figure 6H ) . These results suggest that the coiled phenotype is associated with a higher pollen tube growth rate and hence , that papilla cell walls of ktn1-5 and stage-15 papillae exhibit less resistance to tube penetration . Altogether , these data suggest that KATANIN-dependent mechanical properties of papilla cell wall play a role in pollen tube guidance . Because CMT disorganisation in papillae affects both wall stiffness and mechanical anisotropy , we next investigated the relative contribution of these two parameters in pollen tube growth . Hypocotyl cells of the xxt1 xxt2 double mutant were reported to display CMT orientation defects and reduced stiffness of the cell wall ( Xiao et al . , 2016 ) , features that we found in ktn1-5 papilla cells . However , growing pollen tubes on xxt1 xxt2 stigmas did not coil and grew like on WT stigmas , questioning whether the cell wall stiffness of stigmatic cells was actually affected in the mutant . Using AFM , we found the cell wall of xxt1 xxt2 papillae to be about 30% softer than that of Col-0 papillae , i . e . very similar to that of ktn1-5 ( Figure 6F ) . Despite this similarity , the pollen tube growth rate in xxt1 xxt2 stigmas was identical to Col-0 stigmas at stage 13 ( Figure 6H ) . In addition , contrary to ktn1-5 , pollen tubes were not prominent while growing in xxt1 xxt2 papilla cell walls ( Figure 5—figure supplement 3G ) . These results suggested that although ktn1-5 and xxt1 xxt2 papillae had similar cell wall stiffness as inferred from AFM measurements , they exhibited different mechanical constraints to pollen tube growth . Because of the close relationship between CMT organisation , CMF deposition and cell wall rigidity ( Xiao et al . , 2016; Xiao and Anderson , 2016 ) , we suspected that CMT and CMF organisation might be different between the two mutants and be the possible causal agent of the mechanical differences and related tube coiled phenotype observed in ktn1-5 . To check this possibility , and to gain insight into the predominant cellulose pattern in the stigmatic cell walls , we stained CMFs by using Direct Red 23 . We found cellulose fibres to be highly aligned and slanted to the longitudinal axis of the papilla in stage 13 Col-0 stigmas , whereas ktn1-5 and stage 15 Col-0 papillae both displayed a clearly disordered CMF pattern ( Figure 7A ) with CMFs forming thicker and more spaced bundles ( Figure 7B ) . By contrast , CMFs in xxt1 xxt2 papillae at stage 13 organised in a dense and well oriented pattern of fibres , mostly perpendicular to the papilla long axis , which resembled Col-0 stage 13 CMF organisation ( Figure 7A , B ) . These results suggest that KTN1 loss-of-function alters mechanical properties of the papilla cell wall in a complex manner , lowering both its stiffness to pressure applied perpendicularly to its surface ( e . g . , by the AFM indenter ) and its resistance to pollen tube progression . This latter modification appears as the main cause of the pollen tube coiled phenotype . By contrast , loss of XXT1 and XXT2 functions , by perturbing cell wall stiffness to indenter pressure only , does not induce change in pollen tube growth . Altogether , our data suggest that the mechanical anisotropy of papilla cell walls plays a key role in pollen tube trajectory . Previous work reported that rearrangements of actin microfilaments ( Iwano et al . , 2007; Rozier et al . , 2020 ) and destabilization of CMTs ( Samuel et al . , 2011 ) occur in stigmas following compatible pollination in Brassicaceae species . However , CMT pattern and dynamics during papilla development have never been described . Our data show that during the course of stigma maturation , which is associated with papilla cell elongation , CMT bundles progressively move from perpendicular ( anisotropic ) to the elongation axis at stage 12 to disorganised ( isotropic ) at stage 15 . This correlates with the known CMT dynamics during elongation in plant cells , where CMT arrays are highly anisotropic in young cells and become more isotropic as cells differentiate ( Baskin , 2005; Landrein and Hamant , 2013 ) . During invasion of plant cells by fungal or Oomycete pathogens or upon mechanical stimulation , the network of microtubules is rearranged , generally associated first with the depolymerization of CMTs at the contact site and later with the formation of an array of microtubules surrounding the invading organism or microneedle ( Hardham , 2013 ) . Interestingly , penetration of the pollen tube in the papilla cell wall does not lead to such microtubule reorganisation ( Figure 3 ) . This suggests that at the cellular level , the invading pollen tube does not elicit a defence-like response in the papilla , which accepts and supports pollen tube growth . In addition , we show that when the stigma is manually pressed between slide and coverslip , the mechanical stimulation applied perpendicularly to the wall leads to CMT fragmentation ( Figure 3—figure supplement 1 ) . This indicates that although the papilla microtubule network is sensitive to mechanical forces , the pressure exerted by the pollen tube is not strong enough to perturb CMT organisation . One possible explanation is that the penetration force developed by the growing pollen tube acts mainly in plane to the cell wall surface and hence would only have a little effect on the papilla cell , contrary to a pressure applied perpendicularly to the wall . Our data are in contradiction with the earlier observation that growing pollen tube provokes fragmentation of CMTs in the papilla ( Samuel et al . , 2011 ) . This discrepancy is likely to relate to the microscopic approaches used between the two studies , live imaging and a fluorescent microtubule probe in our case , chemically fixed material and immunostaining of microtubules in Samuel et al . , 2011 . The progressive randomisation of CMT orientation we observed in papilla cells during ageing is accompanied by an increased coiled growth of pollen tubes in papillae . Similarly , when CMTs are destabilized by the microtubule depolymerizing drug oryzalin , coiled pollen tubes are more frequently observed compared with untreated control stigmas . These results reveal a link between the stigmatic CMT cytoskeleton organisation and the trajectory that pollen tube takes while growing in the papilla cell wall , although the effects of oryzalin on cell wall properties remain unknown . The fact that the most striking effect on pollen tube growth was found on ktn1-5 mutant suggests that the coiled phenotype depends not only on the CMT organisation but implicates other factors . Among various phenotypic alterations described in loss-of-function mutants for KTN1 , are the impaired cell mechanical properties and cell elongation , defects in CMT organisation , cell wall composition and CMF orientation ( Burk et al . , 2001; Burk and Ye , 2002; Ryden et al . , 2003 ) . Interestingly , we found that the protuberance of pollen tubes at the surface of ktn1-5 papillae was associated with a faster growth rate and a lower rigidity of the cell wall compared with Col-0 ( Figure 6 ) . This places mechanics of the cell wall as a likely component involved in the coiled phenotype . Surprisingly , of the six cell wall mutants we analysed , including xxt1 xxt2 and prc1 known to exhibit both abnormal CMT organisation and softer cell walls , none induced the coiled phenotype displayed by ktn1-5 papillae . More remarkably , despite the similar stiffness of xxt1 xxt2 and ktn1-5 papilla cell walls as measured by AFM , pollen tubes behaved differently on these two stigma types , pollen tubes not only coiling but also growing faster on ktn1-5 stigmas . This observation is in accordance with the recent findings that changes in indentation properties of plant cell walls do not correlate with changes in tensile stiffness as described in cell-free strips of onion epidermal walls ( Zhang et al . , 2019 ) . In our experiments , indentations tell us about the elasticity of the wall to a pressure normal to the plane of the wall surface ( i . e . out-of-plane mechanics ) , while pollen tube growth rates inform about the resistance of the wall to a force in-plane to the papilla wall ( i . e . in-plane mechanics ) . Several lines of evidence lead us to suggest that mechanical properties of the cell wall are not identical in the two mutants . Previous work showed that cellulose crystallinity is one factor implicated in cell wall mechanics . Indeed , the epidermal cells of the inflorescence stem in the any1 mutant have the relative amount of crystalline cellulose significantly reduced , while the total content of cellulose in the cell wall is unaltered , and this is accompanied by reduced stiffness of the cell wall ( Fujita et al . , 2013; Altartouri et al . , 2019 ) . Our chemical analysis reveals that the cell wall composition is different between xxt1 xxt2 and ktn1-5 stigmas , with this latter having 25% more crystalline cellulose , and hence should exhibit stiffer cell walls ( Figure 5—figure supplement 4 ) . This data is at odds with AFM measurements but a possible explanation would be that the two layers detected by TEM in the papilla cell wall ( Figure 6—figure supplement 1 ) display different mechanical properties , the inner layer being stiffer than the outer one in the ktn1-5 mutant . This may explain the protruding growth of pollen tubes observed in ktn1-5 papillae , the tube pushing away the softer outer layer but not the inner ( Figure 6A–D ) . Interestingly , Zhang et al . , 2019 reported that cellulose networks largely determine in-plane mechanics whereas out-of-plane mechanics depends on both homogalacturonan ( pectin ) and cellulose networks . Cell walls of the qua2-1 mutant are specifically affected in pectins , with a 50% reduction in homogalacturonan content compared with Col-0 without observable changes in other polysaccharides ( Mouille et al . , 2007 ) . In light of these data , it is noteworthy that papilla cells of the qua2-1 mutant do not induce pollen tube turns ( Figure 5—figure supplement 3 ) , which suggests that pectins play no significant role in the pollen tube phenotype and that the main factor contributing to pollen tube guidance is in-plane mechanics , and hence relies on the cellulose network . The orientation of the rigid CMFs is another determinant of cell wall mechanics . The ktn1-5 mutant was described to have a severe reduction in cell length and an increase in cell width in all organs ( Burk et al . , 2001 ) . This cellular phenotype was attributed to the distorted deposition of CMFs correlated with the isotropic orientation of CMTs , whereas in WT cells , CMFs like CMTs are oriented perpendicularly to the elongation axis ( Burk and Ye , 2002 ) . Similarly , we found that orientation of CMFs is also altered in ktn1-5 papillae ( Figure 7 ) and this was associated with larger papilla cells ( Figure 5—figure supplement 5 ) . Hence , we may assume that mechanical anisotropy , described as the cell wall anisotropy made by the orientation of the rigid CMFs ( Sassi et al . , 2014 ) , is impaired in ktn1-5 papillae . Contrary to the ktn1-5 mutant , papillae of the xxt1 xxt2 double mutant display an anisotropic CMF pattern with parallel fibres approximatively oriented transversally to the papilla axis . These data are in agreement with those reported for etiolated hypocotyl cells , where CMFs are largely parallel to one another , straighter than in WT and oriented almost transversely to the long axis of the cell ( Xiao et al . , 2016 ) . In addition , we found the papilla cell shape of xxt1 xxt2 stigmas to be closer to that of Col-0 than that of ktn1-5 papillae , which is consistent with an anisotropic growth of the papillae in the xxt1 xxt2 mutant compared with ktn1-5 . Apart from the similar stiffness of papilla cell walls as deduced from AFM measurements , the main differences between ktn1-5 and xxt1 xxt2 cell walls are the cell wall composition , orientation of CMFs , and possible changes in molecular connections between cell wall components and plasma membrane and/or cytoskeleton proteins . Indeed , a recent proteomic study revealed that loss of KATANIN function is associated with the decrease in abundance of several cytoskeleton proteins , such as profilin 1 , actin-depolymerizing factor 3 and actin 7 ( Takáč et al . , 2017 ) , whereas targeted quantitative RT-PCR showed that several Microtubule-associated protein ( MAP ) and wall signal receptor genes showed lower expression levels in xxt1 xxt2 ( Xiao et al . , 2016 ) . In this latter study , KTN1 expression level was shown to be unchanged compared with Col-0 . Altogether , our study suggests that KTN1 , by maintaining the papilla mechanical anisotropy , has a key function in mediating early pollen tube guidance on stigma papillae . The coiled phenotype was not only observed in ktn1-5 but also in the WT Col-0 papillae at stage 15 . Remarkably , several lines of evidence reveal that papilla cells from stage 15 share common features with ktn1-5 papillae , such as CMT and CMF increased isotropy , decreased stiffness of the cell wall as deduced from AFM and less resistance of the wall to pollen tube growth . Isotropic orientation of the cytoskeleton at stage 15 is likely to relate to cell elongation , which is known to be accompanied by cytoskeleton reorganisation ( Crowell et al . , 2011; Zhang et al . , 2014 ) . At the organ level , it has been suggested that the mechanical anisotropy of the wall restrains organ emergence ( Sassi et al . , 2014 ) . The authors propose that for the same wall stiffness , a cell wall with isotropic properties would lead to larger outgrowth than a wall with anisotropic properties . Our data are consistent with this hypothesis , albeit at the subcellular scale , since large protuberance of papilla wall following pollen tube growth is observed in ktn1-5 papilla cells , exhibiting walls with isotropic properties . It remains unclear how mechanical anisotropy guides pollen tube growth . We can hypothesise that as pollen tube grows inside the wall , it encounters recently deposited CMFs on the inner side of the wall ( facing the cytoplasm ) and older CMFs on the outer side . It is likely that these layers have different mechanical properties related to CMF orientation ( Baskin , 2005 ) . KATANIN , by acting on CMT dynamics and CMF organisation , would fine-tune the mechanical properties of the matrix through which the pollen tube grows . It is worth noting that growth rate of pollen tubes germinated in vitro is slowed down when pollen tubes pass through a microgap of a microfluidic device , the tubes adapting their invasive force to the mechanical constraints ( Sanati Nezhad et al . , 2013 ) . Based on measurements of pollen tube growth rates , our data show that the tube tip , while progressing in the papilla wall , senses the mechanical features of its environment and reacts accordingly . Hence , it reveals some unanticipated internal and hidden properties of the cell wall . Our study shows that mechanics plays a key role in early pollen tube guidance in the papilla cell . This role is mediated by a specific CMT/CMF organisation and mechanical anisotropy of the papilla cell , which both are dependent on KTN1 . Importantly , these specific mechanical properties of the stigmatic cells prevent emerging pollen tubes to grow upward on papillae and straighten pollen tube direction , helping the tube to find its correct path to the stylar transmitting tract . Our findings also raise the question about the biological importance of the mechanical changes that occur during stigma development . Indeed , for aged stigmas , modifications of the mechanical properties of the cell wall , accompanied by an acceleration of pollen tube growth , could be seen as beneficial for the plant by favouring ultimate fertilization and seed set on old flowers , and hence supporting dissemination of the species . This assumption is particularly relevant given that the pistil length increases with ageing and similarly the journey the pollen tube has to travel to reach the ovules . A recent work showed that ageing is associated with decreased fertility due to stigma senescence , which is initiated at stage 16–17 ( Gao et al . , 2018 ) . The faster the pollen tube will grow in old stigmas , the more chance it will have to circumvent papilla cell death and hence to fertilize the ovule . We may suggest that KTN1 , by mediating mechanical anisotropy of stigmatic cells and promoting pollen tube growth in old stigmas , has played an evolutionary role in the success of fertilization . To conclude , we uncovered a yet unexpected function for KTN1 in early pollen tube guidance on the stigma . In addition , our study also clearly unveils that the mechanical properties of one single cell ( e . g . , the stigmatic papilla ) impact the behaviour of its neighbouring cell ( e . g . , the pollen tube ) . Arabidopsis thaliana , ecotype Columbia ( Col-0 ) , Arabidopsis transgenic plants generated in this study and Arabidopsis mutants were grown in soil under long-day conditions ( 16 hr of light/8 hr of dark , 21 °C / 19°C ) with a relative humidity around 60% . ktn1-5 ( SAIL_343_D12 ) , xxt1xxt2 , prc1 . 1 , qua2 . 1 , xyl1 . 4 , kor1 . 1 and any1 mutant lines were described previously ( Cavalier et al . , 2008; Fagard et al . , 2000; Fujita et al . , 2013; Lin et al . , 2013; Mouille et al . , 2007; Nicol et al . , 1998; Sampedro et al . , 2010; Shoji et al . , 2004 ) . All mutants were in Col-0 background except kor1 . 1 which was in WS . All stigmas were analysed at stage 13 of flower development ( Smyth et al . , 1990 ) except when specified . We used the Gateway technology ( Life Technologies , USA ) and two sets of Gateway-compatible binary T-DNA destination vectors ( Hellens et al . , 2000; Karimi et al . , 2002 ) for expression of transgenes in A . thaliana . The DNA fragment containing the Brassica oleracea SLR1 promoter was inserted into the pDONP4-P1R vector . The 165 bp-LTI6B fragment was introduced into the pDONR207 vector . MAP65 gene spanning the coding region from start to stop codons was introduced into the pDONR221 vector . CDS from citrine or GFP were cloned into the pDONP2R-P3 vector . Final constructs , pSLR1::MAP65-citrine and pSLR1::LTI6B-GFP were obtained by a three-fragment recombination system ( Life Technologies ) using the pK7m34GW and the pB7m34GW destination vectors , respectively . We generated a pACT11::RFP construct by amplifying the promoter of the A . thaliana ACTIN-11 gene and cloning it into the pGreenII gateway vector in front of the RFP coding sequence . Transgenic lines were generated by Agrobacterium tumefaciens-mediated transformation of A . thaliana Col-0 as described ( Logemann et al . , 2006 ) . Unique insertion lines , homozygous for the transgene were selected . We introduced the pSLR1::LTI6B-GFP or pSLR1::MAP65-citrine construct in ktn1-5 background by crossing and further selecting the progeny on antibiotic containing medium . Flowers at stage 13 and 15 were emasculated and pollinated on plants with mature pollen from the pACT11::RFP line . Immediately after pollination , stigmas were mounted between two coverslips maintained separated by four grease plugs placed at each coverslip corner . To maintain a constant humidity without adding liquid directly on the stigma surface , we used a wet piece of tissue in contact with the base of the stigma . Pollinated stigmas were observed under a Zeiss microscope ( AxioObserver Z1 ) equipped with a spinning disk module ( CSU-W1-T3 , Yokogawa ) using a 40x Plan-Apochromat objective ( numerical aperture 1 . 1 , water immersion ) . Serial confocal images were acquired in the entire volume of the stigma every 1 µm and every minute . Images were processed with Image J software and pollen tube lengths were measured . Pistils were placed straight in a 2% agar MS medium and 0 . 8% low-melting agarose was added up to a certain level where the papilla cells were maintained well immobilised in agarose while leaving the top accessible to the indenter . This set up allowed accurate measurements of cell wall stiffness on the dome-shaped top of papilla cells . AFM indentation experiments were carried out with a Catalyst Bioscope ( Bruker Nano Surface , Santa Barbara , CA , USA ) that was mounted on an optical microscope ( MacroFluo , Leica , Germany ) equipped with a x10 objective . All quantitative measurements were performed using standard pyramidal tips ( RFESP-190 ( Bruker ) ) . The tip radius given by the manufacturer was 8–12 nm . The spring constant of the cantilever was measured using the thermal tune method and was 35 N/m . The deflection sensitivity of the cantilever was calibrated against a sapphire wafer . All experiments were made in ambient air at room temperature . Matrix of 10 × 10 measurements ( step 500 nm ) was obtained for each papilla , with a 1µN force . The Young’s Modulus was estimated using the Nanoscope Analysis ( Bruker ) software , using the Sneddon model with a < 200 nm indentation . Flowers from stages 12 to 15 were emasculated and pollinated on plants with mature WT pollen . One hour after pollination , pistils were cut in the middle of the ovary , deposited on a SEM platform and observed under Hirox SEM SH-3000 at −20°C , with an accelerating voltage of 15kV . Images were processed with ImageJ software and pollen tube direction was quantified by counting the number of turns made by the tube , only on papillae that received one unique pollen grain . Stage 13 flowers were emasculated and pollinated on plants with mature WT pollen . One hour after pollination , pistils were immersed in fixative solution containing 2 . 5% glutaraldehyde and 2 . 5% paraformaldehyde in 0 . 1 M phosphate buffer ( pH 7 . 2 ) and after 4 rounds of 30 min vacuum , they were incubated in fixative for 12 hr at room temperature . Pistils were then washed in phosphate buffer and further fixed in 1% osmium tetroxide in 0 . 1 M phosphate buffer ( pH 7 . 2 ) for 1 . 5 hr at room temperature . After rinsing in phosphate buffer and distilled water , samples were dehydrated through an ethanol series , impregnated in increasing concentrations of SPURR resin over a period of 3 days before being polymerized at 70°C for 18 hr , sectioned ( 65 nm sections ) and imaged at 80 kV using an FEI TEM tecnaiSpirit with 4 k x 4 k eagle CCD . Flowers from MAP65 lines were emasculated and stigmas were observed under confocal microscope . Images were processed with ImageJ software and quantitative analyses of the average orientation and anisotropy of CMTs were performed using FibrilTool , an ImageJ plug-in ( Boudaoud et al . , 2014 ) . Anisotropy values range from 0 to 1; 0 indicates pure isotropy , and one pure anisotropy . A droplet of Adigor was applied at 2 . 5% ( v/v in water ) on the stigma during 4 hr to facilitate subsequent chemical treatments . As described for hypocotyls ( Landrein and Hamant , 2013 ) , pistils were incubated in 12 . 5% glacial acetic acid for 1 hr and dehydrated in 100% and then 50% ethanol for 20 min each . Pistils were then washed in water during 20 min and stored in 1M KOH during 2 days . Pistils were then stained using 0 . 02% ( w/v ) Direct Red 23 dye ( Sigma-Aldrich ) during 4 hr and washed with distilled water . Pistils were observed with a mRFP filter ( excited at 561 nm ) under a Zeiss LSM 880 confocal microscope using a 63x Plan-Apochromat objective ( numerical aperture 1 . 4 , oil immersion ) . Stigmas were imaged by taking a z-stack of 0 . 2 μm sections in the papilla cells . Image analysis was performed by doing a 3D-projection and measuring the fluorescence intensity longitudinally along the papilla axis using ImageJ software . Flowers from LTI6B lines were emasculated and pollinated with mature pollen from the pACT11::RFP line . 20 min after pollination , stigmas were observed under confocal microscope . Serial confocal images every 1 µm encompassing the entire volume of the stigma were recorded and processed with ImageJ software . Plasma membrane deformation was estimated by choosing the slide from the stack that corresponded to the focus plan of the contact site with the RFP-labelled pollen tube . On the bright field image corresponding to the selected slide , a line was drawn connecting the two ridges of the invagination of the papilla . Two perpendicular lines , one toward the exterior ( ExtD ) of the papilla to the maximum point of deformation , and the other toward the interior ( IntD ) on the GFP image were measured , respectively . To avoid contact of pollen grains with liquid , we performed local applications of oryzalin ( Chemical service , Supelco ) as described ( Sassi et al . , 2014 ) , using lanolin pasted around the style , just under the stigmatic cells , at 833 µg/mL ( DMSO ) , for 4 hr at 21°C . Oryzalin-treated pistils were pollinated with mature WT pollen and 1 hr after pollination observed under SEM . XyG composition of stigmas was performed following the oligosaccharide fingerprinting set up by Lerouxel et al . , 2002 . 10 stigmas per genotype were dissected and kept in 96% ethanol . After ethanol removal , XyG oligosaccharides were generated by treating samples with endoglucanase in 50 mM sodium acetate buffer , pH 5 , overnight at 37°C . Matrix-assisted laser-desorption ionization time of flight mass spectrometry of the XyG oligosaccharides was recorded with a MALDI/TOF Bruker Reflex III using super-DHB ( 9:1 mixture of 2 , 5-dihydroxy-benzoic acid and 2-hydroxy-5-methoxy-benzoic acid; Sigma-Aldrich , sigmaaldrich . com ) as matrix . Alcohol insoluble residues were obtained from around 3 to 5 mg of fresh pistils . They were collected for analysis and fixed in 96% ethanol before incubation for 30 min at 70°C . The pellet was then washed twice with 96% ethanol and twice with acetone . The remaining pellet is called alcohol insoluble residues ( AIR ) and was dried in a fume hood overnight at room temperature . Cell wall monosaccharide content of the non‐cellulosic fraction was determined by hydrolysis of all the AIR obtained from the pistils with 2 M TFA for 1 hr at 120°C . After cooling and centrifugation , the supernatant was dried under a vacuum , resuspended in 200 μl of water and retained for analysis . To obtain the glucose content of the crystalline cellulose fraction , the TFA-insoluble pellet was further hydrolysed with 72% ( v/v ) sulfuric acid for 1 hr at room temperature . The sulfuric acid was then diluted to 1 M with water and the samples incubated at 100°C for 3 hr . All samples were filtered using a 20‐μm filter caps , and quantified by HPAEC‐PAD on a Dionex ICS‐5000 instrument ( ThermoFisher Scientific ) as described ( Fang et al . , 2016 ) . Graph and statistics were obtained with R software or Excel . Statistical tests performed are specified in figure legends .

Flowering plants produce small particles known as pollen that – with the help of the wind , bees and other animals – carry male sex cells ( sperm ) to female sex cells ( eggs ) contained within flowers . When a grain of pollen lands on the female organ of a flower , called the pistil , it gives rise to a tube that grows through the pistil towards the egg cells at the base . The surface of the pistil is covered in a layer of long cells named papillae . Like most plant cells , the papillae are surrounded by a rigid structure known as the cell wall , which is mainly composed of strands known as microfibrils . The pollen tube exerts pressure on a papilla to allow it to grow through the cell wall towards the base of the pistil . Previous studies have shown that the pistil produces signals that guide pollen tubes to the eggs . However , it remains unclear how pollen tubes orient themselves on the surface of papillae to grow in the right direction through the pistil . Riglet et al . combined microscopy , genetic and chemical approaches to study how pollen tubes grow through the surface of the pistils of a small weed known as Arabidopsis thaliana . The experiments showed that an enzyme called KATANIN conferred mechanical properties to the cell walls of papillae that allowed pollen tubes to grow towards the egg cells , and also altered the orientation of the microfibrils in these cell walls . In A . thaliana plants that were genetically modified to lack KATANIN the pollen tubes coiled around the papillae and sometimes grew in the opposite direction to where the eggs were . KATANIN is known to cut structural filaments inside the cells of plants , animals and most other living things . By revealing an additional role for KATANIN in regulating the mechanical properties of the papilla cell wall , these findings indicate this enzyme may also regulate the mechanical properties of cells involved in other biological processes .

2020

KATANIN-dependent mechanical properties of the stigmatic cell wall mediate the pollen tube path in Arabidopsis

To internally reflect the sensory environment , animals create neural maps encoding the external stimulus space . From that primary neural code relevant information has to be extracted for accurate navigation . We analyzed how different odor features such as hedonic valence and intensity are functionally integrated in the lateral horn ( LH ) of the vinegar fly , Drosophila melanogaster . We characterized an olfactory-processing pathway , comprised of inhibitory projection neurons ( iPNs ) that target the LH exclusively , at morphological , functional and behavioral levels . We demonstrate that iPNs are subdivided into two morphological groups encoding positive hedonic valence or intensity information and conveying these features into separate domains in the LH . Silencing iPNs severely diminished flies' attraction behavior . Moreover , functional imaging disclosed a LH region tuned to repulsive odors comprised exclusively of third-order neurons . We provide evidence for a feature-based map in the LH , and elucidate its role as the center for integrating behaviorally relevant olfactory information . To navigate the environment in a way that optimizes their survival and reproduction , animals have evolved sensory systems . These have three essential tasks: First , the external world has to be translated into an internal representation in the form of an accurate neural map . Second , the neural map has to be readable and interpretable , that is , the generated neural code must allow common attributes to be extracted across stimuli to enable the animal to make the best decisions . Third , the animal has to be able to adapt to environmental changes and to form a sensory memory of new stimuli . Many studies have been dedicated to unraveling the primary transformation from a stimulus into an initial neural representation within various sensory systems ( Manni and Petrosini , 2004; Vosshall and Stocker , 2007; Sanes and Zipursky , 2010 ) and to elucidating neuronal plasticity and sensory memory formation in higher-level processing centers ( Heisenberg , 2003; Pasternak and Greenlee , 2005 ) . The ability to extract features and integrate stimulus modalities have so far mainly been studied in the visual system ( Livingstone and Hubel , 1988; Bausenwein et al . , 1992; Nassi and Callaway , 2009 ) . We addressed the question of how stimulus features such as odor valence and intensity are coded and integrated within the olfactory system using the model organism Drosophila melanogaster . The olfactory system of the vinegar fly provides an excellent model system for deciphering olfactory processing mechanisms , since it displays remarkable similarities to the mammalian system but is less complex and highly genetically tractable . Like other sensory systems , the olfactory system employs a spatio-temporal map to translate the variables in chemosensory space into neuronal activity patterns in the brain . This map emerges when the olfactory sensory neurons ( OSNs ) with the same chemosensory receptors converge into one exclusive glomerulus in the antennal lobe ( AL ) which represents the equivalent to the mammalian olfactory bulb ( Hildebrand and Shepherd , 1997; Vosshall et al . , 2000; Vosshall and Stocker , 2007 ) . Glomeruli , the functional and morphological units of the AL , are microcircuits comprising OSNs , multiglomerular local interneurons ( LNs ) and uniglomerular output neurons , so-called excitatory projection neurons ( ePNs ) ( Wilson and Mainen , 2006; Vosshall and Stocker , 2007 ) that convey the olfactory information to higher brain centers , as the mushroom body calyx ( MBc ) and the lateral horn ( LH ) ( Stocker et al . , 1997 ) . The stringent spatial arrangement of OSNs and ePNs in the AL generates a spatial map containing characteristic combinatorial glomerular activity patterns for all odorants ( Fiala et al . , 2002; Wang et al . , 2003a; Couto et al . , 2005; Fishilevich and Vosshall , 2005 ) . The MBc is involved in olfactory memory formation ( Heisenberg , 2003 ) and enables a contextualization of the odor space ( Caron et al . , 2013 ) . By exclusion , the LH is believed to be involved in innate olfactory behavior ( de Belle and Heisenberg , 1994; Jefferis et al . , 2007 ) . Excitatory PNs retain the sensory information encoded in the AL and form glomerulus-dependent , stereotypic axonal terminal fields in the LH ( Marin et al . , 2002; Wong et al . , 2002; Tanaka et al . , 2004 ) . Compartmentalization in the LH has been observed in form of a spatial segregation of ePNs innervating specific glomerular subgroups ( Tanaka et al . , 2004 ) , fruit and pheromone odor information processing ePNs ( Jefferis et al . , 2007 ) as well as ammonia and amine vs carbon dioxide coding ePNs ( Min et al . , 2013 ) . Like many other sensory networks , the olfactory circuit of the fly contains spatially distinct pathways to the higher brain , namely the inner , middle and outer antennocerebral tract ( iACT , mACT and oACT ) ( Stocker et al . , 1990 ) . Notably , the mACT projects from the AL to the LH exclusively and consists of inhibitory PNs ( iPNs ) , which exhibit also uniglomerular but mainly multiglomerular AL innervations ( Ito et al . , 1997; Jefferis et al . , 2007; Lai et al . , 2008; Okada et al . , 2009; Liang et al . , 2013 ) . Both PN populations have been attributed different coding properties: Although both PN populations exhibit odor responses to overlapping odor ligands , iPNs seems to be broader tuned than ePNs ( Wang et al . , 2014 ) . Furthermore , while ePNs encode rather odor identity ( Wang et al . , 2003a; Wilson et al . , 2004; Silbering et al . , 2008 ) , iPNs have been shown to enhance innate discrimination of closely related odors ( Parnas et al . , 2013 ) . Together , these PN populations process information on dual olfactory pathways ( Liang et al . , 2013; Wang et al . , 2014 ) , as do processing mechanisms in other sensory modalities ( Nassi and Callaway , 2009 ) , and most likely accomplish different olfactory behaviors . The mainly multiglomerular AL pattern of iPNs suggests that these neurons extract characteristic stimulus features from the AL code and re-integrate this information into the LH to mediate innate odorant-guided behavior . This assumption is further supported by two recent studies showing that the inhibitory input from the AL to the LH is module-specific , that is , selective for food odors and pheromones ( Liang et al . , 2013; Fisek and Wilson , 2014 ) , while the connectivity in the MBc is rather probabilistic ( Murthy et al . , 2008; Caron et al . , 2013 ) . However , it still remains open if and how different odor features as hedonic valence or intensity are functionally coded and integrated in the LH . In this study , we characterized and dissected the iPN olfactory processing pathway regarding the coding of odor quality and intensity at morphological , functional and behavioral levels . By linking odor-evoked activity patterns in the LH to odor-guided behavior , we provide evidence that iPNs mediate odor attraction . Furthermore , our data demonstrate a feature-based , spatially segregated activity map in the LH comprised of iPNs and third-order neurons and thus expand its role as a center for integrating behaviorally relevant olfactory information . Cell bodies of iPNs are exclusively located in the ventral cell cluster which consists of ∼50 iPNs ( Lai et al . , 2008 ) that project via the mACT to the LH , thereby bypassing the MBc ( Ito et al . , 1997 ) ( Figure 1A , B ) . In contrast , ePN somata are located anterodorsally and laterally of the AL , and their axons project through the iACT or oACT to the MBc and the LH ( Stocker et al . , 1997; Marin et al . , 2002; Wong et al . , 2002; Lai et al . , 2008 ) . To analyze the innervation patterns of iPNs and ePNs , we labeled both PN populations simultaneously in vivo using the enhancer trap lines GH146-QF and MZ699-GAL4 that label the majority of ePNs ( 60% ) and iPNs ( 86% ) , respectively ( Lai et al . , 2008 ) . Double-labeling shows that both PN types innervate overlapping regions in the AL and the LH , while a small posterior-lateral LH area is targeted only by ePNs ( Figure 1A , Figure 1—figure supplement 1 ) . In GH146-positive ( GH146+ ) PNs , immunolabeling reveals GABA production in all ∼6 PNs of the ventral cell cluster ( Wilson and Laurent , 2005 ) , whereas ePNs of this line are exclusively cholinergic ( Shang et al . , 2007 ) . For the ∼45 MZ699-positive ( MZ699+ ) iPNs ( Lai et al . , 2008 ) , GAD1 ( glutamic acid decarboxylase ) in situ hybridizations imply GABA synthesis ( Okada et al . , 2009 ) , which was recently verified via immunostaining ( Liang et al . , 2013; Parnas et al . , 2013 ) . The polarity of both PN populations has been studied in detail , showing that both possess dendritic regions in the AL , indicating the AL as their cholinergic input site , while the LH represents their major output site ( Jefferis et al . , 2001; Okada et al . , 2009; Liang et al . , 2013; Parnas et al . , 2013 ) . 10 . 7554/eLife . 04147 . 003Figure 1 . Detailed glomerular innervations of excitatory and inhibitory projection neurons in the AL . ( A ) Simultaneous labeling of inhibitory projections neurons ( iPNs , labeled by MZ699-GAL4;G-CaMP ) and excitatory projection neurons ( ePNs , labeled by GH146-QF;tdTomato ) in vivo reveals distinct projections to the lateral horn ( LH ) . All iPNs bypass the mushroom body calyx ( MBc ) and innervate the LH exclusively . The MZ699 line labels a few ventrolateral protocerebral neurons ( vlPr neurons ) projecting via the posterior lateral fascicle ( plF ) from the ventrolateral protocerebrum ( vlPr ) to the LH . ( B ) Schematic of the PN connectivity relay from the antennal lobe ( AL ) to higher brain centers ( ePNs in magenta , iPNs in green , and vlPr neurons in orange ) . ( C ) Above , complete glomerular assignment of the AL neuropil ( right AL ) , labeled with elav-n-synaptobrevin:DsRed ( END1-2 ) . Below , glomerular innervations of both PN populations related to in vivo images in Figure 1—figure supplement 2 . Depicted are the ventral level ( ∼−40 µm ) , the medial level ( ∼−20 µm ) and the dorsal view onto the AL . Color annotation: blue glomeruli are not innervated by any of the used GAL4-lines; green glomeruli are innervated by MZ699+ iPNs and magenta by GH146+ ePNs; white glomeruli are innervated by both enhancer trap lines . Scale bar , 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04147 . 00310 . 7554/eLife . 04147 . 004Figure 1—figure supplement 1 . Characterization of excitatory and inhibitory projection neurons . Overlap of ePNs ( QUAS-tdTomato ) and iPNs ( UAS-GCaMP3 . 0 ) in the LH area . The circle indicates the posterior lateral region , which is sparsely innervated by iPNs and dominated by ePN axonal terminal fields . Scale bar , 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04147 . 00410 . 7554/eLife . 04147 . 005Figure 1—figure supplement 2 . Glomerular innervations of ePNs and iPNs . ( A ) Representative in vivo images of glomerular innervations . MZ699- and GH146-GAL4 lines have been reconstructed with the END1-2 background staining ( two upper planes ) and dual labeling via the Q-system and the GAL4-UAS expression system ( lowest plane ) . Scale bar , 20 µm . ( B ) Detailed glomerular AL innervation . Green filled cells indicate innervation by MZ699-GAL4 , magenta GH146-GAL4 innervation , respectively and grey , no innervation by the indicated line . Bottom rows , total number of innervated glomeruli with percentage share indicated below . Merge column: white filled with ‘x’ indicates glomeruli innervated by both lines , grey only one line . Blue filled rows are glomeruli labeled by none of the enhancer trap lines . DOI: http://dx . doi . org/10 . 7554/eLife . 04147 . 005 To further characterize PNs labeled by MZ699-GAL4 and GH146-GAL4 , we analyzed their precise glomerular innervation to unravel how selectively they acquire information in the AL . To allow glomerulus identification in vivo , we employed a transgenic fly carrying elav-n-synaptobrevin:DsRed ( END1-2 ) to express the presynaptically targeted fusion protein under the control of the neuron-specific elav promotor ( Figure 1—figure supplement 2A ) ( Grabe et al . , 2014 ) . The reconstruction and identification of all AL glomeruli provided 53 glomeruli , of which 75% were innervated by MZ699+ iPNs ( 40 ) while 70% ( 37 ) were covered by GH146+ ePNs ( Figure 1C , Figure 1—figure supplement 2B ) . 55% of all glomeruli were innervated by both lines . Notably , dendritic MZ699-GAL4 innervation density was not homogeneous . Certain glomeruli were densely innervated ( e . g . , DM2 , DM4 and DM5 ) , while others did not reveal any postsynaptic sites ( e . g . , DL1 , DL4 and DL5 ) . Hence MZ699+ iPNs target specific glomerular subsets selectively , which suggests that these neurons have a particular function within the olfactory network . Probabilistic synaptic density maps of GH146+ PNs predicted a regionalized neuronal activity in the LH ( Jefferis et al . , 2007 ) . Do iPNs functionally segregate in a comparable way ? To address this question , we expressed the Ca2+-sensitive reporter G-CaMP3 . 0 ( Nakai et al . , 2001; Tian et al . , 2009 ) in iPNs using MZ699-GAL4 and performed functional imaging in the LH ( Figure 2A–C ) . We initially tested three odors with potential relevance for Drosophila at different concentrations: acetoin acetate , an attractive byproduct of the yeast fermentation process , balsamic vinegar , an attractive natural odor mixture , and benzaldehyde , a well-known fly repellant ( Magee and Kosaric , 1987; Keene et al . , 2004; Semmelhack and Wang , 2009 ) . We observed that odor evoked Ca2+ responses separate in certain regions of the LH in an odor-specific and concentration-dependent manner ( Figure 2C ) . Acetoin acetate and balsamic vinegar evoked Ca2+ activity in spatially similar regions . At higher concentrations , an additional region was recruited . Benzaldehyde elicited no response at very low concentrations , but induced clear activity at median and high concentrations in a third region , which was completely separate from the regions activated by the other two odors . Observed patterns were highly reproducible within one animal and stereotypic among different individuals , as shown for the stimulation with 1-octen-3-ol ( Figure 2D ) as well as other odors ( Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 04147 . 006Figure 2 . Odors evoke specific and stereotypic calcium responses in the LH subdivided into three distinct odor response domains . ( A ) Schematic of the olfactory circuit with the investigated area highlighted . ( B ) RAW image of the LH ( top picture ) depicting the recorded area of figures ( C–E ) and the false color image ( bottom picture ) during the solvent application . The ΔF/F scale bar applies for all false color-coded pictures; the alpha-bar for the pixel participation xk of the indicated colors applies for ( E–F ) . ( C ) Representative LH Ca2+ responses ( ΔF/F% ) of acetoin acetate , balsamic vinegar and benzaldehyde at three concentrations . Numbers in the lower right corner indicate individual maxima . ( D ) Odor-evoked Ca2+ responses ( ΔF/F% ) are exemplarily depicted for 1-octen-3ol- at three concentrations in four animals . ( E ) NNMF-extracted LH odor response domains ( ORD ) of four representative animals: three LH ORDs were fully reproducible after being extracted from all measured animals . Domains classified as identical are similarly color-coded: the green ORD is located in the posterior-medial region of the LH ( LH-PM ) ; blue , in the anterior-medial ( LH-AM ) , and red in the anterior-lateral LH area ( LH-AL ) . The alpha-bar for green , blue and red shades is placed in ( B ) . ( F ) Left , schematic outlines of the LH with indicated ORDs . Right , median activity traces of all odors at three concentrations are depicted for each colored ORD . Shadows represent lower and upper quartiles ( n = 6–7 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04147 . 00610 . 7554/eLife . 04147 . 007Figure 2—figure supplement 1 . Odor-evoked activity patterns in the LH are reproducible and stereotypic . Odor-evoked Ca2+ responses ( ΔF/F% ) in the LH for the three odors acetoin acetate , balsamic vinegar and benzaldehyde in four animals are shown as false-color coded images . Two measurements in each animal are given to reveal that the activity patterns are highly reproducible within one animal . Comparison between the patterns among individuals shows that the activity regions are stereotypic . Numbers in the lower right corner indicate individual maxima . The ΔF/F scale bar is shown at the bottom . DOI: http://dx . doi . org/10 . 7554/eLife . 04147 . 00710 . 7554/eLife . 04147 . 008Figure 2—figure supplement 2 . Odor-evoked activity patterns in the LH can be reconstructed with five components . Above , Odor-evoked Ca2+ responses ( ΔF/F% ) in the LH for the three odors acetic acid , balsamic vinegar and 1-octen-3-ol in three animals are shown as false-color coded images . The ΔF/F scale bar is shown at the bottom . Middle , activity patterns were reconstructed using NNMF with five components . Below , residue of the pattern reconstructions with five components ( as shown in the middle panel ) revealing that no stimulus related activity remained . DOI: http://dx . doi . org/10 . 7554/eLife . 04147 . 00810 . 7554/eLife . 04147 . 009Figure 2—figure supplement 3 . Odor-evoked activity patterns in the LH cluster into three components . Hierachichal clustering ( UMPGA ) of the odor response spectra of the NNMF components with a reliable stimulus response ( trial-to-trial correlation >0 . 7 , that is , 28 out of 35 components in seven animals ) . The response spectra segregate into three distinct clusters according to their stimulus response spectra . The corresponding response areas ( left pictures ) are located in similar regions of the LH . DOI: http://dx . doi . org/10 . 7554/eLife . 04147 . 009 Due to the lack of morphological landmarks in the LH , functional data were analyzed using the pattern recognition algorithm Non-Negative Matrix Factorization ( NNMF ) ( Lee and Seung , 1999 ) , which automatically extracts spatial areas possessing a common distinct time-course , further termed LH odor response domains ( ORDs ) . The NNMF analysis extracted three clearly reproducible and spatially robust ORDs ( Figure 2E , see NNMF part in the ‘Materials and methods’ section ) . Notably , ORDs occupying common temporal kinetics exhibited highly stereotypic spatial patterns . We termed the ORDs LH-PM ( LH-posterior-medial ) , LH-AM ( LH-anterior-medial ) and LH-AL ( LH-anterior-lateral ) according to their anatomical positions . To validate our observations , we extended our stimulus array to 11 additional odorants and applied each at three concentrations . Odorants were chosen according to chemical classes , hedonic valence and biological value . Hence , the odor set included acids , lactones , terpenes , aromatics , alcohols , esters , ketones and the natural blend , balsamic vinegar . Remarkably , analysis of the additional odorants revealed neuronal activity exclusively within the three described ORDs ( Figure 2F , Figure 2—figure supplements 2 , 3 ) . Furthermore , median NNMF-extracted Ca2+ response traces with indicated statistical quartiles illustrate very low variability and highly reproducible LH signals . The LH-PM area chiefly revealed robust odor-evoked responses across concentrations , while the LH-AM and LH-AL were mainly activated at very high odor concentrations by distinct odorants . The global responsiveness within separate ORDs in the LH substantiates our finding of a relatively broad AL input to MZ699+ iPNs which converges into three spatially regionalized and stereotypic LH activity domains . We next investigated if the spatially regionalized odor-evoked response patterns are reflected in the axonal terminal fields of MZ699+ iPNs in the LH . To analyze these neurons at the single neuron level , we performed neural tracing by employing a genetically encoded photoactivatable GFP ( PA-GFP ) ( Patterson and Lippincott-Schwartz , 2002; Datta et al . , 2008; Ruta et al . , 2010 ) . The photoconversion of all MZ699+ neurons leaving the AL confirmed the homogeneous distribution of iPN neurites in the LH and the sparse innervation of the posterior-lateral region as mentioned above ( Figure 3A ) . Next we illuminated PA-GFP in single somata to selectively label individual MZ699+ iPNs from the soma up to the farthest axonal terminals in the LH . Individual iPNs were reconstructed and transformed into a reference brain using the END1-2 background ( Grabe et al . , 2014 ) to align neurons of different individuals . Based on their innervation pattern in the LH , MZ699+ iPNs could be assigned to two major morphological classes ( Figure 3B , C ) . As expected from the extracted ORDs , one iPN group diverged to the LH-PM region ( 8/25 of iPNs ) , while a second group extended their axonal terminations within the LH-AM area ( 10/25 of iPNs ) . In order to statistically substantiate our observation , we performed a cluster analysis based on a similarity score ( Kohl et al . , 2013 ) of the target areas of all terminals of each iPN in the LH ( for details see ‘Materials and methods’ ) . The dendrogram of morphological similarity between each individual iPN shows that all , except one iPN , could be clustered according to their target region in either the LH-AM or LH-PM area ( Figure 3D ) which confirms the classification into two major categories . Additionally , we performed a principal component analysis based on the distances of the similarity scores showing that both neuronal classes possess significantly different target areas in the LH ( Figure 3—figure supplement 1A; p < 0 . 001 , one-way ANOSIM ) . 10 . 7554/eLife . 04147 . 010Figure 3 . iPNs can be classified according to their projection pattern in three distinct LH zones . ( A ) Complete population of MZ699+ iPNs labeled using PA-GFP ( left image ) , the posterior-lateral LH region is encircled , arrowhead indicates the final common projection point of iPN axons . Middle image: photoactivation of all vlPr neurons of the MZ699-GAL4 line that project from the LH to the vlPr via the plF . Right image: exemplary single iPN , labeled by photoconverting PA-GFP in a single soma ( arrow ) . Scale bar , 20 µm . ( B ) Framed images: neuronal reconstructions of all iPNs projecting to the LH-PM zone ( n = 8 ) with outlined olfactory neuropils . View from dorsal ( left ) and lateral ( right ) . Right part represents two exemplary registered individual iPNs . ( C ) Neuronal reconstructions of all iPNs projecting into the LH-AM zone ( n = 10 ) , images are arranged as in ( B ) . ( D and E ) Cluster analyses based on the target areas of all terminals of each iPN in the LH ( D ) or based on the innervated glomeruli in the AL ( E ) . The dendrograms are split into colored subclusters . Below each dendrogram , each individual iPN is specified according to the labels in Figure 3—figure supplement 2 . Note , that iPNs can be morphologically clustered according to their target or input regions . ( F ) Neuronal reconstruction of vlPr neurons projecting through the plF to the LH-AL zone . ( G ) Combination of all registered neurons . ( H ) Dual combinations of all registered neurons with their projections in the LH . DOI: http://dx . doi . org/10 . 7554/eLife . 04147 . 01010 . 7554/eLife . 04147 . 011Figure 3—figure supplement 1 . iPNs can be morphologically segregated according to their target and input region . ( A ) Principal component analysis based on the distances of the similarity scores of all terminal points of each individual iPN in the LH ( for details see ‘Materials and methods’ ) . LH-AM iPNs ( blue ) and LH-PM iPNs ( green ) form significantly distinct clusters ( ***p < 0 . 001 , One-Way ANOSIM , Bray–Curtis ) . ( B ) Principal component analysis based on the glomerular innervations of each individual iPN in the AL . Again , LH-AM iPNs ( blue ) and LH-PM iPNs ( green ) form significantly distinct clusters ( ***p < 0 . 001 , One-Way ANOSIM , Bray–Curtis ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04147 . 01110 . 7554/eLife . 04147 . 012Figure 3—figure supplement 2 . Glomerular innervations of individual iPNs . Binary innervation patterns of 25 individually labeled MZ699+ iPNs using PA-GFP . Columns represent innervation patterns of individual neurons which have been grouped according to their innervation properties; rows represent 51 glomeruli in the AL along with the innervation in the specific odor response domains in the LH ( LH-PM , LH-AM ) and/or the mushroom body calyx ( MBc ) . Glomeruli have been sorted according to their iPN innervation . Grey , innervated; white , not innervated . DOI: http://dx . doi . org/10 . 7554/eLife . 04147 . 012 We did not observe any clear panglomerular innervations of individual MZ699+ iPNs that spanned the entire AL , consistent with Liang et al . ( 2013 ) . Instead , MZ699+ iPNs develop mainly oligoglomerular patterns innervating on average 5 . 4 ± 3 . 9 glomeruli ( mean ± SD ) , which are not necessarily in close proximity . It is important to note here , that the glomerular innervations of iPNs are rather sparse in comparison to the innervation of ePNs which complicates the identification of truly innervated glomeruli . After classifying all registered neurons according to their LH zones along with their glomerular innervations , we noted a spatial subdivision of MZ699+ iPN dendritic fields in the AL ( Figure 3—figure supplement 2 ) . Whereas LH-PM iPNs extended dendrites mainly into glomeruli from the ventro- or dorsomedial area of the AL ( e . g . , DM4 , DM2 , VM7 , VM5d ) , iPNs targeting the LH-AM zone innervated glomeruli ranging from the ventro- and dorsoanterior to the dorsocentral region ( e . g . , DC3 , VC1 , VA6 , VL1 ) . We observed that a glomerulus is typically innervated by only LH-PM iPNs or LH-AM iPNs . However , we also found a few cases where a glomerulus can be innervated by both iPN types ( e . g . , glomeruli D and DC2 ) . In order to analyze whether the two categories of iPNs can also be statistically separated according to their glomerular innervations in the AL , we performed a cluster analysis based on the glomeruli innervated by each individual iPN ( Figure 3E ) . Notably , the two iPN classes could be clearly clustered into two groups due to their specific AL innervations . This finding is further supported by a principal component analysis showing that iPNs targeting the LH-PM region innervate a significant different glomerular subset than iPNs that send their axonal terminals to the LH-AM area ( Figure 3—figure supplement 1B; p < 0 . 001 , one-way ANOSIM ) . In accordance with our finding of two major iPN categories is the study by Lai et al . ( 2008 ) who observed several different stereotyped projection patterns of multiglomerular MZ699+ single-cell clones that could be broadly categorized into two groups based on the dendritic and axonal projection patterns . While we observed corresponding innervated areas in the AL , the described target areas in the LH seem to differ . However , due to the lack of 3D reconstruction of the single-cell clone data , the innervation patterns cannot be compared in detail . In addition to the oligoglomerular iPNs , we observed a few uniglomerular MZ699+ iPNs innervating either glomerulus DA1 or VL1 ( 4/25 of iPNs ) , consistent with Lai et al . ( 2008 ) , which target the LH-AM region ( Figure 3—figure supplement 2 ) . Moreover , we identified three other MZ699+ neurons that did not innervate the AL and sent their axons through the mACT to the LH and/or the MBc . Since the MZ699-GAL4 line labels also neurons connecting the LH and the ventrolateral protocerebrum ( vlPr ) ( Ito et al . , 1997; Liang et al . , 2013; Parnas et al . , 2013 ) , we illuminated a small fraction of the posterior lateral fascicle ( plF ) to target these putative third-order neurons ( Figure 3A ) . The plF comprised axons of ventrolateral protocerebral neurons ( vlPr neurons ) , which bifurcated within the LH-AL ( Figure 3F ) . Combinations of all registered neuron types within the assigned zones revealed that iPNs of the LH-AM area and vlPr neurons of the LH-AL region intermingle ( Figure 3G , H ) . To illustrate higher-order connectivity , we labeled the three major neuron types , that is , MZ699+ iPNs , GH146+ ePNs and vlPr neurons , targeting the LH within the olfactory circuitry using PA-GFP ( Figure 4A ) . Since our observed Ca2+ responses in the LH-AL region might reflect activity from vlPr neurons rather than iPNs , we dissected the neuronal contributions within each extracted ORD by conducting transection experiments using two-photon laser-mediated microdissection ( Figure 4B ) . By transecting the mACT , we aimed at abolishing LH-responses deriving from MZ699+ iPNs , while cutting the plF connection should eliminate potential odor-evoked vlPr neuron activity . To achieve unambiguous and comparable results , functional imaging was performed in both brain hemispheres simultaneously . Immediately after the intact brain areas were imaged , the tracts were selectively transected on one brain side each ( Figure 4C ) and the imaging procedure was repeated . We applied a reduced odor set that elicited activity in all ORDs and performed NNMF for pre- and post-lesion recordings . Transecting the mACT significantly reduced responses in the LH-PM and LH-AM region , whereas LH-AL responses were significantly abolished by plF-ablation ( Figure 4D ) . Notably , we observed that LH-AL responses to some odors were significantly increased after mACT transection as a consequence of the suppression of iPN inhibition of vlPr neurons confirming the study by Liang et al . ( 2013 ) . Hence , activity in the LH-PM and LH-AM domain can be assigned to MZ699+ iPNs , while LH-AL activity is mainly evoked by vlPr neurons ( Figure 4E ) . 10 . 7554/eLife . 04147 . 013Figure 4 . Distinct odor response domains in the LH constitute neuronal activity of iPNs and vlPr neurons . ( A ) Representation of all ePNs ( magenta ) and iPNs ( green ) labeled by GH146-GAL4 and MZ699-GAL4 using PA-GFP , respectively . Photoactivation of vlPr neurons ( orange , MZ699-GAL4 ) connecting the LH and the vlPr via the plF . The overlay image depicts a pseudo-merge image of the different GAL4-driver lines . ( B ) Schematic of the olfactory circuit with integrated layout of the transection experiment . After simultaneous Ca2+ imaging of bilateral LHs , the ipsilateral plF and contralateral mACT was transected ( red zigzag line ) with an infrared laser ( dashed red arrow ) . ( C ) Projection images of a 7 µm stack of the LH area prior and post transection . Left images , mACT transected; right image , plF transected . The ablated region is indicated by the dashed red arrow . Scale bar , 20 µm . ( D ) Median time traces displaying percental change of ΔF/F values for indicated ORDs prior to post transection of the mACT ( green , left ) and the plF ( orange , right ) for different odorants . Significant changes of odor-evoked Ca2+ signals due to transection are shown in the column SIG difference . Differences were tested with a two-tailed paired Student's t test ( p < 0 . 05 ) . Color codes are indicated by the corresponding scale bar below , n = 4–5 . Transecting the mACT eliminates Ca2+ signals in the LH-PM and LH-AM domain , while lesioning the plF significantly abolishes LH-AL responses . Notably , the LH-AL domain is significantly stronger activated after mACT transection following application of 1-octen-3-ol and γ-butyrolactone . ( E ) Summarized cartoon of the neuron populations contributing to ORD activity prior and post transection of axons of iPNs or vlPr neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 04147 . 013 We next addressed the behavioral relevance of MZ699+ iPN activity in the LH for innate odor-guided behavior . To precisely target iPN function , we expressed an RNAi construct against glutamic acid decarboxylase 1 ( GADi ) to selectively knock-down the GABA synthesis in MZ699+ iPNs ( Figure 5A ) . We confirmed the reduction in GABA production via immunostaining ( Figure 5B ) . Since vlPr neurons are not GABAergic , they were not affected by the RNAi expression ( Liang et al . , 2013; Parnas et al . , 2013 ) . Using wild-type flies and parental controls , we conducted T-maze assays ( Tully and Quinn , 1985; Chakraborty et al . , 2009 ) with nine of the odorants applied in functional imaging experiments at medium and high concentrations . Notably , flies with silenced MZ699+ iPN GABA production revealed a neutral or aversive behavioral response to attractive odors , while repellent odors evoked an even stronger aversion ( Figure 5C ) . To compare the T-maze data more accurately , we calculated the average change of behavioral response indices ( RIs ) between GADi flies and parental controls ( Figure 5D ) . Indeed , all responses changed in a negative direction , indicating MZ699+ iPNs play a crucial role in mediating attraction behavior . The sole exception involved high concentrations of the most repulsive odor , acetophenone , since this odor had already induced maximum aversion . Overall , these experiments reveal a crucial function of MZ699+ iPNs in mediating attraction behavior by releasing GABA in the LH . 10 . 7554/eLife . 04147 . 014Figure 5 . iPN GABA release in the LH mediates odor attraction behavior . ( A ) Experimental layout: iPN GABA production was selectively silenced via GADi expression in MZ699+ iPNs; ePN and vlPr neuron activity remained unaffected . ( B ) Immunostaining against GABA and GFP within AL somata ( left ) and LH neurites ( right ) of iPNs with intact ( top ) and silenced GABA production ( bottom ) . GADi flies show GABA signals in somata of iPNs labeled by GH146-GAL4 only ( arrowhead ) . The arrow head points to an exemplary GABA-positive bouton in the LH . Scale bar , 20 µm . ( C ) Averaged behavioral response indices ( RIs ) determined with a T-maze assay for wild-type flies ( dark blue ) , parental controls ( light blue ) and experimental animals ( magenta ) for nine odorants at two concentrations . Empty boxes display no response ( Wilcoxon signed-rank test ) . Dunn's Multiple Comparison Test was used for global differences in the dataset followed by a posthoc test for selected pairs ( p* < 0 . 05; **p < 0 . 01; ***p < 0 . 001 ) . Error bars represent SEM . ( D ) RI differences between GADi flies and averaged parental controls . RI differences are negative for all but one odor indicating that GADi expression shifts odor-guided behavior towards aversion . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04147 . 014 The behavioral effect of the iPN knock-down suggests that MZ699+ iPNs encode positive hedonic valences . To correlate the complete ORD pattern array with innate behavioral preferences , we assigned behavioral RIs for all odors at median and high odor concentrations using the T-maze assay as in our previous experiment ( Figure 6A ) . Since extremely low concentrations rarely evoked any behavioral response , we excluded the 10−6 concentration in this analysis . It is important to note here , that different behavioral assays for testing olfactory preferences in flies might lead to contradictory results . However , the majority of odors used here was also tested in two other behavioral paradigms , the trap assay ( Stökl et al . , 2010; Knaden et al . , 2012 ) and the FlyWalk ( Steck et al . , 2012 ) ( pers . comm . M Knaden ) and yielded similar results ( see Figure 6—figure supplement 1 ) . When we plotted median odor-evoked activity in a three-dimensional space defined by the three ORDs , we saw a clear clustering of responses evoked by aversive and attractive odorants ( Figure 6B ) . The LH-AL domain , constituted mainly by vlPr neurons , is coding aversive odors , while attractive odors activated only the LH-PM and LH-AM domains that derive from MZ699+ iPNs . This result is in accordance with our finding that iPNs mediate odor attraction . 10 . 7554/eLife . 04147 . 015Figure 6 . Integration of hedonic valence and odor concentration into ORDs . ( A ) Response indices of wild type flies for all odors at median and high concentrations . Odors are sorted from highly aversive ( −1 , red ) to highly attractive ( +1 , green ) . ( B ) 3D-scatter plot of median Ca2+ responses of all odors based on the three ORDs . Odor-dots are labeled due to their RI shown in ( A ) . Same odors at different concentrations are connected with a line: the dot at the end depicts 10−2 , the centered dot 10−4 , and the end of the line 10−6 . Attractive and aversive odor representations form separate clusters . ( C and D ) Left , schematic LH outlines with colored ORDs corresponding to data on the right . Correlation score r ( upper right corner ) between median activity and measured RI in T-maze experiments or odor concentration , respectively , with significance denoted below . Student's t test , *p < 0 . 05 , ***p < 0 . 001 . ( E ) Complete correlation matrices for Ca2+ response patterns of OSNs in the AL ( left ) and iPNs in the LH ( right ) . The odors are arranged according to single linkage clustering of the LH activity patterns . Heatmap color-code refers to the correlation distance scale bar on the right . Correlation distance is defined as 1 − r , where r is the Pearson correlation coefficient between the response patterns of two odorants . Odor letters are color-coded according to hedonic valence; 10−6 RI values are labeled in grey ( complete list right hand ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04147 . 01510 . 7554/eLife . 04147 . 016Figure 6—figure supplement 1 . Odor valences determined with three different behavioral assays . Odor-evoked behavioral responses of wild type flies for the 14 odors used in this study determined by T-maze assay , trap assay and the FlyWalk . The color denotes an attractive ( green ) , aversive ( red ) or a neutral ( light yellow ) behavioral response . N/T , not tested . The majority of odors yielded similar results independent of the behavioral assay used . In a few cases an attractive odor evoked a neutral response ( i . e . , no response ) , but never induced an aversive response in another assay . DOI: http://dx . doi . org/10 . 7554/eLife . 04147 . 01610 . 7554/eLife . 04147 . 017Figure 6—figure supplement 2 . Calcium responses of OSNs . ( A ) Representative glomerular Ca2+-responses of OSNs in the AL for a subset of odorants at three concentrations . Scale bar to the right . Control ( mineral oil ) recordings are shown additionally as full false-color coded images . ( B ) Glomerular AL atlas used for glomerular identification . ( C ) Median Ca2+-activity traces of all glomeruli for all odorants at the three indicated concentrations . Scale bar and control measurement in the center . Odor application is indicated by the grey bar below the heatmaps ( n = 6–7 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04147 . 01710 . 7554/eLife . 04147 . 018Figure 6—figure supplement 3 . Correlation matrices for odor-evoked responses in the AL and LH . Complete correlation matrices for calcium activity patterns of OSNs in the AL ( left ) and iPNs in the LH ( right ) . The odors are arranged according to single linkage clustering of the AL activity patterns . Heatmap color-code refers to the correlation distance scale bar below each matrix . Odor letters are color-coded according to hedonic valence; 10−6 RI values are labeled in grey ( complete list right hand ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04147 . 018 We next correlated ORD activity to odor valence separately for all ORDs . This evaluation enabled us to analyze iPN and vlPr neuron coding properties apart from each other ( Figure 6C ) . As expected , the analysis revealed a significant correlation between positive valence and the LH-PM domain , whereas Ca2+ responses in the LH-AL were strongly negatively correlated to hedonic valence . The LH-AM domain exhibited a positive but not significant correlation for odor valence . Remarkably , activity within the LH-PM was totally independent of concentration , whereas activity in both anterior domains was significantly correlated to odor intensity ( Figure 6D ) . Hence , MZ699+ iPNs integrate odor attraction information into the LH-PM domain independent of odor intensity , confirming behavioral experiments . Intensity coding is in turn conducted separately by distinct iPNs within the LH-AM domain . In contrast , putative third-order vlPr neurons projecting into the LH-AL area code both negative valence and odor intensity . Finally , we wondered if this valence-specific LH representation is already reflected at the primary level of olfactory processing . The odor-evoked responses in iPNs are generally similar to those in OSNs ( Wang et al . , 2014 ) , indicating a straight forward transduction of cholinergic OSN responses . We therefore performed functional imaging of odor-evoked Ca2+ dynamics at the AL input level by expressing G-CaMP3 . 0 in OSNs using Orco-GAL4 ( Larsson et al . , 2004 ) ( Figure 6—figure supplement 2 ) . In order to compare the activity patterns at both processing levels , we calculated correlation distances for all pair-wise combinations of odor-evoked response patterns and plotted these with respect to maximal ORD pattern similarity in the LH ( Figure 6E ) . As expected , odor representations in the LH clearly clustered within three separated parts of the matrix , reflecting our observed ORDs . However , this coding similarity could not predict AL activity patterns , even if the correlation matrix was sorted with respect to pattern similarity in the AL ( Figure 6—figure supplement 3 ) . We augment our present understanding of the Drosophila olfactory circuitry by elucidating the coding properties of a parallel and behaviorally relevant higher-order processing pathway to the LH . Morphological , functional and behavioral approaches provide strong evidence for a functional subdivision of iPNs into neurons coding either odor attraction or odor intensity . Our behavioral experiments reveal that inhibitory properties of iPNs are necessary for innate odor-guided attraction . In addition , we characterize a third neural pathway coding odor repellence . Do MZ699+ iPNs fulfill anatomical requirements to constitute a distinct processing channel in addition to ePNs ? A remarkable anatomical feature of MZ699+ iPNs is their glomerular innervation pattern in the AL . Whereas GH146+ ePNs are uniglomerular and retain the topographic code in their axonal arrangement ( Marin et al . , 2002; Wong et al . , 2002; Jefferis et al . , 2007 ) , most MZ699+ iPNs possess oligoglomerular innervations suggesting that these neurons might not convey precise odor-identity information . In addition , MZ699+ iPNs in the AL diverge only into specific glomerular subsets , and so might be pre-determined to selectively extract common features of distinct odors . We have previously shown that the AL map at the PN level exhibits a spatial segregation of valence representation ( Knaden et al . , 2012 ) . Certain glomeruli , which have been classified as aversion coding at the GH146+ ePN level , are omitted by MZ699+ iPNs , whereas most glomeruli classified as attraction coding are particularly densely innervated . These results suggest that within the MZ699+ iPN population , mainly positive odor traits are extracted , whereas the odor information of negative valence is neglected . This conclusion is consistent with the recent finding that one type of LH neurons is receiving input from PNs that mainly innervate glomeruli coding fruity-smelling acetates ( Fisek and Wilson , 2014 ) which represent attractive odor cues ( Knaden et al . , 2012 ) . We furthermore demonstrate that the MZ699+ iPN population is split into two major morphological classes possessing a clear spatial segregation in the AL which is strictly maintained within the LH . It has to be kept in mind that we do not cover all iPNs by using MZ699-GAL4 . Further experiments characterizing the ∼6 missing MZ699– iPNs , which are labeled by GH146-GAL4 ( Wilson and Laurent , 2005; Lai et al . , 2008 ) , will elucidate if our assumptions apply for the whole iPN population . So far only a handful neuroanatomical studies targeting GH146+ ePNs have dealt with the question of how olfactory information is integrated and read out by higher brain structures , in particular the LH ( Marin et al . , 2002; Wong et al . , 2002; Tanaka et al . , 2004; Jefferis et al . , 2007 ) . A recent study that traced the projection pattern of PNs coding ammonia and amines as attractive stimuli and carbon dioxide and acids as repulsive signals suggests that sensory stimuli of opposing valence are represented in spatially distinct areas within the LH ( Min et al . , 2013 ) . In addition the study by Liang et al . ( 2013 ) showed that MZ699+ iPNs selectively suppress the activity of vlPr neurons to food odors , while pheromone responses were not affected verifying the assumption that different odor features are processed separately . However , functional evidence for a feature-based , spatially segregated activity map in the LH was so far missing . To unravel the coding properties of MZ699+ iPNs within the LH , we conducted Ca2+ imaging experiments of MZ699+ iPNs in the LH to odorants having different hedonic valences and intensities , and could classify the LH into three functional ORDs . Our neuronal tracing and transection experiments validated the LH segmentation into two medial domains that derive from MZ699+ iPNs , and the LH-AL domain formed by vlPr neurons . In line with our observations are morphological studies on ePNs and third-order LH neurons revealing a similarly tight constriction into three zones within the LH ( Tanaka et al . , 2004 ) , while single-cell labeling combined with image registration resulted in five ePN target zones ( Jefferis et al . , 2007 ) . However , the ePN terminal zones do not necessarily correspond to the target domains of iPNs , since it has recently been shown that MZ699+ iPNs do not inhibit odor responses of GH146+ ePNs ( Liang et al . , 2013 ) and that the presynaptic sites of iPNs are spatially separated from those of ePNs ( Wang et al . , 2014 ) . Hence both PN populations represent parallel processing pathways that most likely accomplish distinct processing tasks analogous to the honeybee olfactory system which possesses dual olfactory pathways to the higher brain that accomplish parallel processing of similar odors ( Brill et al . , 2013 ) . Silencing MB function revealed that the LH alone is sufficient for basic olfactory behavior ( de Belle and Heisenberg , 1994; Connolly et al . , 1996; Heimbeck et al . , 2001 ) . Our behavioral results demonstrate that selectively silencing MZ699+ iPNs severely reduced the flies' odor attraction behavior . Hence our results suggest that MZ699+ iPNs are capable of extracting specific features from the combinatorial code emerging in the AL . A behavioral study revealed that silencing MBc neurons impairs odor attraction but not repulsion ( Wang et al . , 2003b ) . The authors drew the conclusion that the LH is involved in mediating innate repulsion rather than attraction . These results are not necessarily contradictory to ours since some ePNs might activate the LH-AL domain exclusively ( i . e . , vlPr neurons ) . On the other hand , Wang et al . ( 2003b ) did not include highly concentrated attractive odors . Therefore it is possible that in their experiments , the odor detection threshold was simply reduced , so that only highly concentrated odors , which induced odor aversion , could be distinguished . Our behavioral results , in contrast , revealed the constant influence of the MZ699+ iPNs in mediating attraction for odorants over a range of concentrations . Our data suggests that odors with opposing hedonic valences are encoded by an interplay of distinct processing pathways . The study by Liang et al . ( 2013 ) showed that GABA release from MZ699+ iPNs directly inhibits responses of vlPr neurons to food odors as mentioned above . This finding fits well to our observations that iPNs are activated mainly by attractive odors while vlPr neurons are not , likely due to the inhibitory input from iPNs . VlPr neurons are , on the other hand , almost solely activated by repellent odors , which do hardly activate iPNs and therefore do not induce a strong inhibition to vlPr neurons . Repellent odors most likely activate vlPr neurons via ACh release of ePNs which is supported by immunostainings with pre- and postsynaptic markers indicating that vlPr neurons receive input in the LH , while the vlPr represents their major output region ( Parnas et al . , 2013 ) . The vlPr is supposedly also a target of visual neurons from the optic lobe ( Tanaka et al . , 2004 ) implying that a certain integration of different sensory modalities takes place at this central processing relay . Given that iPNs are inhibitory neurons , the underlying mechanism of odor attraction behavior might therefore be an inhibition of aversive neuronal circuits from the LH to the vlPr that are mainly composed of vlPr neurons . However , this assumption needs to be verified with further experiments elucidating if vlPr neurons are sufficient and necessary to mediate odor aversion . What is known about odor coding in the LH in other insect species ? Notably , in locusts it has been shown that LH neurons receiving convergent PN input appeared to encode stimulus intensity in their net firing rates and in the phases of their spikes ( Gupta and Stopfer , 2012 ) . Hence these results support the idea that within the LH , general stimulus features such as odor intensity are extracted , which is well in line with our observation of the anterior LH domains whose activity is also significantly correlated to odor intensity . Also in line with our results is a study from honeybees which shows that the representation of different pheromone types is spatially segregated in the LH ( Roussel et al . , 2014 ) , indicating that odors eliciting specific behaviors are coded according to their biological values . In conclusion , our study provides an important step in unraveling higher olfactory processing mechanisms that are crucial for mediating innate behaviors in Drosophila . We provide functional evidence for a feature-based spatial arrangement of the LH decoding opposing hedonic valences and odor intensity . The role of the LH as a center for integrating biological values towards innate decisions by computing conveyed information of two processing pathways is thus expanded . All fly stocks were maintained on conventional cornmeal-agar-molasses medium under L:D 12:12 , RH = 70% and 25°C . For wild-type controls D . melanogaster of the Canton-S strain was used . Transgenic lines were obtained from Bloomington Stock Center ( http://flystocks . bio . indiana . edu/ ) and Vienna RNAi stock center ( http://www . vdrc . at ) . Other fly stocks were kindly provided by Kei Ito ( MZ699-GAL4 ) and Maria Luisa Vasconcelos ( UAS-C3PA ) . The END1-2 fly line is published in Grabe et al . ( 2014 ) . Whole-mount immunofluorescence staining was carried out as described ( Laissue et al . , 1999; Vosshall et al . , 2000 ) . Initially brains were dissected in Ringer's solution ( 130 mM NaCl , 5 mM KCl , 2 mM MgCl2 , 2 mM CaCl2 , 36 mM saccharose , 5 mM HEPES , [pH 7 . 3] ) ( Estes et al . , 1996 ) and fixed in 4% PFA in PBS-T ( PBS , 0 . 2–1% Triton-X ) . After washing with PBS-T brains were blocked with PBS-T , 5% normal goat serum ( NGS ) . Primary antibodies were diluted in blocking solution or PBS-T and incubated at 4°C for 2–3 days . Secondary antibody incubation lasted 1–2 days . Brains were mounted in VectaShield ( Vector Labs , Burlingame , CA ) . The following primary antibodies were used: rabbit α-GABA ( 1:500 ) ( Sigma ) , mouse α-GFP ( 1:500 ) ( Invitrogen ) . The following secondary antibodies were used: Alexa Fluor 488 , goat anti-mouse ( 1:500 ) ; Alexa Fluor 546 , goat anti-rabbit ( 1:500 ) ; ( all IgG Invitrogen ) . Fly preparation and functional imaging of the AL was conducted as previously described ( Stökl et al . , 2010; Strutz et al . , 2012 ) . LH imaging was conducted similarly , except for the higher resolution achieved with a 60× water immersion objective ( LUMPlanFl 60×/0 . 90 W Olympus ) . The optical plane was ∼30 µm below the most dorsal entrance point of the iPN tract into the LH . Binning on the CCD-camera chip resulted in a resolution of 1 pixel = 0 . 4 × 0 . 4 µm . For bilateral LH imaging during transection a 20× water immersion objective ( NA 0 . 95 , XLUM Plan FI , Olympus ) was employed . All recordings lasted 10 s with a frame rate of 4 Hz . Odors included acids ( propionic acid , acetic acid ) , lactones ( γ-butyrolactone ) , terpenes ( linalool ) , aromatics ( acetophenone , methyl salicylate , benzaldehyde , phenylacetic acid ) , alcohols ( 1-octen-3-ol ) , esters ( acetoin acetate , cis-vaccenyl acetate , 2-phenethyl acetate ) , ketones ( 2 , 3 butanedione ) and balsamic vinegar diluted in mineral oil ( all from Sigma Aldrich ) . Odors were applied during frame 8–14 ( i . e . , after 2 s , lasting for 2 s ) . Flies were imaged for up to 1 hr , with a minimum inter-stimulus interval of one minute . We selected conventional widefield Ca2+ imaging as the method of choice , since we were able to obtain single bouton resolution with this technique . Calcium imaging data of AL were analyzed with custom-written IDL software ( ITT Visual Information Solutions ) provided by Mathias Ditzen as previously described ( Stökl et al . , 2010; Strutz et al . , 2012 ) . Regarding the Ca2+ imaging data in the LH , we repeated recordings of each odor at each concentration two to three times to ensure the reliability of the extracted domain information . To execute NNMF analysis ( see below ) , at least 6–7 valid measurements , that is , animals with repeated identical recordings , were collected for each odor and employed for the analysis . Individual odor measurements were aligned using ImageJ ( Fiji ) to correct movement artifacts . Fluorescence changes ( ∆F/F ) for each odor were calculated in relation to background fluorescence using frames 0–6 ( i . e . , 2–0 . 5 s before odor application ) . A Gaussian low-pass filter ( ó = 1px ) was applied to compensate for remaining movement artifacts and pixel noise . To reduce the computational load , the frame rate was averaged by two consecutive frames , and recordings were spatially down-sampled by a factor of two . The resulting concatenated time-series of the recordings is denoted as measurement matrix Y with element Yt , p being the tth observed value of pixel p . In contrast to the AL , which consists of highly ordered glomerular subunits , the LH comprises a mainly homogenous neuropil which does not provide spatial or functional landmarks . Therefore , we used the automatic method NNMF to extract Ca2+ signals that exhibit common spatial or temporal features . NNMF , like other matrix factorization techniques ( e . g . , Principal Component Analysis ( PCA ) and Independent Component Analysis ( ICA ) ) , decompose the measurement matrix Y into k components , Y=∑kxk∗akT+R . The time-course ak of each component contains a common underlying time-courses of all pixels and each pixel participation xk declares how strongly each pixel is involved in this time-course . The residual matrix R contains the unexplained data . In order to perform NNMF , we implemented the HALS algorithm in Python including a spatial smoothness constraint ( asm = 0 . 1 ) ( Cichocki and Phan , 2009 ) and an additional spatial decorrelation constraint ( ade = 0 . 1 ) ( Chen and Cichocki , 2005 ) . In PCA decomposition is performed such that either timecourses ak or pixel participation xk are uncorrelated , whereas ICA aims for timecourses ( temporal ICA ) or pixel participation ( spatial ICA ) to be independent . Although spatial ICA is able to segregate signals into functional similar neuropils ( Reidl et al . , 2007 ) , we chose the NNMF approach , because it is known to achieve even a better parts-based representation compared to the more holistic results of PCA or ICA ( Lee and Seung , 1999 ) . In contrast to PCA and ICA , NNMF constrains both the extracted time-courses and pixel participations to be positive . Positive pixel participation enabled us to make a straightforward physiological interpretation , reading the participation values as the contribution strength of an underlying physiological domain . The restriction to positive time-courses reflects the fact that we did not observe any significant decrease of fluorescence in response to an odor in the original measurement data . For each animal we performed decomposition into k = 5 components . This was sufficient to explain most of the data's variance ( 88% + 8% , error is standard deviation across individuals ) . The remaining variance in the residual matrix R contained no additional domains but rather reflected remaining movement artifacts of the measurements ( Figure 2—figure supplement 2 ) . Of the five components extracted by NNMF , three stood out prominently ( Figure 2—figure supplement 3 ) : First , they were extracted in all animals at very clearly defined anatomical positions . Second , their responses to stimuli repetitions were highly reproducible in contrast to the other two components , that is , they exhibited a significant ( p < 2*10−8 , t test ) higher trial-to-trial correlation of 0 . 72 ± 0 . 20 in contrast to 0 . 52 ± 0 . 26 for the remaining components; third , the odorant spectra of their responses were characteristic across animals . Though we cannot completely rule out that the remaining components of the factorization are ORDs of their own , there are several indications that they are not . On the one hand , they exhibit a lower trial-to-trial correlation than the three selected components . Second , those components did not consistently appear at similar anatomical position . Third , they were spatially overlapping with the selected three components . Instead of independent ORDs , these regions might convey fluorescence changes independent of odor stimulation or an overlapping region of two of the reliable ORDs . A validation of our NNMF-based results with spatial ICA yielded very similar , but slightly worse results . Whereas the three reliable ORDs from NNMF were also extracted in spatial ICA , the two remaining components exhibited much higher variability than when obtained with NNMF . Hence , we conclude that the LH area comprising MZ699+ neurons consists of three ORDs . We labeled those three components according to the anatomical position of their pixel participation within the LH . To determine the coding properties of extracted odor response domains ( ORDs ) , we calculated the mean response of each animal within a time window of 1–4 s after stimulus onset . Hence , median responses over all animals defined the standard stimulated response rORDo of an ORD to an odor o . Initially , regions were evaluated individually , and correlations were calculated between standard response spectra and the behavioral response index ( RI ) , or odor concentration , respectively , using the ‘linregress’ function of the Python scipy . stat module . To analyze the combined ORD representations of odor patterns po= ( rPMo , rAMo , rALo ) we calculated for all odor pairs the pattern similarity as correlation distance do1 , o2=1−corr ( po1 , po2 ) . In order to visualize the correlation matrix in a comprehensible way , we then arranged odors according to the single linkage clustering of the Python scipy . cluster . hierarchy module . To compare the representation in the LH to those of the AL , we applied the same procedure to the dorsal glomerular odor activation pattern . For in vivo photoactivation experiments , 1–6 day old flies ( genotype: END1-2 , UAS-C3PA;MZ699-GAL4 ) were dissected as in the imaging experiments except that tracts of the salivary glands were cut to prevent movement . Photoactivation was accomplished via continuous illumination with 760 nm for 15–25 min . After a 5-min break to permit full diffusion of the photoconverted molecules , 925 nm z-stacks of the whole brain were acquired and subsequently used for neuronal 3D-reconstruction . For all 3D reconstructions , the segmentation software AMIRA 4 . 1 . 1 & 5 . 3 . 3 ( FEI Visualization Sciences Group , Burlington , MA ) was used . Neurons of different individuals were embedded into the reference brain using a labelfield registration as previously described ( Rybak et al . , 2010 ) . Briefly , segmented labels of brain neuropils ( AL , MBc , LH ) were registered onto a reference brain image using affine registration followed by elastic warping . In a second step , the calculated transformation matrix was applied to the respective neuron morphology that was then aligned to the reference brain image . For morphological analysis of reconstructed iPNs , we first determined all terminal points of each iPN in the LH area . For each combination of terminals we calculated a similarity score ( s ) in analogy to ( Kohl et al . , 2013 ) as follows:s ( t1 , t2 ) =√e−Δ ( t1 , t2 ) 2/2σ2 , where t is the terminal position , Δ ( t1 , t2 ) is the Euclidean distance and σ is a free parameter that determines how close in space terminal points must be to be considered similar; analogue to Kohl et al . ( 2013 ) we set this parameter to 3 µm . Finally we calculated the pairwise similarity score between two neurons as their average all-to-all terminal similarity scores , normalized to their self-scores as follows:S ( n1 , n2 ) =∑t1 , t2s ( t1 , t2 ) /∑t1s ( t1 , t1 ) ∗∑t2s ( t2 , t2 ) Effectively this quantifies the relative overlap of the target area of all pairs of iPNs . For clustering , the similarity scores were converted to distances ( i . e . , 1-S ) and a hierarchical clustering was performed using UPGMA method . Principal component analysis and one-way ANOSIM was performed using the statistical software PAST 3 . x ( Paleontological statistics software package for education and data analysis ) . Transections of either the plF tract or the mACT were conducted in one brain hemisphere , each of the same fly . The target area was monitored with 925 nm and chosen to be close to the LH but distant enough not to affect neurites ramifying in the LH neuropil . For both tracts , lesioned areas had an average size of 34 µm and were illuminated with short pulses of 710 nm every 40 ms for 250 ms in 60 ( plF ) to 80 ( mACT ) cycles in a single focal plane . After a fast z-stack with 925 nm to confirm complete lesion , a 5-min neuronal recovery interval followed before continuing the imaging procedure . Data were analyzed using NNMF . Photoactivation and transection procedures as well as image acquisition following immunohistochemistry were accomplished with a 2-photon confocal laser scanning microscope ( 2PCLSM , Zeiss LSM 710 NLO ) equipped with a 40× ( W Plan-Apochromat 40×/1 . 0 DIC M27 , Zeiss ) or 20× ( W N-Achroplan 20×/0 . 5 M27 , Zeiss ) . The 2PCLSM was placed on a smart table UT2 ( Newport Corporation , Irvine , CA , USA ) and equipped with an infrared Chameleon Ultra diode-pumped laser ( Coherent , Santa Clara , CA , USA ) . Z-stacks were performed with argon 488 nm and helium-neon 543 nm laser or the Chameleon Laser 925 nm ( BP500-550 for G-CaMP and LP555 for DsRed/tdTomato ) and had a resolution of 1024 or 512 square pixels . The maximum step size for immuno-preparations or single neuron projections was 1 µm and for AL reconstructions 2 µm . Flies carrying P[GAD1-RNAi];P[MZ699-GAL4] were crossed just before the experiment to prevent dosage compensation effects . T-maze experiments were performed as described ( Stensmyr et al . , 2012 ) . WT , parental controls ( P[GAD1-RNAi] or P[MZ699-GAL4] ) and test flies carrying both insertions were tested separately under identical conditions . The response index ( RI ) was calculated as ( O-C ) /T , where O is the number of flies in the odor arm , C is the number of flies in the control arm , and T is the total number of flies used in the trial . Hence , the RI ranges from −1 ( complete avoidance ) to 1 ( complete attraction ) . Each experiment was carried out on 30 flies and was repeated 12 times . Dunn's Multiple Comparison Test was used for global differences in the dataset . Whenever the Multiple Comparison Test was significant ( i . e . , p < 0 . 05 ) , a posthoc test for selected pairs was performed , that is , between the GADi-flies and the other three control lines as we were not interested in differences among the different control lines . All RI were tested against 0 ( no response ) by using the Wilcoxon-rank-sum test .

Organisms need to sense and adapt to their environment in order to survive . Senses such as vision and smell allow an organism to absorb information about the external environment and translate it into a meaningful internal image . This internal image helps the organism to remember incidents and act accordingly when they encounter similar situations again . A typical example is when organisms are repeatedly attracted to odors that are essential for survival , such as food and pheromones , and are repulsed by odors that threaten survival . Strutz et al . addressed how attractiveness or repulsiveness of a smell , and also the strength of a smell , are processed by a part of the olfactory system called the lateral horn in fruit flies . This involved mapping the neuronal patterns that were generated in the lateral horn when a fly was exposed to particular odors . Strutz et al . found that a subset of neurons called inhibitory projection neurons processes information about whether the odor is attractive or repulsive , and that a second subset of these neurons process information about the intensity of the odor . Other insects , such as honey bees and hawk moths , have olfactory systems with a similar architecture and might also employ a similar spatial approach to encode information regarding the intensity and identity of odors . Locusts , on the other hand , employ a temporal approach to encoding information about odors . The work of Strutz et al . shows that certain qualities of odors are contained in a spatial map in a specific brain region of the fly . This opens up the question of how the information in this spatial map influences decisions made by the fly .

2014

Decoding odor quality and intensity in the Drosophila brain

γ-secretase is responsible for the proteolysis of amyloid precursor protein ( APP ) into short , aggregation-prone amyloid-beta ( Aβ ) peptides , which are centrally implicated in the pathogenesis of Alzheimer’s disease ( AD ) . Despite considerable interest in developing γ-secretase targeting therapeutics for the treatment of AD , the precise mechanism by which γ-secretase produces Aβ has remained elusive . Herein , we demonstrate that γ-secretase catalysis is driven by the stabilization of an enzyme-substrate scission complex via three distinct amino-acid-binding pockets in the enzyme’s active site , providing the mechanism by which γ-secretase preferentially cleaves APP in three amino acid increments . Substrate occupancy of these three pockets occurs after initial substrate binding but precedes catalysis , suggesting a conformational change in substrate may be required for cleavage . We uncover and exploit substrate cleavage preferences dictated by these three pockets to investigate the mechanism by which familial Alzheimer’s disease mutations within APP increase the production of pathogenic Aβ species . Alzheimer’s disease ( AD ) is the most common form of dementia and currently the sixth leading cause of death in the United States , with no disease-modifying therapeutics available . A central and pathological hallmark of AD is the deposition of amyloid-beta ( Aβ ) plaques in the brain ( Hardy and Selkoe , 2002 ) . These plaques are comprised of aggregates of Aβ peptides , which are formed by the sequential cleavage of the membrane embedded amyloid precursor protein ( APP ) by two proteases—β-secretase first removes the ectodomain of APP , then γ-secretase cleaves the remaining C-terminal fragment within its transmembrane domain ( TMD ) to liberate Aβ from cellular membranes . Via its proteolytic component presenilin ( Li et al . , 2000; Wolfe et al . , 1999 ) , γ-secretase processes the TMD of APP into Aβ peptides of differing lengths , mostly producing a more benign Aβ species 40 amino acids in length , termed Aβ40 , as well as lesser amounts of a longer , more aggregation prone and pathogenic species , Aβ42 . The total amount of Aβ42 , as well as the ratio of Aβ42/40 , are thought to be key mediators of Aβ pathogenesis , a hypothesis strongly supported by the fact that nearly all the 200-plus autosomal-dominant familial Alzheimer’s disease ( FAD ) mutations in presenilin-1 , -2 and APP increase the Aβ42/40 ratio ( see www . alzforum . org/mutations ) . Due to the strong link between γ-secretase catalyzed Aβ formation and AD pathogenesis , the development of γ-secretase targeting therapeutics has been of high interest over the past two decades ( Golde et al . , 2013; De Strooper , 2014 ) . Both γ-secretase inhibitors ( GSIs ) and a γ-secretase modulator ( GSM ) , the latter working by an unknown mechanism to influence γ-secretase to produce shorter , presumably less pathogenic Aβ species , have failed in recent clinical trials . The failure of GSIs is due at least in part to toxicities from cleavage inhibition of other γ-secretase substrates such as Notch ( Doody et al . , 2013; Golde et al . , 2013 ) . Little is known about how γ-secretase recognizes the transmembrane domain of substrates , given that no consensus amino acid cleavage motif has been identified for the more than one hundred γ-secretase substrates discovered to date ( Haapasalo and Kovacs , 2011 ) . Unfortunately , the further development of safe and effective γ-secretase targeting therapeutics has been held back by a fundamental lack of understanding of how γ-secretase recognizes and cleaves the TMDs of its many substrates , especially APP . Elucidation of this basic mechanism should at the very least add to our understanding of how Aβ is produced and may also aid in the development of safe and effective disease-modifying therapeutics . Mass spectrometry studies have identified a complex mixture of products generated from γ-secretase’s cleavage of the transmembrane domain of APP ( Matsumura et al . , 2014 ) . Looking at the formation of these products over time , it is apparent that the TMD of APP is mostly processed via two major pathways . γ-secretase predominantly initiates endoproteolysis at a so-called epsilon ( ε ) cleavage site—after Leu49 or Thr48 , generating Aβ49 or Aβ48 and two different APP intracellular domains ( AICD ) , AICD 50–99 or AICD 49–99 , respectively ( Kakuda et al . , 2006; Sato et al . , 2003 ) . Aβ49 and Aβ48 are then sequentially cleaved in increments of three amino acids to produce mostly Aβ40 and Aβ42 , respectively ( Takami et al . , 2009 ) . The two major pathways are therefore Aβ49 → Aβ46 → Aβ43 → Aβ40 and Aβ48 → Aβ45 → Aβ42 ( Figure 1A ) ( Fernandez et al . , 2014; Takami et al . , 2009 ) . There are , however , other Aβ species generated by γ-secretase through usually minor and sometimes overlapping , alternative pathways ( Matsumura et al . , 2014; Olsson et al . , 2014 ) . Importantly a shorter peptide , Aβ38 , can be formed from both major pathways , originating from Aβ42 or Aβ43 ( Okochi et al . , 2013 ) . Additionally , a third , sparingly used site of ε cleavage can lead to the production of Aβ47 , which rather than being processed to Aβ44 is instead mostly cleaved to Aβ43 , subsequently generating Aβ40 ( Aβ47 → Aβ43 → Aβ40 ) ( Matsumura et al . , 2014 ) . 10 . 7554/eLife . 17578 . 003Figure 1 . Tripeptide fragments of APP inhibit γ-secretase . ( A ) Schematic diagram of the major sequential cleavage pathways of the transmembrane domain of APP ( Aβ49 → Aβ46 → Aβ43 → Aβ40 in red and Aβ48 → Aβ45 → Aβ42 in blue ) . Mutations causing Familial Alzheimer’s disease are below the APP TMD in blue . ( B ) IC50 curves from the inhibition of γ-secretase activity by APP product tripeptide fragments . Mean ± SD , n = 2 . ( C ) Noncompetitive inhibition of γ-secretase with VIV tripeptide , R2 = 0 . 98 . ( D ) Yonetani-Theorell plot for the mutually exclusive binding of VIV and the noncompetitive transition-state analog inhibitor III-31-C , R2 = 0 . 98 . ( E ) Cartoon representation of the three S’ pockets of presenilin ( PSEN ) along with three P’ amino acids of substrate and the transition-state analog L685 , 458 . ( F ) IC50 curves from the inhibition of γ-secretase activity with FAF and AFA synthetic tripeptides . Mean ± SD , n = 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 17578 . 003 Normally , γ-secretase uses the Aβ49 → Aβ40 and the Aβ48 → Aβ42 pathways to produce mostly Aβ40 and Aβ42 via a stepwise , tripeptide cleavage process . The mechanism that dictates this preferred tripeptide cleavage ( and thus the driving force behind γ-secretase catalysis and Aβ formation ) is completely unknown . In this study , we report that γ-secretase tripeptide cleavage is driven by three S’ pockets within the active site of the enzyme . We identify specific substrate cleavage preferences dictated by the three S’ pockets and exploit these preferences to determine the predominant mechanism of each FAD mutation within the transmembrane domain of APP , including a novel mechanism in which final cleavage products are uncoupled from initial ε pathway preference . When studying enzyme catalysis much focus is appropriately placed on determining how an enzyme interacts with its substrate . However , oftentimes the manner in which an enzyme interacts with product ( in the form of product inhibition ) can be equally informative with regard to its catalytic mechanism . To this end , we asked whether the naturally produced tripeptide fragments of APP are inhibitors of γ-secretase . We found that all five tripeptides produced from the TMD of APP are indeed capable of inhibiting γ-secretase activity , albeit rather weakly with IC50 values ranging from ~150 μM to several mM ( Figure 1B ) . Although these binding affinities are too low for the tripeptides to be involved in any form of biologically relevant feedback inhibition , we imagined the manner in which they inhibit γ-secretase could be instructive in elucidating the basic cleavage mechanism of the protease . We characterized the mode of inhibition of the most potent of the tripeptides , VIV , finding that these data fit well to a noncompetitive inhibition model , with a global R2 of 0 . 98 ( Figure 1C ) . Given that the tripeptide segments of the TMD of APP must occupy the active site of γ-secretase during catalysis , we hypothesized that VIV may compete for the same binding site on the enzyme as transition-state analogs . An inhibitor cross-competition analysis reveals this is likely true , with a series of parallel lines resulting from a Yonetani-Theorell plot demonstrating mutually exclusive binding of VIV and the transition-state analog III-31-C ( Figure 1D ) . We noticed that nearly all of the γ-secretase targeting transition-state analog inhibitors developed to date ( e . g . L685 , 458 , III-31-C ) contain essentially a tripeptide fragment C-terminal to the transition-state-mimicking hydroxyl isostere ( Figure 1E ) . Structure-activity relationship ( SAR ) studies have demonstrated that transition-state analog inhibitors containing only two amino acids here are relatively weak inhibitors compared to those comprised of three amino acids , while adding a fourth amino acid does not achieve additional potency ( Esler et al . , 2004 ) . This suggests that there are three , and only three , putative S’ pockets in the presenilin active site that contribute to inhibitor binding . Of note , the SAR studies also suggest that while the putative S1’ and S3’ pockets are large and can accommodate amino acids of varying size , the S2’ pocket is small and inhibitors with an aromatic amino acid ( phenylalanine ) at this position have decreased potency by two orders of magnitude compared to those with less bulky aliphatic amino acids ( Esler et al . , 2004 ) . We imagined that the tripeptide APP TMD products may be binding these three putative S’ pockets to achieve inhibition . In agreement with this hypothesis , a synthetic tripeptide of the predicted optimal binding sequence ( FAF , to fit in the large-small-large S1’-S2’-S3’ pockets ) was a more potent inhibitor of γ-secretase activity than any of the naturally occurring APP-derived tripeptides , while a peptide predicted to clash with this binding site ( AFA ) is a very weak inhibitor , even at mM concentrations ( Figure 1F ) . Based on these results , we reasoned that the three putative S’ pockets within the γ-secretase active site are likely occupied by substrate prior to hydrolysis of the scissile bond as a means for γ-secretase to stabilize a transition-state-like scission complex with substrate . This would provide a simple mechanism for the preferred cleavage of APP in three amino acid increments , as well as provide an explanation for why γ-secretase mostly sticks to each major pathway , producing Aβ40 or Aβ42 after initiating cleavage of APP at either the L49 or T48 ε sites , respectively . Additionally , given that substrate movement in the form of helical unwinding is thought to be a required step in the poorly defined intramembrane protease cleavage mechanism ( Akiyama et al . , 2015; Dickey et al . , 2013; Fluhrer et al . , 2012; Moin and Urban , 2012; Urban and Freeman , 2003; Ye et al . , 2000 ) , the three S’ pockets could potentially provide a means for γ-secretase to stabilize its helical substrate in a more cleavable conformation , thereby lowering the activation energy required for catalysis . To test this hypothesis , we took advantage of the fact that the putative S2’ pocket of γ-secretase is apparently small and has a reduced ability to accommodate a bulky amino acid such as phenylalanine . We predicted that we should be able to selectively decrease cleavage of either the Aβ49 → 40 pathway or the Aβ48 → 42 pathway simply by placing an aromatic amino acid at the P2’ position of APP at the initial ε cut site . In other words , an aromatic amino acid placed at V50 should reduce ε cleavage after T48 , thereby lowering the amount of Aβ42 produced , thus decreasing the Aβ42/40 ratio . Conversely , an aromatic amino acid at M51 should reduce ε cleavage after L49 , lowering the amount of Aβ40 produced and therefore increase the Aβ42/40 ratio ( Figure 2A ) . 10 . 7554/eLife . 17578 . 004Figure 2 . Selective blocking of the Aβ40 or Aβ42 pathways with aromatic amino acids placed at the P2’ position of ε cleavage . ( A ) Schematic diagram of the TMD of APP with pathway blocking aromatic amino acid mutations at the P2’ position for the T48 or L49 ε cleavage . ( B ) In vitro Aβ42/40 ratios with Phe or Trp mutations at V50 or M51 . Aβ measured using Aβ40 and Aβ42 ELISA kits from Invitrogen . Mean ± SD , n = 3 , t-test **<0 . 01 , ****<0 . 0001 . ( C ) MALDI/TOF MS of the AICD fragments generated from in vitro cleavage of C100: WT ( AICD 50–99 , expected mass: 6905 . 6 , observed mass: 6907 . 4; AICD 49–99 , expected mass: 7018 . 8 , observed mass: 7021 . 3 ) . V50F ( AICD 50–99 , expected mass: 6953 . 8 , observed mass: 6949 . 8; AICD 48–99 , expected mass: 7167 . 9 , observed mass: 7163 . 5 ) . M51F ( AICD 49–99 , expected mass: 7034 . 8 , observed mass: 7030 . 1; AICD 48–99 , expected mass: 7135 . 8 , observed mass: 7131 . 8 ) . ( D ) Aβ42/40 ratios measured from the media of HEK cells transfected with V50 mutants . Aβ levels measured by 6E10 ELISA . Mean ± SD , n = 3 , t-test **<0 . 01 , ***<0 . 001 , ****<0 . 0001 . ( E ) Aβ42/40 ratios measured from the media of HEK cells transfected with M51 mutants . Aβ levels measured by 6E10 ELISA . Mean ± SD , n = 3 , t-test ***<0 . 001 , ****<0 . 0001 . ( F ) Aβ40 and Aβ42 levels for aromatic substitutions at V50 and M51 normalized to WT . Aβ levels measured by 6E10 ELISA . Mean ± SD , n = 3 , t-test *<0 . 05 , ***<0 . 001 , ****<0 . 0001 . ( G ) Total secreted Aβ levels ( see Materials and methods ) from the aromatic mutations at V50 and M51 . DOI: http://dx . doi . org/10 . 7554/eLife . 17578 . 004 In an in vitro assay using purified γ-secretase to cleave recombinant C100-FLAG APP-based substrate , V50F and V50W both decreased the Aβ42/40 ratio , while the same substitutions at M51 increased the Aβ42/40 ratio as predicted ( Figure 2B ) . MALDI/TOF mass spectrometry ( MS ) analysis of the corresponding AICD fragment revealed the complete elimination of AICD 49–99 for V50F and of AICD 50–99 for M51F ( Figure 2C ) . Interestingly , in addition to the expected AICD fragments , both V50F and M51F are also cleaved to a minor extent after I47 , producing AICD 48–99 . The reason for this is unknown , although γ-secretase may be compensating for reduced cleavage through one of the two major pathways . Previous MS studies have demonstrated that the majority of Aβ47 is eventually processed to Aβ40 , through an Aβ43 intermediate ( Matsumura et al . , 2014 ) . We obtained similar results measuring secreted Aβ after transiently transfecting HEK cells with full-length APP containing mutations at V50 or M51 . Here , all mutations tested at V50 caused a reduction in the Aβ42/40 ratio , although none as robustly as the bulky aromatic amino acids Phe , Tyr and Trp ( Figure 2D ) . And while mutations of smaller amino acids at M51 caused modest reductions in the Aβ42/40 ratio , aromatic substitutions here all caused substantial increases in the Aβ42/40 ratio ( Figure 2E ) . As predicted by our model , there was a significant reduction in Aβ42 production for the V50 aromatic substitutions , with little change in the Aβ40 levels ( Figure 2F ) . There was a similar predicted reduction in Aβ40 for the M51 aromatic mutants; however , we also see an increase in Aβ42 levels here ( Figure 2F ) , likely because the normally less used Aβ48 → 42 pathway is compensating for the reduced flux through the preferred Aβ49 → 40 pathway , as these mutants all produced roughly equivalent amounts of total Aβ ( Figure 2G ) . Because each tripeptide cleavage event requires the reading of three amino acids of substrate at a time ( as dictated by the three S’ pockets ) , we reasoned we should be able to predictably shift the Aβ42/40 ratio by placing a Phe in the P2’ position at each tripeptide cleavage event along the two major pathways ( Figure 3A ) . As expected , Phe substitutions at V44 and I47 decreased the Aβ42/40 ratio , while Phe mutations at I45 and T48 increased the Aβ42/40 ratio ( Figure 3B and C ) . The predicted Aβ42/40 shifts were nearly identical whether Aβ levels were measured from an in vitro assay ( Figure 3B ) or from a cell-based assay ( Figure 3C ) . These results are of particular note , as I45F is a known FAD mutation ( Guerreiro et al . , 2010 ) , likely indicating that the mechanism of this mutation is the Phe positioned in the P2’ position at the Aβ43 cut site , blocking cleavage of APP through the less pathogenic Aβ49 → 40 pathway and favoring the pathogenic Aβ48 → 42 pathway . The Phe at I45 also lies in the S3’ position for Aβ42 cleavage , likely making the precursor to Aβ42 production a better substrate for γ-secretase through a favorable P3’-S3’ interaction . Mutating A42 to Phe completely blocked the formation of Aβ40 in agreement with our model ( Figure 3D ) , while still allowing for the production of other Aβ species ( Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 17578 . 005Figure 3 . Phenylalanine mutations at P2’ positions predictively shift the Aβ42/40 ratio . ( A ) Schematic diagram of Phe mutations at P2’ positions at cut sites within the TMD of APP and the expected Aβ42/40 changes compared to WT . ( B ) In vitro Aβ42/40 ratios from γ-secretase cleavage of recombinant C100-FLAG substrates . Aβ measured using Aβ40 and Aβ42 ELISA kits from Invitrogen . Mean ± SD , n = 3 , t-test *<0 . 05 , **<0 . 01 . ( C ) Aβ42/40 ratios from Aβ secreted from HEK cells . Aβ levels measured by 6E10 ELISA . Mean ± SD , n = 3 , t-test **<0 . 01 , ***<0 . 001 , ****<0 . 0001 . ( D ) Aβ40 levels measured from the conditioned media of HEK cells transfected with WT or A42F APP . Aβ levels measured by 6E10 ELISA . Mean ± SD , n = 3 . ( E ) Western blot analysis of the AICD fragments generated from V44F , I45F , I47F and T48F in vitro . Total AICD was measured with anti-FLAG M2 antibody ( green ) . Aβ49 → Aβ40 pathway preference was measured with an antibody specifically recognizing the N-terminus of AICD 50–99 fragment ( red ) . ( F ) MALDI/TOF MS confirmation of the AICD fragments measured in ( E ) : V44F ( AICD 49–99 , expected mass: 7018 . 8 , observed mass: 7020 . 8 ) , V45F ( AICD 50–99 , expected mass: 6905 . 6 , observed mass: 6906 . 2; AICD 49–99 , expected mass: 7018 . 8 , observed mass 7020 . 1 ) , I47F ( AICD 49–99 , expected mass: 7018 . 8 , observed mass: 7019 . 5 ) , T48F ( AICD 50–99 , expected mass: 6905 . 6 , observed mass: 6907 . 4; AICD 49–99 , expected mass: 7018 . 8 , observed mass: 7019 . 5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17578 . 00510 . 7554/eLife . 17578 . 006Figure 3—figure supplement 1 . Total Aβ from the A42F mutant . ( A ) Total secreted Aβ was measured by ELISA from transiently transfected HEK cells with full length WT and A42F APP . Mean ± SD , n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 17578 . 006 Although we expected the Phe substitutions at V44 , I45 , I47 and T48 to be acting independently of the pathway chosen at initial ε cleavage , an alternative explanation for the above results is that these mutations instead influence ε cleavage in favor of the final Aβ products measured . To investigate this possibility , we utilized an antibody that specifically recognizes the free N-terminus of AICD 50–99 ( Chávez-Gutiérrez et al . , 2012 ) , and therefore the initiation of the Aβ49 → 40 pathway . Surprisingly , all four of these mutants actually caused a shift in ε cleavage toward the initiation of the opposite pathway . Although V44F and I47F cause reductions in the Aβ42/40 ratio , they shifted initial ε cleavage away from the Aβ49 → 40 pathway—nearly completely eliminating AICD 50–99 . And while I45F and T48F increased the Aβ42/40 ratio , they shifted ε cleavage more in favor of AICD 50–99 compared to WT ( Figure 3E ) . These AICD species were confirmed by mass spectrometry ( Figure 3F ) . To explore further the apparent ability of tripeptide cleavage preference to dissociate the normal connection between initial ε cleavage and final γ cleavages , we made double Phe mutants in which the ε cut site is controlled with a Phe at either V50 or M51 , while placing a conflicting Phe at V44 , I45 , I47 or T48 in the initial pathway . Measuring Aβ secreted from transfected HEK cells , we clearly saw the double mutants behave almost identically to the single point mutants N-terminal to the ε cut site ( Figure 4A and B ) . MS of the AICD fragments revealed the expected and complete blocking of AICD 49–99 or AICD 50–99 for the V50F and M51F containing double mutants , respectively ( Figure 4C ) . Together these data suggest final γ cleavages can be completely uncoupled from initial ε cleavages . 10 . 7554/eLife . 17578 . 007Figure 4 . Phenylalanine mutations in the P2’ position of the last read tripeptide segment dictates final pathway preference . ( A ) Aβ42/40 ratios from HEK cells of V44F-M51F and I47F-M51F double mutants behave like single Phe mutants V44F and I47F , respectively . Aβ levels measured by 4G8 ELISA . Mean ± SD , n = 3 , t-test **<0 . 01 , ***<0 . 001 . ( B ) Aβ42/40 ratios from HEK cells of I45F-V50F and T48F-V50F double mutants behave like single Phe mutants I45F and T48F , respectively . Aβ levels measured by 4G8 ELISA . Mean ± SD , n = 3 , t-test *<0 . 05 , **<0 . 01 , ***<0 . 001 , ****<0 . 0001 . ( C ) MALDI/TOF MS conformation of the elimination of AICD 49–99 and AICD 50–99 for the V50F and M51F containing double Phe mutants , respectively . V44F-M51F ( AICD 49–99 , expected mass: 7034 . 8 , observed mass: 7030 . 7; AICD 47–99 expected mass: 7249 . 0 , observed mass: 7253 . 7 ) , I45F-V50F ( AICD 50–99 , expected mass: 6953 . 8 , observed mass: 6950 . 1; AICD 48–99 , expected mass: 7167 . 9 , observed mass: 7164 . 6 ) , I47F-M51F ( AICD 49–99 , expected mass: 7034 . 8 , observed mass: 7032 . 2; AICD 47–99 , expected mass: 7283 . 0 , observed mass: 7280 . 1 ) , T48F-V50F ( AICD 50–99 , expected mass: 6953 . 8 , observed mass: 6949 . 4; AICD 48–99 , expected mass: 7214 . 0 , observed mass: 7209 . 4 ) . ( D ) Aβ42/40 ratios from HEK cells transfected with double Phe mutations in tandem . Aβ levels measured by 6E10 ELISA . Mean ± SD , n = 3 , t-test **<0 . 01 , ***<0 . 001 , ****<0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 17578 . 007 Single Phe mutations at V46 and L49 both caused modest increases in the Aβ42/40 ratio . We predicted that since these mutations do not occupy the S2’ pocket for either major pathway , V46F and L49F when paired with a neighboring Phe mutation should behave like its neighboring Phe mutant alone . Indeed , L49F-V50F had a reduced Aβ42/40 ratio compared to WT like V50F alone , while T48F-L49F had an elevated Aβ42/40 ratio comparable to T48F ( Figure 4D ) . Likewise , V46F-I47F displayed a reduced Aβ42/40 ratio similar to I47F alone , while I45F-V46F had a drastically increased Aβ42/40 ratio like I45F alone ( Figure 4D ) . To this point , aromatic amino acids placed in the P2’ position at each tripeptide cleavage site within the TMD of APP caused a predictable outcome without exception . We therefore reasoned that placing two phenylalanines in tandem , such that there is a Phe in the P2’ position at both major ε cut sites , should reduce overall cleavage . As predicted , a double mutant of V50F-M51F caused a sharp reduction in total AICD formation compared to WT and the V50F and M51F single point mutations alone ( Figure 5A ) . However , AICD formation from V50F-M51F cleavage was not completely abolished; rather , its rate of production was markedly reduced ( Figure 5B ) . This suggests γ-secretase has a means by which to overcome two aromatic amino acids in a row , which would seemingly conflict with its cleavage preferences . 10 . 7554/eLife . 17578 . 008Figure 5 . Phenylalanine blocking mutations at both ε cleavage sites reduces APP cleavage but not binding to γ-secretase . ( A ) Western blot of γ-secretase cleavage of WT , V50F , M51F and V50F-M51F C100-FLAG . Duplicates from each substrate represent separate independent data points . * denotes a degradation product which co-purified with the substrate . ( B ) Cleavage of WT and V50F-M51F C100-FLAG over time . ( C ) Co-immunoprecipitation of Myc-tagged WT or V50F-M51F C100 substrate . Duplicates are from separate pull-down experiments . * antibody light chain . ( D ) Competitive cleavage of WT C100-FLAG by WT C100-Myc or V50F-M51F C100-Myc . ( E ) Aβ42/40 ratio of the V50F-M51F double mutant . Mean ± SD , n = 3 , t-test , ****<0 . 0001 . ( F ) MALDI/TOF MS of the AICD fragment from the V50F-M51F mutant: ( AICD 51–99 , expected mass: 6822 . 5 , observed mass: 6817 . 1; * unknown peak , observed mass: 7030 . 4; AICD 48–99 , expected mass: 7183 . 9 , observed mass: 7179 . 2; AICD 47–99 , expected mass: 7297 . 1 , observed mass: 7292 . 5 , AICD 46–99 , expected mass: 7396 . 2 , observed mass: 7396 . 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17578 . 008 The observation that both tripeptide cleavage products of APP and transition-state analog inhibitors are noncompetitive inhibitors of γ-secretase suggests that the subsites on γ-secretase for initial substrate binding and subsequent catalysis are spatially separate and distinct , meaning substrate movement is likely required after initial substrate binding but prior to catalysis . This would be in agreement with previous reports suggesting there may be an exosite on γ-secretase to which substrate initially binds prior to translocating to the active site for cleavage ( Kornilova et al . , 2003 , 2005 ) . It would also agree with studies of other intramembrane cleaving proteases , which have proposed substrate movement in the form of helical unwinding is a prerequisite for catalysis ( Akiyama et al . , 2015; Dickey et al . , 2013; Fluhrer et al . , 2012; Moin and Urban , 2012; Urban and Freeman , 2003; Ye et al . , 2000 ) . Given that the S’ pockets within the active site are likely the second binding site for substrate , we reasoned that although the V50F-M51F mutant cannot be hydrolyzed as efficiently as WT , it should still be effectively bound by γ-secretase in an initial docking site . Co-IP of WT and V50F-M51F C100 in complex with γ-secretase reveals equal amounts of substrate bound , suggesting both substrates have a similar binding affinity for the enzyme ( Figure 5C ) . Furthermore , myc-tagged V50F-M51F was just as effective as myc-tagged WT C100 at competing for γ-secretase cleavage of FLAG-tagged WT C100 ( Figure 5D ) , again indicating V50F-M51F and WT substrate initially interact with γ-secretase in a similar manner . Together this suggests that the two Phe mutations in V50F-M51F do not affect initial recognition or binding of the C100 substrate , but rather may reduce the stabilization of a cleavable intermediate enzyme-substrate complex by clashing with the S’ pockets in the presenilin active site . We noticed that in addition to being processed more slowly , V50F-M51F also produced a sharply lower Aβ42/40 ratio compared to WT ( Figure 5E ) . To determine why this was occurring , we performed MS on the AICD fragment . Surprisingly , we found that ε cleavage was initiated almost exclusively after I47 ( the very next available cleavage site ) , generating a 48–99 AICD fragment ( Figure 5F ) , indicating γ-secretase is capable of skipping the two phenylalanines to initiate cleavage , albeit at a slower rate . Aβ47 is primarily processed to Aβ40 ( Aβ47 → Aβ43 → Aβ40 ) ( Matsumura et al . , 2014 ) , accounting for the decreased Aβ42/40 ratio . Although γ-secretase usually cleaves APP in increments of three amino acids to produce predominantly Aβ40 and Aβ42 in the mechanism outlined in this study , a seemingly unusual cleavage deviates from the tripeptide preference to produce Aβ38 in appreciable quantities ( Okochi et al . , 2013; Takami et al . , 2009 ) . To determine if production of Aβ38 occurs through the same three S’ pockets used by γ-secretase to achieve tripeptide cleavage , we transiently transfected HEK cells with mutant APP containing a single Phe point mutant at V39 , V40 , I41 or A42 . Measuring secreted Aβ38 revealed that V40F almost completely eliminated Aβ38 production ( Figure 6A ) , demonstrating occupancy of the three S’ pockets is required for Aβ38 formation . V39F and I41F both sharply increased Aβ38 , a result we interpret as the phenylalanine mutations making the precursor to Aβ38 a better substrate for γ-secretase , as the S1’ and S3’ pockets are large and prefer phenylalanine over smaller amino acids ( Esler et al . , 2004 ) . A42F produces about as much Aβ38 as WT , indicative of the lack of a contributory S4’ pocket . 10 . 7554/eLife . 17578 . 009Figure 6 . γ-Secretase preferentially cleaves APP near the helix-destabilizing Gly-Gly motif . ( A ) Aβ38 levels from HEK cells transiently transfected with V39F , V40F , I41F or A42F APP . Aβ levels measured by 4G8 ELISA . Mean ± SD , n = 3 . ( B ) Aβ42/40 ratios from V44F-I47F-V50F and I45F-T48F-M51F triple mutants from transiently transfected HEK cells . Aβ levels measured by 4G8 ELISA . Mean ± SD , n = 3 , t-test , ****<0 . 0001 . ( C ) Aβ38 , Aβ40 and Aβ42 levels from HEK cells transfected with V44F-I47F-V50F and I45F-T48F-M51F triple mutants , the V44F-I45F double mutant and the hexa-mutant V44F-I45F-I47F-T48F-V50F-M51F . Aβ levels measured by 4G8 ELISA . Mean ± SD , n = 3 . ( D ) Schematic diagram of sequential Phe mutants in the TMD of APP . ( E ) Aβ38 + 40 + 42 secreted from HEK cells transiently transfected with the mutants from ( D ) . Aβ levels measured by 4G8 ELISA . Mean ± SD , n = 3 , t-test ***<0 . 001 , ****<0 . 0001 . ( F ) Aβ42/40 ratios from ( E ) . Aβ levels measured by 4G8 ELISA . Mean ± SD , n = 3 , t-test *<0 . 05 , ****<0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 17578 . 00910 . 7554/eLife . 17578 . 010Figure 6—figure supplement 1 . HEK cell expression and Aβ production from consecutive phenylalanine APP mutants . ( A ) Aβ38 , Aβ40 and Aβ42 levels secreted from HEK cells after transient transfection with APP mutants containing stretches of consecutive phenylalanines ( see Figure 6D ) . Aβ levels measured by 4G8 ELISA . Mean ± SD , n = 3 . ( B ) Expression levels of consecutive phenylalanine APP mutants from Figure 6D . DOI: http://dx . doi . org/10 . 7554/eLife . 17578 . 010 During the course of this study , we found that three phenylalanines sequentially mutated in the P2’ positions of each major pathway ( V44F-I47F-V50F and I45F-T48F-M51F ) caused a very strong reduction or elevation in the Aβ42/40 ratio in the predicted direction ( Figure 6B ) . As expected , the shifts were caused by the near complete elimination of Aβ42 for the V44F-I47F-V50F mutant and of Aβ40 for the I45F-T48F-M51F mutant ( Figure 6C ) . Interestingly , in each case , Aβ38 levels were produced in amounts comparable to or greater than WT . This demonstrates that Aβ38 is capable of being produced from both the Aβ49 → 40 and Aβ48 → 42 pathways , in perfect agreement with recent MS studies that identified the precursors to Aβ38 as being either Aβ43 or Aβ42 ( Matsumura et al . , 2014; Olsson et al . , 2014 ) . Surprisingly , when we blocked the production of both Aβ43 and Aβ42 at the same time with a V44F-I45F double mutant we not only did not prevent the production of Aβ38 , but rather Aβ38 levels were drastically increased ( Figure 6C ) . Even after blocking the first six major cleavage sites ( V44F-I45F-I47F-T48F-V50F-M51F ) of APP , we still observe elevated Aβ38 compared to WT , demonstrating γ-secretase is fully capable of traversing multiple phenylalanines within APP to find the especially labile amide bond between G38 and V39 . It is likely that the helix-destabilizing Gly-Gly motif at G37 and G38 is the reason for this observation , making the G38-V39 bond particularly accessible for cleavage . Next , we attempted to determine how many phenylalanines in a row γ-secretase was capable of skipping by taking advantage of the fact that the GG motif apparently allows for γ-secretase to deviate from normal sequential tripeptide cleavage . Astonishingly , even after increasing the number of phenylalanines in a row to eight , γ-secretase was still able to produce Aβ above mock transfected levels ( Figure 6D , E ) . The V44F⇒M51F mutant produced mostly Aβ38 ( Figure 6—figure supplement 1 ) , in what may be a single endoproteolytic cleavage event , although we have not been able to obtain enough AICD for MS confirmation . Predictably , the Aβ42/40 ratios for these mutants follow a pattern expected if γ-secretase cleaves at the next available site after each additional Phe ( Figure 6F ) . Using our newfound knowledge of the basic cleavage mechanism of γ-secretase , and our ability to precisely control it with phenylalanine mutations , we next decided to investigate the mechanism of FAD mutations within the TMD of APP , which all increase the Aβ42/40 ratio to different degrees . There are more than a dozen missense FAD mutations targeting this region of APP ( Figure 1A ) , with the majority being located N-terminal and downstream of ε cleavage . To date , the currently accepted explanations for how these mutations increase the Aβ42/40 ratio are: 1 ) affecting the positioning/helical stability of the γ-secretase bound APP TMD such that initial ε cleavage is shifted toward T48 , thus favoring the production of Aβ42 ( Chávez-Gutiérrez et al . , 2012; Chen et al . , 2014; Dimitrov et al . , 2013 ) ; or 2 ) influencing the reaction kinetics of γ-secretase’s processing of APP , leading to incomplete carboxy-trimming and therefore increased Aβ42 over the more processed Aβ38 ( Chávez-Gutiérrez et al . , 2012 ) ( Figure 7A ) . As demonstrated above with the I45F FAD mutant , we now show that a third possible mechanism exists , in that sequence-specific cleavage preferences can uncouple initial ε cleavage and final γ cleavages of APP ( Figure 7A ) . 10 . 7554/eLife . 17578 . 011Figure 7 . The tripeptide cleavage mechanism of γ-secretase and the effect of APP transmembrane domain FAD mutations . ( A ) The three mechanisms by which FAD mutations within the TMD of APP increase the Aβ42/40 ratio . 1 ) Mutations shift initial ε cleavage towards the Aβ42 pathway . 2 ) Mutations reduce cleavage of the third cleavage event , producing more Aβ42 over the more processed Ab38 . 3 ) Cleavage specific preferences cause switching from the Aβ40 to the Ab42 pathway , as exemplified by the I45F FAD mutant . ( B ) The Aβ42/40 ratio of each FAD mutation with and without an additional V50F mutation to control the ε cleavage site . The majority of mutations are rescued by the V50F substitution suggesting that these FAD mutations increase the Aβ42/40 ratio by influencing ε cleavage and/or affecting carboxy-trimming . I45F , I45T and T48P retain significantly elevated ratios , indicating these mutants dissociate initial ε and final γ cleavages . Aβ levels measured by 4G8 ELISA . Mean ± SD , n = 3 , t-test ***<0 . 001 , ****<0 . 0001 . ( C ) The tripeptide cleavage mechanism of γ-secretase . After initial substrate binding , we speculate that the helical TMD of substrate unwinds into the active site of presenilin ( PSEN ) where it is stabilized by the three S’ pockets in the catalytic pocket prior to cleavage . Successive carboxy tripeptide trimming occurs until the eventual release of Aβ peptide . ( D ) γ-secretase cleavage of the transmembrane domain of Notch from Okochi et al , 2002 . DOI: http://dx . doi . org/10 . 7554/eLife . 17578 . 01110 . 7554/eLife . 17578 . 012Figure 7—figure supplement 1 . APP FAD mutant panel measured by 6E10 ELISA . ( A ) The samples from Figure 7B were measured by ELISA with the 6E10 detection antibody , showing similar results to those of the 4G8 ELISA . Mean ± SD , n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 17578 . 01210 . 7554/eLife . 17578 . 013Figure 7—figure supplement 2 . AICD fragments for the three I45 FAD mutants determined by western blot using the AICD 50–99 specific antibody . ( A ) Western blot of the AICD 50–99 fragment for I45F , I45V and I45T FAD mutations . ( B ) Quantification of western blot bands from ( A ) . Mean ± SD , n = 3 , t-test **<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 17578 . 01310 . 7554/eLife . 17578 . 014Figure 7—figure supplement 3 . Secreted Aβ levels from V40F , A42F and V44F . ( A ) Aβ38 , Aβ40 and Aβ42 levels measured from V40F , A42F and V44F transfected HEK cells respectively , as well as mock transfected and WT . Aβ levels measured by 6E10 ELISA . Mean ± SD , n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 17578 . 014 To determine the prevalence of the two previously proposed mechanisms and to possibly identify additional pathway uncoupling mutants , we made each FAD mutant within the APP TMD alone and as a double mutant with V50F . The V50F mutation would be predicted to block ε cleavage after T48 independent of the FAD mutant’s affect on subsequent cleavage events . We would therefore predict that if an FAD mutant causes an increase in the Aβ42/40 ratio by influencing initial ε cleavage or by affecting subsequent carboxy-trimming along the Aβ48 → 42 pathway , then that same FAD mutant when paired with V50F should produce a reduced Aβ42/40 ratio compared to WT , similar to V50F alone . Conversely , if an FAD mutant affects subsequent cleavage events independent of ε cleavage , causing uncoupling of initial ε cleavage and final γ cleavages , then the FAD-V50F double mutant should retain an elevated Aβ42/40 ratio , similar to the FAD mutant alone . Screening nearly all the FAD mutants within the TMD of APP by this method , we found that the majority were either completely or nearly completely rescued when paired with V50F ( Figure 7B , Figure 7—figure supplement 1 ) , resulting in 42/40 ratios significantly less than WT . This suggests the predominant mechanism of elevating the Aβ42/40 ratio by FAD mutants within the TMD of APP is by shifting the preference of initial ε cleavage from the Aβ40 to the Aβ42 pathway , and/or by influencing carboxy-trimming . This may have been predicted , given that I45F is the only aromatic amino acid mutation to fall in the S2’ pocket of one of the major cleavage pathways . The only other FAD mutation containing an aromatic amino acid , V46F , falls within the S1’ and S3’ pockets for the Aβ42 and Aβ40 pathways , respectively , therefore never clashing with the S2’ pocket and not influencing the Aβ42/40 ratio as a major pathway blocker . However , we found that in addition to I45F , there are two other mutants , I45T and T48P , which appear to dissociate the normal connection between initial pathway preference and final cleavage products . V50F partially rescues the Aβ42/40 ratio when paired with I45T , but remains significantly elevated compared to WT , indicating the I45T mutant both influences initial ε cleavage and uncouples ε from γ cleavages . The change in ε cleavage preference of I45T was verified using the AICD 50–99 specific antibody ( Figure 7—figure supplement 2 ) , showing a small reduction in AICD 50–99 . Interestingly , the I45T mutant reduces the amount of AICD 50–99 comparable to I45V , even though these two mutants display very different Aβ42/40 ratios . This again suggests I45T dissociates cleavage downstream of ε to achieve such a high Aβ42/40 ratio . Exactly how I45T does this is currently unknown and requires further investigation . The T48P-V50F double mutant behaves identically to T48P alone . Given that proline is a helix-breaking amino acid , the T48P mutant may not undergo normal ε cleavage after P48 or L49 . Determining how T48P increases the 42/40 ratio and overcomes the ε controlling V50F mutant will require further investigation . In this study , we identify that three S’ amino acid binding pockets guide the productive positioning of substrate into the γ-secretase active site , providing the mechanism behind the enzyme’s preferred tripeptide cleavage of APP and pathogenic Aβ production . Based on the data reported herein ( discussed further below ) and numerous other studies of γ-secretase ( Das et al . , 2003; Kornilova et al . , 2003 , 2005 ) and other intramembrane proteases ( Akiyama et al . , 2015; Dickey et al . , 2013; Fluhrer et al . , 2012; Moin and Urban , 2012; Urban and Freeman , 2003; Ye et al . , 2000 ) , we speculate that after initial binding to γ-secretase , substrate must undergo a translocation and/or conformational change in order to bind the three S’ pockets within the active site and subsequently be cleaved by the enzyme ( Figure 7C ) . After initial endoproteolysis , the three S’ pockets guide further carboxy-trimming of the retained Aβ species until it is short enough to dissociate from the complex , producing predominantly Aβ38 , Aβ40 and Aβ42 . More than a decade ago , SAR studies of γ-secretase-targeting transition-state analogs putatively assigned three S’ pockets to the active site of the enzyme ( Esler et al . , 2004 ) . Transition-state analogs containing only two amino acids for S’ pocket binding inhibited γ-secretase less effectively , indicating that occupancy of the S3’ pocket is required for a strong interaction . This likely provides the reason why γ-secretase only cleaves APP in segments of three or more amino acids , but never two . The preference for cleaving three amino acids originates from the lack of a contributory fourth S’ pocket . A fourth amino acid ( no matter the size ) added to a transition state inhibitor neither increased nor decreased the inhibitors potency ( Esler et al . , 2004 ) , suggesting a fourth S’ pocket doesn’t contribute to γ-secretase-inhibitor or -substrate interactions . To support our proposed model , we exploited the fact that the second of the three S’ pockets is apparently too small to readily accommodate an aromatic amino acid . We generated dozens of substrates containing aromatic amino acid substitutions in the P2’ positions at each cleavage site along the Aβ40 or Aβ42 pathways , selectively blocking each individual cleavage event . Without exception we were able to predict the shift in the Aβ42/40 ratio . This was accomplished both in vitro with purified γ-secretase and recombinant C100 substrate as well as in a cell-based assay , transiently transfecting mutant full-length APP in HEK cells and measuring secreted Aβ . The same predicted Aβ42/40 ratio changes were observed in vitro and in the cell-based assay whether we measured Aβ ( 6E10 detection antibody ) or Aβ plus p3 products ( 4G8 detection antibody ) by ELISA . Together this demonstrates we are probing the fundamental mechanism by which γ-secretase cleaves APP , irrespective of mutant effects on cellular localization , γ-secretase’s interaction with activity-modulating proteins/lipids within the cell or any artifacts that may arise from more artificial in vitro assays . Although we are unable to measure every cleavage product from the ~70 mutant forms of APP we generated , we do note that every time a Phe was placed in the P2’ position of a cleavage product that we could readily and directly measure , there were almost negligible amounts of that product formed . For example , in the cell-based assay , V40F , A42F and V44F generated levels of Aβ38 , Aβ40 and Aβ42 , respectively , that were actually less than mock transfected levels ( Figure 7—figure supplement 3 ) . These Aβ levels are orders of magnitude less than that from WT APP transfected cells , although we cannot say whether these low levels of Aβ species were produced from endogenous HEK cell APP or from our transfected mutants . Similarly , we were unable to detect by MS any AICD products containing a Phe in the P2’ position . Together , these data suggest that aromatic amino acids may be completely excluded from the S2’ pocket and further demonstrates that substrate occupancy of the three S’ pockets is an absolute requirement for catalysis . It is likely that phenyalanine substitutions at various positions along the transmembrane domain of APP influence the general structure and/or helical stability of the substrate . This could affect the manner in which these substrates interact with γ-secretase . Given that we are able to accurately predict the Aβ42/40 ratio without exception for the dozens of mutants used in this study , we expect that cleavage preferences dictated by the presence of aromatic amino acids in the P2’ position are overriding any affect these mutations have on substrate helical structure/stability and any altered manner in which these mutant substrates initially interact with the enzyme . This is directly supported in Figure 3 where V44F , I45F , I47F and T48F all shift initial ε cleavage in favor of the opposite pathway relative to the final cleavage products measured and originally predicted by our model . We present several lines of evidence suggesting that substrate movement and/or a substrate conformational change after initial enzyme binding is an important step in γ-secretase’s catalytic mechanism . We identify that like transition-state analogs , tripeptide cleavage products are noncompetitive inhibitors of γ-secretase , albeit very weak inhibitors . By definition this means the binding sites on γ-secretase for initial substrate binding and subsequent catalysis are spatially separate , requiring substrate movement after initial binding to be an integral part of γ-secretase’s catalytic mechanism . For rhomboid , a recent study demonstrates that product-mimicking peptide aldehydes are non-competitive inhibitors of this serine intramembrane protease ( Cho et al . , 2016 ) , exactly like tripeptide products and transition-state analog inhibitors are for γ-secretase , suggesting a common two-step mechanism between these two intramembrane proteases . Additionally , the double Phe mutant V50F-M51F , which is predicted by our model to sterically clash with the three S’ pockets in the active site , is still efficiently bound to γ-secretase even though it is processed less efficiently than WT . Given that this mutant effectively competes for γ-secretase processing of other substrates , it must bind to the initial docking site on the enzyme for substrate . This suggests that binding to the S’ pockets is the final step in substrate recognition and positioning within the enzyme prior to catalysis . Furthermore , we are unable to prevent the formation of Aβ38 by specifically blocking the production of the known Aβ38 precursors Aβ42 and Aβ43; instead , paradoxically the V44F-I45F double mutant increases Aβ38 production . It is likely that the helix-destabilizing Gly-Gly motif at G37 and G38 is the reason for this observation . Local helical unwinding around this position probably makes the amide bond between G38 and V39 particularly accessible for cleavage , and this may be the reason for γ-secretase’s normal deviation from the preferred tripeptide cleavage for Aβ38 production . However , further investigation , including quantification of the intramembrane peptide products by mass spectrometry , will be required to prove this . In the absence of a γ-secretase—substrate co-complex structure , it will not be possible to definitively prove the existence of a partially unwound substrate intermediate . Conceivably , it should be possible to capture such an intermediate with a transition-state analog covalently linked to the C-terminus of an APP- or notch-based helical substrate . Given that tripeptide cleavage is dictated by the three S’ pockets in the active site of presenilin , we expect that other γ-secretase substrates will be similarly cleaved preferentially in increments of three amino acids , while skipping aromatic amino acids that fall in the S2’ pocket along the way , and with cleavage occurring preferentially in helix-destabilized regions . This is important to note given that many of γ-secretase’s substrates naturally contain aromatic amino acids . The TMD of notch , for example , naturally contains three phenylalanines . We know from a previous MS study that these phenylalanines are skipped by γ-secretase in a pattern consistent with our model ( Figure 7D ) ( Okochi et al . , 2002 ) . In a recent study , we have demonstrated that substrates with large ectodomains have a reduced binding affinity for γ-secretase due to steric clashing with the nicastrin component of the γ-secretase complex ( Bolduc et al . , 2016 ) . Based on this , and results from our current study , we might expect that γ-secretase can chose its substrates through a complex interplay between ectodomain length , helical TMD stability and the sequence of amino acids ( specifically aromatic amino acids ) within substrate TMD . There may even exist non-substrates containing short ectodomains but having stable helices that are further protected from cleavage by sequential stretches of aromatic amino acids . Whether such non-substrates exist , and the relationship between ectodomain size , helical stability and amino acid sequence will require further investigation . We show that the majority of FAD mutants within the TMD of APP primarily increase the Aβ42/40 ratio by changing ε cleavage , favoring the Aβ48 → 42 pathway . This is mostly in agreement with previous studies ( Chávez-Gutiérrez et al . , 2012; Chen et al . , 2014; Dimitrov et al . , 2013; Quintero-Monzon et al . , 2011 ) . However , we also identify a new mechanism by which certain FAD mutations can increase the production of pathogenic Aβ species . Here , final γ cleavages are uncoupled from initial pathway preference determined by ε cleavage . In the case of the I45F FAD mutation , the bulky Phe sterically clashes with the S2’ pocket of γ-secretase at the Aβ43 cleavage site , blocking its cleavage and the subsequent production of Aβ40 . It is likely that the positioning of the Phe in the S3’ pocket of the Aβ42 cleavage also enhances γ-secretase proteolysis at this position through a favorable S3’-P3’ interaction . These two interactions likely combine to account for the fact that no other mutation within the TMD of APP produces as much Aβ42 as I45F . This is the most severe APP FAD mutation , with an onset of clinical AD at 31 years of age ( Guerreiro et al . , 2010 ) . There are at least two additional FAD mutations , I45T and T48P , that appear to dissociate the ε and γ cleavages . How these mutations accomplish this is currently unknown . Our data also help explain previous observations in the literature . Prior to the identification of presenilin as being the protease responsible for γ-secretase activity ( Wolfe et al . , 1999 ) , Lichtenthaler et al . performed a phenylalanine scanning study of the TMD of APP . In their study , they observed the exact same Aβ42/40 shifts identified here for several of the same mutants ( Lichtenthaler et al . , 1999 ) . Later , Sato et al . found that γ-secretase was unable to cleave through stretches of 3–5 consecutive tryptophans inserted into the TMD of APP ( Sato et al . , 2005 ) . These observations are now explained by the determination that the small S2’ pocket of γ-secretase cannot accommodate aromatic amino acids . With the identification of the key role the three S’ pockets play in γ-secretase’s cleavage mechanism , several important new questions are raised . The exact locations of the three S’ pockets in presenilin are currently unknown . Additionally , the identity of the initial substrate-binding site on presenilin is unknown . The resolution of enzyme—substrate and/or -inhibitor co-complexes by cryo-EM will be informative in this regard . All enzymatic assays in this study utilized purified γ-secretase complex containing presenilin-1 . It will be interesting to see if presenilin-2 has similar substrate cleavage preferences . A major unsolved question pertains to the mechanism by which substrate TMD is repositioned within γ-secretase after each cleavage event in order to be close enough to the active site for the next round of catalysis to occur . Is this a ratcheting or sliding motion ? Is this an active process or based on Brownian motion ? At present , we do not know how γ-secretase is capable of skipping stretches of several phenylalanines in a row . Determining how γ-secretase accomplishes this may help elucidate how substrate moves during normal sequential cleavage . Now that we have identified several key aspects of γ-secretase’s substrate recognition and cleavage mechanisms , as well as provide valuable new tools for future structural and biochemical studies , we should be able design experiments to elucidate some of γ-secretase’s remaining unanswered functional questions . The answers from which should have further broad implications for our understanding of Alzheimer’s disease and the development of safe and effective therapeutics targeting γ-secretase function . The following antibodies were used: α-Myc ( 9E10 Santa Cruz #sc-40 ) , α-Flag ( M2 Sigma #F3165 ) , α-Nct ( Cell Signaling 3632S ) , α-GAPDH ( Cell Signaling Technology , ab125247 ) , α-APP ( C7 ) , α-AICD 50–99 ( Rb ) was a kind gift from Philip Szekeres at Eli Lilly , α-Ms 800nm ( Licor Bio 926–32212 ) and α-Rb 680nm ( Licor Bio 926–68021 ) . Tripeptides were synthesized by Anaspec corp . Total brain lipid extract was from Avanti Polar Lipids ( #131101 ) . The following Aβ ELISA kits were used: 4G8 ( Meso Scale Diagnostics , K15199E ) or 6E10 ( Meso Scale Diagnostics , K15200E ) for cell-based assays; Aβ40 ( #KHB3482 ) and Aβ42 ( #KHB2442 ) ELISA kits from Invitrogen for in vitro assays . All mutant forms of C100 FLAG or full-length APP were generated by site directed mutagenesis of either C100-FLAG in pET22b or full-length WT APP in the pCMV695 plasmid . Adherent HEK cells were cultured in complete growth media: Dulbecco’s Modified Eagle’s Medium ( DMEM ) supplemented with 10% fetal bovine serum ( FBS ) , 2 mM L-glutamine , 10 Units/mL penicillin , and 10 µg/mL streptomycin . For transfection , adherent HEK cells were seeded in six-well dishes at a density of 5x105 cells per well . Transfection was carried out with Lipofectamine 3000 reagent in serum-free conditions with Opti-MEM I . Cells were incubated for 24 hr , at which time conditioned media was harvested for ELISA and cells were harvested for western blot . Suspension HEK cells were cultured in 100 mL of unsupplemented Freestyle 293 media ( Life Technologies , 12338-018 ) with shaking at 125 rpm , and passaged at a density of 2 × 106 cells/mL . For transfection , suspension HEK cultures were grown to a density of 2 × 106 cells/mL . Media was replaced with fresh Freestyle 293 media . 5 mL Freestyle 293 , 150 µg of γ-secretase vector containing presenilin 1 ( provided by Yigong Shi ) , and 450 µg of 25-kDa linear polyethylenimines ( PEI ) was mixed and incubated for 30 min at room temperature . The DNA/PEI solution was then added to the HEK culture and cells were grown for ~60 hrs prior to harvesting . γ-secretase was purified as previously described ( Fraering et al . , 2004; Osenkowski et al . , 2009 ) C100-FLAG substrates were expressed in BL21 E . coli for 3 hrs at 37°C after induction with 1 mM IPTG . Cells were then pelleted and lysed by French press in 50 mM HEPES pH 7 . 0 , 1% Triton X-100 detergent . FLAG-tagged substrates were then isolated by immunoprecipitation for 3 hrs at 4°C with anti-FLAG M2 beads from Sigma . Substrates were then eluted from the beads with 100 mM glycine pH 2 . 5 , 0 . 25% NP40 prior to being neutralized with tris buffer and stored at -80°C . Purified γ-secretase was incorporated into vesicles by first dissolving total brain lipid extract ( 1 . 25 mM final ) in 50 mM HEPES pH 7 . 0 , 150 mM NaCl , 0 . 25% CHAPSO . γ-Secretase ( 5–30 nM final concentration ) was then added to the solution and detergent removed by mixing SM-2 biobeads ( 62 mg/mL ) ( Bio-Rad ) with the lipid/detergent/enzyme solution for two hrs at 4°C . Biobeads were removed from the newly formed proteoliposomes and reactions were initiated with the addition of purified recombinant substrate C100-FLAG substrate . Reactions were quenched with SDS loading dye for western blot or centrifuged for ELISA or mass spectrometry on Aβ or AICD fragments , respectively . For inhibition studies , tripeptide fragments or inhibitors were dissolved in DMSO prior to being diluted into the reaction buffer . The concentration of C100-FLAG used in all in vitro assays was 500 nM unless otherwise stated . Conditioned media from HEK cells transfected with WT or mutant APP was assayed for Aβ by 4G8 or 6E10 Aβ ELISA kits from Meso Scale Diagnostics . Aβ levels were measured by both 6E10 and 4G8 ELISA for each data point , yielding nearly identical results . In vitro assay Aβ was measured using Aβ40 and Aβ42 ELISA kits from Invitrogen . All ELISAs were performed according to the manufacture’s protocols . Purified γ-secretase ( 5 nM final concentration ) was preincubated in assay buffer ( 50 mM HEPES pH 7 . 0 , 150 mM NaCl , 0 . 25% CHAPSO , 0 . 1% DOPC and 0 . 025% DOPE , 2% BSA ) in the presence of 2 μM III-31C for 1 hr at room temperature . Purified WT or mutant C100-Myc ( 20 nM final concentration ) was then incubated with γ-secretase for 1 hr prior to pull down with anti-HA magnetic affinity beads for 4 hr with mixing at room temperature . The immunoprecipitated complex was then washed three times and eluted with SDS loading buffer prior to western blot with an anti-Myc antibody . Tripeptide IC50 inhibition was fit to:vivo=1/ ( 1+[ I ]IC50 ) where vi is the initial velocity in the presence of inhibitor at concentration [I] and vo is the initial velocity in the absence of inhibitor . Tripeptide inhibition was globally fit to the following noncompetitive equation:v= Vmax[S][ S ] ( 1+[ I ]Kii ) +Km ( 1+[ I ]Ki ) where , v is the initial rate , Ki is dissociation constant for inhibitor binding to free enzyme , Kii is the dissociation constant for inhibitor binding to the enzyme-substrate complex . Inhibitor cross-competition was globally fit to:1vij=1/v0 ( 1+[ I ]Ki+[ J ]Kj+[ I ][ J ]αKiKj ) where , vij is the initial rate in the presence of inhibitors , v0 is the initial rate in the absence of inhibitor , Ki and Kj are the dissociation constants for inhibitors I and J , respectively . α = ∞ for inhibitors which bind in a mutually exclusive fashion , while α = 1 for inhibitors which have distinct binding sites . Conditioned media was collected from transiently-transfected adherent HEK cells . Each well of an uncoated 96-well multi-array plate ( Meso Scale Discovery , #L15XA-3 ) was coated with 30 µL of a PBS solution containing 3 µg/mL of 266 capture antibody ( Elan ) , and incubated at room temperature overnight . A detection antibody solution was prepared with 3D6B detection antibody ( Elan ) , 100 ng/mL Streptavidin Sulfo-TAG ( Meso Scale Discovery , #R32AD-5 ) , and 1% MSD Blocker A ( #R93BA-4 ) in wash buffer ( #R61TX-1 ) . Following overnight incubation , 25 µL/well of sample , followed by 25 µL/well of detection antibody solution were incubated for 2 hr at room temperature with shaking at >300 rpm , washing wells with wash buffer between incubations . Plate was read and analyzed according to manufacturer protocol . Following an in vitro proteoliposome activity assay , AICD-FLAG products were isolated by immunoprecipitiation with anti-FLAG M2 magnetic beads from Sigma . Completed reactions were incubated with 50 µL of M2 beads in 10 mM MES pH 6 . 5 , 10 mM NaCl , 0 . 05% DDM detergent in 500 µL volumes overnight at 4°C . AICD was then eluted from the beads with acetonitrile:water ( 1:1 ) with 0 . 1% trifluoroacetic acid . MALDI/TOF mass spectrometry was performed with sinapinic acid matrix on a calibrated ultraflextreme MALDI/TOF/TOF from Bruker in linear mode .

Individuals with Alzheimer’s disease generally have deposits known as “amyloid plaques” in the brain . These plaques are made up of a mixture of molecules called amyloid beta peptides that clump together and are thought to be a key cause of the disease . The amyloid beta peptides vary in size; the larger peptides tend to be more prone to forming clumps than the smaller ones and are thus more toxic to the brain . An enzyme called gamma-secretase makes amyloid beta peptides by cutting up a protein called APP . Proteins are made of chains of building blocks called amino acids and studies using a technique called mass spectrometry show that gamma-secretase cuts APP in segments of three amino acids at a time . The size of the amyloid beta peptides produced is determined by the positions in APP that gamma-secretase selects to cut . Therefore , understanding how the enzyme works could provide new opportunities for developing drugs to treat Alzheimer’s disease . Here , Bolduc et al . found that the human gamma-secretase enzyme has sites that amino acids in APP can bind to that help to guide the enzyme to cut APP by three amino acids at a time . These binding sites control where the enzyme cuts APP and therefore determines which amyloid peptides are produced . Previous studies have linked several naturally occurring mutations in the gene encoding APP to inherited forms of Alzheimer’s disease . Bolduc et al . now reveal that several of these mutations affect the places that gamma-secretase cuts APP to produce amyloid peptides . These findings may be helpful for developing drugs that could manipulate gamma-secretase to produce smaller , less harmful amyloid peptides . Gamma-secretase can cut many other proteins , and so a future challenge will be to find out if the enzyme cuts these other proteins in the same way that it cuts APP .

2016

The amyloid-beta forming tripeptide cleavage mechanism of γ-secretase

In the hippocampus , the inhibitory neurotransmitter GABA shapes the activity of the output pyramidal neurons and plays important role in cognition . Most of its inhibitory effects are mediated by signaling from GABAB receptor to the G protein-gated Inwardly-rectifying K+ ( GIRK ) channels . Here , we show that RGS7 , in cooperation with its binding partner R7BP , regulates GABABR-GIRK signaling in hippocampal pyramidal neurons . Deletion of RGS7 in mice dramatically sensitizes GIRK responses to GABAB receptor stimulation and markedly slows channel deactivation kinetics . Enhanced activity of this signaling pathway leads to decreased neuronal excitability and selective disruption of inhibitory forms of synaptic plasticity . As a result , mice lacking RGS7 exhibit deficits in learning and memory . We further report that RGS7 is selectively modulated by its membrane anchoring subunit R7BP , which sets the dynamic range of GIRK responses . Together , these results demonstrate a novel role of RGS7 in hippocampal synaptic plasticity and memory formation . Signaling through G protein-coupled receptors for the inhibitory neurotransmitter GABA ( GABABR ) has been recognized to play key roles in mood , nociception , memory , reward , and movement ( Bowery , 2006; Padgett and Slesinger , 2010 ) . In the hippocampus , activation of postsynaptic GABABR on pyramidal neurons produces slow inhibitory postsynaptic currents ( sIPSCs ) , which counteract the excitatory influence of ionotropic glutamate receptors to shape neuronal output ( Ulrich and Bettler , 2007; Luscher and Slesinger , 2010 ) . As a result , GABABR signaling profoundly affects hippocampal synaptic plasticity and has marked effects on memory formation ( Davies et al . , 1991; Wagner and Alger , 1995; Schuler et al . , 2001 ) . A large share of the postsynaptic inhibitory effect of GABABR stimulation in the hippocampus is mediated by activation G protein-gated inwardly-rectifying K+ ( GIRK/Kir3 ) channels , which inhibit neuronal excitability via hyperpolarizing K+ efflux ( Luscher and Slesinger , 2010 ) . In the hippocampus , GIRK channels are predominantly formed by GIRK1 and GIRK2 subunits , which co-localize and may interact directly with GABABR protomers ( Koyrakh et al . , 2005; Fajardo-Serrano et al . , 2013 ) . Activation of GABABR releases G protein βγ subunits , which bind to GIRK channels and increase channel gating ( Padgett and Slesinger , 2010 ) . Blockade of GABABR or GIRK channels by either pharmacological manipulations or genetic knockout ablates the slow IPSC , and blunts a form of hippocampal synaptic plasticity known as depotentiation ( Luscher et al . , 1997; Chung et al . , 2009 ) . Conversely , enhanced GABABR-GIRK signaling seen in a mouse model of Down syndrome disrupts both excitatory and inhibitory synaptic plasticity , and is linked to cognitive impairment ( Kleschevnikov et al . , 2004; Cramer et al . , 2010; Cooper et al . , 2012 ) . GABABR-GIRK signaling is negatively modulated by the Regulators of G protein Signaling ( RGS ) proteins , which accelerate G protein inactivation ( Hollinger and Hepler , 2002; Padgett and Slesinger , 2010 ) . Among more than 30 RGS genes found in mammalian genomes , the R7 family of RGS proteins ( R7 RGS ) stands out for its prominent roles in a range of fundamental neuronal processes , from vision to motor control to reward-related behavior ( Anderson et al . , 2009 ) . The four members of this group ( RGS6 , RGS7 , RGS9 and RGS11 ) form heterotrimers with two subunits ( Gβ5 and R7BP ) , and these interactions regulate the localization and/or expression of the complexes ( Anderson et al . , 2009; Jayaraman et al . , 2009 ) . Previous studies have shown that Gβ5 serves as a central scaffold that bridges the catalytic ( RGS ) and targeting ( R7BP ) subunits and ensures the stability of R7 RGS protein ( Cheever et al . , 2008; Sandiford et al . , 2010; Masuho et al . , 2011 ) . Elimination of Gβ5 also resulted in dramatic slowing of GIRK channel deactivation kinetics , prolongation of synaptically-evoked slow IPSCs in hippocampal pyramidal neurons , and increased behavioral sensitivity to GABABR stimulation ( Xie et al . , 2010 ) . However , the identity of the RGS isoform that modulates GABABR-GIRK signaling in hippocampus , as well as the relative impact of the R7BP subunit , are unknown . Furthermore , the relevance of RGS-dependent modulation of GIRK-dependent signaling to hippocampal circuit function , plasticity , and behavior remain unclear . In this study , we examined the importance of RGS and R7BP subunits to hippocampal physiology and hippocampal-dependent behavior . We report that ablation of Rgs7 results in alterations of GABABR-GIRK signaling , disrupts synaptic plasticity in the hippocampus , and impairs contextual learning and memory . Moreover , the function of the RGS7/Gβ5 complex is fine-tuned by R7BP , which sets the sensitivity range of GABABR-GIRK signaling . We began by characterizing the expression of RGS complex subunits in the mouse hippocampus . We detected robust expression of RGS6 , RGS7 , R7BP and Gβ5 by western blotting ( Figure 1A ) . To understand the contribution of individual subunits to GABABR-GIRK signaling , we studied the effects of selective knockout of Rgs6 , Rgs7 , and R7bp in mice . Elimination of RGS7 dramatically reduced levels of Gβ5 and R7BP in the hippocampus ( Figure 1A , B ) . In contrast , elimination of RGS6 had no significant effect on the expression of R7BP or Gβ5 . Similarly , loss of one RGS protein did not affect the expression of the other , or the GIRK channel subunit GIRK2 . Given the interdependence of subunit expression in R7 RGS complexes ( Chen et al . , 2003; Anderson et al . , 2007; Grabowska et al . , 2008 ) , these results suggested that RGS7 was likely the dominant catalytic subunit in the hippocampus . 10 . 7554/eLife . 02053 . 003Figure 1 . RGS7 and R7BP modulate GABABR-GIRK signaling in cultured hippocampal neurons . ( A ) Western blot analysis of protein expression in hippocampi extracted from wild-type ( WT ) mice , or mice lacking RGS7 ( Rgs7−/− ) , RGS6 ( Rgs6−/− ) or R7BP ( R7bp−/− ) . ( B ) Quantification of Western blotting data , with protein levels arrayed as a function of genotype , ***p<0 . 01 , One-Way ANOVA n = 3 mice . ND-undetectable . ( C ) Representative normalized traces of GIRK currents evoked by a saturating concentration of the GABABR agonist baclofen ( 100 µM ) . ( D–E ) Rgs6 , Rgs7 , and R7bp knockouts do not affect GIRK current amplitudes ( D ) , desensitization ( E ) and current onset kinetics ( F ) evoked by 100 μM Baclofen . p>0 . 05 , One-Way ANOVA , n = 8–29 cells for each genotype . ( G ) Current deactivation rate following removal of baclofen was slower in neurons from Rgs7−/− and R7bp−/− mice as compared to WT controls , *p<0 . 05 and **p<0 . 01 vs WT , †††p<0 . 001 Rgs7−/− vs R7bp−/− . One-Way ANOVA , Bonferroni’s post hoc test , n = 8–29 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 02053 . 003 We next compared GABABR-GIRK responses in hippocampal pyramidal neurons from Rgs6 , Rgs7 , and R7bp knockout ( −/− ) and wild-type ( WT ) mice ( Figure 1C–F ) . Application of a saturating concentration ( 100 μM ) of the GABABR agonist baclofen elicited currents with similar maximal amplitudes in neurons of all genotypes ( Figure 1D ) . We also measured the activation and deactivation kinetics of the baclofen-induced currents . At this saturating concentration of baclofen , we observed no significant differences between genotypes in the activation phase of the response ( Figure 1C , F ) . Desensitization of current during the timeframe of agonist application was negligible and similar across genotypes , suggesting that the response at the steady-state is not compounded by the GIRK channel inactivation ( Figure 1C , E ) . While no change in the current deactivation kinetics was observed in Rgs6−/− neurons , elimination of RGS7 markedly slowed response deactivation ( Figure 1C , G ) . The rate of GIRK current deactivation in R7bp–/– neurons was also slower than in wild-type neurons , but the effect was substantially smaller than seen in neurons from Rgs7−/− mice ( Figure 1C , G ) . Importantly , there were no detectable changes in any of the measured response parameters in Rgs6−/− neurons as compared to wild-type , arguing that GABABR-GIRK signaling in hippocampal neurons is modulated by RGS7 and R7BP , but not RGS6 . Given the observed changes in the kinetics of GIRK channel modulation by GABABR , we next sought to determine the impact of ablating Rgs7 and R7bp on the sensitivity of the GIRK response to GABABR activation . Increasing baclofen concentrations caused a progressive enhancement in GIRK-mediated currents in all genotypes ( Figure 2A ) . However , pronounced differences in the concentration-response relationship were evident ( Figure 2B ) . First , in both Rgs7−/− and R7bp−/− neurons , curves were shifted to the left relative to wild-type neurons , indicating that ablation of RGS7 or R7BP increased GABABR-GIRK coupling sensitivity . Second , while RGS7 ablation resulted in a largely parallel leftward shift of the curve , its shape in R7bp−/− neurons was markedly steeper . Indeed , response amplitudes resembled those seen in wild-type neurons at lower agonist concentrations , but at higher concentrations , the responses more resembled those seen in Rgs7−/− neurons . 10 . 7554/eLife . 02053 . 004Figure 2 . Timing and sensitivity of GABABR-GIRK signaling is differentially controlled by RGS7 and R7BP . ( A ) Representative traces of GIRK currents evoked by increasing concentrations of baclofen . ( B ) Dose-response curves fit by Hill equation . EC50 values are 1 . 48–2 . 14 μM for WT , 0 . 35–0 . 58 μM for Rgs7−/− , and 0 . 65–0 . 81 μM for R7bp−/− ( 95% CI ) . Hill coefficients were 1 . 48 ± 0 . 17 in WT , 1 . 23 ± 0 . 21 in Rgs7−/− , and 2 . 04 ± 0 . 28 in R7bp−/− . p<0 . 0001 for difference in EC50 or each curve; p=0 . 04 for difference in Hill coefficients , Extra sum-of-squares F test , n = 10–22 cells . ( C ) Rgs7−/− cultured hippocampal neurons show slower current activation at lower concentrations of baclofen . ***p<0 . 001 vs WT , †††p<0 . 001 for Rgs7−/− vs R7bp−/− , Two-Way ANOVA , Bonferroni’s post hoc test , n = 10–22 cells . ( D ) Dependence of GIRK current deactivation kinetics on agonist concentration . *p<0 . 05 and ***p<0 . 001 vs WT; ††p<0 . 01 , †††p<0 . 001 for Rgs7−/− vs R7bp−/− , Two-Way ANOVA , Bonferroni’s post hoc test , n = 10–22 cells . ( E–G ) Differences in the lag times before the onset of response deactivation ( Lag Deactivation ) . ( E ) Representative traces of currents evoked by 100 µM baclofen show Lag time measurement as time between the onset of agonist removal and the point at which 10% of the current deactivated: 1560 , 4900 and 3000 ms for WT , Rgs7−/− , R7bp−/− , respectively . ( F ) Semi-log plot of lag time dependence on concentration . The data were fitted with a linear regression , R2 = 0 . 66 , 0 . 67 , and 0 . 84; and slopes 619 ± 69 , 1440 ± 151 , and 952 ± 84 for WT , Rgs7−/− , R7bp−/− , correspondingly . p<0 . 0001 for differences in slopes , two way ANOVA , n = 10–22 cells . ( G ) Comparison of lag times at baclofen concentrations that generated equivalent responses . Lag times were significantly different in genotypes at saturating ( EC90 ) but not submaximal ( EC60 ) concentrations , *p<0 . 05 and ***p<0 . 001 vs WT , ††p<0 . 01 Rgs7−/− vs R7bp−/− , Two-Way ANOVA , Bonferroni’s post hoc test , n = 10–22 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 02053 . 004 We next tested whether GABABR-GIRK current kinetics exhibited a similar dependence on agonist concentration . Comparing activation rates across different concentrations of baclofen revealed that onset kinetics were affected by RGS7 ablation only at low baclofen concentrations , whereas no effect of R7BP elimination was seen at any level of GABABR stimulation ( Figure 2C ) . Since GIRK activation kinetics could be influenced by the G protein deactivation cycle ( Doupnik et al . , 1997; Lambert et al . , 2010 ) , we did not put significant emphasis on the analysis of the differences in the raising phases of the response . The impact of RGS7 ablation on deactivation rates was consistent at different agonist concentrations ( Figure 2D ) . In contrast , R7BP ablation significantly affected deactivation kinetics only at higher agonist concentrations , consistent with the effects on response sensitivity ( Figure 2D ) . Analysis of the responses also revealed that elimination of RGS7 and R7BP resulted in a significant delay between agonist removal and the beginning of current deactivation ( Figure 2E ) . This lag time showed an exponential dependence on agonist concentration that continued to develop past the saturation point for the maximal GIRK response ( ∼10 μM baclofen ) in all genotypes ( Figure 2F ) . While there was no difference in lag time between genotypes at non-saturating baclofen concentrations ( ∼EC60 ) , it became pronounced as the current response reached saturation ( ∼EC90 ) ( Figure 2G ) . Combined with the observations that GABABR continues to increase the amount of activated G proteins past the saturation point of the GIRK channel response ( Hensler et al . , 2012 ) , these data suggest that the lag time reflects the clearance of free βγ subunit produced above the stoichiometric level relative to the GIRK channel . To provide an independent evidence that increase in the lag time reflects changes in stoichiometry of Gβγ subunits relative to their effector molecule , we utilized a bioluminescence resonance energy transfer ( BRET ) -based approach that monitors interactions of Gβγ with a reporter derived from an effector ( GRK3 ) upon reconstitution in transfected cells ( Figure 3A ) . In this assay , we induced production of free Gβγ subunits via GABABR activation and then measured delay time between antagonizing GABABR and the onset of signal decay , while changing the Gβγ to GRK3 effector ratio ( Figure 3B ) . In agreement with the electrophysiological recordings of GIRK channel activity , BRET experiments showed that the increase in Gβγ stoichiometry over an effector results in prolongation of the response deactivation lag ( Figure 3C , D ) . 10 . 7554/eLife . 02053 . 005Figure 3 . Changing the ratio of Gβγ to an effector affects response deactivation lag in a reconstituted system . ( A ) Schematic representation of the assay principle . The assay measures interaction of YFP-tagged Gβγ with its effector reporter GRK , tagged with N-luc , producing BRET signal . Gβγ subunits are released upon GABABR stimulation with GABA increasing BRET signal . Inactivation of GABABR with an antagonist CGP 54626 results in dissociation of Gβγ from GRK and re-association with Gαo to form inactive heterotrimer . ( B ) Time course of changes in BRET signal upon stimulation of cells with GABA and subsequent deactivation by CGP 54626 . Cells were transfected with varying amounts of constructs encoding Gβγ and GRK reporter ( from black to green ) . ( C ) Deactivation phase of the response showing kinetics of signal decay . The lag deactivation time ( dotted line for the trace in black ) is defined as the time that it takes to quench the BRET signal by 10% from its steady state value in the presence of an agonist . ( D ) Quantification of a lag deactivation time as a function of Gβγ/GRK ratio . Error bars are SEM values , n = 4 per condition . DOI: http://dx . doi . org/10 . 7554/eLife . 02053 . 005 In transfected cells , R7BP is essential for the membrane localization of RGS7 ( Drenan et al . , 2005; Narayanan et al . , 2007 ) . Furthermore , we previously reported that knockout of R7bp resulted in a reduction of the total membrane-bound RGS7 protein in hippocampal tissue ( Panicker et al . , 2010 ) . To analyze the effect of R7BP on RGS7 localization in hippocampal pyramidal neurons , we performed high-resolution immunoelectron microscopy . Consistent with the earlier findings , immunoparticles for RGS7 were abundant on the extrasynaptic plasma membrane of dendritic spines and dendritic shafts of pyramidal cells , as well as at intracellular sites ( Fajardo-Serrano et al . , 2013 ) . In the hippocampus of R7bp−/− animals , immunoparticles for RGS7 were distributed similarly to the wild-type , but they were more frequently observed just beneath the plasma membrane and more broadly distributed in somata ( Figure 4 ) . Indeed , quantitative analysis indicated that in R7bp−/− mice , RGS7 was less frequently detected in the plasma membrane and tended to accumulate within 100 nm of the plasma membrane . We also detected a significant increase in RGS7 labelling in the rough endoplasmic reticulum ( rER ) in the soma of R7bp−/− pyramidal neurons ( 2489 vs 3371 immunoparticles in wild-type and R7bp−/− neurons , respectively ) . These findings indicate that knockout of R7BP caused a modest but significant redistribution of RGS7 away from the plasma membrane . 10 . 7554/eLife . 02053 . 006Figure 4 . Change in subcellular localization of RGS7 in the hippocampus of the R7bp−/− mice . Electron micrographs of the stratum radiatum of the hippocampal CA1 region showing immunoparticles for RGS7 , as detected using a pre-embedding immunogold method . Dendritic spines ( s ) and axon terminals ( at ) are marked . Arrows indicate locations of immunoparticles at the plasma membrane , while arrowheads identify RGS7 immunoparticles found just below the membrane . Quantitative analysis showed that RGS7 is less frequently detected in the plasma membrane , and accumulates within the first 100 nm from the plasma membrane , samples from R7bp−/− mice , *p<0 . 05 , One-way ANOVA followed by the Bonferroni’s post hoc test , n = 3 mice . Scale bar: 0 . 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 02053 . 006 The slower kinetics and increased sensitivity of GIRK channels to GABABR activation suggested that there was a net up-regulation of GIRK-dependent inhibition in hippocampal neurons lacking RGS7 . Since GIRK channels significantly contribute to setting neuronal excitability ( Chen and Johnston , 2005 ) , we next determined how deletion of RGS7 influences the excitability of CA1 pyramidal neurons . To characterize intrinsic electrophysiological properties of different genotypes , the responses of CA1 neurons to somatic current injections ranging from −150 pA to +300 pA were measured ( Figure 5A ) . While membrane resistance was similar in both genotypes ( Rin = 120 ± 10 MΩ vs 105 ± 11 MΩ , p=0 . 31 , in wild-type and Rgs7−/− cells , respectively ) , resting membrane potential ( RMP ) was significantly hyperpolarized in Rgs7−/− neurons ( −64 . 7 ± 0 . 5 mV vs −68 . 5 ± 0 . 7 mV in wild-type and Rgs7−/− , respectively; Figure 5B ) . In addition , the current required to elicit action potentials ( APs ) was significantly higher in RGS7−/− neurons ( I = 104 . 3 ± 8 . 8 pA vs 146 . 4 ± 17 . 0 , in WT and Rgs7−/− cells , respectively; Figure 5C ) . Finally , Rgs7−/− neurons fired significantly fewer action potentials in response to depolarizing current as compared to wild-type controls ( Figure 5D ) . Collectively , these observations argued that hippocampal pyramidal neurons from Rgs7−/− mice were less excitable than wild-type counterparts . 10 . 7554/eLife . 02053 . 007Figure 5 . Altered intrinsic excitability and excitatory transmission in CA1 hippocampal neurons from Rgs7−/− mice . ( A ) Representative traces of responses elicited by current injections of −150 , −50 , 0 , +50 , and +300 pA in WT and Rgs7−/− CA1 neurons . ( B ) Hyperpolarized resting membrane potential ( RMP ) in Rgs7−/− neurons , ***p=0 . 0001 , t test , n = 14–23 cells . ( C ) Current required to evoke an action potential ( firing threshold ) is higher in Rgs7−/− neurons , *p=0 . 02 , t test , n = 14–23 cells . ( D ) Lower intrinsic excitability in Rgs7−/− neurons . **p<0 . 01 and *p<0 . 05 , two-way ANOVA with Bonferroni’s posttest , n = 14–23 cells . ( E ) Representative traces of slow excitatory synaptic currents ( sEPSCs ) in WT and Rgs7−/− cells . ( F ) Cumulative distribution and mean values for sEPSCs amplitudes ( 14 . 9 ± 1 . 9 vs 12 . 8 ± 0 . 9 pA in WT and Rgs7−/− correspondingly , p=0 . 3 , unpaired t test; n = 11 cells and 1100 events for each genotype ) . ( G ) Cumulative distribution and mean values for sEPSCs and frequencies ( 3 . 4 ± 0 . 6 vs 3 . 1 ± 0 . 6 Hz in WT and Rgs7−/− correspondingly , p=0 . 7 , unpaired t test , n = 11 cells and 1100 events for each genotype ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02053 . 007 We further examined whether decreased intrinsic excitability of Rgs7−/− neurons is caused by dysregulated net excitatory transmission . For this purpose , we measured spontaneous EPSCs ( sEPSC ) in CA1 pyramidal neurons ( Figure 5E–G ) . There were no significant differences in either amplitudes or frequencies of sEPSC events between Rgs7−/− and wild-type neurons ( Figure 5F–G ) . These observations indicate that excitatory input into CA1 pyramidal neurons is unchanged by elimination of RGS7 and suggest that the observed decrease in neuronal excitability is caused by changes in the intrinsic membrane properties of CA1 neurons , likely stemming from enhanced activity of the postsynaptic GIRK channel . Previous studies have implicated GIRK channels in several forms of hippocampal synaptic plasticity , including long-term potentiation ( LTP ) ( Cramer et al . , 2010 ) , long-term depression ( LTD ) ( Cooper et al . , 2012 ) and depotentiation ( DP ) ( Chung et al . , 2009; Cooper et al . , 2012 ) . Because GABAB receptors play key roles in these processes and since GABAB-GIRK signaling is severely dysregulated in Rgs7−/− neurons , we next examined the impact of RGS7 ablation on hippocampal synaptic plasticity . A high-frequency stimulation protocol of 2 tetanized stimuli ( TS , 100 Hz for 1s each ) produced robust LTP in both genotypes ( Figure 6A , D ) . The extent of the potentiation in Rgs7−/− slices was not significantly different from that measured in wild-type slices , arguing that elimination of RGS7 has no effect on LTP . A low-frequency stimulation ( LFS , 2 Hz for 10 min , 1200 pulses ) elicited LTD in both wild-type and Rgs7−/− hippocampal slices ( Figure 6B , D ) . However , the extent of inhibition was significantly smaller in Rgs7−/− slices . Interestingly , the magnitude of the first fEPSP after the train of LFS stimulation was significantly reduced in knockout mice ( 51 ± 6% and only 72 ± 6% in wild-type and Rgs7−/− slices correspondingly , *p=0 . 03 , t test , n = 6 ) , suggesting that the LTD impairment in Rgs7−/− is due to a deficit in an induction mechanism ( s ) . The depotentiation produced by consecutive application of TS and LFS was also significantly impaired in Rgs7−/− mice ( Figure 6C , D ) . 10 . 7554/eLife . 02053 . 008Figure 6 . RGS7 ablation disrupts hippocampal synaptic plasticity . Field EPSP ( fEPSP ) slope change following induction of: ( A ) LTP in Rgs7−/− , 166 ± 9% vs wild-type , 155 ± 6% slices; p=0 . 33 , t test , n = 6–10; ( B ) LTD in Rgs7−/− , 90 ± 2% vs wild-type , 81 ± 2%; *p=0 . 01 , t test , n = 6–7; ( C ) depotentiation ( DP ) in Rgs7−/− , 111 ± 3% vs wild-type , 128 ± 6%; *p=0 . 013 , t test , n = 6–11 . Insets show representative fEPSP traces at baseline and 1 hr following induction protocol in WT ( black and grey ) and Rgs7−/− ( red and pink ) slices . ( D ) Quantification of the EPSP slope change 55–60 min following induction of each form of plasticity after normalization to pre-induction baseline . ( E ) Paired Pulse Ratio ( PPR ) dependence on the inter stimulus interval for wild-type ( WT ) and Rgs7−/− . Significant inter-stimulus interval but not genotype effect was observed , p<0 . 0001 and p=0 . 5 , correspondingly . Two-way ANOVA , n = 4 slices per genotype . ( F–H ) Basal synaptic transmission properties in Rgs7−/− slices . Dependence of fEPSP slope ( F ) and FV amplitude ( G ) on stimulus intensity . ( H ) Linear regression plot of fEPSP slope dependence on FV amplitude . The data were fitted with a linear regression , R2 = 0 . 67 and 0 . 53; and slopes 1 . 9 ± 0 . 1 and 1 . 5 ± 0 . 2 for WT and Rgs7−/−correspondingly . p=0 . 1 for differences in slopes , two way ANOVA , n = 12–13 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 02053 . 008 We further investigated the impact of RGS7 ablation on paired-pulse facilitation ( PPF ) , a form of short-term plasticity that results from the enhancement of presynaptic vesicle release in response to two closely-spaced stimuli ( Zucker , 1989; Dobrunz and Stevens , 1997 ) . We found that fEPSP facilitation was similar between genotypes across the examined range of 10–1000 ms interpulse intervals ( Figure 6E ) , indicating that presynaptic mechanisms are not likely to be the cause of the observed alterations in synaptic plasticity . We next compared the dependence of fEPSP slopes ( Figure 6F ) and presynaptic fibre volley ( FV ) amplitudes ( Figure 6G ) on stimulus intensity . Analysis of the relationship between FV amplitudes and fEPSP slopes by linear fitting reveals their nearly perfect correspondence between wild-type and Rgs7−/− slices , indicating preservation of basal synaptic transmission properties ( Figure 6H ) . Together , these results suggest that RGS7 elimination selectively impaired LTD and DP forms of synaptic plasticity , likely by a post-synaptic mechanism . Changes in the intrinsic excitability of CA1 pyramidal neurons , together with deficits in hippocampal synaptic plasticity , suggested that RGS7 complexes play a role in spatial learning and memory . To test this possibility , we first studied the impact of RGS or R7BP ablation in contextual fear conditioning , a test that requires hippocampal processing for memory formation ( Maren , 2001 ) . We found significant deficits in context recognition 24-hr after training in Rgs7−/− mice , but not in Rgs6−/− or R7bp−/− mice ( Figure 7A–C ) . Importantly , we observed no difference in baseline freezing behavior before associative training between the genotypes . Furthermore , there were no significant differences between wild-type mice and mutant mice in amygdala-dependent cue recognition . 10 . 7554/eLife . 02053 . 009Figure 7 . RGS ablation affects hippocampal-dependent learning and memory . ( A–C ) Evaluation of mouse behavior in fear conditioning paradigm . ( A ) Rgs7−/− mice show deficits in hippocampal-dependent contextual , but not cue , memory . *p<0 . 05 , t test; n = 12 per genotype . ( B ) Normal contextual and cue memory of Rgs6−/− mice in the fear conditioning test as compared to wild-type ( WT ) littermates . n = 8 per genotype . ( C ) R7bp−/− mice showed the same contextual and cue memory in fear conditioning test as compared to WT littermates , n = 8 per genotype . ( D–F ) Evaluation of mouse behavior in Morris water maze . Rgs7−/− mice ( n = 13 ) and their WT littermates ( n = 12 ) were trained for 6 d with 4 trials/d with an inter-trial interval of approximately 15 min . Performance during the acquisition phase was monitored and plotted as average time ( D ) or success rate ( E ) to reach the hidden platform . Mice showed improvement with training . There was a significant effect of a genotype in both escape latencies ( *p<0 . 05 ) and success rates ( *p<0 . 05 ) using two-way ANOVA analysis . Post hoc comparison revealed a significant impairment of Rgs7−/− mice during the last two acquisition days ( *p<0 . 05 , Tukey’s test ) . ( F ) Results of a probe trial given 24 hr after 6 d of training . The latency to the first crossing of the former location of the platform and the total number of crossing are shown , *p<0 . 05 , t test . ( G ) Evaluation of mouse behavior in novel object recognition paradigm . Rgs7−/− mice showed significant impairment during the test trial with a novel object in comparison with wild-type littermates . *p<0 . 05 , t test; n = 12–13 . DOI: http://dx . doi . org/10 . 7554/eLife . 02053 . 009 To clarify the impairment of hippocampus dependent learning and memory in Rgs7−/− mice , we conducted other tests that rely on hippocampal function . In the Morris water maze test , Rgs7−/− mice showed delayed escape latencies and reduced success rates during acquisition trials ( Figure 7D , E ) . In the probe trial , when the platform was removed 24 hr after training , it took Rgs7−/− mice significantly longer to reach the target area where the platform was located during training as compared to their wild-type littermates ( Figure 7F ) . Mice lacking RGS7 also made fewer crossings over the target area during the probe trial . Importantly , despite a pronounced spatial impairment , Rgs7−/− mice performed normally in the visible platform version of water maze . In the novel object recognition test , wild-type littermates spent significantly more time exploring novel objects over familiar objects ( Figure 7G ) . However , Rgs7−/− mice spent approximately equal time exploring familiar and novel objects . Furthermore , there was a significant difference in time exploring novel objects between Rgs7−/− mice and their wild-type littermates . No genotype differences in object exploratory behavior were observed during the training day , when the mice were first introduced to identical objects . Taken together , these results indicate that elimination of RGS7 leads to a disruption of spatial learning and memory in mice . The results of this study , together with prior investigations , establish the RGS7/Gβ5/R7BP complex as an essential regulator of GABABR-GIRK signaling in the hippocampus . In hippocampal neurons , RGS7 is closely co-localized with both GABABR and GIRK2-containing channels ( Fajardo-Serrano et al . , 2013 ) . Furthermore , in transfected cells , RGS7 can directly interact with GIRK channel subunits as demonstrated by both co-immunoprecipitation and bioluminescence energy transfer approaches ( Xie et al . , 2010; Zhou et al . , 2012 ) . The complex may further involve GABABR as well as G protein subunits ( Kovoor and Lester , 2002; Fajardo-Serrano et al . , 2013 ) , supporting the contention that components of the entire pathway are scaffolded into a larger macromolecular assembly . The interaction between RGS7 with GIRK channels is mediated by the Gβ5 subunit , which mimics signal-transducing Gβγ subunits in binding to the channel ( Xie et al . , 2010; Zhou et al . , 2012 ) . Knockout of Gβ5 greatly slows GABABR-GIRK response deactivation rates and dramatically increases agonist sensitivity of the signaling pathway ( Xie et al . , 2010 ) . In addition to mediating interactions with the channel , Gβ5 is essential for the expression of all four R7 RGS proteins ( Chen et al . , 2003 ) . With several R7 RGS proteins expressed in hippocampus , the identity of the exact isoform that participates in GIRK channel regulation in the region , through complex formation with Gβ5 , remained unknown . The elimination of RGS7 , reported in this study , largely phenocopies the loss of Gβ5 , both in terms of kinetics and agonist sensitivity . This argues that in hippocampal pyramidal neurons , GIRK signaling is regulated by RGS7 and not by other R7 RGS proteins . Indeed , while RGS6 regulates GIRK signaling in cerebellar neurons ( Maity et al . , 2012 ) and sinoatrial pacemaking cells ( Posokhova et al . , 2010; Yang et al . , 2010 ) , RGS6 ablation had no effect on any GIRK response parameters in hippocampal pyramidal neurons . The function of R7 RGS proteins is regulated by the membrane anchor R7BP , which augments the ability of R7 RGS proteins to deactivate G protein signaling in reconstituted systems ( Drenan et al . , 2006; Masuho et al . , 2013 ) . Consistent with these observations , we show that knockout of R7BP reduces plasma membrane localization of RGS7 and slows GABABR-GIRK current deactivation in hippocampal pyramidal neurons . Our results also agree well with a recent study reporting similar deceleration of GIRK deactivation kinetics in R7bp−/− hippocampal neurons ( Zhou et al . , 2012 ) . However , the consequences of RGS7 and R7BP elimination are different in two important ways . First , relative to the loss of RGS7 or Gβ5 , elimination of R7BP had a very modest effect on GABABR-GIRK deactivation kinetics , and was seen only at high GABABR agonist concentrations . This suggests that the RGS7/Gβ5 complex can regulate GABABR-GIRK signaling without R7BP . Secondly , loss of R7BP and RGS7 has different effects on GABABR-GIRK coupling sensitivity . While elimination of RGS7 ( or Gβ5 ) results in a parallel leftward shift in the concentration-response relationship , the concentration-response measured in R7bp−/− neurons shows markedly greater cooperativity . Thus , at low levels of GABABR stimulation , R7BP is dispensable and its elimination fails to affect response amplitude . As the response reaches saturation , R7BP becomes indispensable and its elimination has the same effect on response amplitude as loss of RGS7 . These observations are inconsistent with the proposed role for R7BP as a critical factor in the assembly of RGS7-GIRK complexes ( Zhou et al . , 2012 ) , which postulates the equivalence of R7BP effects across agonist concentrations and a similar impact of R7BP and RGS7 elimination on GIRK channel kinetics , and certainly not an increase in cooperativity of GIRK channel activation . Based on our findings , we propose an alternative model , where R7BP sets GABABR-GIRK coupling efficiency . In this model , at low levels of GIRK activation , when released Gβγ subunits do not saturate the GIRK channel , deactivation kinetics are primarily mediated by RGS7/Gβ5 directly associated with the GIRK subunits . Under these conditions , there are virtually no genotype-dependent differences in the lag time of GIRK deactivation upon agonist removal . With higher agonist concentrations , as GABABR produces more Gβγ subunits , the amount of available free Gβγ exceeds that of activatable GIRK channels , evidenced by a delay that precedes GIRK current deactivation upon agonist removal . Under these conditions , R7BP elimination affects response sensitivity as much as elimination of RGS7 . While response deactivation kinetics reflect Gβγ inactivation after it had a chance to interact with the channel , response sensitivity likely relates to the efficiency of the Gβγ reaching the GIRK channel . Considering that R7BP affects both of these parameters only when free Gβγ is produced in excess , we think that the RGS7/Gβ5 complex exists in two states: ( 1 ) anchored to the GIRK complex where it affects deactivation kinetics , and ( 2 ) anchored via R7BP to the plasma membrane outside of GIRK complex where it primarily determines response sensitivity . In agreement with this model , we observe an increase in cooperativity of GIRK channel activation upon R7BP loss that suggests that RGS7/Gβ5 alone , but not in complex with R7BP , promotes more productive association of Gβγ with the channel . On an intuitive level , this may reflect larger proportion of Gβγ subunits that reach GIRK while being deactivated in its vicinity by RGS7/Gβ5 . In contrast , complexes of RGS7/Gβ5 with R7BP may predominantly act elsewhere on the plasma membrane , thereby decreasing the number of Gβγ subunits capable of reaching the channel . By physically binding to the GIRK channel , RGS7/Gβ5 may further act to promote productive interactions of Gβγ ( and/or Gα ) with the channel , thus increasing cooperativity of its activation . This model is also consistent with the observed decrease in RGS7 on the plasma membrane but largely preserved localization at specific postsynaptic sites where it might be anchored via complex formation with GIRK channels . Thus , the main role of R7BP appears to be in tuning the sensitivity of the response by endowing RGS7/Gβ5 complexes with an ability to deactivate G proteins before they reach the GIRK channel . RGS proteins are potent negative regulators of both the extent and duration of neurotransmitter signaling via G protein-coupled receptors ( Sjogren , 2011; Xie and Martemyanov , 2011 ) . As a result , knockout of individual RGS genes in mice is frequently associated with augmented GPCR signaling . This makes the analysis of changes in sensitivity of behavioral or cellular reactions to neurotransmitter actions associated with elimination of individual RGS proteins a powerful strategy that allows for establishing physiological receptor-RGS pairings . Using this strategy , two RGS proteins have been previously implicated in controlling GABABR signaling in vivo: RGS6 in cerebellar neurons ( Maity et al . , 2012 ) and RGS2 in the ventral tegmental area ( Labouebe et al . , 2007 ) . Now , our findings establish that in hippocampal neurons , GABABR signaling is negatively regulated by RGS7 . These observations complement earlier findings with RGS14 , the only other RGS protein implicated in hippocampal synaptic plasticity and spatial learning ( Lee et al . , 2010 ) . However , in contrast to RGS7 that acts in the CA1 region , RGS14 functions in CA2 synapses and its ablation augments LTP and improves spatial learning in mice . The major forms of synaptic plasticity ( LTP , LTD and depotentiation ) in the hippocampus have been implicated in learning and memory ( Martin et al . , 2000 ) . Recent models propose that all three forms cooperate to affect distinct aspects of spatial information storage ( Kemp and Manahan-Vaughan , 2007 ) . In relevance to this thinking , we found a selective disruption of LTD and depotentiation in RGS7−/− mice , but normal LTP . There is a mounting evidence suggesting an active and selective role of LTD in object recognition and creation of spatial representation of memory ( Kemp and Manahan-Vaughan , 2007 ) . For example , LTD but not LTP was found to be important for spatial memory consolidation and memory enhancement for novelty acquisition ( Ge et al . , 2010; Dong et al . , 2012 ) . Consistent with this , electrophysiological recordings in behaving animals show facilitation of LTD during novelty exploration ( Xu et al . , 1997 , 1998; Manahan-Vaughan and Braunewell , 1999; Dong et al . , 2012 ) . Disruption of LTD via genetic mutations often results in memory deficits , particularly affecting behavioral flexibility ( Nicholls et al . , 2008 ) , while LTD enhancement can lead to improvement in spatial reversal learning ( Duffy et al . , 2008 ) . In particular , our findings resemble the phenotype of SRF−/− mice , which exhibited a selective disruption of LTD in hippocampal CA1 neurons paralleled by an inability to learn hippocampus-dependent tasks ( Etkin et al . , 2006 ) . While specific knowledge on how depotentiation contributes to memory is still lacking , it has been observed that exposure to novel environment and spatial exploration depotentiates previously-induced LTP ( Xu et al . , 1998 ) . Together , our observations in mice lacking RGS7 reinforce the idea that normal inhibitory synaptic plasticity ( LTD and depotentiation ) is required for hippocampal-dependent learning and memory . Although , it has never been systematically addressed in a single study , GIRK channel elimination was reported to enhance LTP ( Cramer et al . , 2010 ) . Given that the Girk2−/− neurons are relatively depolarized , the enhanced LTP likely results from a general increase in excitability . Conversely , GIRK2 over-expression in transgenic models of Down syndrome leads to reduced LTP ( Kleschevnikov et al . , 2004; Siarey et al . , 2005 ) , but enhanced LTD ( Siarey et al . , 2005; Cooper et al . , 2012 ) . Interestingly , GIRK2 over-expression and ablation both resulted in decreased depotentiation ( Chung et al . , 2009; Cooper et al . , 2012 ) . Similar to Down syndrome mouse models , Rgs7−/− mice also exhibited augmented GABABR-GIRK signaling . However , in the case of Rgs7−/− mice , this augmentation resulted in yet another distinct phenotype: selective deficits in LTD and depotentiation , with normal LTP . Several lines of evidence argue that these changes in synaptic plasticity are driven by a postsynaptic mechanism whereby RGS7 controls inhibitory GIRK-dependent signaling in the CA1 pyramidal neurons . First , CA1 pyramidal neurons lacking RGS7 are hyperpolarized and less excitable relative to wild-type neurons , a property that is significantly shaped by the GIRK channels . Second , basal excitatory transmission is unaltered in Rgs7−/− slices . Third , direct measurements of events that reflect excitatory presynaptic function ( e . g . , FV and PPR ) reveal no changes caused by the elimination of RGS7 . Given these observations , we propose that GIRK signaling sets the general excitability of postsynaptic pyramidal neurons , which in turn determines a sliding scale window for the induction of different forms of synaptic plasticity . By adjusting signaling strength in the hippocampal GABABR-GIRK pathway , the RGS7/Gβ5/R7BP complex influences the range of neuronal responses necessary for memory formation . While dysregulation of GABABR-GIRK signaling in hippocampus may be sufficient for explaining the effects of RGS7 on learning an memory , at this point we cannot rule out a contribution of other brain regions and/or signaling pathways in the process . In any event , we believe that the results of this study establish RGS7 complex as an important molecule for understanding and/or correcting the pathology of neuropsychiatric disorders associated with disruptions in synaptic plasticity and imbalances in inhibitory signaling . All studies were carried out in accordance with the National Institute of Health guidelines and were granted formal approval by the Institutional Animal Care and Use Committee of the Scripps Research Institute . The generation of Rgs6−/− ( Posokhova et al . , 2010 ) , Rgs7−/− ( Cao et al . , 2012 ) , R7bp−/− ( Anderson et al . , 2007 ) mice were described earlier . All animals used for comparing genotypes were littermates derived from heterozygous breeding pairs . Mice were housed in groups on a 12 hr light–dark cycle with food and water available ad libitum . Males and females ( 2–5 months ) were used for all experiments . Lysates were prepared by homogenizing hippocampal tissue from age-matched littermates by sonication in the lysis buffer ( 1 × PBS , 150 mm NaCl , 1% Triton X-100 , protease inhibitors ) followed centrifugation at 14 , 000×g for 10 min . The resulting extract was used for protein concentration determination by the BCA protein assay ( Pierce , Rockford , IL ) . The lysates were adjusted to equalize total protein content by adding lysis buffer and 2x SDS sample buffer . Samples were boiled for 5 min , resolved on SDS-PAGE gels , transferred onto PVDF membrane and subjected to western blot analysis using HRP conjugated secondary antibodies and ECL West Pico ( Pierce ) detection system . Signals were captured on film and scanned by densitometer , and band intensities were determined using NIH ImageJ software . Rabbit anti-R7BP ( TRS ) and rabbit Gβ5 ( ADTG ) were generous gifts from Dr William Simonds ( NIDDK/NIH ) . Anti-RGS6 was generated and used previously ( Posokhova et al . , 2010 ) . Chicken IgY anti-RGS7 antibody was from Thermo Scientific ( Waltham , MA ) , anti-GIRK2 was purchased from Alomone labs ( Jerusalem , Israel ) and anti-β-actin was from Sigma ( St . Lois , MO ) . Primary cultures of hippocampal neurons were prepared using a modified version of a published protocol ( Xie et al . , 2010 ) . Briefly , hippocampi were extracted from neonatal ( P1-3 ) pups and placed into an ice-cold HBSS/FBS solution: Hank’s Balanced Salt Solution ( Sigma; St . Louis , MO ) , 4 . 2 mM NaHCO3 , 1 mM HEPES , and 20% FBS . The tissue was washed twice with HBSS/FBS , and then three times with HBSS alone . Hippocampi were digested at room temperature for 5 min with 10 mg/ml Trypsin Type XI ( Sigma; St . Louis , MO ) in a solution that contained ( in mM ) : 137 NaCl , 5 KCl , 7 Na2HPO4 , and 25 HEPES ( pH 7 . 2 ) . The tissue was washed twice with HBSS/FBS and three times with HBSS alone , and then hippocampi were mechanically-dissociated in HBSS ( supplemented with 12 mM MgSO4 ) using Pasteur pipettes of decreasing diameter . The neurons were pelleted by centrifugation ( 600×g for 10 min at 4°C ) and plated onto 8-mm glass coverslips pre-treated with Matrigel ( BD Biosciences; San Jose , CA ) in 48-well plate . Neurons were allowed to adhere for 30 min prior to adding 0 . 3 ml of pre-warmed culture medium consisting of Neurobasal A ( Life Technologies; Carlsbad , CA ) , 2 mM GlutaMAX-I ( Life Technologies , Carlsbad , CA ) , 2% B-27 supplement , and 5% FBS . After 4–12 hr , culture media was completely replaced with the same media without FBS . Neurons were incubated at 37°C/5% CO2 , and half of the medium was replaced with fresh medium on each of the first 3 days of culture . Neurons were kept in culture for 10–14 days prior to experiments . Coverslips containing neurons were transferred to a chamber containing a low-K+ bath solution ( in mM ) : 145 NaCl , 4 KCl , 1 . 8 CaCl2 , 1 MgCl2 , 5 . 5 D-glucose , 5 HEPES/NaOH ( pH 7 . 4 ) . Borosilicate patch pipettes ( 3–5 MΩ ) were filled with ( in mM ) : 130 KCl , 10 NaCl , 1 EGTA/KOH ( pH 7 . 2 ) , 0 . 5 MgCl2 , 10 HEPES/KOH ( pH 7 . 2 ) , 2 Na2ATP , 5 phosphocreatine , 0 . 3 GTP . Baclofen ( R- ( + ) -b- ( aminomethyl ) -4-chlorobenzenepropanoic acid hydrochloride ) was purchased from Sigma ( St . Louis , MO ) . Baclofen-induced currents were measured at room temperature using a high-K+ bath solution ( in mM ) : 120 NaCl , 25 KCl , 1 . 8 CaCl2 , 1 MgCl2 , 5 . 5 D-glucose , 5 HEPES/NaOH ( pH 7 . 4 ) . The high-K+ bath solution ( +/− baclofen ) was applied directly to the soma and proximal dendrites with an SF-77B rapid perfusion system ( Warner Instruments , Inc . ; Hamden , CT ) . Current responses to the application of the high-K+ solution ( +/− baclofen ) were measured at a holding potential of −80 mV . Membrane potentials and whole-cell currents were measured in large neurons ( >75 pF ) with hardware ( Axopatch-700B amplifier , Digidata 1440A ) and software ( pCLAMP v . 10 . 3 ) from Molecular Devices ( Sunnyvale , CA ) . All currents were low-pass filtered at 2 kHz , sampled at 5 kHz , and stored on computer hard disk for subsequent analysis . Peak and steady-state current amplitudes were measured for each experiment . Current activation rates were extracted from a standard exponential fit of the current trace corresponding to the onset of drug effect and the peak evoked current , while deactivation rates were extracted from an exponential fit of the trace corresponding to the return of current to baseline following removal of drug ( Clampfit v . 10 . 3 software ) . Current desensitization was defined as % change in steady state current from the maximal baclofen-evoked response amplitude during 10 s of continuous drug application . Only experiments where access resistances were stable and low ( <20 MΩ ) were included in the analysis . Immunohistochemical reactions were carried out using the pre-embedding immunogold method as described earlier ( Lujan et al . , 1996 ) . Briefly , after blocking with 10% serum for 1 hr at room temperature free-floating sections were incubated for 48 hr with anti-RGS7 antibodies ( 1–2 mg/ml ) . Sections were washed and incubated for 3 hr with goat anti-rabbit IgG coupled to 1 . 4 nm gold ( Nanoprobes Inc ) at 1:100 dilution . Sections were washed , postfixed in 1% glutaraldehyde and processed for silver enhancement of the gold particles with an HQ Silver kit ( Nanoprobes Inc . ) . The reacted sections were treated with osmium tetraoxide ( 1% in 0 . 1 M PB ) , block-stained with uranyl acetate , dehydrated in graded series of ethanol and flat-embedded on glass slides in Durcupan ( Fluka ) resin . Regions of interest were cut at 70–90 nm on an ultramicrotome ( Reichert Ultracut E; Leica ) . Staining was performed on drops of 1% aqueous uranyl acetate followed by Reynolds’s lead citrate . Ultrastructural analyses were performed in a Jeol-1010 electron microscope . To establish the relative the abundance of RGS7 immunoreactivity along the plasma membrane of pyramidal cells , we used 60-µm coronal slices processed for pre-embedding immunogold immunohistochemistry . The procedure was similar to that used previously ( Lujan et al . , 1996 ) . Briefly , for each of three animals from different postnatal ages and adult , three samples of tissue were obtained for preparation of embedding blocks ( totalling nine blocks for each age ) . To minimize false negatives , electron microscopic serial ultrathin sections were cut close to the surface of each block , as immunoreactivity decreased with depth . We estimated the quality of immunolabelling by always selecting areas with optimal gold labelling at approximately the same distance from the cutting surface . Randomly selected areas were then photographed from the selected ultrathin sections and printed with a final magnification of 45 000X . Quantification of immunogold labelling was carried out in reference areas totalling approx . 1 , 800 µm2 for each age . Immunoparticles identified in each reference area and present in different subcellular compartments ( dendritic spines , dendritic shafts and somata ) were counted . We measured the radial distance of each immunoparticle to the plasma membrane , being 0 for those just located in the plasma membrane . The data was expressed as percentage of immunoparticles along the radial distance from the plasma membrane expressed in nanometers . Mice were sacrificed under isoflurane anesthesia , and brains were rapidly removed and placed in ice-cold artificial cerebrospinal fluid ( aCSF ) without CaCl2 , composed of ( mM ) : 124 NaCl , 3 KCl , 24 NaHCO3 , 1 . 25 NaH2PO4 , 1 MgSO4 , and 10 D-Glucose , equilibrated with 95% O2 and 5% CO2 . The tissue was cut in 350–400 µm thick sections with a Vibrating microtome ( Leica VT1200S , Germany ) . The slices were warmed to 35°C for 25–45 min in aCSF supplemented with 2 mM CaCl2 , and equilibrated with 95% O2 and 5% CO2 . Then slices were maintained in gassed aCSF at room temperature until being transferred to submerged-type recording chambers of volume ∼1 . 5 ml . Here , the slices were constantly superfused ( 1–2 ml/min ) with warmed ( 30–31°C ) , gassed aCSF . All measurements were performed by an experimenter blind to genotype . CA1 neurons were visually identified in the hippocampal transverse slices of 350 µM thickness using Scientifica SliceScope system . Glass microelectrodes with an open-tip resistance of 3 . 5–6 . 5 MΩ were used . The following internal solution was used ( mM ) : 130 K-Gluconate , 20 KCl , 10 K-HEPES , 0 . 2 EGTA , 0 . 3 Na-GTP and 4 Mg-ATP ( pH 7 . 3 ) . To determine intrinsic cellular properties such as resting membrane potential , input resistance and spike numbers , 500 ms , 50 pA , 10-step hyperpolarizing and depolarizing current injections were delivered every 10 s . Cells with series resistance >20 MΩ or resting membrane potentials > −55 mV were excluded from analysis . Liquid junction potential was −14 mV . Spontaneous EPSCs ( sEPSCs ) were measured by holding the cells at −70 mV in normal aCSF . At least 100 events , which are above 4 pA , were obtained during sEPSC measurements in each cell . Field excitatory postsynaptic potentials ( fEPSPs ) were elicited by a concentric bipolar stimulating electrode ( inner diameter [ID]: 25 µm; outer diameter [OD]: 125 µm , FHC Inc . , Bowdoin , ME ) connected to a constant current isolated stimulator unit ( A-M Systems; Carlsborg , WA ) and recorded with low resistance ( 3–5 MΩ ) glass pipettes ( ID: 1 . 16 mm , OD: 1 . 5 mm , Harvard Apparatus , Holliston , MA ) filled with aCSF . The electrodes were placed in the stratum radiatum of the CA1 area of the dorsal hippocampus slices . Stimulation frequency was set to 0 . 05 Hz . Input-output curves were generated by adjusting the stimulus intensity in increments of 10 µA , from 0 to 100 µA . Paired pulse ratio ( PPR ) was assessed using a succession of paired pulses separated by time intervals ranging from 25 to 1000 ms , delivered every 20 s . The degree of facilitation was determined by taking the ratio of the initial slope of the second fEPSP to the initial slope of the first fEPSP . A PPR >1 was considered to reflect facilitation . For synaptic plasticity experiments , a stable baseline for at least 30 min was achieved prior to induction . LTP and LTD were recorded for 1 hr after HFS ( 2 tetanized stimuli ( TS ) of 100 Hz for 1s each ) or LFS ( 2 Hz for 10 min , 1200 pulses ) was applied . Depotentiation was achieved by applying HFS followed by LFS after 1-2 min interval . Spatial learning and memory were evaluated in the Morris water maze using a video tracking systems ( EthoVision XT , Noldus Information Technology , Wageningen , Netherlands ) . Mice first received 4 cued trials ( visible but variable platform location ) on the first day to determine if non-associative impairment in other behavioral responses affecting performance in this task , such as exploratory activity , motor coordination , vision and motivation . After completing the cued trials , spatial learning acquisition was evaluated during the place condition ( hidden platform , constant location; 4 trials/day , 6 consecutive days ) . Escape latency and success rates in finding hidden platform were calculated for all place trials . Retention performance was evaluated during probe trials ( platform removed ) , which were conducted 24 hr after the last cued trial on the sixth day . Latency to the first target platform area crossing and the number of crossings in the probe trial ( 1 min ) served as the dependent variables . For fear conditioning experiments , mice were habituated in individual conditioning chambers to obtain freezing baseline on the first day . On the second day , mice were trained in individual conditioning chambers ( Med Associates , St . Albans , VT ) . Video images were recorded via video tracking systems ( EthoVision XT , Noldus Information Technology , Wageningen , Netherlands ) . In context A , visible light was turned on , a stainless steel grid floor inserted , and the chambers were cleaned with 70% ethanol prior to conditioning . In context B , white plastic inserts were placed inside the chamber to change the shape , size and texture of the wall and the floor . A small weight boat with orange extract is placed behind the wall inserts to provide a novel smell for the chambers . The chambers were cleaned with 70% isopropanol . During the training sessions , mice were placed in context A and allowed to explore for 3 min prior to the delivery shocks ( 0 . 75 mA , 1 s ) . We used auditory tone ( 85 dB , 30 s duration ) as CS and an electric foot shock ( delivered through the grid floor , 0 . 75 mA AC current , 1 s duration ) as US . Mice received 3 tone-shock stimuli ( with 1–2 min interval ) during the training day . In every session , tone always co-terminated with electric shock . Mice were removed from the conditioning chamber 30 s after the last shock . Contextual test ( 5 min ) was done in context A 24 hr after training . Cue test was performed 1 hr after contextual test in context B . The animals were brought into the test room in a different covered mover with different bedding inside to avoid contextual reminders . The animal was immediately placed in the contextual B chamber and allowed to explore for 3 min prior to the delivery the auditory tone ( 85 dB ) for another 3 min . The freezing response was measured with the automated tracking and analyzed offline . For the novel object recognition test , mice were habituated to the arena for two consecutive days , where they were placed in a white plastic open-box of dimensions 43 . 2 cm2 by 30 . 5 cm height for 10 min each . The acquisition phase of the object recognition assay involved placing each individual mouse in the test arena for 8 min in the presence of two identical objects ( either silver or black objects in different shape , size and texture , the order of which alternated between mice ) . The video was recorded via a camera . The time spent investigating both identical objects was recorded by automatic tracking system ( EthoVision XT , Noldus Information Technology , Wageningen , Netherlands ) . 24 hr after the acquisition phase , each mouse was reintroduced to the test arena for 8 min . During this phase ( retention testing phase ) the mouse was presented with one copy of the object that it was exposed to during the acquisition phase ( familiar object ) and one novel object ( either silver or black objects , depending on which was presented during the acquisition phase ) . The time spent investigating each object was recorded . All objects and the testing box were thoroughly cleaned with 70% methanol between mice to remove any odor cues . Agonist-dependent cellular measurements of bioluminescence resonance energy transfer ( BRET ) between Venus-Gβ1γ2 and its effector fragment masGRK3ct-Nluc were performed upon reconstitution in living cells as previously described with slight modification ( Hollins et al . , 2009; Masuho et al . , 2013 ) . The masGRK3ct-Nluc construct contained amino acids G495-L688 of bovine GRK3 ( NP_776925 ) , preceded by a myristic acid attachment peptide ( mas; MGSSKSKTSNS ) . The stop codon of GRK3 was replaced with a GGGS linker , which was followed by the NanoLuc ( Nluc ) ( Hall et al . , 2012 ) . Briefly , GABAB1R , GABAB2R , GαoA , Venus156-239-Gβ1 , and Venus1-155-Gγ2 constructs were transfected into HEK293T/17 cells at a 1:1:2:1:1 ratio with increasing masGRK3ct-Nluc from 0 . 125 to 6 in ratio . 5 µg total DNA was delivered per 4 × 106 cells in a 6-cm-dish . 16–24 hr post transfection cells were stimulated with 100 µM GABA followed by treatment with 100 µM CGP 54626 . The BRET signal is determined by calculating the ration of the light emitted by the Venus-Gβ1γ2 ( 535 nm ) over the light emitted by the masGRKct-Rluc8 ( 475 nm ) . The average baseline value recorded prior to agonist stimulation was subtracted from BRET signal values , and the resulting difference ( ΔBRET ) was obtained . Statistical analyses were performed using Prism ( GraphPad Software , Inc . ; La Jolla , CA ) . Data are presented throughout as the mean ± SEM . Student t test , one-way or two-way ANOVA , followed by Bonferroni’s post hoc test were used as appropriate . The minimal level of significance was set at p<0 . 05 .

Neurons communicate with one another at junctions called synapses . The arrival of an electrical signal known as an action potential at the first cell causes molecules known as neurotransmitters to be released into the synapse . These molecules diffuse across the gap between the neurons and bind to receptors on the receiving cell . Some neurotransmitters , such as glutamate , activate cells when they bind to receptors , thus making it easier for the second neuron to ‘fire’ ( i . e . , to generate an action potential ) . By contrast , other neurotransmitters , such as GABA , usually make it harder for the second neuron to fire . Many of the effects of GABA involve a type of receptor called GABAB . When GABA binds to one of these receptors , a molecule called a G-protein is recruited to the receptor . This activates the G-protein , triggering a cascade of events inside the cell that lead ultimately to the opening of potassium ion channels , which as known as GIRKs , in the cell membrane . Positively charged potassium ions then leave the cell through these channels , and this makes it more difficult for the cell to fire . Now , Ostrovskaya et al . have revealed that a complex of three proteins regulates the interaction between GABAB receptors and GIRK channels . In neurons that lack either of these proteins , the receptors have less influence on GIRKs than in normal cells . Moreover , mice that lack one of the proteins ( called RGS7 ) perform less well in various learning and memory tests: for example , they take longer than normal animals to learn the location of an escape platform in a water maze , or to retain a memory of a fearful event . By identifying the proteins that regulate the interaction between GABAB receptors and GIRKs , Ostrovskaya et al . have helped to unravel a key signaling cascade relevant to cognition . Given that GIRK channels have recently been implicated in Down’s syndrome , these insights may also increase understanding of cognitive impairments in neuropsychiatric disorders .

2014

RGS7/Gβ5/R7BP complex regulates synaptic plasticity and memory by modulating hippocampal GABABR-GIRK signaling

Metabolite exchange among co-growing cells is frequent by nature , however , is not necessarily occurring at growth-relevant quantities indicative of non-cell-autonomous metabolic function . Complementary auxotrophs of Saccharomyces cerevisiae amino acid and nucleotide metabolism regularly fail to compensate for each other's deficiencies upon co-culturing , a situation which implied the absence of growth-relevant metabolite exchange interactions . Contrastingly , we find that yeast colonies maintain a rich exometabolome and that cells prefer the uptake of extracellular metabolites over self-synthesis , indicators of ongoing metabolite exchange . We conceived a system that circumvents co-culturing and begins with a self-supporting cell that grows autonomously into a heterogeneous community , only able to survive by exchanging histidine , leucine , uracil , and methionine . Compensating for the progressive loss of prototrophy , self-establishing communities successfully obtained an auxotrophic composition in a nutrition-dependent manner , maintaining a wild-type like exometabolome , growth parameters , and cell viability . Yeast , as a eukaryotic model , thus possesses extensive capacity for growth-relevant metabolite exchange and readily cooperates in metabolism within progressively establishing communities . All living cells possess a system for biochemical reactions , the metabolic network , which supplies cells with their necessary molecular constituents . The reactions participating in this network are highly conserved , so much so that all life is made up of a markedly similar set of metabolites ( Braakman and Smith , 2013; Caetano-Anolles et al . , 2009 ) . The functionality of the metabolic system is bound to a series of transport reactions that facilitate the uptake of metabolites from the environment , as well as metabolite export . Metabolite export primarily occurs for the purpose of maintaining balance of the metabolic system ( 'overflow metabolism' ) and to maintain chemical and physical integrity of the metabolic network . This includes indiscriminate metabolite export through non-specific multi-drug transporters required in removal of toxic metabolites for cells ( Paczia et al . , 2012; Piedrafita et al . , 2015 ) . Co-growing cells can uptake the released metabolites and exploit their presence . Indeed , to a lower extent , metabolite export of metabolites can specifically occur for the purpose of establishing inter-cellular metabolic interactions in biosynthetic metabolism ( Nigam , 2015; Paczia et al . , 2012; Yazaki , 2005 ) and can lead to mutualistic situations in which cells profit from coexistence ( Foster and Bell , 2012; Oliveira et al . , 2014 ) . In between species , mutually positive interactions can readily establish when exchange concerns an overflow metabolite , exemplified by yeast–algae interactions that can form on the basis of a CO2 and sugar exchange ( Hom and Murray , 2014 ) , or between different cells of the same species or tissue , exemplified in tumours , when lactate produced in excess by one cell type is metabolised by another ( Bonuccelli et al . , 2010 ) , or between neurons and glial cells that exchange sugar metabolites ( Bélanger et al . , 2011; Volkenhoff et al . , 2015 ) . It is more difficult to assess whether metabolite exchange is indicative of non-cell-autonomous metabolism , when exchange concerns metabolites that are needed by both exchange partners , amino acids , and nucleobases for instance . Exchange of costly intermediates is associated with a significant risk , as exported metabolites can be lost through diffusion , chemical damage , or cheating ( Dobay et al . , 2014; Oliveira et al . , 2014; Wintermute and Silver , 2010 ) . Despite these constraints , exchange of intermediates is frequently observed within bacterial microbial communities . Many bacterial species lose essential biosynthetic pathways , disabling them from living autonomously , which may explain why more than 90% of bacteria cannot be cultivated in the absence of a community environment ( Costerton et al . , 1994; Johnson et al . , 2012 ) . The energetic benefit and selective advantage associated with non-autonomous cellular metabolism is often not clear but might involve , for example , the ability to reduce genome size which would in turn facilitate faster proliferation . This may explain why bacteria frequently appear to cooperate in the biosynthesis of more costly and biosynthetically complex metabolites , such as aromatics ( Mee et al . , 2014 ) . While metagenomics has boosted the knowledge of metabolite exchange strategies in bacteria ( Blaser et al . , 2013; Manor et al . , 2014; Zelezniak et al . , 2015 ) , relatively little is known about eukaryotic species . This includes yeast , a popular single cellular eukaryotic model organism , whose metabolic capacities are regularly exploited in biotechnology . Yeast cells are known to participate in multi-species communities ( e . g . on human skin ( Findley et al . , 2013 ) ) , but as wild yeast isolates usually maintain similar prototrophic genomes ( Jeffares et al . , 2015; Liti et al . , 2009 ) , metagenomic data are not conclusive about yeast's metabolite exchange strategies . In laboratory experiments , yeast cultures were , however , not effective in supporting co-growth of auxotrophs that have complementary defects in amino acid and nucleotide metabolism , unless they were genetically modified to increase metabolite export ( Müller et al . , 2014; Shou et al . , 2007 ) . This contrasts with analogous studies in bacterial species , in which such growth experiments regularly show that co-cultured cells can overcome complementary metabolic deficiencies ( Foster and Bell , 2012; Oliveira et al . , 2014; Pande et al . , 2014; Vetsigian et al . , 2011 ) . In the absence of quantitative metabolite data , this observation has triggered the conclusion that co-growing prototrophic yeast cells produce amino acid and nucleotide metabolites predominantly for themselves and export them at insufficient quantities to support growth of co-growing cells ( Momeni et al . , 2013; Shou et al . , 2007 ) . Conflicting with this interpretation , we here report that yeast colonies maintain a rich exometabolome and that cells exploit this metabolic pool preferentially over their own biosynthetic capacities , which implies that metabolite exchange establishes as a natural property of yeast growth . To test whether yeast indeed possesses the capacity for metabolite exchange at growth relevant quantities , we established an alternative method to co-culture experiments . We exploited the stochastic segregation of episomes to randomly and progressively introduce metabolic auxotrophies into a yeast population which self-establishes from an initially prototrophic cell . This strategy enabled co-growing auxotrophs to enter an efficient state of metabolic cooperation , named self-establishing metabolically cooperating communities ( SeMeCos ) . Despite an auxotrophic cell composition of up to 97% , SeMeCos achieve metabolic efficiency , growth parameters , and cell viability similar to that of genetically prototrophic cells , revealing a natural capacity of yeast to exchange metabolites at growth relevant quantities . In a SeMeCo that possesses auxotrophies in histidine , leucine , uracil and methionine metabolism , we distinguish up to eight cell types , each of which is unable to survive on its own , or in co-culture studies , however , could adapt effectively and cooperatively to overcoming metabolic deficiencies , once self-established in a community structure . Communities have a stable population composition as well as distinct spatial heterogeneity , which was , however , not essential for metabolite exchange , as SeMeCos maintain growth in liquid suspension . Self-establishing , complex communities thus demonstrate that yeast is not only exchanging metabolites , but is also able to do so at growth relevant quantities , to facilitate growth on the basis of a non-cell-autonomous metabolism . Histidine , leucine , uracil , and methionine biosynthetic pathways were chosen for our study as ( i ) they can be interrupted by deletion of a single , non-redundant gene , which has been reported not to cause compensatory mutations and ( ii ) because cells possess efficient uptake mechanisms for these nutrients ( Mülleder et al . , 2012; Pronk , 2002; Teng et al . , 2013 ) . Paired combinations of histidine ( his3Δ ) , leucine ( leu2Δ ) , uracil ( ura3Δ ) , or methionine ( met15Δ ) auxotrophs were unable to sustain growth in the absence of supplementation required for both individual cell types ( Figure 1A ) . A similar result was obtained by co-culturing flocculating yeast cells ( Figure 1—figure supplement 1 ) , which are able to maintain biofilm-like physical contact ( Smukalla et al . , 2008 ) , and in Schizosaccharomyces pombe ( Figure 1B ) , indicating evolutionary conservation of this observation in yeast species . 10 . 7554/eLife . 09943 . 003Figure 1 . Yeast auxotrophs do not compensate for metabolic deficiencies upon co-culturing , yet export the relevant metabolites and prefer metabolite uptake over self-synthesis . ( A ) Complementary pairs of Saccharomyces cerevisiae auxotrophs do not overcome metabolic deficiencies upon co-culturing . his3∆ , leu2∆ , met15∆ , and ura3∆ yeasts were combined in complementary pairs and spotted on corresponding selective media . No pairs exhibited co-growth together . ( B ) A complementary pair of Schizosaccharomyces pombe auxotrophs does not overcome metabolic deficiencies upon co-culturing . leu1Δ and ura4Δ yeasts were combined in a complementary pair and spotted on corresponding selective media . No co-growth occurred . ( C ) The concentration of metabolites in the S . cerevisiae colony exometabolome obtained from 1 . 3e08 YSBN5 cells grown in a colony on synthetic minimal agar media ( SM ) and quantified by LC-MS/MS . Abbreviations: single letter IUPAC amino acid codes , O = ornithine , CIT = citrulline . n = 3 , error bars = ± SD . ( D ) ( i ) Metabolites quantified as in ( C ) , comparing intracellular ( total cell extracts ) and extracellular metabolite concentrations in YSBN5 . n = 3 , error bars = ± SD . Dashed line: linear regression fit , grey band shows 95% confidence region . ( ii ) Metabolites quantified as in ( C ) , comparing extracellular metabolite concentrations of YSBN5 and BY4741-pHLUM yeast colonies grown on minimal media . H , L , U , and M are highlighted in red circles . n = 3 , error bars = ± SD . Dashed line: linear regression fit , grey band shows 95% confidence region . Abbreviations IUPAC codes; H = histidine , L = leucine , U = uracil , M = methionine . ( E ) Consumption of uracil , histidine , leucine , and methionine in yeast batch cultures in synthetic complete ( SC ) media as measured by LC-MS/MS . Uracil , histidine , leucine , and methionine prototrophic cells consume these metabolites at rates and quantities comparable to the corresponding auxotrophic strains . ( F ) ( i ) Deletion of URA3 ( Orotidine-5'-phosphate decarboxylase ) causes accumulation of the Ura3p substrate orotidine-5'-phosphate ( OMP ) , when cells are supplemented with 20 mg/L uracil ( fold change of OMP abundance , relative to URA3 without uracil supplementation ) , as determined by LC-MS/MS . Error bars = ± SD . ( ii ) Uracil supplementation of wild-type cells alters their metabolite profile to resemble ura3∆ cells , which obtain uracil solely from the growth media . Heatmap scaling ( [0 , 1] and min , max per metabolite ) was based on median concentration . The dendrogram was constructed by comparing euclidean distance ( dissimilarity ) between samples . DOI: http://dx . doi . org/10 . 7554/eLife . 09943 . 00310 . 7554/eLife . 09943 . 004Figure 1—figure supplement 1 . Flocculation does not enable Saccharomyces cerevisiae cells to establish viable co-cultures . ( i ) ( left ) The FLO+ phenotype in yeast cells transforms their typical cell suspension ( right ) into a physiological state reminiscent of biofilms ( left ) ( Smukalla et al . , 2008 ) . Cultures were grown to stationary phase in rich media ( YPD ) and flocculation was detected via an inability of cells to re-suspend following repeated tube inversion . ( ii ) Complementary pairs of flocculating S . cerevisiae auxotrophs do not overcome metabolic deficiencies upon co-culturing . his3∆ , leu2∆ , met15∆ , and ura3∆ yeasts were combined in complementary pairs and spotted on corresponding selective media . No pairs exhibited co-growth together . DOI: http://dx . doi . org/10 . 7554/eLife . 09943 . 00410 . 7554/eLife . 09943 . 005Figure 1—figure supplement 2 . Uracil biosynthetic genes in the uracil prototroph ( URA3 ) and auxotroph ( ura3Δ ) remain expressed in ( uracil supplemented ) SC media , as determined by RNA sequencing . Abbreviations: RPKM = reads per kilobase per million . n = 3 , error bars = ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 09943 . 005 In two previous studies , leucine/tryptophan and adenine/lysine auxotrophic cell pairs , respectively ( Müller et al . , 2014; Shou et al . , 2007 ) , could co-grow upon removing metabolic feedback control . Feedback resistance renders cells metabolite over-exporters , leading to the conclusion that wild-type yeast cells produce intermediates primarily for themselves , at quantities that are not sufficient for growth relevant metabolite exchange ( Momeni et al . , 2013; Shou et al . , 2007 ) . In a detailed analysis of the intra-colony exometabolome , using an ultra-sensitive mass spectrometry method , the intra-colony fluid showed however to contain a plethora of metabolites , with the amino acids glutamine , glutamate , and alanine being the most highly concentrated ( Figure 1C ) . Furthermore , histidine , leucine , methionine , and uracil all showed to be part of this exometabolome ( Figure 1C ) . These measurements were obtained from cells in exponential growth phase , where apoptosis and necrosis are negligible . Comparing extracellular metabolite concentrations to intracellular levels ( the endometabolome ) we observed a general trend of correlation between the highest and lowest concentrated metabolites ( r2 = 0 . 517; Figure 1Di ) , but overall extracellular metabolite concentrations do not replicate the corresponding endometabolome . Tryptophan , phenylalanine , proline , and valine , for instance , were over-proportionally more concentrated inside the cell , whereas uracil , serine , tyrosine , and glycine were relatively over-represented in the extracellular fluid ( Figure 1Di ) . Instead , highly similar exometabolome concentration values ( r2 = 0 . 971 ) were observed in the related yeast strain BY4741 upon complementing its auxotrophies with the centromere-containing single-copy vector ( a minichromosome ) , 'pHLUM' , which contains all four marker genes ( Mülleder et al . , 2012 ) ( Figure 1Dii ) . Metabolite concentrations in the exometabolome between these two related yeast strains are hence substantially more similar than the endo- versus exometabolome in the same strain , implying that the intra-colony exometabolome is a distinct metabolite pool . A second requirement to establish metabolite exchange is that cells need to be able to sense extracellular metabolites and to exploit them as a nutrient source . Yeast is known to uptake amino acids when they are available extracellularly ( Stahl and James , 2014 ) . We tested how extensive this uptake was by comparing the uptake rates between auxotrophs and prototrophs . Remarkably , prototrophic cells consumed histidine , leucine , methionine , and uracil at a comparable rate to the genetic auxotrophs , who depend 100% on external metabolite pools ( Figure 1E ) . This demonstrated that yeast cells completely shift from de novo synthesis to uptake in the presence of each of the four metabolites . Studying the URA3 genotype in greater detail confirmed the preference of uptake over self-synthesis . Enzymes involved in uracil biosynthesis remained expressed in both the URA3 and the ura3Δ strains under fully supplemented conditions ( Figure 1—figure supplement 2 ) , but uracil biosynthesis-related intermediates shifted to similar concentrations both in the wild-type strain and in the ura3Δ strain once uracil was supplemented ( Figure 1F ) . The only exception was the direct substrate of the URA3 enzyme ( orotidine-5'-phosphate decarboxylase ) , orotidine-5'-phosphate ( OMP ) , which accumulated upon uracil supplementation once its metabolising enzyme ( URA3 ) was deleted ( Figure 1Fi ) . In summary , yeast cells do not compensate for metabolic deficiencies in co-culture experiments consistently as others reported previously ( Müller et al . , 2014; Shou et al . , 2007 ) , but they ( i ) export the relevant metabolites even when grown on minimal media and ( ii ) take up histidine , leucine , uracil , and methionine at similar rates to auxotrophs if supplementation is available . At least for uracil , ( iii ) the biosynthetic enzymes and majority of biosynthetic intermediates in the supplemented wild-type cell resemble those of the corresponding auxotroph . In light of these results , we speculated that the inability to cooperate could be found in the nature of the co-culturing experiment . To establish an alternative method , we made use of a , in other circumstances disadvantageous , property of yeast plasmids , their occasional , stochastic loss from cells ( segregation ) . Segregation is observed for both popular replication types , centromeric 'cen' and 2µ , at a rate of 2–4% expressed per cell division ( Christianson et al . , 1992 ) . This property allowed us to randomly and progressively introduce auxotrophies into a developing yeast community starting from a single , initially prototrophic , cell: when a plasmid carries a gene that complements for an auxotrophy , a newly budded cell re-gains the metabolic deficiency according to the segregation of its plasmid . We transformed plasmids from the classic pRS and p400 series which express HIS3 , LEU2 , MET15 , or URA3 genes under the respective S . cerevisiae promoters ( Christianson et al . , 1992; Mumberg et al . , 1995; Sikorski and Hieter , 1989 ) into the standard laboratory strain BY4741 , deficient in these markers ( Brachmann et al . , 1998 ) ( Figure 2Ai ) . As expected , the transformed cells grew competently in the absence of histidine , leucine , uracil , and methionine supplementations . We then quantified plasmid segregation and confirmed earlier literature values ( Figure 2Aii and Figure 2—source data 1 ) ( Christianson et al . , 1992; Ghosh et al . , 2007 ) . 10 . 7554/eLife . 09943 . 006Figure 2 . A self-establishing yeast community can cooperatively compensate for progressive loss of prototrophy on minimal media . ( A ) ( i ) Schematic illustration of BY4741 carrying four plasmids to complement its auxotrophies in histidine ( his3Δ1 ) , leucine ( leu2Δ0 ) , methionine ( met15Δ0 ) , and uracil ( ura3Δ0 ) . ( ii ) Plasmid segregation rates ( probability of plasmid loss per cell division ) of BY4741 carrying four plasmids encoding HIS3 ( p423 ) , LEU2 ( pRS425 ) , URA3 ( p426 ) , and MET15 ( pRS411 ) ( y-axis ) compared to BY4741 carrying one plasmid at a time ( x-axis ) . n = 3 , error bars = ± SD . Dashed line: linear regression fit . ( B ) Schematic illustration of the segregate strain composition over time on rich or complete media where no cooperation is necessary for cells to survive . Sequential plasmid loss leads to an increase in auxotrophy , with loss of up to four plasmids leading to the formation of 16 cell types with varying metabolic capacity ( metabotypes ) . ( C ) Three possible outcomes for BY4741 carrying four segregating plasmids , when establishing a colony on minimal media; ( i ) no cooperation , only cells carrying four plasmids grow , ( ii ) no cooperation but plasmid segregation is faster than the growth rate of cells carrying four plasmids leading to no growth capacity . Finally ( iii ) , cells cooperate , wherein cells that have obtained auxotrophy continue growth by sharing metabolites with neighbouring cells in the colony . ( D ) Auxotrophy of BY4741 colonies carrying single plasmids encoding HIS3 ( p423 ) , LEU2 ( pRS425 ) , URA3 ( p426 ) , and MET15 ( pRS411 ) on selective media after approximately 33 doublings . The number of plasmid-free cells ( % auxotrophy abundance ) was measured by replica plating . n = 3 , error bars = ± SD . ( E ) Mathematical simulation of segregation over time , starting from 100% cells carrying four plasmids , based on the experimentally measured segregation rates . Highlighted is the situation after 57 doublings ( achieved in dashed line ) where >99 . 9% of cells have segregated >1 plasmid . ( F ) Segregation over time in a colony on rich media ( no selection to maintain the plasmids ) ; starting from a micro-colony of four-plasmid prototrophic cells on minimal media , cells were transferred to rich ( YPD ) media and established as a giant colony , segregation was followed by replica plating . Biomass gain is counted from the single cell . ( G ) Giant colonies established for 57 biomass doublings on minimal media are composed of ( left ) 73 . 3% auxotrophic cells , ( centre ) contain a mixed number of auxotrophies and ( right ) a non 1:1 ratio of auxotrophy types . ( n = 542 genotyped cells ) . Colony growth is achieved , despite the majority of cells possessing one or more auxotrophies . DOI: http://dx . doi . org/10 . 7554/eLife . 09943 . 00610 . 7554/eLife . 09943 . 007Figure 2—source data 1 . Plasmid segregation rates . DOI: http://dx . doi . org/10 . 7554/eLife . 09943 . 00710 . 7554/eLife . 09943 . 008Figure 2—figure supplement 1 . Experimentally obtained colony composition , compared to the composition expected if segregation continued without selective pressure to maintain cells able to synthesise leucine , uracil , methionine and histidine . ( i ) ( left ) Colonies established for 57 biomass doublings on minimal media ( SM ) are composed of 73 . 3% auxotrophic cells ( n = 542 genotyped cells ) . ( right ) Uninterrupted segregation would lead to 99 . 9% auxotrophic cells . ( ii ) ( left ) Auxotrophy number ( 0 to 4 ) for cells within the synthetic metabolically cooperating colony ( SeMeCo ) . ( right ) Result when there is uninterrupted segregation ( theoretical composition ) . ( iii ) ( left ) Composition of SeMeCo in terms of auxotrophy type ( histidine , leucine , uracil and methionine ) . ( right ) Result when there is uninterrupted segregation ( theoretical composition ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09943 . 00810 . 7554/eLife . 09943 . 009Figure 2—figure supplement 2 . Schizosaccharomyces pombe , like Saccharomyces cerevisiae , are also able to establish SeMeCo colonies . The four possible metabotypes resulting from a combination of uracil and leucine auxotrophies are found within S . pombe SeMeCos , despite growing colonies on selective media ( n = 3 ) . Separately established populations ( >90 cells ) were genotyped per replicate . DOI: http://dx . doi . org/10 . 7554/eLife . 09943 . 00910 . 7554/eLife . 09943 . 010Figure 2—figure supplement 3 . Complementary pairs of auxotrophs , re-isolated from established SeMeCo colonies , do not overcome metabolic deficiencies upon co-culturing , similar to the original strains . his3∆ , leu2∆ , met15∆ , and ura3∆ yeasts isolated from SeMeCo colonies were combined in complementary pairs and spotted on corresponding selective media . No pairs exhibited co-growth together , indicating that SeMeCo metabotypes did not acquire secondary mutations to overcome metabolic deficiencies , while establishing a cooperating community . DOI: http://dx . doi . org/10 . 7554/eLife . 09943 . 01010 . 7554/eLife . 09943 . 011Figure 2—figure supplement 4 . Different auxotrophy combinations do not enable metabolic cooperation . ( left ) Quadruple mixed cultures of Saccharomyces cerevisiae auxotrophs do not overcome metabolic deficiencies upon co-culturing . his3Δ , leu2Δ , met15Δ , and ura3Δ yeasts were combined together in mixed ratios and spotted as a co-culture on corresponding selective media . ( right ) A similar outcome upon co-culturing the four genotypes over night in rich media prior to transferring to minimal media . DOI: http://dx . doi . org/10 . 7554/eLife . 09943 . 011 Cells having lost prototrophy can only continue growth if they obtain the relevant nutrient from the environment ( Figure 2B ) . Transferred to minimal media , the lack of nutrient supplementation leads to three possible outcomes ( Figure 2C ) : First , if the cooperative potential would not suffice to overcome the increasing content of metabolically deficient cells , colony growth would only be explained by cells maintaining all four plasmids ( Figure 2C left ) . Alternatively , if the segregation is faster than the growth rate of cells carrying four plasmids , the colonies would not be able to grow ( Figure 2C centre ) . Finally , the third outcome is that colony growth continues , despite an increasing auxotrophic composition , facilitated by cells exchanging histidine , leucine , uracil , and methionine at growth relevant quantities ( Figure 2C right ) . First , we observed that upon approximately 33 biomass doublings , segregation for HIS3 and URA3 had continued until less than 50% of cells were prototrophs ( Figure 2D ) , even though a 1:1 co-culture of the same auxotrophs was not able to co-grow ( Figure 1A ) . Then , we assayed for the formation of a heterogeneous yeast community , starting from the four-plasmid ( 4P ) strain , that can give rise to the emergence of 16 complementary auxotrophic genotypes ( Figure 2B ) . In the 4P strain , individual plasmid segregation rates were similar but not identical to yeast carrying one plasmid at a time and were in linear correlation , indicating that no specific interaction between the plasmids occurred ( Figure 2Aii ) . With a total segregation rate of 11% , 4P cells regain auxotrophy rapidly so that only 21 cell divisions ( doublings ) would result in >90% of cells losing prototrophy ( Figure 2E ) . The continuous loss of prototrophy from the 4P strain was experimentally confirmed on rich ( YPD ) media; In the presence of rich supplementation , only 45 biomass doublings resulted in 96% of cells losing prototrophy ( Figure 2F ) . Testing whether cells can maintain growth by cooperating in the biosynthesis of histidine , leucine , uracil , and methionine , colonies were grown over 7 days on minimal media agar through dilution and re-spotting once giant colonies had formed ( every 48 hr ) , so that the continuous gain in biomass necessitates constant de novo synthesis of intermediate metabolites . The experiment yielded viable colonies and from the obtained biomass , we calculated that 57 doublings had occurred . Fifty-seven doublings would have been sufficient for >99% of cells to segregate ( Figure 2E ) . Replica plating revealed a predominantly auxotrophic composition of the obtained colonies . These were composed of 73 . 3 ± 3 . 7% auxotrophic cells ( Figure 2G left ) , of which 39 . 9% had lost one plasmid , 20 . 6% two , 6 . 6% three , and 6 . 1% had lost all four markers ( Figure 2G centre ) . No auxotrophies were in a 1:1 ratio with each other ( 36 . 9% for uracil , 27 . 7% leucine , 23 . 7% histidine , and 11 . 7% methionine ( Figure 2G right ) , despite the segregation rates predicting a relatively equal distribution , implying that selection pressure for certain metabotypes affected colony composition ( Figure 2G right and Figure 2—figure supplement 1 ) . We also confirmed that S . pombe is capable of forming similar communities , indicating conservation in these evolutionary distant yeast species ( Figure 2—figure supplement 2 ) . As additional controls , we ( i ) re-isolated the auxotrophs from the established colonies , and repeated the co-culture experiment , after having grown the cells for 48 hr in supplemented media , as with the original strains ( Figure 1A ) . Even when isolated from a functional cooperating colony , complementary auxotrophic cells did not complement each other's deficiencies upon co-culturing ( Figure 2—figure supplement 3 ) , ruling out the possibility that new mutations altering metabolite exchange capacities could explain the formation of the cooperating community . We also mixed all four auxotrophs together , both in a 1:1 mixture , as well as in the ratio observed from the community and performed co-culturing both with and without co-cultivation before spotting; These attempts did not result in successful co-growth either ( Figure 2–figure supplement 4 ) . Hence , by exploiting plasmid segregation to overcome culturing and allowing the community to self-establish , heterogeneous yeast colonies were formed , which could sustain exponential growth under nutrient limitation , despite the majority of cells being auxotrophic for at least one metabolite . These co-growing cells could therefore overcome metabolic deficiencies through cooperative metabolism , demonstrating that yeast possesses metabolite exchange capacities at a growth relevant quantity . The obtained colonies were viable on minimal media and showed no apparent growth defects , despite containing a content of 73% auxotrophs , each of which were non-viable in co-culture studies ( Figure 1A , B , Figure 1—figure supplement 1 , Figure 2—figure supplement 3 , 4 ) . To characterise the properties of this community , we started with LC-MS/MS to compare its exometabolome against prototrophic yeast strains ( YSBN5 , BY4741-pHLUM ) , and the unpassaged strain carrying the four plasmids ( 4P ) ; ( Figure 3A , B and Figure 3—source data 1 ) . SeMeCo colonies possessed similar extracellular metabolite concentrations to prototrophic controls ( Figure 3Bi ) . Of particular note are the extracellular concentrations of H , L , U , and M . Aside from a statistically non-significant trend towards a lower leucine concentration , only uracil ( U ) was significantly affected . To our surprise , however , the concentration of this metabolite was increased , indicating that SeMeCo had adapted by maintaining a higher level of uracil in its exometabolome ( Figure 3Bii ) . 10 . 7554/eLife . 09943 . 012Figure 3 . Growth and physiological parameters of the self-established metabolically cooperating yeast community 'SeMeCo' . ( A ) Schematic illustration of colonies derived from the genomically prototrophic yeast strain YSBN5 , the single-vector complemented BY4741-pHLUM ( 'pHLUM' ) , BY4741 complemented with four plasmids ( 'FourP' ) , and the self established yeast population ( SeMeCo; self-established metabolically cooperating yeast community ) ; ( from left to right ) . ( B ) ( i ) Extracellular concentrations of metabolites in colonies of YSBN5 , pHLUM and SeMeCo growing exponentially on minimal media as determined by LC-MS/MS , n = 3 . Histidine ( H ) , leucine ( L ) , methionine ( M ) , and uracil ( U ) are highlighted in red circles . ( ii ) Detailed extracellular concentration values of uracil , leucine , methionine , and histidine as determined by LC-MS/MS . n = 3 , error bars = ± SD . ( C ) ( left ) Growth curve of YSBN5 , pHLUM , FourP , and SeMeCo as determined by measuring optical density ( OD595 ) . n = 3 , error area = ± SD . ( centre ) Dry biomass collected from 100 mL batch cultures after three days growth in minimal media , 30°C , n = 3 , error bars = ± SD . ( right ) Maximum specific growth rate ( µmax ) as determined from OD595 growth curves using a model-richards fit ( Kahm et al . , 2010 ) . n = 3 , error bars = ± SD . ( D ) The ratio of colony-forming units ( CFUs ) to number of cells used for plating , for YSBN5 and SeMeCo . n = 3 , error bars = ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 09943 . 01210 . 7554/eLife . 09943 . 013Figure 3—source data 1 . Absolute quantification of amino acids and uracil in yeast strains YSBN5 , pHLUM and SeMeCo , absolute concentration values . DOI: http://dx . doi . org/10 . 7554/eLife . 09943 . 013 SeMeCo also maintained wild-type like growth efficiency and biomass-forming capacities under nutrient limitation ( Figure 3C ) . Comparing SeMeCo against prototrophic yeast strains ( Figure 3A ) , dry biomass formation did not vary significantly ( Figure 3C centre; p-values = 0 . 27 , 0 . 70 , and 0 . 09 for FourP , pHLUM , and YSBN5 , respectively ) . In liquid media , lag phase was prolonged , and the maximum specific growth rate ( µmax ) was slightly reduced compared to the genetically prototrophic YSBN5 or pHLUM cells ( 0 . 17 OD595/hr vs 0 . 20 and 0 . 21 OD595/hr , respectively ) ( Figure 3C right ) . However , this difference appeared more as a cost of plasmid segregation , as both lag phase and µmax did not vary significantly between SeMeCo and the FourP strain ( 0 . 17 and 0 . 16 OD595/hr , respectively ) ( Figure 3C right ) . Finally , we tested to what extent cell death occurs in the cooperating community . Both a wild-type ( YSBN5 ) and a SeMeCo culture were grown to exponential phase and cells were counted . Then , the cultures were plated on SC media and the number of colony-forming units ( CFUs ) determined . The CFU count was nearly equal between SeMeCo and YSBN5 , and similar to a 1:1 relationship to the cell count measured prior to spotting ( Figure 3D ) . This indicates that cells in SeMeCo have a comparable colony-forming capacity to that of exponentially growing wild-type cells , and in both populations , virtually every cell can form a new colony . To establish if cells cooperating in SeMeCo are distributed in a random or organised manner , we analysed colony spatial structure using confocal fluorescence microscopy ( Figure 4A ) . For this , the community was re-established with alternative plasmids that express the fluorescent protein markers CFP ( cyan fluorescent protein ) , Venus ( yellow fluorescent protein ) , Sapphire ( a UV-excitable green fluorescent protein [Sheff and Thorn , 2004] ) , and mCherry ( red fluorescent protein ) coupled to the auxotrophic markers HIS3 , URA3 , LEU2 , and MET15 , respectively ( Bilsland et al . , 2013 ) . Segregation of the labelled plasmids were within the same range , although not identical to the original pRS and p400 plasmids ( Figure 4—figure supplement 1 ) . Images were acquired with a SP5 on a DMI6000 inverted microscope ( Leica , Wetzlar , Germany ) and show the underside of a live two day micro-colony which had , prior to imaging , been growing on minimal media ( SM ) . In our hands , the Sapphire-LEU2 fluorescence was also visible under the imaging conditions used to visualise Venus-URA3 . For this reason , the Venus-URA3 channel was removed from the colony image . The spatial heterogeneity of fluorescent markers in the micro-colony revealed that cells in SeMeCo unequally distribute over the macroscopic structure , and form regions where the biosynthesis of a particular metabolite dominates ( Figure 4A ) . A prediction in truly cooperating communities is , however , whether complementary cells maintain physical proximity to each other , to oppose diffusion of exchanged metabolites ( Müller et al . , 2014 ) . Using computational image analyses of colony micrographs , we find that even when the most stringent cut-off was applied , complementary metabotypes across the community maintained an average distance ( 6 . 86 μm ) of less than two cell diameters ( Figure 4B ) . Cells are hence most likely to exchange the majority of metabolites with those maintaining close proximity . Despite these results , SeMeCo could , however , continue growth after disruption of this spatial structure in liquid media ( Figure 3C ) . To verify this assumption , SeMeCo was replicated for 7 days in liquid minimal media , as previously , with re-dilution every 2 days . Indeed , the liquid culture maintained a similar content of auxotrophs as obtained with colony grown SeMeCos ( Figure 4—figure supplement 2 ) . The capacity of SeMeCos to overcome metabolic deficiencies through metabolic cooperation is hence not in essence bound to colonial growth . 10 . 7554/eLife . 09943 . 014Figure 4 . Spatial organisation of SeMeCo . ( A ) Spatial organisation of metabolically cooperating yeast micro-colony on minimal agar media ( SM ) . SeMeCo was established with plasmids expressing fluorescent protein coupled to each auxotrophic marker; LEU2 , MET15 , and HIS3 cells are coloured green , red , and blue , respectively . Cells containing more than one marker are coloured as a product of the additive RGB colour model . Two–day-old live and growing micro-colony is visualised from underneath . ( B ) Minimum , mean and maximum distances between leucine , histidine , and methionine auxotrophs and their corresponding prototrophs in a SeMeCo colony . Maximum distance between auxotroph and prototroph for 90% of cells shows an average distance of 6 . 86 μm , using the highest cut-off . Despite the heterogeneous macroscopic colony composition , complementary auxotrophs are maintained in physical proximity to each other . DOI: http://dx . doi . org/10 . 7554/eLife . 09943 . 01410 . 7554/eLife . 09943 . 015Figure 4—source data 1 . Segregation rates of fluorescent protein plasmids from the yEp , pRS and p400 series . DOI: http://dx . doi . org/10 . 7554/eLife . 09943 . 01510 . 7554/eLife . 09943 . 016Figure 4—figure supplement 1 . Plasmid segregation rates of fluorescent protein plasmids ( %; probability of plasmid loss per cell division ) of BY4741 carrying plasmids encoding HIS3 ( yEpCFP_HIS ) , LEU2 ( yEpSapphire_LEU ) , URA3 ( yEpVenus_URA ) , and MET15 ( pRS411-GPDpr-mCherry ) respectively , compared to BY4741 carrying all four at the same time . n = 3 , error bars = ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 09943 . 01610 . 7554/eLife . 09943 . 017Figure 4—figure supplement 2 . SeMeCos continue growth in minimal ( SM ) liquid culture . As in the colony growth experiment , cells were transferred from a micro-colony on SM agar , and then grown for 7 days with re-diluting every 2 days , however , here cells were in shaking batch liquid culture ( 25 mL ) . Auxotrophy abundance is 70 . 3 ± 2 . 5% despite cells growing in minimal media . DOI: http://dx . doi . org/10 . 7554/eLife . 09943 . 017 To determine not only the spatial but also the population structure , we switched back to the non-fluorescent SeMeCo to avoid confounding effects of fluorescent protein expression , and quantified by replica plating the colony contribution of all 16 possible metabotypes , resulting from all possible combinations of the four auxotrophies ( Figure 2B ) . These experiments revealed that within SeMeCo , 95 . 6% of cells belonged only to 8 of the 16 possible metabolic combinations ( Figure 5A ) . We questioned whether this composition was the result of a stochastic event , however , the dominance of the same metabotypes establish three times independently . Moreover , the eight successful metabotypes contributed to SeMeCo at comparable percentages ( Figure 5A inset ) . Using our segregation rate model as well as growth rate data , we could rule out this colony composition being a result of ( i ) varying plasmid segregation rate , ( ii ) the number or type of auxotrophy , or ( iii ) differences in growth rates . First , a community composition calculated on the basis of the experimentally determined plasmid segregation values ( Figure 2E , Figure 2—figure supplement 1 ) showed zero correlation with the actual population composition ( r² = 0 . 051 ) ( Figure 5B ) . Second , all histidine , leucine , uracil , or methionine auxotrophies , as well as all plasmid numbers ( 1 to 4 ) were found amongst both the frequent and rare metabotypes . For instance , while single uracil ( HIS3 , LEU2 , MET15 , ura3∆; 19 . 7% ) or leucine ( HIS3 , URA3 , MET15 , leu2∆; 11 . 1% ) auxotrophs were amongst the most frequent cells , their methionine-deficient counterparts ( HIS3 , URA3 , LEU2 , met15∆; 0 . 4% ) were among the most rare ( Figure 5A ) . Also , the high frequency of the dual auxotrophs LEU2 , MET15 , his3∆ , ura3∆ ( 8 . 1% ) and HIS3 , MET15 , leu2∆ , ura3∆ ( 10 . 1% ) , contrasts with the rareness of the other dual auxotrophs ( HIS3 , URA3 , leu2∆ , met15∆ ( 1 . 1% ) , MET15 , URA3 , leu2∆ , his3∆ ( 0 . 2% ) , URA3 , LEU2 , his3∆ , met15∆ ( 0 . 5% ) , and LEU2 , HIS3 , ura3∆ , met15∆ ( 0 . 6% ) ) . Thus , the number of plasmids or type of auxotrophy a cell had did not indicate whether a cell-type would be rare or frequent ( Figure 5A ) . Finally , the growth rate of 16 strains , carrying the same marker and supplement combination that replicates the 16 metabotypes ( Mülleder et al . , 2012 ) did not distinguish the depleted from the selected metabotypes either ( Figure 5C ) . 10 . 7554/eLife . 09943 . 018Figure 5 . The community composition is distinct and dynamic . ( A ) Frequency of the 16 metabotypes that result from combination of histidine , leucine , methionine , and uracil auxotrophies as found within SeMeCo colonies . Separately established populations ( n=3 ) were genotyped ( >180 cells per colony ) and eight metabotypes showed to dominate in the populations ( inset ) . Frequency of the 16 metabotypes in independently established cell populations . The eight metabotypes of low frequency , which were depleted in all experiments , are highlighted with a red circle . ( B ) No correlation shown between the frequency of the 16 metabotypes in SeMeCo versus a segregation rate-predicted colony composition . Coloured points correspond to the experimentally observed eight frequent metabotypes . Dashed line: linear regression fit . ( C ) Maximum specific growth rate ( µmax ) in supplemented minimal media ( red dots ) , of the 16 strains carrying HIS3 , LEU2 , URA3 , and MET15 plasmids in all combinations obtained from ( Mülleder et al . , 2012 ) , relative to the frequency of the specific metabotype in SeMeCo ( light blue ) . Dashed line indicates average µmax for the eight most and least frequent metabotypes within SeMeCo colonies . ( D ) SeMeCo re-established on minimal media supplemented with uracil . After 7 days of growth ( with re-spotting every two days ) , SeMeCo adapted with an entirely different composition of metabotypes ( green box plot ) compared to original SeMeCo colony composition ( grey bars ) . Uracil producing cells decline , including the FourP genotype , so that 97% of cells are cooperating auxotrophs . DOI: http://dx . doi . org/10 . 7554/eLife . 09943 . 01810 . 7554/eLife . 09943 . 019Figure 5—figure supplement 1 . Abundance over time of a fluorescent labelled frequent ( HIS3 , LEU2 , MET15 , ura3∆ ) and rare ( his3∆ , leu2∆ , URA3 , MET15 ) genotype spiked into SeMeCo , as measured by FACS . ( Metabotype frequency determined from SeMeCo colony , Figure 5A ) . ( left ) % Fluorescence of the frequent and rare metabotype established individually as a colony shows frequent and rare abundance respectively ( unlabelled prototroph control is BY4741-pHLUM ) . ( right ) Frequent and rare metabotypes spiked into pre-established SeMeCo shows depletion of both cell types after approx . 48 hr . n = 3 , error bars = ± SD . FACS: Fluorescence-activated cell sortingDOI: http://dx . doi . org/10 . 7554/eLife . 09943 . 019 As growth potential and segregation parameters did not explain the population architecture of SeMeCo , we conclude that this community was selected for on its ability to cooperate effectively . If this interpretation is correct , it would imply that a different pressure to cooperate would result in a different SeMeCo composition . To test this hypothesis , we focussed on uracil , as mass spectrometry had detected an increase in uracil concentration in the SeMeCo colony exometabolome , indicating that uracil is the most limiting metabolite ( Figures 2G right and 3Bii ) . SeMeCos established on uracil adapted a different composition , resulting from a dramatic decline in cells needed to produce uracil . Importantly , this included the prototroph with its total content in the community decreasing from 26 . 7% in the original SeMeCo to solely 3 . 0% , so that 97 . 0% of cells were cooperating auxotrophs ( Figure 5D ) . Hence , SeMeCo colonies establish a population that is dynamic to changes in the external metabolite pool , and can persist in a state with virtually all cells being genetically auxotrophic for at least one essential metabolite . Metabolic exchange interactions occur frequently among cells that grow in proximity to one another , as metabolites are constantly released from cells for different reasons , such as overflow metabolism , metabolite repair , as well as export to facilitate metabolite exchange . In bacteria , a subset of such metabolite exchanges are of a cooperative nature in the sense that all exchange partners profit from this situation ( Oliveira et al . , 2014 ) , whereas for the majority of eukaryotic organisms , metabolite exchange strategies remain unclear . Despite yeast auxotrophs being viable in supplemented and rich media ( Mülleder et al . , 2012 ) , in the absence of amino acid supplementation , they fail to complement metabolic deficiencies in several pairs or higher order co-culture experiments ( Figure 1A , B , [Müller et al . , 2014; Shou et al . , 2007] ) , a clear difference to bacterial studies , where similar experiments were effective ( Foster and Bell , 2012; Freilich et al . , 2011; Harcombe , 2010; Pande et al . , 2014; Ramsey et al . , 2011; Vetsigian et al . , 2011 ) . This led to speculations that yeast might , in contrast to many bacterial species , lack the required export capacities to enable growth relevant exchange of intermediary metabolites such as amino acids and nucleobases ( Shou et al . , 2007 ) . Analysing the intra-colony exometabolome we could , however , detect the required metabolites; in fact , we found that cells within a colony are surrounded by a rich exometabolome . We also found that yeast would efficiently exploit the nutrients when available , to the extent that they solely rely on these extracellular metabolites . This result implied that metabolite exchange among co-growing yeast cells is frequent by nature; the lack of complementation in the co-culture experiments could thus reflect a limit of the experiment itself , and not represent the metabolite exchange capacities of yeast cells . To circumvent combining two or more cultures , we chose an approach of synthetic biology and exploited the stochastic loss of plasmids to progressively introduce the metabolic deficiencies in random combination from an initially single cell . The progressive loss of prototrophy allowed cells to maintain cell growth on the basis of metabolite exchange , resulting in a community with 73% auxotrophy , which increased to 97% upon supplementation with the most limiting metabolite , uracil . Despite its dominant auxotrophic composition , the SeMeCo community could maintain a wild-type like exometabolome , metabolic efficiency , as well as cell viability , implying that this type of cooperation is a robust physiological property . Hence yeast's natural metabolite export and import capacities are wholly sufficient to support co-growth on the basis of metabolite exchange . The establishment of SeMeCos was not facilitated by mixing the auxotrophs in a higher order combination either . This is consistent with the notion that losing more metabolic genes reduces biochemical capabilities and does not add new ones . The key of the SeMeCo system is instead to allow the progressive self-establishment of the community starting from the single cell ( Figure 2 ) . Metabolic feedback regulatory systems therefore do not inhibit metabolite export in general but prevent cooperative co-growth when already pre-established co-cultures are mixed ( Müller et al . , 2014; Shou et al . , 2007 ) . A possible role of these mechanisms could perhaps prevent the spread of foreign , potentially cheating , cells that derive from a competing yeast colony . We could replicate behaviour which is in favour of such an assumption; By spiking into SeMeCo a cell culture possessing the same genotype as a frequent ( HIS3 LEU2 ura3Δ MET15 ) and rare ( his3Δ leu2Δ URA3 MET15 ) genotype ( Figure 5A ) , we observed that both genotypes were rapidly depleted from the pre-established SeMeCo , irrespective of the frequency of the respective genotype in SeMeCo ( Figure 5—figure supplement 1 ) . Studying the genotypic composition of SeMeCo implied that there are a defined set of interactions underlying the properties of the cooperating community , which maintained a similar population composition involving eight reproducibly concentrated metabotypes when established independently . This indicates that this quantitatively defined community composition was most effective in metabolic cooperation . This finding may close an important gap in the understanding of the evolution of multicellularity; If a defined composition is most effective in cooperative growth , a selection advantage could be provided by any sort of physical bonding which can maintain cooperation partners in the defined equilibria , and would in addition , provide additional protection against the invasion of cheating cell types . Indeed , the exometabolome data implies that the number of metabolite exchange interactions among co-growing cells could be significant . The colony exometabolome contained a vast array of biomolecules , including the majority of amino acids ( Figure 1C ) ( Castrillo et al . , 2007; Paczia et al . , 2012; Silva and Northen , 2015 ) . The finding that yeast cells prefer uptake over synthesis of amino acids and uracil , even when they are genetically prototrophic , shows that exchange interactions will readily establish once the cellular environment has acquired a critical concentration of metabolites . This has implications for the interpretation of metagenomic studies and cheating/benefactor experiments , as for this reason , it cannot be concluded from the genetic presence or absence of a single metabolic pathway , or from following the synthesis of a single metabolite , how many other metabolites are being exchanged as well . Even without selective pressure , both wild-type yeasts and SeMeCo established an amino-acid-rich exometabolome on minimal media ( Figure 1C ) and engaged in metabolite uptake when nutrients were available ( Figure 1E ) . This implies that these features are a natural property of yeast and raises the question of why natural yeast communities are not composed of co-growing , genetic , auxotrophs . To answer this question , one needs to keep in mind that possessing metabolic genes in the genome is not equal to the pathway being constantly active; Indeed , prototrophs can flexibly switch from self-synthesis to amino acid uptake ( Figure 1 ) . Being genetically prototrophic hence gives a higher level of metabolic flexibility , as prototrophic cells can re-activate a synthetic pathway when required . Unlike in SeMeCo , the genotype of a cell in a natural community is not in essence reflecting its metabotype or its metabolic role in the community . Additionally , the natural life cycle of yeast involves the formation of endospores , which are important for enduring starvation , and to spread between habitats . Without a prototrophic genotype , a single spore can no longer establish a colony on its own as genetic auxotrophy would interrupt the yeast life cycle . Second , only a fraction of the natural yeast life cycle occurs under exponential growth that requires abundant carbohydrate and nitrogen supply . The maintenance of a prototrophic genotype both in S . pombe and in S . cerevisiae wild isolates ( Jeffares et al . , 2015; Liti et al . , 2009 ) is hence fully compatible with the presence of elaborate amino acid and nucleotide exchange mechanisms . The finding that these cells fully shift from self-synthesis to uptake for histidine , leucine , methionine , and uracil , once these metabolites are provided , has direct implications on research using yeast , a primary eukaryotic model organism in genome-scale studies . A majority of yeast genetic experiments are conducted in auxotrophic strains , requiring amino-acid supplemented or rich media compositions . Important parts of biosynthetic metabolism ( amino acid biosynthesis can account for up to 50% of metabolic flux towards biomass ) may have thus stayed silent in a significant amount of functional genomics experiments . The effects of metabolic–genetic interactions on cellular physiology could thus substantially exceed our current knowledge and could be discovered upon switching to minimal nutrient supplementations . In this context , SeMeCos are simple to handle , establish rapidly and are easy to analyse , and therefore represent an effective and broadly applicable eukaryotic model system to study both cooperativity and effects of metabolism in the laboratory . In summary , using histidine , leucine , methionine , and uracil as model metabolic pathways for exchangeable metabolites , we found that S . cerevisiae cells prefer these nutrients' uptake over their self-synthesis and maintain an amino-acid-rich exometabolome in the extracellular colony space , indicators of ongoing inter-cellular metabolite exchange . Although yeast is known to fail in compensating for auxotrophy in pairwise and higher order co-culture experiments , the cells did successfully enter a state of metabolic cooperative growth upon exploiting stochastic plasmid segregation so that a single cell could progressively develop into a complex heterogeneous community . Composed of auxotrophic cell types that are non-viable on their own , SeMCo communities were able to overcome metabolic deficiencies and maintain metabolite concentrations and robustness similar to wild-type cells . Additionally , cooperation had imposed different metabolic roles on contributing cells . Progressive community formation thus reveals that yeast possesses full capacity to exchange anabolic metabolites at growth relevant quantities and readily establishes a non-cell-autonomous metabolism within complex but defined community structures . Yeast cells were grown under standard conditions on synthetic minimal ( SM or EMM ) , SC and rich ( YPD or YES ) media . Plasmid segregation was calculated according to Christianson et al . ( 1992 ) , by monitoring plasmid retention after cells are shifted from non-selective to selective media , and by expressing the number of cells that have lost the marker as a function of generation time . Metabolites were quantified after quenching using an online UPLC-coupled 6460 ( Agilent Technologies , Waldbronn , Germany ) triple quadrupole mass spectrometer . Confocal fluorescence microscopy was conducted with a SP5 confocal on a DMI6000 inverted microscope ( Leica ) using a 10x/0 . 3 HC PL Fluotar Air objective . All experiments involved , unless otherwise indicated , used BY4741 yeast strain ( his3Δ1 , leu2Δ0 , ura3Δ0 , met15Δ0 ) ( Brachmann et al . , 1998 ) with prototrophy restored by complementation either with vectors p423 ( HIS3 ) , pRS425 ( LEU2 ) , p426 ( URA3 ) , and pRS411 ( MET15 ) ( Christianson et al . , 1992; Mumberg et al . , 1995; Sikorski and Hieter , 1989 ) , with the centromeric vector ( minichromosome ) pHLUM ( Mülleder et al . , 2012 ) ; Addgene number: 40276 ) , or with the fluorescent protein vectors yEpCFP_HIS ( HIS3 ) , yEpSapphire_LEU ( LEU2 ) , yEpVenus_URA ( URA3 ) ( Bilsland et al . , 2013 ) , and pRS411-GPDpr-mCherry ( MET15 ) ( Table 1 ) . Cloning was conducted according to standard procedures; oligonucleotides are listed in Table 2 . 10 . 7554/eLife . 09943 . 020Table 1 . Strains and plasmids used in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 09943 . 020NameDescriptionReferenceStrainsBY4741MATa , his3∆1 leu2∆0 met15∆0 ura3∆0 ( ATCC 201388 ) ( Brachmann et al . , 1998 ) BY4741 FLO+ Derived from tetrad dissection after crossing and sporulating a flocculating BY4741 strain derived from the knock out collection ( ∆tpo1 ) with BY4742 and isolating a FLO+ TPO1 wild-type progeny This studyYSBN5MATa , FY3 ho::Ble ( Canelas et al . , 2010 ) ED666h+ ade6-M210 ura4-D18 leu1-32Bioneer Cat . No . M-3030HPlasmids p423GPD2 µ vector with HIS3 marker ( Mumberg et al . , 1995 ) pRS4252 µ vector with LEU2 marker ( Christianson et al . , 1992 ) p426GPD2 µ vector with URA3 marker ( Mumberg et al . , 1995 ) pRS411Yeast centromeric vector with MET15 marker ( Brachmann et al . , 1998 ) pHLUMYeast centromeric vector with HIS3 , URA3 , LEU2 and MET15 markers ( minichromosome ) . ( Addgene number: 40276 ) ( Mülleder et al . , 2012 ) pFS118Yeast high-copy vector with endogenous promoter forura4+ ( Addgene number: 12378 ) ( Sivakumar et al . , 2004 ) pREP41-MCS+Yeast high-copy vector with endogenous promoter for LEU2 . ( Addgene number: 52690 ) A gift from Michael Nick Boddyp416GPDYeast centromeric vector with endogenous promoter for URA3 ( Mumberg et al . , 1995 ) pHS12-mCherryYeast vector with mCherry fluorescent tag and LEU2 marker . ( Addgene number: 25444 ) A gift from Benjamin Glickp426-GPDpr-mCherryYeast 2 µ vector with endogenous promoter for URA3 marker and a GPD promoter for mCherry fluorescent tagThis study . Derived from p416GPD , pHS12-mCherry and p426GPDpRS411-GPDpr-mCherryYeast 2 µ vector with endogenous promoter for MET15 marker and a GPD promoter for mCherry fluorescent tagThis study . Derived from p426-GPDpr-mCherry and pRS411yEpVenus_URAYeast 2 µ vector with TDH3-promoter-driven Venus ( YFP ) and URA3 marker ( Bilsland et al . , 2013 ) yEpCFP_HISYeast 2 µ vector with TDH3-promoter-driven CFP and HIS3 marker ( Bilsland et al . , 2013 ) yEpSapphire_LEUYeast 2 µ vector with TDH3-promoter-driven Sapphire ( a UV-excitable GFP ) and LEU2 marker ( Bilsland et al . , 2013 ) pHLM-GPDpr-mCherryYeast centromeric vector with a GPD promoter for mCherry fluorescent tagThis study . Derived from pHLUMpUM-GPDpr-mCherryYeast centromeric vector with a GPD promoter for mCherry fluorescent tagThis study . Derived from pHLUM10 . 7554/eLife . 09943 . 021Table 2 . Oligonucleotides used to create expression plasmid p426-GPDpr-mCherry . DOI: http://dx . doi . org/10 . 7554/eLife . 09943 . 021NameSequencemCherry_Bam_Sac_fwAAGAAGAGCTCAAAAGGATCCGGGATGGTGAGCAAGGGCGAGGmCherry_Xho_rvCCTTTTCTCGAGCTTGTACAGCTCGTCCATGC All experiments involving wild-type yeast strains were carried out using as indicated , and using YSBN5 , a prototrophic haploid variant of S . cerevisiae S288c ( Canelas et al . , 2010 ) . For microscopy analyses of colony spatial organisation , BY4741 had prototrophy restored by complementation with the above fluorescent protein vectors ( Table 1 ) . For fluorescence-activated cell sorting ( FACS ) , BY4741 was used with mCherry labelled derivatives of pHLUM , pHLM-GPDpr-mCherry , and pUM-GPDpr-mCherry ( Table 1 ) . For S . pombe experiments , ED666 yeast strain ( ade6-M210 ura4-D18 leu1-32 ) was used that had uracil and leucine prototrophy restored by complementation with vectors pFS118 ( ura4+ ) and pREP41-MCS+ ( LEU2 ) ( Table 1 ) . Yeast was cultivated if not otherwise indicated at 30ºC , in rich ( YPD; 1% yeast extract [Bacto] , 2% peptone [Bacto] or YES; [Formedium; 35 . 25 g/L] ) , complete supplemented synthetic media ( SC; CSM complete supplement mixture [MP Biomedicals; 0 . 56 g/L] , YNB , yeast nitrogen base [Sigma; 6 . 8 g/L] ) , or minimal supplemented synthetic media ( SM; YNB [Sigma; 6 . 8 g/L] or EMM; [Formedium; 32 . 3 g/L] ) , with 2% glucose ( Sigma ) as the carbon source . Media recipes and amino acid compositions for S . cerevisiae were used as previously published ( Mülleder et al . , 2012 ) . For S . cerevisiae , auxotrophic derivatives of prototrophic BY4741 ( Mülleder et al . , 2012 ) were cultured alone or mixed in combination with other auxotrophs , and 1 . 1e05 cells of individual or mixed auxotrophs were spotted on respective selective media . Growth was then documented after 2 days incubation at 30°C . For the flocculation experiments , a FLO+ derivative of BY4741 was obtained by back-crossing and tetrad dissection of a tpo1Δ ( YLL028W ) strain obtained from Euroscarf ( Frankfurt , Germany ) . For S . pombe , 1 . 9e04 cells of auxotrophic derivatives of ED666 h+ , prototrophic for leucine and uracil , were spotted alone or mixed together on corresponding selective media . Growth was then documented after 2 days incubation at 30°C . All proteogenic amino acids ( except for cysteine ) and uracil , citrulline , and ornithine were analysed by selective reaction monitoring ( SRM ) using an online coupled UPLC ( 1290 Infinity , ( Agilent ) ) / triple quadrupole mass spectrometer ( 6460 , ( Agilent ) ) system . The compounds were separated by hydrophilic interaction chromatography on an ACQUITY UPLC BEH amide column ( 2 . 1 mm × 100 mm , 1 . 7 µm ) by gradient elution . Solvent A consisted of 95:5:5 acetonitrile:methanol:water , 10 mM ammonium formate , 0 . 176% formic acid , and solvent B of 50:50 acetonitrile:water , 10 mM ammonium formate , 0 . 176% formic acid . The gradient conditions were 0–07 min 85% B , 0 . 7–27–2 . 55 min 85–585–5% B , 2 . 55–255–2 . 75 5% B , 2 . 75–275–2 . 8 min 5–855–85% B and 2 . 8–38–3 . 25 min 85% B at a constant flow rate of 0 . 9 mL/min and 25°C column temperature . SRM ( Q1/3 settings ) are given in Table 3 . Metabolite signals were automatically integrated using Masshunter ( Agilent ) corrected after manual inspection and quantified by external calibration . 10 . 7554/eLife . 09943 . 022Table 3 . SRM transitions for quantification of amino acids and uracil . DOI: http://dx . doi . org/10 . 7554/eLife . 09943 . 022Compound nameCompound abbreviationSRM transitionFragmentor ( V ) Collision energy ( V ) PolarityUracilU111 . 0 > 42 . 1629-−PhenylalanineF166 . 1 > 1201009+LeucineL132 . 1 > 86808+TryptophanW205 . 1 > 188855+IsoleucineI132 . 1 > 86808+MethionineM150 . 1 > 104408+TaurineTau126 > 44 . 111016+ValineV118 . 1 > 71 . 910010+ProlineP116 . 1 > 70 . 110013+TyrosineY182 > 165905+AlanineA90 > 44 . 1508+ThreonineT120 . 1 > 74809+GlycineG76 > 30 . 1505+GlutamineQ147 . 1 > 845016+GlutamateE148 . 1 > 84 . 17510+SerineS106 > 60409+AsparagineN133 . 1 > 74809+AspartateD134 . 1 > 748010+HistidineH156 . 1 > 110 . 28012+ArginineR175 . 1 > 7010015+LysineK147 . 1 > 845016+CitrullineCit176 > 159604+OrnithineO133 > 709010+ Cells were spotted on SM solid media and incubated at 30°C in FLUOstar OPTIMA plate reader ( BMG LABTECH , Aylesbury , United Kingdom ) to establish giant colony . Cells were collected at 26 hr ( exponential phase ) and re-suspended in H2O . Cells were then gently centrifuged , and pellet ( intracellular ) and supernatant ( extracellular ) fractions were separated . Metabolites were extracted from both fractions using 75% boiling ethanol containing l-taurine as an internal standard . Here , samples were left to incubate with extraction solvent in water bath ( 80°C ) for 2 min then mixed vigorously . Incubation and vigorous mixing step was then repeated . Solvent was evaporated using a Concentrator plus Speed Vac ( Eppendorf , Hamburg , Germany ) and samples were reconstituted in 50 µL 80% ethanol with intracellular fraction diluted 1:5 with 80% ethanol . All samples were submitted to LC-MS/MS and metabolite identification , and quantification was then performed as in 'LC-MS/MS based quantification of amino acids and uracil' . Data was illustrated following correction to the internal standard of amino acid abundances from both intra- and extracellular fractions . S . cerevisiae strains were transferred from cryo-preserved cultures to SC solid media , grown for 2 days and selected on SM solid media , supplemented only with required amino acids/ nucleobases for 1 day . Pre-cultures were inoculated in 1 . 5 mL SM containing the minimal supplementation and cultured O/N at 30°C . Main cultures were started at an OD595 of 0 . 15 in 1 . 5 mL of SC media in deep well 96-well plates and cultured in a Titramax ( Heidolph , Schwabach , Germany ) for 30 hr ( 950 rpm , 30°C , 4 mm stirring bead/ well ) . Samples of 50 µL were harvested every 3 hr , where cells were removed by centrifugation ( 3000 g , 5 min ) and the supernatant diluted 1:20 in absolute ethanol . Then , 1 µL of supernatant was used for quantification of extracellular metabolites by LC-MS/MS . ‘gcFitModel’ function from ‘grofit’ R package ( Kahm et al , 2010 ) was used to estimate the uptake rate of histidine , leucine , uracil , and methionine in different auxotrophic strains . After O/N pre-culture S288c MATa yeast without auxotrophies and ura3Δ yeast ( S288c , MATa ) ( Mülleder et al . , 2012 ) were grown in 30 mL SM in shake flasks at 30°C , 300 rpm . The media contained either ( i ) no additives , ( ii ) uracil ( 20 mg/L ) ( iii ) or uracil ( 20 mg/L ) , leucine ( 60 mg/L ) , methionine ( 20 mg/L ) and histidine ( 20 mg/L ) . During mid-exponential growth ( OD595 between 0 . 7 and 1 . 2 ) , 1 mL samples of the cultures were quenched in 4 mL -40°C 60% methanol , 10 mM NH4-acetate . After centrifugation ( -9°C , 4500 g ) , the cell pellet was stored at -80°C until extraction . Prior to extraction , 13C-yeast internal standard was spiked into the cell pellets , which were subsequently extracted with 1 mL 80°C 75% ethanol , 10 mM NH4-acetate for 3 min . During extraction , the suspension was vortexed on a 0 . 5–15–1 min time interval . After extraction , the suspension was centrifuged ( -9°C , 4500 g ) and the supernatant , hence the extract , was dried in a vacuum centrifuge before being stored at -80°C until measurement . For LC-MS measurements , the dried extracts were dissolved in 50–10050–100 µL H2O . The metabolites were separated with reversed phase ion-pairing chromatography on a Acquity UPLC ( Waters , Cheshire , United Kingdom ) with a Waters Acquity T3-endcapped column ( 150 mm , 2 . 1 mm , 18 µm ) as described in ( Buescher et al . , 2010 ) . Subsequently , the metabolites were analysed with a TSQ quantum ultra triple quadrupole mass spectrometer ( Thermo Fisher Scientific , Waltham , MA ) ( Buescher et al . , 2010 ) . Specifically , the metabolites were ionised with an electro spray ( ESI ) and the mass spectrometer was run in negative mode with SRM . The SRM transitions used are described in ( Buescher et al . , 2010 ) and for orotidine-monophosphate , where no standard was available , we used the phosphate fragment ( m/z 367 → 79 ) trajectory ( Horai et al . , 2010 ) . The obtained data was integrated with an in-house software and normalised to 13C-internal standard and OD595 , hence biomass . The median value of different replicates were then scaled and used to illustrate data . Plasmid stability ( segregation ) of vectors p423 ( HIS3 ) , pRS425 ( LEU2 ) , p426 ( URA3 ) , and pRS411 ( MET15 ) was determined according to Christianson et al . ( 1992 ) . BY4741 ( his3Δ1 , leu2Δ0 , ura3Δ0 , met15Δ0 ) either transformed with one or all four plasmids , respectively , for either the four non-fluorescent or fluorescent vectors , were cultured in 25 mL of YPD media for 48 hr then plated at 1:100 , 000 dilution on YPD solid media . Plasmid retention was then calculated by replica plating CFUs from YPD solid media onto selective solid media . Number of doublings ( g ) and segregation rate ( m ) were calculated as in ( Christianson et al . , 1992 ) . Segregation rates of p423 ( HIS3 ) , pRS425 ( LEU2 ) , p426 ( URA3 ) , and pRS411 ( MET15 ) were simulated over generation time in R by iterative cycling ( looping ) . The script is given in Supplementary file 1 . Plasmid abundances were binned by plasmid number ( 0 to 4 ) and illustrated with R package 'ggplot2' in terms of auxotrophy . Unless otherwise indicated , cells were first spotted and grown for 2 nights on SM solid media to establish a giant colony . The colony was then re-suspended in H2O and diluted to 3 . 4e03 cells in 200 μL SM , and their optical density ( OD595 ) was recorded in a FLUOstar OPTIMA plate reader ( BMG LABTECH ) every 20 min for 40 hr at 30ºC . Both maximum specific growth rate ( μmax ) and lag phase were determined from growth curves using a model-richards fit from the R ‘grofit’ package ( Kahm et al , 2010 ) . For determining dry biomass , colony was re-suspended in H2O and normalised to 1 . 1e07 cells in 100 mL SM , then incubated for 72 hr at 30°C and pelleted . Pellets were dried for 5 days at 50°C and then weighed to obtain dry biomass . μmax for individual metabotypes was determined in batch SM culture ( 50 mL ) and supplemented accordingly for the different auxotrophic requirements . Cells from giant colonies of SeMeCo and YSBN5 were grown to exponential growth phase in SC media , and cell number was then measured with a CASY Model TTC ( Roche Innovatis , Switzerland ) cell counter . Cells were then diluted and plated on solid SC media to establish individual CFUs and the number of CFUs with initial cell number were compared . A micro-colony of BY4741 with prototrophy restored by complementation with the fluorescent protein vectors yEpCFP_HIS ( HIS3 ) , yEpSapphire_LEU ( LEU2 ) , yEpVenus_URA ( URA3 ) ( Bilsland et al . , 2013 ) , and pRS411-GPDpr-mCherry ( MET15 ) ( Table 1 ) was grown for 2 nights on SM . Prior to imaging , colony was embedded in 2% agarose ( Type I-B; Sigma ) and gently transferred to a μ-slide glass bottom ( ibidi ) . Cells were imaged with a DMI6000 inverted Leica SP5 confocal microscope , using a 10×/0 . 3 HC PL Fluotar Air objective , running LAS AF software ( version 2 . 7 . 3 . 9723 ) . Fluorescence for each marker was separated by excitation ( CFP: 458 nm , Sapphire: 405 nm , Venus: 514 nm and mCherry: 561 nm ) . In our hands , the Sapphire-LEU2 was also visible under the imaging conditions used to visualise the Venus-URA3 . For this reason , we removed Venus-URA3 channel from the colony image . A look-up table was applied to each channel post-acquisition to allow visualisation of the different channels together using ImageJ software . For the microscopy image showing fluorescent-labelled cells in a colony , prototrophs and auxotrophs were identified as being present or absent using several cut-offs for fluorescent signal intensity . Based on these different cut-off values ( 0 . 1 , 0 . 2 , 0 . 3 , and 0 . 4 ) , the monochrome fluorescence microscopy images of each individual marker ( HIS3 , LEU2 and MET15 ) were recognised as metabolite producing ( prototrophic ) or requiring ( auxotrophic ) pixels . The auxotrophic pixels were marked black , prototrophic pixels white , and the background was illustrated grey to separate it from the colony pixels . For each auxotrophic pixel , the distance to the next prototrophic pixel was calculated and distance values for the minimum , mean , overall maximum , and maximum for 90% of cells were taken for each marker . Genotypes depicting SeMeCo frequent ( HIS3 , LEU2 , MET15 , ura3Δ ) and rare ( his3Δ , leu2Δ , URA3 , MET15 ) metabotypes were reconstructed by transforming mCherry labelled derivatives of pHLUM ( pHLM-GPDpr-mCherry and pUM-GPDpr-mCherry , respectively ) into BY4741 . Strains were spiked into established SeMeCo in 1:10 ( frequent or rare metabotype: SeMeCo ) ratio taken from their established giant colonies on selective media ( where plasmid segregation is ongoing ) . Abundance of fluorescent cells was monitored immediately after mixing and approximately 48 hr after re-establishment of giant colony on minimal solid media with a BD LSRFortessa cell analyser . Data analysis was performed with FlowJo .

Life is sustained by an array of chemical reactions that is collectively referred to as metabolism . Some of these reactions break down complex substances to release energy and vital compounds , while others make new molecules from smaller building blocks . Bacterial communities are regularly composed of heterogeneous species , several of which have lost one or more essential metabolic pathways . Nevertheless , these cells can still survive by making use of metabolic products released by their neighbouring cells . Yeast are single-celled fungi that also form colonies and , as eukaryotes , they possess cells that are more similar to our own . However , in the laboratory , complementary metabolically deficient yeast cells do not survive when mixed together . It was presumed this is because yeast cells make only enough of each essential metabolite for themself , and so can’t replace those that are missing from their neighbouring cells . Campbell et al . now challenge this view by finding that yeast cells release a variety of metabolites , they use these released metabolites in preference to making their own , and possess the capacity to grow on the basis of a non-cell-autonomous metabolism . This discovery came with the design of a new experimental test to study metabolite exchange interactions . This method uses yeast cells that have one or more of their own metabolic genes disabled , and instead have a copy of these genes on small circular DNA 'mini-chromosomes' ( called plasmids ) . The gene on the plasmid can compensate for the yeast having its own gene missing , and allows the cell to still make the metabolic product it needs to survive . However , as a single cell divides to form a colony , cells randomly lose these plasmids , leaving some of the cells deficient for a particular metabolite . These cells can only survive if the other cells in the colony export the missing metabolite in the quantity needed for growth . Using this test , Campbell et al . found that yeast cells can export missing metabolites at levels needed to sustain these emerging metabolic mutants . Additionally , these yeast communities could grow at levels comparable to other yeast without metabolic deficiencies . The resulting colonies also feature one of several different genetic and metabolic profiles , which change in response to the metabolite that is missing . These findings demonstrate that yeast cells can exchange high amounts of metabolites , sufficient to form cooperative colonies , and as metabolite concentrations are not altered compared to normal cells , it is likely that exchange of metabolites is ongoing between neighbours in yeast communities . The additional discovery that yeast stop making metabolites when they can obtain them from neighbouring cells has implications for research . This is because many yeast genetic studies use metabolically deficient strains that are supplemented in culture with metabolites . Future work could address whether such supplementation has kept certain functions of metabolism hidden .

2015

Self-establishing communities enable cooperative metabolite exchange in a eukaryote

Feeding is critical for survival , and disruption in the mechanisms that govern food intake underlies disorders such as obesity and anorexia nervosa . It is important to understand both food intake and food motivation to reveal mechanisms underlying feeding disorders . Operant behavioral testing can be used to measure the motivational component to feeding , but most food intake monitoring systems do not measure operant behavior . Here , we present a new solution for monitoring both food intake and motivation in rodent home-cages: the Feeding Experimentation Device version 3 ( FED3 ) . FED3 measures food intake and operant behavior in rodent home-cages , enabling longitudinal studies of feeding behavior with minimal experimenter intervention . It has a programmable output for synchronizing behavior with optogenetic stimulation or neural recordings . Finally , FED3 design files are open-source and freely available , allowing researchers to modify FED3 to suit their needs . Feeding is critical for survival , and dysregulation of food intake underlies medical conditions such as obesity and anorexia nervosa . Quantifying food intake is necessary for understanding these disorders in animal models . However , it is challenging to accurately measure food intake in rodents due to the small volume of food that they eat . Researchers have devised multiple methods for quantifying food intake in rodents , each with advantages and drawbacks ( Ali and Kravitz , 2018 ) . Manual weighing of a food container is a simple and widely used method for quantifying food intake in a rodent home-cage . Yet this is time consuming to complete , is subject to error and variability , and does not allow for fine temporal analysis of consumption patterns ( Acosta-Rodríguez et al . , 2017; Reinert et al . , 2019 ) . Automated tools have been developed for measuring food intake in home-cages with high temporal resolution , although most require modified caging , powdered foods , or connected computers which limit throughput ( Ahloy-Dallaire et al . , 2019; Farley et al . , 2003; Moran , 2003; Yan et al . , 2011 ) . These include automated weighing ( Hulsey and Martin , 1991; Meguid et al . , 1990; Minematsu et al . , 1991 ) , pellet dispensers ( Aponte et al . , 2011; Gill et al . , 1989; Oh et al . , 2017 ) , or video detection-based systems ( Burnett et al . , 2016; Jhuang et al . , 2010; Salem et al . , 2015 ) . In addition to measuring total food intake , understanding neural circuits involved in feeding requires exploring why animals seek and consume food . Has their motivation for a specific nutrient changed ? Has their feeding gained a compulsive nature that is insensitive to satiety signals ? These questions can be answered with operant tasks , where rodents receive food contingent on their actions ( Curtis et al . , 2019; Mourra et al . , 2020; O'Connor et al . , 2015; Skinner , 1938; Thorndike , 1898; Wald et al . , 2020 ) . Typically , operant behavior is tested in dedicated chambers for a few hours each day . Commercial systems for testing operant behaviors typically comprise a dedicated arena equipped with levers , food or liquid dispensers , tones or spotlights , and video cameras . Programmable software interfaces allow experimenters to control this equipment to train animals on behavioral tasks such as fixed-ratio or progressive ratio responding . Some chambers are also equipped with video cameras and tracking software , so the location of the animal can be used as a trigger to control task events ( London et al . , 2018 ) . Training in dedicated operant chambers has limitations: tasks can take weeks for animals to learn , animals may be tested at different phases of their circadian cycle due to equipment availability , and food restriction can be necessary to get animals to seek food outside of their home-cage , which can confound feeding studies . To mitigate these issues , several researchers have begun to test operant behavior in rodent home-cages , resulting in both fewer interventions from the researcher and faster rates of learning ( Balzani et al . , 2018; Francis and Kanold , 2017; Lee et al . , 2020 ) . Here , we present a new solution for monitoring food intake and testing operant behavior in rodent home-cages: the Feeding Experimentation Device version 3 ( FED3 ) . Our goal was to develop a device for measuring food intake in rodent home-cages with high temporal resolution , while also measuring food motivation via operant behavior . FED3 is a stand-alone device that contains a pellet dispenser , two ‘nose-poke’ sensors for operant behavior , visual and auditory stimuli , and a screen for experimenter feedback . FED3 is compact and battery powered , fitting in most commercial vivarium home-cages without any connected computers or external wiring . FED3 also has a programmable output that can control other equipment , for example to trigger optogenetic stimulation after a nose-poke or pellet removal , or to synchronize feeding behavior with electrophysiological or fiber-photometry recordings , or with in vivo calcium imaging . Finally , FED3 is open source and was designed to be customized and re-programmed to perform novel tasks to help researchers understand food intake and food motivation . To this end , we have written a user-friendly Arduino library to facilitate custom behavioral programming of FED3 . Here , we describe the design and construction of FED3 and present several experiments that demonstrate its functionality . These include measuring circadian patterns of food intake over multiple days , performing meal pattern analysis , automated operant training , closed-economy motivational testing , and optogenetic self-stimulation . FED3 extends existing methods for quantifying food intake and operant behavior in rodents to help researchers achieve a deeper understanding of feeding and feeding disorders . FED3 is controlled by a 48 mHz ATSAMD21G18 ARM Cortex M0 microprocessor ( Adafruit Feather M0 Adalogger ) and contains two nose-pokes , a pellet dispenser , a speaker for auditory stimuli , eight multi-color LEDs for visual stimuli , a screen for experimenter feedback , and a programmable analog output ( Figure 1A , B ) . When mice interact with FED3 , the timing of each poke and pellet removal are logged to an internal microSD card for later analysis and summary data is displayed on the screen for immediate feedback to the researcher ( Figure 1B ) . FED3 is small ( ~10 cm × 12 cm × 9 cm ) , battery powered , and completely self-contained , so it fits in most vivarium home-cages without modification or introducing wiring to the cage ( Figure 1C ) . It is powered by a rechargeable battery that lasts ~1 week between charges ( exact battery life depends on the behavioral program being run and number of pellets dispensed ) . FED3 also has magnetic mounts to facilitate wall-mounting on any plastic box to mimic a traditional operant setup ( see Video 1 ) . We provide example programs for running rodents on multiple common behavioral paradigms , including free feeding , time-restricted free-feeding , fixed-ratio 1 ( FR1 ) , and progressive ratio ( PR ) operant tasks , and have written a FED3 Arduino library to ease development of custom programs . FED3 also has an auxiliary I2C port for hardware customization and a programmable BNC output connection that allows synchronization with external equipment for aligning behavioral events with fiber-photometry recordings ( London et al . , 2018; Mazzone et al . , 2020 ) , electrophysiological recordings ( London et al . , 2018 ) , or other equipment such as video tracking systems ( Krynitsky et al . , 2020; Li et al . , 2019 ) . This BNC output can report task events or control external hardware with delays of less than 200 µs , which compares favorably with computers that process information through Universal Serial Bus ( USB ) , which can introduce delays of several milliseconds . Finally , we provide a graphical analysis package ( FED3VIZ ) written in Python that enables users to generate detailed plots from FED3 data ( Figure 2F ) . FED3 is open-source and freely available online , including 3D design files , printed circuit board ( PCB ) files ( Figure 1—figure supplement 1A , B ) , build instructions ( Figure 1D ) , and code ( https://github . com/KravitzLabDevices/FED3 copy archived at swh:1:rev:ff8fc79d288a440d91566f4c6cd80011956f4be0 ) . Ten FED3 devices were set to the ‘free-feeding’ program , which dispenses a pellet and monitors its presence in the pellet well ( Figure 2A ) . In this paradigm , each time the pellet is removed , FED3 logs the date and time to the storage card and dispenses another pellet . This allows for the reconstruction of detailed feeding records over multiple days . Assuming calorie content of the pellets is known , total caloric intake can also be tracked over time . Devices were placed with ten singly housed mice for six consecutive days with no additional food source . We observed the expected circadian rhythm in food intake ( Figure 2B , C ) . To test accuracy , we confirmed that the daily change in weight of the FED3 device correlated with the number of pellets removed multiplied by the weight of each pellet ( 20 mg pellets , Figure 2D ) . The coefficient of determination ( R2 ) for the regression was 0 . 97 , indicating an error rate of ~3% between manual weighing and counting of pellets by FED . Finally , we confirmed that the mice obtained all of their necessary daily calories from FED3 , evidenced by a stable body weight across these 6 days ( Figure 2E ) . In other experiments , we have run FED3 for >1 month without observing weight loss . FED3 can be used for multi-week experiments like this as it has a low rate of jamming . This is facilitated by a unique angled pellet hopper and dispensing path that has fewer places where pellets can jam than traditional horizontal pellet dispensers ( Figure 2—figure supplement 1 ) , as well as ‘jam clearing’ movements in the code that vibrate and rotate the dispensing disk to dislodge stuck pellets when they are detected . These are improvements over the first generation of this device ( Nguyen et al . , 2016 ) . As a point on rodent safety , although FED3 is resistant to jamming it still must be checked for function at least once per day when being used as the only food source . FED3 creates a large amount of data , particularly when run over multiple days . To facilitate analysis of FED3 data , we created FED3VIZ , an open-source Python-based analysis program ( https://github . com/earnestt1234/FED3_Viz ) . FED3VIZ offers plots for visualizing different aspects of FED3 data , including pellet retrieval , poke accuracy , pellet retrieval time , delay between consecutive pellet earnings ( inter pellet intervals ) , meal size , and progressive ratio breakpoint . These metrics can be plotted for single files or group averages . FED3VIZ’s averaging methods provide options for aggregating data recorded on different days , while preserving time-of day or phase of the light–dark cycle . Furthermore , circadian activity patterns can also be visualized with the FED3VIZ ‘Chronogram’ and ‘Day/Night Plot’ functions ( see Figure 7 ) , which segment data based on a user-defined light cycle . The processed data going into each plot can also be saved and used to create new visualizations or compute statistics in other programs , promoting reproducible , sharable analysis and visualization of FED3 data . FED3 records the date and time that each pellet is removed , which can be used to quantify feeding patterns including the size and quantity of meals , eating rate within meals , and timing between meals . Different parameters have been used by different groups to define meals , often based on numerical cut-offs ( Farley et al . , 2003; Kanoski et al . , 2013; Melhorn et al . , 2010 ) . To complement these approaches , we aimed to develop an unbiased approach for understanding meal patterns , based on the distribution of time intervals between each pellet consumed . For the 10 mice in the above experiment , we plotted the inter-pellet time interval histogram ( Figure 3A ) for both the dark and light cycles , based on an approach that was previously described in rats ( Cottone et al . , 2007 ) . We observed a large peak at <1 min between pellets , showing that the majority of pellets were consumed within 1 min of another pellet . We used this distribution to classify pellets eaten within the same minute as belonging to the same meal , and pellets with larger intervals between them belonging to different meals . This approach revealed that animals eat fewer meals during the light cycle ( Figure 3B ) but eat similar number of pellets within each meal and obtain a similar fraction of their pellets within meals in each cycle . Differences in meal patterning have been linked to obesity ( Farley et al . , 2003; Wald and Grill , 2019 ) , so this analytical approach may assist in understanding obesity and other disorders of feeding . A unique feature of FED3 is that it is small and wire-free , allowing it to fit inside of traditional vivarium home-cages . This facilitates quantification of circadian rhythms in both operant and feeding behavior ( Figure 4A ) . To demonstrate operant capability , we quantified how nose-poking for pellets varied over the circadian cycle , by running mice on a FR1 task for six consecutive days . Mice learned to nose-poke on the active port for a single pellet . We tracked active ( rewarded ) and inactive ( unrewarded ) pokes . As expected , both active and inactive pokes were higher during the dark cycle ( Figure 4B–D ) . Surprisingly , however , poke efficiency ( % of total pokes on the active port ) was slightly higher during the light cycle ( Figure 4E ) , suggesting that operant responding is more efficient during the light cycle . Multiple research groups currently use FED3 . We therefore asked how mouse operant learning varies across different laboratories . To do this , we obtained operant data from the first overnight FR1 session from seven laboratories , resulting in data from 122 mice ( mix of males and females , Figure 5A ) . When comparing pellets earned in an overnight session across laboratories , we observed a significant effect of group , linked to significant differences in 4 of 21 post hoc comparisons ( Figure 5B ) . When viewed as a single distribution ( Figure 5B inset ) , the distribution of pellets earned from all groups was consistent with a single Gaussian distribution ( p>0 . 05 , Shapiro-Wilk test for normality ) . This highlights how FED3 enables high-throughput studies of operant behavior , and the power of large sample sizes to achieve adequate sampling of a distribution of behavior ( Figure 5B ) . A unique capability of FED3 is that it has a sensor for detecting the presence of the pellet in the feeding well . This enables time-stamping of when each pellet was removed for constructing feeding records , as well as measuring the time between the pellet dispense and removal , which we termed ‘retrieval time’ . Changes in retrieval time may reflect learning , as this measure progressively decreased as animals gained experience with FED3 ( Figure 5E ) . Accordingly , active active count and poke efficiency ( % of pokes on the active port ) increased over time , demonstrating that retrieval time decreased on the same time scale that mice learned which poke resulted in a pellet ( Figure 5C ) . To compare learning rates in overnight training to traditional daily operant sessions , we ran a new group of eleven mice for sixteen daily 1 hr FR1 sessions with FED3 , returning them to the colony for the remainder of each day . Learning rates did not differ between mice exposed to 16 daily 1 hr sessions or one 16 hr overnight session ( Figure 5F ) . This suggests that the acquisition of this FR1 task may depend most strongly on the cumulative time mice interact with FED3 and that overnight training can greatly speed up task acquisition by giving them many hours of experience in one night . Based on prior literature ( Steinhauer et al . , 1976 ) , we predicted that the speed of acquisition of an FR1 task would be further increased by prior magazine training . To test this , we exposed 24 mice to magazine training using the free feeding FED3 paradigm for 1–3 days prior to FR1 training ( Figure 6A ) . This paradigm dispenses a pellet into the feeding well of FED3 and replaces it whenever it is removed , providing an automatic method for magazine training animals . In this way , magazine training allows mice to learn the location of the food source and associate food with the sound of the pellet dispenser operating and the pellet being dispensed , before commencing with FR1 training . We compared the performance of the 24 magazine trained mice to 122 mice from Figure 5 ( which had not been magazine trained ) and found that magazine training resulted in significantly higher ( ~2 ) levels of pellet acquisition in the first night with the FR1 task ( Figure 6B , C ) . Therefore , magazine training is recommended to speed up acquisition of nose-poking tasks . FED3 can also be used for closed-economy feeding studies , in which all food is earned by operant nose-poking ( Figure 7 ) . Closed-economy operant tasks have been used to quantify the economic demand for food ( Bauman , 1991; Chaney and Rowland , 2008 ) , as well as changes in economic demand due to manipulation of the dopamine system ( Beeler et al . , 2010; Mourra et al . , 2020 ) . We tested seven mice on a repeating progressive ratio task ( adapted from Mourra et al . , 2020 ) , in which the nose-poking requirement began on FR1 and increased each time a pellet was earned . When mice refrained from poking on either port for 30 min the ratio reset to FR1 ( Figure 7A , B ) . Although pellets earned from this task were their only food source , mouse weights remained stable across days , demonstrating that mice earned their entire caloric need with the task ( Figure 7C ) . Total pellets earned did not differ between light and dark cycles ( Figure 7D ) . However , nose-poking rates were higher during the dark phase of their daily cycle ( Figure 7E ) , meaning mice performed more pokes to earn each pellet in this phase ( Figure 7F ) . This suggests that motivation for food pellets varies over the circadian cycle such that mice are willing to work harder during the dark cycle vs . the light cycle . When the data was visualized hourly across the circadian cycle , we further noted a ~3 hr window of enhanced nose-poking rates starting shortly after the onset of the dark cycle ( Figure 7F ) . This time coincides with a circadian peak in extracellular dopamine levels in the striatum ( Ferris et al . , 2014 ) , and a peak in activity levels in mice ( Matikainen-Ankney et al . , 2020 ) . Therefore , this task may be used to assay changes in accumbal dopamine function across the circadian cycle . FED3 was also designed to synchronize with other experimental equipment through its programmable output port . To demonstrate this capability , we programmed FED3 to act as a function generator and control an LED for an intra-cranial self-stimulation task . The dorsal striatum contains two main classes of output neurons , differentiated by their expression of the dopamine D1 or D2 receptors ( Gerfen et al . , 1990 ) . Dorsal striatal neurons that express the D1 receptor are highly reinforcing when optogenetically stimulated ( Kravitz et al . , 2012 ) . To demonstrate how FED3 can be used to perform an optogenetic self-stimulation experiment , we used Cre-dependent viral expression to target channelrhodopsin-2 ( ChR2 ) to D1R-expressing neurons in the dorsal striatum ( Figure 8A ) . FED3 was programmed to generate and trigger a 1 s train of 20 Hz pulses of 475 nm light upon each active nose-poke , and the session continued until they received 75 trains of stimulation ( Figure 8B ) . Three mice were run on this task , resulting in significantly greater active vs inactive pokes ( Figure 8C , D ) , and demonstrating that FED3 can be used for optogenetic self-stimulation studies . The programmable output is attached to an on-board digital-to-analog converter chip , meaning that it can be programmed to output pulses of any voltage from 0 to 3 V , allowing FED3 to represent multiple behavioral events with one output line . The timing of the BNC output port is regulated by the SAMD21 microprocessor that has a documented pulsing precision of <1 µs . Quantification of food intake is necessary to understand animal models of feeding disorders , as well as other medical conditions . To quantify food intake in rodent home-cages , we previously published the FED version 1 , an open-source , stand-alone feeding device that can fit in a rack-mounted home-cage ( Nguyen et al . , 2016 ) . FED records a time series of pellets removed , which can be used to reconstruct feeding records . FED has been used by multiple research groups to understand feeding ( Brierley et al . , 2020; Burnett et al . , 2019; Chen et al . , 2020; Li et al . , 2019; London et al . , 2018 ) . Since its original publication , we have redesigned the FED twice and here present the third version of FED , which revamps our original design and adds the ability to measure operant behavior , as well as a unique ‘angled’ pellet dispenser mechanism that is resistant to jamming , which enables long term studies . We distributed the design for FED3 online , and it has been used to study how specific neural circuit manipulations alter food motivation ( Mazzone et al . , 2020; Sciolino et al . , 2019; Vachez et al . , 2021 ) , how weight-loss alters food motivation ( Matikainen-Ankney et al . , 2020 ) , and how food motivation is altered in a stress-susceptible mouse population ( Rodriguez et al . , 2020 ) . FED3 is a stand-alone solution for home-cage operant training , enabling researchers to understand not just total food intake , but also motivation for and learning about food rewards . FED3 has several unique benefits over comparable systems including: ( 1 ) FED3 is low cost . The FED3 electronics cost ~$200 and the housing is 3D printed . Even if 3D printing is done professionally this is >10× cheaper than most commercial solutions for measuring food intake or testing operant behavior; ( 2 ) FED3 is self-contained and fits within traditional vivarium caging , allowing for measures of true ‘home-cage’ feeding without modifying the cage . Due to its small size , it can also be placed inside of other equipment where wires might be impractical , such as within an indirect calorimetry system; ( 3 ) FED3 has a programmable output that allows it to easily synchronize with other equipment . Due to wiring , connecting FED3 to external hardware likely requires FED3 to be used outside of the home-cage , or to modify the home-cage system . This has been done by multiple labs to synchronize the output of FED3 with fiber-photometry recordings ( London et al . , 2018; Mazzone et al . , 2020 ) , electrophysiological recordings ( London et al . , 2018 ) , optogenetic stimulation ( Vachez et al . , 2021 ) , and video tracking ( Krynitsky et al . , 2020; Li et al . , 2019 ) ; and ( 4 ) FED3 is open source and all design files and codes are freely available online . This enables users to modify the code and hardware to achieve new functionality . The purpose of this manuscript is to demonstrate the utility of FED3 for feeding research . To this end , we demonstrate experiments that measured total food intake , operant responding , and optogenetic stimulation . We further highlight how the high temporal resolution enables meal pattern analysis across multiple days . Finally , we coordinated with six other research groups to compile a dataset of 122 mice across seven research sites , all running the same experimental FR1 program . We observed similar patterns of acquisition across all sites , demonstrating that FED3 can be used for multi-site research studies on feeding . Due to its low cost and open-source nature , we believe that multi-site studies with FED3 are more feasible than with other systems . While FED3 has many strengths , it also has limitations . One limitation is that FED3 uses internal microSD cards to store data . While microSD cards are convenient , they are not ideal for large numbers of devices , where removing multiple cards can be cumbersome . Wireless data logging is a potential solution to this problem , although there are challenges to implementing this in rodent cages . A second limitation is that animals can ‘hoard’ pellets from FED3 , as animals can remove food without consuming it . In our experience , this seemed to be a rare trait that specific mice engaged in ( <10% of mice hoard pellets ) . Unfortunately , FED3 has no way to determine whether a mouse actually consumes each pellet after removal so we recommend checking for pellet hoarding and accounting for this in experimental conclusions . A third limitation is that granular bedding can be kicked into the pellet well and interfere with pellet detection . To avoid this , we recommend using ‘iso-pad’ bedding , or bedding pellets that are large enough to avoid this possibility . A fourth limitation is that FED3 is not waterproof and would not be resistant to flooding , though the 3D printed housing is sufficient to protect electronics from typical wear associated with placement in a rodent cage . A final limitation of FED3 is that it has no way of identifying individual mice in group housed environments , so feeding records must be collected in singly housed mice . However , FED3 is open source , so it can be modified and improved with innovations from our group and the feeding community , and future iterations of the device may include methods for identifying multiple mice using radio-frequency identification ( RFID ) tags . Being able to run studies on group housed mice would greatly increase throughput and allow for the study of interactions between social behavior and feeding . We published the FED3 design as open-source and look forward to community contributions and modifications to enable new functionality and overcome these limitations . The FED3 device is open-source , and design files and code are freely available online at: https://github . com/KravitzLabDevices/FED3 . In addition , we have made all data and analysis code for the figures in this paper available at https://osf . io/hwxgv/ . One hundred and fifty-nine C57Bl/6 mice were housed in a 12 hr light/dark cycle with ad libitum access to food and water except where described . Mice were provided laboratory chow diet ( 5001 Rodent Diet; Lab Supply , Fort Worth , TX ) . All procedures were approved by the Animal Care and Use Committee at Washington University in St Louis , the National Institutes of Health , Williams College , Virginia Tech , and Monash University . Tutorial videos and other information on assembling FED3 are available at: https://github . com/KravitzLabDevices/FED3/wiki . We demonstrate multiple operational modes for FED3 to highlight a range of functionality and applications . The following programs are included with the standard FED3 code: Viral infections of male and female Drd1-Cre mice were conducted under 0 . 5–2 . 5% anesthesia on a stereotaxic apparatus . Using a Nanoject injector , 500 nL of AAV2-DIO-hSyn-ChR2 was injected bilaterally ( 500 nL/hemisphere ) in the dorsomedial striatum . Optical fibers were then implanted in the same region . After letting ChR2 express for 4 weeks , animals were pre-trained on an FR1 schedule for pellets overnight in their home-cage . Two days later , mice were placed in a box with a FED3 device connected to an LED driver . The active poke for this paradigm was on the opposite side to the active poke of the pre-training session . During the self-stimulation session , when the mouse poked on the active nose-poke , it received a 1 s train of 1 mW 475 nm light at 20 Hz . Sessions were run for 60 min in a 550 cm2 arena . CSV files generated by FED3 were processed and plotted with custom python scripts ( Python , version 3 . 6 . 7 , Python Software Foundation , Wilmington , DE ) . All data and scripts are available on Open Science Framework ( https://osf . io/hwxgv/ ) . Visualization was also completed using FED3VIZ GUI to generate plots . FED3VIZ was written in Python’s standard library for developing GUIs ( tkinter ) . FED3VIZ is a custom open-source graphical program for analyzing FED3 data . FED3VIZ code , version history , installation instructions , and user manual are available on GitHub ( https://github . com/earnestt1234/FED3_Viz ) . FED3VIZ offers plotting and data output for visualizing different aspects of FED3 data , including pellet retrieval , poke accuracy , pellet retrieval time , delay between consecutive pellet earnings ( inter-pellet intervals ) , meal size , and progressive ratio breakpoint . Based on inter-pellet interval histograms ( Figure 3 ) , we defined meals as pellets eaten within 1 min of each other . In addition , we defined a minimum size of 0 . 1 g ( five pellets ) to be counted as a meal . Bartlett’s test for equal variances was performed; one- or two-way ANOVAs with Tukey post-tests were used to compare groups with equal variance ( p>0 . 05 ) where appropriate . Linear mixed effects models were used to analyze groups with significantly difference variances ( p<0 . 05 ) . p-value<0 . 05 was considered significant . Statistical tests to compare means were run using statsmodels module in python ( Seabold and Perktold , 2010 ) . Data sets are presented as mean ± SEM . Numbers of animals per experiment is listed as n=number of animals . Linear regression was used to determine correlative relationships . T-tests or Mann–Whitney U tests were used to compare the means of two groups for parametric or nonparametric data distributions , respectively .

Obesity and anorexia nervosa are two health conditions related to food intake . Researchers studying these disorders in animal models need to both measure food intake and assess behavioural factors: that is , why animals seek and consume food . Measuring an animal’s food intake is usually done by weighing food containers . However , this can be inaccurate due to the small amount of food that rodents eat . As for studying feeding motivation , this can involve calculating the number of times an animal presses a lever to receive a food pellet . These tests are typically conducted in hour-long sessions in temporary testing cages , called operant boxes . Yet , these tests only measure a brief period of a rodent's life . In addition , it takes rodents time to adjust to these foreign environments , which can introduce stress and may alter their feeding behaviour . To address this , Matikainen-Ankney , Earnest , Ali et al . developed a device for monitoring food intake and feeding behaviours around the clock in rodent home cages with minimal experimenter intervention . This ‘Feeding Experimentation Device’ ( FED3 ) features a pellet dispenser and two ‘nose-poke’ sensors to measure total food intake , as well as motivation for and learning about food rewards . The battery-powered , wire-free device fits in standard home cages , enabling long-term studies of feeding behaviour with minimal intervention from investigators and less stress on the animals . This means researchers can relate data to circadian rhythms and meal patterns , as Matikainen-Ankney did here . Moreover , the device software is open-source so researchers can customise it to suit their experimental needs . It can also be programmed to synchronise with other instruments used in animal experiments , or across labs running the same behavioural tasks for multi-site studies . Used in this way , it could help improve reproducibility and reliability of results from such studies . In summary , Matikainen-Ankney et al . have presented a new practical solution for studying food-related behaviours in mice and rats . Not only could the device be useful to researchers , it may also be suitable to use in educational settings such as teaching labs and classrooms .

2021

An open-source device for measuring food intake and operant behavior in rodent home-cages

Nematodes and insects are the two most speciose animal phyla and nematode–insect associations encompass widespread biological interactions . To dissect the chemical signals and the genes mediating this association , we investigated the effect of an oriental beetle sex pheromone on the development and behavior of the nematode Pristionchus pacificus . We found that while the beetle pheromone is attractive to P . pacificus adults , the pheromone arrests embryo development , paralyzes J2 larva , and inhibits exit of dauer larvae . To uncover the mechanism that regulates insect pheromone sensitivity , a newly identified mutant , Ppa-obi-1 , is used to reveal the molecular links between altered attraction towards the beetle pheromone , as well as hypersensitivity to its paralyzing effects . Ppa-obi-1 encodes lipid-binding domains and reaches its highest expression in various cell types , including the amphid neuron sheath and excretory cells . Our data suggest that the beetle host pheromone may be a species-specific volatile synomone that co-evolved with necromeny . An understanding of ecological interactions and evolutionary history can be beneficial to the understanding of the cellular and developmental processes of an organism . For example , the genes responsible for fruit and bacterial odor preferences in Drosophila melanogaster and Caenorhabditis elegans are most informative when considering the natural ecologies in rotting fruits and vegetation ( Bargmann et al . , 1993; Bargmann , 2006; Hallem and Carlson , 2006; Kiontke and Sudhaus , 2006 ) . In Pristionchus pacificus , an emerging model organism for the study of development and behavior ( Hong and Sommer , 2006b ) , a systematic effort to identify the natural ecology of Pristionchus nematodes revealed that P . pacificus host preferences include the oriental beetle ( Exomala orientalis ) found in Japan and northeastern United States ( Herrmann et al . , 2007 ) . Unlike several fructivorous Caenorhabditis species , Pristionchus species are considered to be necromenic nematodes that have species-specific beetle host preferences and feed on the microorganisms that emerge from the beetle carcass ( Herrmann et al . , 2006a , 2006b; Brown et al . , 2011 ) . Pristionchus species are members of the Diplogasteridae family of nematodes whose common ancestor with the Rhabditidae family ( e . g . , C . elegans ) diverged ∼300 million years ago ( Dieterich et al . , 2008 ) . Given that beetles ( Coleoptera ) represent a group with the most described species , the possibility that most beetle species can harbor specific entomophilic nematode associations underscores the vast number of nematode species remaining to be described . Although biologists have been aware of such species-specific insect–nematode associations for some time , the genetic basis for such interactions remains largely unknown . Previous studies suggest that the beetle host specificity of Pristionchus species is in part due to nematode attraction to various plant volatile compounds released during beetle feeding , as well as to beetle aggregation and sex pheromones during mating ( Hong and Sommer , 2006a; Hong et al . , 2008b , 2008a ) . For example , Pristionchus maupasi associates primarily with the European May beetle Melolontha spp . and is attracted to the blend of the green leaf alcohol from plants and the phenol pheromone from beetles ( Hong et al . , 2008b ) . Thus , insect and plant odors can synergize to provide nematodes with information on host presence , sex , and maturity . By contrast , C . elegans attractants such as 2 , 3-butanedione ( diacetyl ) , 2 , 3-pentanedione , and isoamyl alcohol are significantly less attractive to P . pacificus ( Bargmann et al . , 1993; Hong and Sommer , 2006a ) . This piggybacking of intra-species communication for inter-species interactions between Pristionchus and beetles can also evolve rapidly at the population level , as exhibited by the natural variation of P . pacificus odor preference for insect pheromones ( Hong et al . , 2008a; McGaughran et al . , 2013b ) . In particular , P . pacificus Washington strain ( PS1843 ) is attracted to the lepidopteran sex pheromone E-11-tetradecenyl acetate ( ETDA ) and the oriental beetle sex pheromone z-7-tetradece-2-one ( ZTDO ) ( Herrmann et al . , 2007 ) , but the wild-type reference strain from California ( PS312 ) requires exogenous cGMP treatment to become attracted to these insect pheromones ( Hong et al . , 2008a ) . These natural variations in insect pheromone attraction , combined with regional-specific P . pacificus host preferences , may offer support for the geographic mosaic theory of co-evolution ( Thompson , 2005 ) . Although the distinct chemosensory profiles of P . pacificus populations suggest insect pheromone attraction to be an important and rapidly evolving trait , other studies on insect–nematode interactions show that there are other potential roles host chemicals can play in coordinating host–parasite development and nematode dispersal behavior . For example , the pinewood nematode Bursaphelenchus xylophilus is intimately associated with its Monochamus beetle vector , and fatty acid ethyl esters exuded during beetle eclosion promote the formation of the dispersal fourth larval stage B . xylophilus ( Zhao et al . , 2013 ) . Similarly , the species-specific phoretic association between the burrower bug Parastrachia japonensis and the nematode Caenorhabditis japonica may be mediated by the beetle cuticular compounds that not only attract C . japonica but also assist infestation by inhibiting C . japonica dispersal ( Okumura et al . , 2013 ) . P . pacificus populations have been found on various beetle hosts worldwide , but the association with the oriental beetle was the first to be identified and the only one whose sex pheromone is a strong attractant ( Herrmann et al . , 2007; McGaughran et al . , 2013b ) . Therefore , the oriental beetle sex pheromone represents the single most ecologically relevant semiochemical by which we could repeatedly interrogate the P . pacificus genes involved for host sensing and other unexplored developmental roles . In this study , we show that although the beetle pheromone , ZTDO , is attractive to P . pacificus adults , the beetle host pheromone may moderate nematode infection by arresting embryonic development , inducing larval paralysis , and inhibiting dauer exit . We leveraged the ability to use exogenous cGMP to potentiate the California reference strain for ZTDO attraction to find additional genes important for ZTDO sensitivity by performing an unbiased forward genetic screen for mutants with altered responses to ZTDO . We found that a putative lipid-binding protein Ppa-OBI-1 mediates both beetle pheromone attraction as well as a previously unknown role of the beetle pheromone to regulate P . pacificus development . Our results suggest that the host pheromone may play a complex role in reducing infection or coordinating development for the dispersal cycle of the necromenic lifestyle . P . pacificus and C . elegans have diametrically opposed odor preference profiles . Pristionchus species are generally attracted to plant volatiles , insect pheromones , and cuticular hydrocarbons , but each Pristionchus species can have its own chemosensory profile reflective of its preferred beetle hosts ( Hong and Sommer , 2006a; McGaughran et al . , 2013a ) . P . pacificus populations found on the oriental beetle E . orientalis in Japan and northeastern United States are attracted to the sex pheromone of the beetle , z-7-tetradecen-2-one ( ZTDO ) ( Zhang et al . , 1994; Herrmann et al . , 2007 ) . In contrast , the compost-dwelling C . elegans is less attracted to or even repelled by plant and insect odors , including ZTDO , but is strongly attracted to small compounds released by microbes ( Bargmann et al . , 1993; Hong and Sommer , 2006a ) . To identify genes required to mediate ZTDO attraction in a pheromone-insensitive reference strain , California PS312 , we performed a behavior-based genetic screen for mutants that are insensitive toward ZTDO following a brief 1-hr treatment with the cell-permeable 8-br-cGMP . We isolated two Oriental Beetle pheromone Insensitive mutants , Ppa-obi-1 ( tu404 ) and Ppa-obi-3 ( tu405 ) . The chemosensory phenotype of Ppa-obi-1 and Ppa-obi-3 is strongly specific for ZTDO and does not affect the chemoattraction towards other insect pheromones ETDA and myristate nor plant volatiles β-caryophyllene and nicotine ( Figure 1 ) . Odor avoidance in obi mutants towards 1-octanol , a repellant common between P . pacificus and C . elegans , also resembles wild type . These results indicate that both obi mutants are specifically defective in sensing a beetle pheromone rather than more generally defective in chemosensation . Ppa-obi-1 ( tu404 ) and Ppa-obi-3 ( tu405 ) are non-allelic recessive mutants that complement each other in their physical and behavioral phenotypes . We proceeded to characterize Ppa-obi-1 ( tu404 ) based on its altered response to ZTDO attraction and its unexpected function in protecting larvae . 10 . 7554/eLife . 03229 . 003Figure 1 . Chemosensory behavior of oriental beetle pheromone insensitive mutants . Ppa-obi-1 and Ppa-obi-3 mutant animals are specifically defective for sensing the host oriental beetle sex pheromone , Z-7-tetradece-2-one . The ZTDO structure is depicted below the beetle . In contrast , obi mutants' response is wild type towards another cGMP-dependent attractant , E-11-tetradecenyl acetate ( ETDA , a lepidopteran pheromone ) , as well as myristate ( methyl tetradecanoate , an insect allomone ) , plant volatiles beta-caryophyllene and nicotinic acid , and the repellent 1-octanol . N ≥ 15 replicates in ≥3 trials . Significant differences to wild type are indicated by p values ( Dunnett's test ) and error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 03229 . 003 Necromeny occupies a midpoint in the continuum of nematode–insect interactions that span from mutualism to pathogenicity . To address possible developmental effects of long-term exposure to ZTDO , we attempted to grow worms in the presence of the ZTDO by placing eggs onto OP50 bacteria in plates containing 10 µl of 0 . 5% ZTDO suspended under the plate lids , which is approximately a third of the amount used in chemotaxis assays . Surprisingly , very few of the synchronized eggs from wild type and Ppa-obi-1 ( tu404 ) hatched after 36 hr , compared to ∼94% of the untreated eggs ( Figure 2A ) . ZTDO therefore may function not just as a volatile attractant but also as an antagonistic host-derived developmental regulator . 10 . 7554/eLife . 03229 . 004Figure 2 . The oriental beetle pheromone ZTDO inhibits P . pacificus development and induces paralysis . ( A ) Ppa-obi-1 mutant is hypersensitive to 0 . 10% ZTDO compared to wild type PS312 ( California ) . However , PS312 is more sensitive to 0 . 5% ZTDO than another strain , PS1843 ( Washington ) . PS312 ( n = 372 , 82 , 90 , 320 ) ; PS1843 ( n = 309 , 88 , 92 , 237 ) ; Ppa-obi-1 ( n = 332 , 69 , 71 , 389 ) ; C . elegans wild-type N2 ( n = 316 , 63 , 84 , 174 ) . ( B ) Both wild-type PS312 and Ppa-obi-1 mutant J2 larvae are more sensitive to ZTDO than PS1843 . PS312 ( n = 154 , 115 ) ; PS1843 ( n = 110 , 102 ) ; Ppa-obi-1 ( n = 104 , 100 ) ; C . elegans wild-type N2 ( n = 23 , 73 ) . ( C ) Ppa-obi-1 J4 larvae is uniquely sensitive to ZTDO compared to PS312 ( p < 0 . 0001 ) . PS312 ( n = 209 , 290 ) ; PS1843 ( n = 102 , 99 ) ; obi-1 ( n = 244 , 237 ) ; C . elegans wild-type N2 ( n = 69 , 109 ) . ( D ) 90 min exposure of a Ppa-obi-1 ( tu404 ) J4 larvae to ethanol control compared to ( E ) 0 . 50% ZTDO resulted in a rigid fully paralyzed posture . ( F ) Adults are not paralyzed by ZTDO . PS312 ( n = 108 , 113 ) ; PS1843 ( n = 114 , 107 ) ; Ppa-obi-1 ( n = 109 , 109 ) ; C . elegans wild-type N2 ( n = 62 , 72 ) . ( G ) 0 . 01% ZTDO prevents resumption of post-dauer development in approximately 40% of wild-type animals and nearly all Ppa-obi-1 ( tu404 ) mutants . This ZTDO inhibition of dauer exit is dependent on the cGMP-dependent PKG , Ppa-egl-4 . PS312 ( n = 384 , 654 , 267 ) ; PS1843 ( n = 195 , 139 , 101 ) ; Ppa-obi-1 ( n = 339 , 343 , 146 ) ; Ppa-egl-4 ( tu374 ) ( n = 123 , 181 , 77 ) ; Ppa-obi-1;Ppa-egl-4 ( n = 132 , 144 , 63 ) ; N2 ( n = 131 , 109 , 115 ) . ( H ) Exposure to 0 . 50% ETDA moth pheromone from Spodoptera littoralis does not paralyze Ppa-obi-1 mutant and wild-type J4 larvae . PS312 ( n = 40 , 103 ) ; PS1843 ( n = 40 , 64 ) ; Ppa-obi-1 ( n = 40 , 193 ) . p values indicate Dunnett's multiple comparisons test to wild-type PS312 . ***p < 0 . 001; **p < 0 . 01 indicate difference between ethanol control and ZTDO exposed groups ( pairwise unpaired t test ) . Error bars are SEM and n is the number of nematodes tested . DOI: http://dx . doi . org/10 . 7554/eLife . 03229 . 004 Detailed examination revealed that Ppa-obi-1 ( tu404 ) eggs had significantly lower hatching rate in 0 . 10% ZTDO compared to wild-type PS312 . Interestingly , the Washington strain that has natural cGMP-independent attraction to ZTDO was more resistant to 0 . 25–0 . 50% ZTDO than the ZTDO-insensitive California strain . Although ZTDO exposure had no noticeable effect on egg laying output by adult hermaphrodites , eggs continuously exposed to ZTDO for 2 days at an early stage remain arrested at the pre-comma globular embryonic stage , before gastrulation . The embryocidal effect of ZTDO was concentration-dependent and species-specific , since wild-type C . elegans eggs were completely unaffected ( Figure 2A ) . Ppa-obi-1 ( tu404 ) hypersensitivity to the developmental and paralyzing effects of ZTDO also continued into the subsequent J2 and J4 larval stages ( Figure 2B , C ) . Both wild-type PS312 and Ppa-obi-1 ( tu404 ) J2 larvae exhibited ZTDO-induced paralysis , while the Washington strain was significantly more resistant to paralysis . However , by the J4 larval stage , only Ppa-obi-1 ( tu404 ) was sensitive to ZTDO compared to both wild-type P . pacificus strains ( Figure 2D–E ) . Ppa-obi-1 ( tu404 ) J4 larvae exposed to ZTDO displayed initial agitation , followed by full body paralysis between 90 and 120 min . In contrast , wild-type C . elegans J4 larvae displayed only minute sensitivity to ZTDO . To determine if the Ppa-obi-1 ( tu404 ) mutation renders it susceptible to paralysis from other insect pheromones , we also tested the moth pheromone ETDA and found that the J4 larvae of Ppa-obi-1 ( tu404 ) and the wild-type strains were not affected ( Figure 2H ) . All strains , including Ppa-obi-1 ( tu404 ) , were insensitive to ZTDO-induced paralysis when they reach adulthood ( Figure 2F ) , the dominant stage used in our chemotaxis assays . Taken together , our findings show that ZTDO is capable of arresting wild-type P . pacificus embryogenesis and severely paralyzing J2 larvae , indicating that ZTDO is a potential host-specific developmental regulator . We speculate that wild-type Ppa-OBI-1 may serve to protect P . pacificus embryos and reproductive larvae against the volatile anesthetic-like effects of ZTDO from the adult beetle host . The dauer larva ( DL ) is an alternative , non-reproductive , long-lasting stage in nematodes capable of withstanding starvation , as well as the infective stage in parasitic nematodes ( Lee , 2002 ) . To explore the possibility that ZTDO also acts to restrain the dispersive stage of P . pacificus development after hatching , we asked if ZTDO can also regulate DL exit in the presence of food using culturing plates containing 0 . 001–0 . 01% ZTDO and bacteria . Nearly all untreated wild-type P . pacificus DL developed into J4 larvae after 2 days on food , whereas ∼40% of the wild-type worms remained as active DL on 0 . 01% ZTDO with OP50 food that would have otherwise promoted reproductive development ( Figure 2G ) . However for Ppa-obi-1 DL , this dauer exit defect is exacerbated by ZTDO due to paralysis . Approximately 80% of the developmentally blocked Ppa-obi-1 ( tu404 ) DL show immobility on 0 . 001% ZTDO , and nearly 95% of the Ppa-obi-1 ( tu404 ) DL became paralyzed on 0 . 01% ZTDO . Although DL is the primary infective stage in many entomophilic nematodes found on live beetles ( Weller et al . , 2010 ) , we were unable to assess DL's chemotaxis response to ZTDO because of its immobilizing effect on DL . The negative regulation of DL exit by ZTDO is specific to P . pacificus and barely affected C . elegans even at 0 . 01% ZTDO . These results show that 0 . 01% ZTDO can block active wild-type P . pacificus DL from progressing to the reproductive stage without causing paralysis , but 0 . 001–0 . 01% ZTDO significantly reduced dauer exit in Ppa-obi-1 DL by primarily inducing paralysis and death . Since Ppa-obi-1 ( tu404 ) has a defect in DL exit that is further enhanced by exposure to ZTDO , we asked if DL arrest in Ppa-obi-1 ( tu404 ) by ZTDO is similarly dependent on Ppa-EGL-4 function . Egl-4 mutants are defective in a cGMP-dependent protein kinase in C . elegans , which affect dauer formation as well as chemosensation ( Daniels et al . , 2000; L'Etoile et al . , 2002 ) . We found that while the Ppa-egl-4 ( tu374 ) loss-of-function mutation did not significantly affect the ability to exit the dauer stage , the Ppa-egl-4 ( tu374 ) mutation in the Ppa-obi-1 ( tu404 ) background did partially suppress the inhibitory effect of ZTDO , such that Ppa-egl-4;Ppa-obi-1 post-dauer development on 0 . 001% ZTDO is higher than Ppa-obi-1 ( tu404 ) ( Figure 2G ) . The Ppa-egl-4 ( tu374 ) mutation did not , however , reduce the ZTDO inhibition of DL exit in the presence of a higher amount of 0 . 01% ZTDO . This outcome suggests that the susceptibility of Ppa-obi-1 ( tu404 ) DL to ZTDO is partially dependent on Ppa-egl-4 . While the loss of Ppa-obi-1 alone can cause reduced rate of DL exit on food independent of ZTDO , the aggravated DL arrest by ZTDO in Ppa-obi-1 ( tu404 ) is partially mediated by Ppa-egl-4 . Although wild-type P . pacificus adult animals are attracted to ZTDO , the host beetle sex pheromone regulates P . pacificus development by inhibiting embryogenesis , restricting DL exit , and inducing paralysis in larval stages . This ZTDO sensitivity is the most acute at the embryonic stage and is likely to be buffered by wild-type Ppa-OBI-1 in later developmental stages as the strongest hypersensitivity to ZTDO was observed in the J4 stage of the Ppa-obi-1 ( tu404 ) mutant . By contrast , the PS1843 Washington strain that is naturally attracted to ZTDO is more resistant to ZTDO in non-dauer stages compared to the PS312 California wild type . Further studies in other natural populations are needed to determine the correlation between ZTDO attraction and resistance to ZTDO-induced paralysis in P . pacificus . In addition to the specific chemosensory defect in Ppa-obi-1 ( tu404 ) mutants , we also noticed a linked phenotype in its body morphology . Detailed measurements indicate that Ppa-obi-1 ( tu404 ) animals are significantly longer and thinner than wild type ( Figure 3 ) , resulting in a slim body phenotype with adult Ppa-obi-1 ( tu404 ) hermaphrodites ∼20% longer than wild-type animals . The slim body phenotype may also contribute to Ppa-obi-1 ( tu404 ) 's moderate egg-laying defect ( egl; Table 1 ) and reduction in hatching rate similar to Ppa-egl-4 ( tu374 ) mutants , possibly due to defects in vulva formation and cuticle development . Using molecular markers and the slim phenotype , we were able to delineate the Ppa-obi-1 lesion to an ∼138 kb region on Supercontig 1 of chromosome I between markers SSLP21 and SSLP17 ( Figure 4A ) . We subsequently identified all polymorphic sites between PS312 and Ppa-obi-1 ( tu404 ) in the fine mapping interval by Illumina whole genome sequencing and focused on disruptions of predicted open reading frames . We found a single base pair deletion in a predicted coding region orthologous to C06G1 . 1 in C . elegans and confirmed this tu404 deletion by PCR amplification and Sanger sequencing of both wild type and Ppa-obi-1 ( tu404 ) genomic and cDNA . The tu404 allele contains a single base pair deletion in exon 9 . The resulting frame shift and nonsense mutation are predicted to stop translation half way into the reading frame to result in a hypomorphic allele . In the two Ppa-obi-1 splice forms , the longer transcript contains a VKDDDPR peptide that may be the result of an alternative splice donor site . Both splice variants were isolated from the California and Washington strains in equal abundance by sub-cloning RT-PCR products ( data not shown ) , and no differences in nucleotide sequence were found in the coding region of these two strains . Therefore , splice form variation in Ppa-OBI-1 is unlikely to contribute to the difference in insect pheromone attraction between these two strains . The region encoding for VKDDDPR is also absent in all the Ppa-obi-1 orthologs we have examined in available databases ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 03229 . 005Figure 3 . Ppa-obi-1 mutants are long and slim . ( A and B ) Wild-type J4 larva and young adult . ( C and D ) Ppa-obi-1 J4 larva and young adult . ( E and F ) The body lengths and widths of wild-type ( dotted red line ) versus Ppa-obi-1 mutants ( solid black line ) recorded over 4 days post J4 stage ( μm ) ; n = 15–30 animals for each time point . Error bars are SEM . Scale bars denote 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03229 . 00510 . 7554/eLife . 03229 . 006Table 1 . Ppa-obi-1 affects hatching rate and brood sizeDOI: http://dx . doi . org/10 . 7554/eLife . 03229 . 006Hatching rate ( % ) Brood size ( 3 days ) nWild-type ( PS312 ) 95 ± 3156 ± 1618Ppa-obi-1 ( tu404 ) 85 ± 3103 ± 171510 . 7554/eLife . 03229 . 007Figure 4 . Ppa-obi-1 gene structure and protein comparisons . ( A ) The order of mapping markers is depicted on top ( not to scale ) . The Ppa-obi-1 mRNA is SL1 trans-spliced and contains 19 exons spanning ∼19 kb . The 50 bp 5′ UTR and 341 bp 3′ UTR are indicated in light blue . The tu404 allele contains a single base pair deletion in exon 9 that results in a frameshift , converting a codon to a stop codon ( inverted triangles ) . The red dot represents the splice variant . ( B ) A diagram of the secretion and two BPI lipid-binding domains ( PFAM and SMART ) . ( C–E ) Transgenic rescue of the Ppa-obi-1 ( tu404 ) . ( C ) A 4 . 3 kb promoter driving Ppa-obi-1 cDNA completely rescued cGMP-dependent ZTDO attraction in Ppa-obi-1 ( tu404 ) mutants . PS312 ( N = 21 ) ; obi-1 ( N = 15 ) ; obi-1 rescue ( N = 12 ) . The Ppa-obi-1 transgenic rescue ( D ) restored the body length defect in Ppa-obi-1 ( tu404 ) mutants , PS312 ( n = 15 ) ; obi-1 ( n = 15 ) ; obi-1 rescue ( n = 19 ) as well as ( E ) significantly reduced ZTDO-induced paralysis in Ppa-obi-1 ( tu404 ) J4 larvae . PS312 ( N = 11 ) ; obi-1 rescue rfp+ ( N = 12 ) ; obi-1 rescue rfp− ( N = 5 ) . **p < 0 . 01; ***p < 0 . 001; ****p < 0 . 0001 ( Tukey's multiple comparisons test ) . Error bars represent SEM . N is the number of assays performed and n is the number of individual worms scored for each time point . DOI: http://dx . doi . org/10 . 7554/eLife . 03229 . 00710 . 7554/eLife . 03229 . 008Figure 4—figure supplement 1 . An alignment of putative Ppa-OBI-1 protein orthologs . Caenorhabditis elegans ( C06G1 . 1 ) , C . briggsae ( CBP03891 ) , C . remanei ( RP35181 ) , C . japonica ( JA26805 ) , Brugia malayi ( BM32564 ) , and a human Bactericidal Permeability Increasing protein ( NP_001716 ) . The predicted protein secretion signal peptides for Ppa-OBI-1 ( PP33427 ) and C06G1 . 1 ( CE27054 ) are 17–18 amino acids long at the N terminus ( red bar ) . The longest predicted protein splice forms are used in the alignment , with orthologs in Caenorhabditis species containing ‘GS’ ( red boxed ) and Ppa-OBI-1 in P . pacificus containing ‘VKDDDPR’ peptides ( red boxed with a red dot ) . The single base pair deletion in Ppa-obi-1 ( tu404 ) converts the leucine codon to a stop codon ( inverted red triangle ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03229 . 00810 . 7554/eLife . 03229 . 009Figure 4—figure supplement 2 . A maximum likelihood relationship of lipid-binding proteins in P . pacificus and C . elegans . The ‘PP’ and ‘CE’ prefixes denote P . pacificus and C . elegans proteins , respectively ( www . wormbase . org ) . 10 orthologous protein pairs are circled . Ppa-OBI-1 ( PP33427 ) and the 12 other paralogs are also represented in the P . pacificus proteome ( Pristionchus . org ) . A human BPI protein ( GenBank NP_001716 . 2 ) was used as outgroup . Bootstrap values are indicated along branches . DOI: http://dx . doi . org/10 . 7554/eLife . 03229 . 00910 . 7554/eLife . 03229 . 010Figure 4—figure supplement 3 . Chemosensory behavior of C . elegans mutants C06G1 . 1 and nrf-5 . A mutation in C06G1 . 1 ( vc20357 ) , the ortholog of Ppa-obi-1 , does not alter chemotaxis behavior towards 10% ZTDO , 1% 2 , 3-pentanedione , 1% 2 , 3-butanedione ( diacetyl ) , and 1% isoamyl alcohol . However , nrf-5 ( sa513 ) mutants have reduced attraction to 1% 2 , 3-butanedione and 1% isoamyl alcohol . nrf-5 ( sa513 ) was not tested for response to 2 , 3-pentanedione . Significant differences to wild-type are indicated by p values ( two sampled t test ) . N ≥ 10 replicates . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 03229 . 01010 . 7554/eLife . 03229 . 011Figure 4—figure supplement 4 . Reduction of C06G1 . 1 function produce a slimmer body phenotype . ( A ) RNAi using two dsRNA fragments of C06G1 . 1 fed to C . elegans rrf-3 ( pk1426 ) resulted in a longer and thinner body proportion . ( B ) Genetic mutation in C06G1 . 1 ( VC20357 ) resulted in significantly slimmer body proportion ( increase length/width ) similar to RNAi against C06G1 . 1 and the Ppa-obi-1 ( tu404 ) mutant . n ≥ 13 animals . Note that y-axes do not begin at zero . Significant differences to wild-type are indicated by p values ( two sampled t-test ) . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 03229 . 011 To confirm that Ppa-obi-1 is indeed responsible for the lack of ZTDO attraction in adults and for hypersensitivity to ZTDO-induced paralysis in the J4 larvae , we made transgenic lines containing a 4 . 3 kb Ppa-obi-1 promoter driving the Ppa-obi-1 full-length cDNA ( Schlager et al . , 2009 ) . The introduction of the wild-type Ppa-obi-1 transgene restored the cGMP-dependent ZTDO attraction in Ppa-obi-1 ( tu404 ) adults compared to Ppa-obi-1 ( tu404 ) adults and sibling animals that did not segregate for the rescue construct ( Figure 4C ) . Furthermore , the Ppa-obi-1 rescue transgene significantly shortened the body length of Ppa-obi-1 ( tu404 ) adults ( Figure 4D ) , confirming that both the ZTDO chemosensory and the long body length phenotype used for subsequent positional mapping were due to the same mutation in Ppa-obi-1 ( tu404 ) . Lastly , we tested the ability of the wild-type Ppa-obi-1 transgene to reduce the ZTDO hypersensitivity and found that it significantly reduced the percentage of ZTDO-induced paralysis in Ppa-obi-1 ( tu404 ) J4 larvae ( Figure 4E ) . Ppa-obi-1 therefore mediates ZTDO chemosensation , body dimensions , and ZTDO susceptibility . Sequence analyses of coding regions show that Ppa-obi-1 and C06G1 . 1 encode for proteins that represent a class of highly conserved but functionally diverse group of secreted lipid-transfer/lipid-binding proteins present in almost all metazoans except Drosophila ( Beamer et al . , 1998 ) , including lipopolysaccharide-binding proteins ( LBP ) ( Tobias et al . , 1997 ) , bactericidal permeability-increasing proteins ( BPI ) ( Weiss et al . , 1978 ) , phospholipid transfer proteins ( PLTP ) ( Gautier and Lagrost , 2011 ) , and cholesteryl ester transfer proteins in mammals ( CETP ) ( Kawano et al . , 2000 ) . The LBP/BPI/CETP/PTLP family in humans is known to play a role in immunological responses to bacterial pathogens and mediate the transfer of cholesteryl ester and triglycerides in the blood plasma ( Bruce et al . , 1998; Kawano et al . , 2000; Schultz and Weiss , 2007 ) . Ppa-OBI-1 and C06G1 . 1 each contain two predicted lipid-binding domains as well as a predicted secretion peptide sequence in the N-terminal regions ( PFAM and SMART; SignalP 4 . 0 Server ) ( Figure 4—figure supplement 1 ) . To identify all members of the LBP family in the sequenced genome and transcriptome of P . pacificus , we performed reciprocal BLASTP searches ( WormBase WS221 ) , which indicated 12 members predicted to be encoded in the P . pacificus genome . Maximum likelihood phylogenetic analysis of the OBI-1/C06G1 . 1 LBP protein family indicates 10 closely related orthologous pairs in the two species . Interestingly , the lengths of the branches suggest that all the orthologs between P . pacificus and C . elegans share similar divergence times , with the exception of the OBI-1/C06G1 . 1 pair that seemed to have undergone faster divergence relative to other paralogs ( Figure 4—figure supplement 2 ) . Consistent with their roles in immune response in mammals , a member of the LBP family in C . elegans , F44A2 . 3 , was found to be up-regulated in response to several species of pathogenic bacteria ( Wong et al . , 2007 ) . Another C . elegans LBP member , NRF-5 , is partially necessary for nose contraction in response to the anti-depressant fluoxetine ( Prozac ) ( Choy et al . , 2006 ) , and for mediating phosphatidylserine transfer from apoptotic cells to engulfing cells during cell corpse removal ( Zhang et al . , 2012 ) . We found that nrf-5 ( sa513 ) mutants are also defective in attraction to 2 , 3-butanedione ( diacetyl ) , and isoamyl alcohol , although the C06G1 . 1 ( vc20357 ) allele with a missense mutation did not show chemosensory defects ( Figure 4—figure supplement 3 ) . Thus LBP members may have a conserved previously uncharacterized role in chemosensation in diverse nematodes . To determine if C06G1 . 1 can also regulate body morphology in C . elegans , we performed C06G1 . 1 RNAi by feeding . C06G1 . 1 RNAi was previously observed to generate a long body phenotype in a large-scale RNAi screen in C . elegans ( Simmer et al . , 2003 ) . We used this fragment ( Fragment 1 ) , along with another C06G1 . 1 fragment more proximal to the 3′ end ( Fragment 2 ) , to quantify the effect of C06G1 . 1 RNAi in the enhanced RNAi rrf-3 ( pk1426 ) strain . Although C06G1 . 1 RNAi did not significantly change the absolute body length of adult worms , we found RNAi against C06G1 . 1 using both fragments resulted in a significantly slimmer body proportion ( length/width ratio ) similar to the Ppa-obi-1 ( tu404 ) mutant ( Figure 4—figure supplement 4 ) . A missense mutation in C06G1 . 1 ( VC20357 ) also resulted in a similar slim body phenotype . Thus both Ppa-OBI-1 and its C . elegans ortholog also share a conserved function in regulating body proportions . To characterize which tissue types Ppa-OBI-1 functions in , we sought to determine the Ppa-obi-1 expression pattern using a Ppa-obi-1p::gfp transcriptional reporter gene containing a 4 . 3 kb promoter fragment upstream of the Ppa-obi-1 ATG initiation codon . Our analysis was based on a stably integrated transgenic line csuIs01 and parsimonious inferences to homologous C . elegans cell positions and known cellular functions . Detailed analyses of all developmental stages show that the earliest Ppa-obi-1 expression occurs in at least one cell in the pre-comma embryonic stage but soon disappears until weak expression reappears in the hyp7 and seam cells in J4 larvae ( Figure 5 ) . The most robust and diverse Ppa-obi-1 expression was found in mid-J4 hermaphrodites , which is consistent with real-time quantitative PCR results indicating that transcript levels in J2 and J3 larvae are less than in J4 larvae ( data not shown ) . In the mid-J4 stage , Ppa-obi-1 is strongly expressed in the major hypodermal syncytium and the lateral hypodermal cells that cover most of the body ( hyp7 and seam cells ) as well as the ventral tail ( hyp 8 and 9 ) ( Figure 5 ) . In contrast to young adult C . elegans , which have 16 seam cells on each lateral surface ( H0–H3 , V1–V6 , T descendants; The WormAtlas ) , we counted only 15 seam cells per side in P . pacificus J4 larvae ( n = 20 ) . Ppa-obi-1 is also strongly expressed in the P ( 5–7 ) . p cell descendants that form the vulva lumen ( vulC , D , E ) during the mid-J4 stage until just prior to vulval eversion as well as in the vulva muscles ( likely vm1 , 2 ) ( Figure 5B inset; Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 03229 . 012Figure 5 . Ppa-obi-1 and C06G1 . 1 expression in their respective nematode species . ( A–D ) An integrated Ppa-obi-1 promoter::gfp transgene ( csuIs01 ) in P . pacificus . ( A ) Lateral view of a single animal showing prominent expression in an amphid sheath cell , seam cells , and the syncytial hypodermal cell ( hyp7 ) in a J4 hermaphrodite . ( B ) Medial view of a transgenic animal with expression in the putative dorsal and ventral inner labial ( IL ) or outer labial ( OL ) neuronal support cells ( socket or sheath ) , the duct and excretory cells on the ventral side , the amphid sheath cell on the dorsal side ( arrow head ) , and the vulval muscles . The inset shows a ventral view of the young adult vulva in another animal . Autofluorescence is visible in the intestine . ( C ) A DiI-labeled neuron ( red arrow ) anterior to a Ppa-obi-1p::gfp expressing amphid sheath cell ( white arrow ) is dorsal to an excretory cell ( arrow head ) . A putative IL or OL socket or sheath cell ( arrow ) is also visible . ( D ) DIC and fluorescent overlay image of expression in early P . pacificus embryo . ( E–I ) C06G1 . 1 promoter::gfp transgenes ( csuEx02 , csuEx03 ) in C . elegans . ( E ) Lateral view of a seam cell expression in a L3 hermaphrodite . ( F ) Medial view of putative inner or outer labial neuron support cells ( arrow ) and excretory cells ( arrowhead ) . No amphid sheath shows gfp expression . ( G ) Ventral view of head region shows prominent expression in excretory cells and canal ( arrow ) . ( H ) Ventral view of young adult vulva . ( I ) Embryonic expression in C . elegans embryos appears later , stronger , and in more cells than in Ppa-obi-1 . Anterior is left and the scale bars represent 50 μm . Each panel is representative of ≥50 animals . DOI: http://dx . doi . org/10 . 7554/eLife . 03229 . 01210 . 7554/eLife . 03229 . 013Figure 5—figure supplement 1 . Ppa-obi-1 is expressed in putative amphid socket cells , amphid sheath cells , and vulval cells . Dorsal and ventral views of Ppa-obi-1p::gfp expression in roller prl-1 ( tu92 ) background with merged DIC and fluorescence ( A , C , E ) and corresponding fluorescence images ( B , D , F ) . ( A and B ) Dorso-lateral view of head region shows possible Ppa-obi-1 expression in a socket cell ( top ) , an IL sheath cell ( bottom ) , two seam cells ( asterisks ) , and two amphid sheath cells ( arrows ) . ( C and D ) Ventral view of a mid-J4 larvae vulva shows Ppa-obi-1p::gfp expression in P ( 5–7 ) . p vulval epidermal cell descendants , a putative vulval muscle ( triangle ) , and a hyp7 cell ( asterisk ) . ( E and F ) Ventral view of a young adult vulva ( top left ) shows Ppa-obi-1p::gfp expression in P ( 5–7 ) . p vulval epidermal cell descendents and four vulval muscles ( triangles ) and a lateral view of a mid-J4 vulva ( bottom right ) in another animal . Anterior is left and the scale bars represent 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03229 . 01310 . 7554/eLife . 03229 . 014Figure 5—figure supplement 2 . Targeted misexpression of Ppa-obi-1 in C . elegans . L2 larvae show amphid sheath cell expression using ( A ) Cel-T02B11 . 3 promoter::Ppa-obi-1:gfp ( csuEx29 ) and ( B ) Cel-F16F9 . 3 promoter::Ppa-obi-1:gfp ( csuEx30 ) . The co-injection marker myo-2p::DsRED is visible in the anterior pharyngeal bulb due to bleeding over into the GFP spectrum . Arrows indicate amphid sheath cell bodies . ( C ) Chemotaxis response to the ZTDO beetle pheromone remains unchanged towards 2% or 10% ZTDO . ( D ) ZTDO-induced paralysis in L4 larvae shows that transgenic Cel-F16F9 . 3p::Ppa-obi-1::gfp are better protected than wild-type animals in a 16 hr assay . N2 ( N = 8 ) , T02B11 . 3p::Ppa-obi-1 ( N = 6 ) , F16F9 . 3p::Ppa-obi-1 ( N = 6 ) . Anterior is left and the scale bar represents 25 μm . p value indicates Dunnett's multiple comparisons test to wild-type N2 . Error bars are SEM and N is the number of assays performed . DOI: http://dx . doi . org/10 . 7554/eLife . 03229 . 014 Most notably , Ppa-obi-1 expression was observed in the two bilaterally symmetrical specialized epithelial cells that envelope the chemosensory neurons , known as the amphid sheath cells ( Figure 5A–B , arrows ) . The amphid sheath cells are support cells for the sensory amphid neurons and completely enclose the ciliated endings of the AWA , AWB , and AWC chemosensory neurons in C . elegans . Ppa-obi-1p::gfp expression in the putative amphid sheath cells lie dorsal–lateral and posterior to the DiI-labeled amphid neurons ASK , ADL , and ASI ( Figure 5C ) . DiI staining of amphid neurons in Ppa-obi-1 mutants is superficially wild type , suggesting that Ppa-OBI-1 is unlikely to be involved in the development of the amphid neurons ( data not shown ) . We also noticed expression in four unknown cell types anterior to the metacarpus ( anterior pharyngeal bulb ) , which could be putative neuronal support cells for the inner labial or outer labial quadrant sensilla ( IL or OLQ ) ( Figure 5—figure supplement 1 ) . Ppa-obi-1 is also expressed in putative duct and excretory cells on the ventral anterior side that form part of the excretory system ( Figure 5B , arrowhead ) . Based on conserved DiI labeling of P . pacificus amphid neurons and cell positions , the Ppa-obi-1 expression profile and secretion peptide sequence suggest that Ppa-OBI-1 is likely an extracellular protein involved in sensory neuron support , epithelial cell development , and osmoregulation . Although Ppa-OBI-1 and its ortholog C06G1 . 1 in C . elegans share 61% amino acid identity , P . pacificus and C . elegans have opposite responses to the pheromone ZTDO . While small changes in sequence can lead to large changes in protein function , the absence of expression of the C . elegans obi-1 ortholog in amphid sheath cells suggests that changes in gene expression pattern may also play a role in determining ZTDO sensitivity . We therefore sought to determine if the Ppa-obi-1 ortholog C06G1 . 1 have shared tissue expression with Ppa-obi-1 . We examined endogenous C06G1 . 1p::gfp expression in C . elegans in five independent transgenic lines with extrachromosomal arrays and found strong mid-L4 larvae expression in the seam cells , two tail hypodermal cells , four cells in the excretory duct system ( duct , excretory , gland cells ) , vulval muscles , and possible IL or OLQ sheath or socket neuronal support cells similar to the expression pattern observed for Ppa-obi-1 ( Figure 5E–H ) . In contrast to the earliest Ppa-obi-1 expression in pre-comma stage , C06G1 . 1 expression is detectable during late embryogenesis in putative seam cells ( Figure 5I ) . This seam cell expression is earlier than Ppa-obi-1p::gfp in the L1 stage , and is maintained until the young adult stage . A reason for this earlier seam cell expression could be that the equivalent P . pacificus J1 larvae remain inside the egg while C . elegans eggs hatch as L1 larvae ( Fürst von Lieven , 2005 ) . Like Ppa-obi-1 , C06G1 . 1 expression is largely absent in all tissue types when the transgenic animals become 1-day old adult hermaphrodites , coinciding with the fusion of seam cells into a syncytium . This observation indicates that C06G1 . 1 expression in the excretory and epithelial system may be coordinated and involved in seam cell development . Our expression analysis of obi-1 orthologs in the their respective nematode species revealed that the most striking difference is the lack of expression in the amphid sheath cells in C . elegans , suggesting that unlike Ppa-obi-1 , C06G1 . 1 is not likely to function directly with chemosensory neurons in C . elegans . In addition to differences in gene expression pattern , the possibility that there is a ZTDO receptor that interacts with Ppa-OBI-1 in P . pacificus but not in C . elegans may also help to explain how Ppa-obi-1 orthologues confer different ZTDO sensitivities in the two species during certain developmental stages . To explore yet another possibility that changes in both protein function and expression may lead to differences in ZTDO sensitivity , we expressed the Ppa-obi-1:cDNA under the C . elegans amphid sheath cell-specific promoters T02B11 . 3 and F16F9 . 3 in C . elegans ( Bacaj et al . , 2008; Kramer et al . , 2010; Oikonomou et al . , 2011; Procko et al . , 2011 ) . Both T02B11 . 3p::Ppa-obi-1:gfp ( csuEx29 ) and F16F9 . 3p::Ppa-obi-1:gfp ( csuEx30 ) transgenic animals showed strong GFP expression in the amphid sheath cells ( Figure 5—figure supplement 2 ) as well as phasmid sheath cells in the tail region ( not shown ) . While the targeted misexpression of Ppa-obi-1 in C . elegans did not alter the strong avoidance behavior towards either 2% or 10% ZTDO , Ppa-obi-1 expression in C . elegans did reduce the paralytic effects of ZTDO ( Figure 5—figure supplement 2 ) . Because wild-type C . elegans N2 is not susceptible to ZTDO-induced paralysis at 0 . 5% ZTDO that P . pacificus is , we increased the stringency of the assay by using 10% ZTDO , incubating for 16 hr , and exposing the animals in the absence of OP50 food . Under this enhanced ZTDO paralysis assay , we found that F16F9 . 3p::Ppa-obi-1:gfp ( csuEx30 ) transgenic L4 larvae but not T02B11 . 3p::Ppa-obi-1:gfp ( csuEx29 ) exhibited a small but statistically significant increase in protection from ZTDO-induced paralysis compared to wild-type N2 ( Figure 5—figure supplement 2C , D ) . Because we could not detect apparent differences in GFP expression patterns between the two amphid promoters , the ability of F16F9 . 3p::Ppa-obi-1:gfp to reduce ZTDO paralysis may be due to subtle transient expression differences . These results suggest that while the expression of the Ppa-OBI-1 protein in the C . elegans amphid sheath cells is not sufficient to reduce avoidance to ZTDO , it can contribute to the obi-1-dependent protection from ZTDO-induced paralysis . Sensitivity to ZTDO-induced paralysis/arrest is Ppa-obi-1-dependent predominantly during the J4 larval stage and , to a lesser degree , in the embryos and dauers . Given the diversity of tissues expressing Ppa-obi-1 in the J4 larvae , it is not apparent which tissues require Ppa-obi-1 expression for proper chemosensation and if the same tissues are involved with protection from ZTDO-induced paralysis in wild-type animals ( Figure 6 ) . Our gene expression studies show that despite coding highly similar proteins , P . pacificus Ppa-obi-1 is uniquely expressed in the amphid sheath cells compared to its C . elegans ortholog C06G1 . 1 . Based on this species-specific expression pattern , we are tempted to speculate that these chemosensory glial cells , the IL and amphid sheath cells , play a dominant role in mediating ZTDO sensitivity . However , additional experiments will be needed to test the possibility that epidermal cells such as the vulva , hypodermis , and seam cells , or cells in the excretory system may also mediate ZTDO sensing . 10 . 7554/eLife . 03229 . 015Figure 6 . Summary and model of the oriental beetle pheromone ZTDO on P . pacificus development and lifecycle . ( Top ) Peak Ppa-obi-1 expression in the wild-type J4 larvae coincides with the least sensitive developmental stage to the paralytic effects of the beetle pheromone ZTDO ( at 0 . 5% ) . Ppa-obi-1 mutants are most sensitive to ZTDO-induced paralysis in the J4 stage . Our working model on Ppa-OBI-1 functions in the J4 larvae hypothesizes that Ppa-OBI-1 proteins in the extracellular space , such as the lumen of the amphid sheath cell , may be required to remove ZTDO bound to odor receptors on neurons . ( Bottom ) This working model shows both P . pacificus adults and dauer larvae from non-beetle populations associate with the host oriental beetle through attraction to the beetle's sex pheromone ZTDO , whereby the progeny born on the beetle are arrested as eggs or dauer larvae . Death of the host allows microorganism to grow , which then encourages dauer larvae to exit and resume reproductive development . Several iterations of asynchronous nematode populations likely complete each dispersal cycle . DOI: http://dx . doi . org/10 . 7554/eLife . 03229 . 015 Given the growing evidence that the amphid sheath cells in C . elegans are critical for proper neuronal development and function of the 12 pairs of chemosensory amphid neurons ( Shaham , 2006 ) , we propose the following models on how the Ppa-OBI-1 protein can act in the sheath cells to directly or indirectly modulate ZTDO chemosensation . In C . elegans , many proteins have enriched expression in the amphid sheath , such as DAF-6 and the secreted venom allergen-like VAP-1 , and these secretions are corroborated by dense populations of vesicles in amphid sheath cells under electron microscopy ( Bacaj et al . , 2008; Doroquez et al . , 2014 ) . However , it is not yet known which amphid proteins , if any , are required in processing odors using behavioral assays . Interestingly , DAF-6 is a Patched-related protein with expression pattern that overlaps with Ppa-obi-1 expression ( amphid lumen , excretory canal , vulva lumen ) and is required for chemotaxis and dauer formation ( Perens and Shaham , 2005 ) . Similarly , Ppa-obi-1 expression in the putative sheath or socket cells of the inner labial neurons may also be involved in ZTDO sensing , as recent ultrastructural electron micrograph reconstructions indicate that BAG ( gas sensing ) and FLP ( nociceptive ) neurons in C . elegans associate closely with the socket cells of the inner labial sensillae ( Doroquez et al . , 2014 ) . One possible mode of action for ZTDO may be its ability to change cell membrane permeability , thereby indirectly affecting the functions of the glia-associated neurons . Hence , Ppa-OBI-1 may function to counter the cell membrane permeability changes brought about by ZTDO by interacting with cell membrane lipids . Alternatively , Ppa-OBI-1 could mediate pheromone binding directly . In lepidopterans such as the wild silk moth Antheraea polyphemus and the cotton leafworm Spodoptera littoralis , pheromone-degrading and odorant-degrading enzymes ( PDEs and ODEs ) are esterases specifically expressed in the male antennal sensillar lymph analogous to the amphid lumen ( Ishida and Leal , 2005; Durand et al . , 2011 ) . The termination of the signal by removing odors that have already triggered a response is an important process for preventing odor adaptation and increasing odor sensitivity . Thus , it is possible that Ppa-OBI-1 acts as a solubilizing factor for the lipid pheromones in the excretion-filled amphid sheath lumen ( Figure 6 ) . The weak ability of amphid sheath-specific expression of Ppa-OBI-1 to confer ZTDO protection in C . elegans suggests that both expression pattern and protein function could be important factors in ZTDO sensing in P . pacificus . Future work will be necessary to determine if Ppa-obi-1 expression restricted to the amphid sheath is sufficient to rescue the ZTDO attraction defect in P . pacificus and if the amphid sheath cells are required to mediate ZTDO sensing . Given that Ppa-obi-1-independent susceptibility to ZTDO occurs during the J2 larval stage when Ppa-obi-1 is expressed in the seam and hypodermal cells rather than in the amphid sheath that is only active in the J4 larvae , ZTDO-induced paralysis likely acts through the extracellular matrix of the nematode cuticle . P . pacificus adults , whose cuticle may be compositionally different from the larval cuticle , are not susceptible to ZTDO . However in P . pacificus embryos , the susceptibility of ZTDO may be due to ZTDO limiting gas or ion exchange with the environment rather than direct interaction between ZTDO and cellular components , given that nematode egg shells are considered to have very limited permeability accessible only to small molecular weight solutes and gas molecules ( Chitwood and Chitwood , 1974 ) . Since ZTDO exposure does not lead to embryonic arrest and paralysis in C . elegans , ZTDO must also interact with other non-conserved P . pacificus-specific molecules that have yet to be identified . There are very few known nematocides from natural compounds . One rare example comes from the oyster mushroom Pleurotus ostreatus , which secretes droplets that can immobilize rhabditid nematodes in an hour , though its effects on embryos have not been investigated ( Barron and Thorn , 1987; Stadler et al . , 1994 ) . Future work on the different susceptibility to ZTDO may help to answer if there is a wider pattern of negative correlation between ZTDO attraction and ZTDO-induced paralysis in other natural isolates , as observed in the California and Washington strains . Such antagonistic traits may reveal possible co-evolution between host finding and host tolerance in necromenic associations . We further speculate that in the natural ecology of P . pacificus , ZTDO may enable the beetle host to limit the nematode number and to thus synchronize the developmental stage of infective P . pacificus by arresting embryogenesis and lethally paralyzing J2 larvae . The unexpected role of a P . pacificus attractant to act as a developmental regulator against the younger larval stages , particularly the dauer larvae ( DL ) , suggests that the ZTDO-resistant adult stage is a potential dispersal stage . This attenuated antagonistic interaction also revealed that nematode larvae and adults may be susceptible to distinct selective pressures . P . pacificus DL does not disperse readily , likely due to the secretion of the wax ester nematoil that enables P . pacificus to form dauer towers when potential hosts are nearby ( Penkov et al . , 2014 ) . Therefore under natural conditions , lower concentrations of ZTDO than the concentrations we have tested may serve as a close-range host dwelling strategy by inhibiting DL dispersal to form dauer aggregates . The developmental role of ZTDO may be part of a mechanism to limit nematode dispersal until the beetles mature . In these respects , the promotion of a non-DL dispersal stage and inhibition of DL dispersal by ZTDO resemble other known host-derived compounds found in the entomophilic nematode B . xylophilus and the phoretic nematode C . japonica ( Okumura et al . , 2013; Zhao et al . , 2013 ) . Attraction and paralysis by ZTDO may have co-evolved in the necromenic association to allow successive populations of founder P . pacificus DL and adults to infect beetles , culminating in arrested DL remaining on sexually mature beetles during the dispersal cycle . The pheromone arrested DL state may also help to account the observation that Pristionchus nematodes emerge several days later than rhabditids from the same species of host beetles . Our findings that a lipid-binding protein has a role in coordinating chemosensation , body proportions , and protection from a host pheromone encourages fresh inquiries on additional genes involved in ZTDO sensory signaling and its mechanism of action . All nematodes were raised on NGM media containing OP50 E . coli at ∼23°C as previously described ( Hong and Sommer , 2006a; Hong et al . , 2008a ) . The following strains were used: P . pacificus: California ( PS312 ) wild type , Washington ( PS1843 ) , Ppa-obi-1 ( tu404 ) I; Ppa-egl-4 ( tu374 ) IV ( null ) ; Ppa-obi-3 ( tu405 ) ; csuIs01 [Ppa-obi-1p::gfp; Ppa-egl-20p::rfp; PS312 gDNA]X; rlh49 ( csuIs01; tu374 ) ‘Ppa-obi-1p::gfp; Ppa-egl-4’; rlh50 ( csuIs01; tu92 ) ‘Ppa-obi-1p::gfp; roller’; rlh51 ( csuIs01; tu404 ) ‘Ppa-obi-1p::gfp; Ppa-obi-1’ . C . elegans: Bristol ( N2 ) wild type , nrf-5 ( sa513 ) , csuEx02 [C06G1 . 1p::gfp; Cel-myo-2::DsRED]; csuEx03 [C06G1 . 1p::gfp; pRF4 roller] , csuEx29 [T02B11 . 3p::Ppa-obi-1:gfp; Cel-myo-2::DsRED] , and csuEx30 [F16F9 . 3p::Ppa-obi-1:gfp; Cel-myo-2::DsRED] . The whole-genome sequenced strain vc20357 was outcrossed to him-5 ( e1467 ) three times and the C06G1 . 1 locus was sequenced to confirm that the slim body phenotype is linked to the missense gk306193 mutation ( D48N ) . We X-ray mutagenized wild-type P . pacificus PS312 ( California ) at 325 Roentgen for 25 min with radioactive beryllium and picked 40 Po onto individual plates . We subsequently picked 1067 F2s from 400 F1s and screened 713 viable F3 populations for insensitivity to 150 nl of oriental beetle pheromone , Z-7-tetradecen-2-one ( 99% pure; Bedoukian Research , CT ) after 1 hr treatment with 1 mM exogenous 8-Bromo-cGMP ( Sigma–Aldrich , MO ) ( Hong et al . , 2008a ) . 100 mM nicotinic acid ( Sigma–Aldrich , MO ) was dissolved in water . Population chemotaxis assays were conducted for ∼16 hr at 23°C , and individual candidate F3 from plates showing insensitivity were rescued onto fresh plates by mouth pipetting . We repeated the cGMP-treated chemotaxis assay from the progeny of the candidate mutant lines two more times with at least two duplicate assay plates per line . Only those lines that failed the third round of chemotaxis assay were considered Oriental Beetle pheromone Insensitive obi alleles . Ppa-obi-1 maintained long body morphology and obi-3 displayed more coiled body posture after 3× outcrossing to wild-type PS312 and selecting the ZTDO insensitive phenotype . Because both the chemotaxis and morphological phenotypes co-segregate after 5× outcross , we subsequently used the long body phenotype as a visible marker for Ppa-obi-1 during positional mapping . Population chemotaxis assays were performed as previously described based on modified protocol for C . elegans ( Bargmann et al . , 1993; Hong and Sommer , 2006a ) . Well-fed J4 and adult worms were washed in M9 buffer twice before assay . For a more consistent response , C . elegans was washed in spring water rather than M9 buffer before the ZTDO avoidance assay . For the exogenous cGMP treatment , each washed population was split for mock or 1 hr treatment with 1 mM 8-Bromo-cGMP ( in water ) . For each condition , at least 10 worms had to reach the odor sources in each assay and 12 assays ( sample size N ) were performed over three sessions . Chemotaxis assays with transgenic obi-1 rescue worms were scored by rfp expression in the tail . 20 live worms or eggs from non-starved cultures were placed on NGM agar plates with fresh OP50 . For consistency , eggs were synchronized by picking ∼50 adults onto OP50 and allowing them to lay eggs for 3 hr . The eggs were then transferred to different plates and counted . 10 µl of ZTDO or ETDA diluted in ethanol was added to the lid of the plate , immediately sealed with Parafilm , and incubated at room temperature . Larvae were scored 3 hr later or the next day by counting the number of paralyzed and active worms . A worm was scored as paralyzed if it did not move at least one pharynx length after tapping on the outside of the plate . Eggs were scored 2 days later by counting the number of unhatched eggs . For dauer larvae exit assays on NGM plates containing none , 0 . 001% or 0 . 01% ZTDO ( vol/vol ) , 25 dauer larvae from the same 2- to 3-week old starved plate were picked onto each type of plate with OP50 and scored for arrested/active dauer larvae , J3 larvae , or J4 larvae after 2 days at ∼23°C . A single library was paired-end sequenced twice in separate flow cell lanes with 2 × 76 bp that yielded 61 million and 65 million reads after preprocessing ( quality control , adapter removal ) . More than 90% of all reads could be mapped to the P . pacificus genome ( www . pristionchus . org ) with BWA ( version 0 . 5 . 9 ) for each library . We then employed ACCUSA for SNP detection ( Froehler and Dieterich , 2010 ) . Briefly , we performed a head-to-head comparison of the mutant Ppa-obi-1 and wild-type PS312 mapped libraries . The algorithm represents read stacks ( base calls and qualities thereof ) as Dirichlet distributions . For any given genome position , we require a read coverage of at least three in both samples . The smaller read stack serves as reference distribution from which the larger stack is drawn . A probability density for this draw is computed , and a SNP is called if the log-density is lower than −30 . Stable transgene expression in P . pacificus requires co-injection of endogenous genomic DNA and co-injection markers containing compatible cohesive ends with the target gene as previously described ( Schlager et al . , 2009; Cinkornpumin and Hong , 2011 ) . A 4 . 3 kb Ppa-obi-1 promoter fragment was amplified from PS312 gDNA using RHL121/RHL136 ( primary ) and RHL124/RHL149 ( secondary ) primers . The Ppa-obi-1 promoter was then fused with the gfp- Ppa-rpl-23-3′UTR by PCR ( Hobert , 2002 ) . PS312 animals were injected with a DNA mix containing a final concentration of 10 ng/μl of the SpeI/SalI digested 4 . 3 kb Ppa-obi-1 promoter::NLS-gfp-SalI fusion PCR product from three independent reactions; 10 ng/μl SpeI digested Ppa-egl-20p::rfp linearized plasmid; and 60 ng/μl SpeI/SalI digested PS312 genomic DNA . Individual F1s expressing Ppa-egl-20p::rfp in the hermaphrodite tail were selected for Ppa-obi-1 expression analysis . Two stable transgenic lines were initially generated with similar expression patterns . We used the stably transmitted reporter line csuIs01[Ppa-obi-1p::gfp; Ppa-egl-20p::rfp; PS312 gDNA]X for detailed expression analyses . csuIs01 is likely to be the result of a spontaneous integration into the X chromosome based on the observation that transgenic males expressing the co-injection rfp marker crossed to Ppa-unc-1 ( ChrIV ) hermaphrodites gave rise to all F1 males lacking rfp expression but all F1 hermaphrodites expressing rfp . More than 50 individuals at all developmental stages were analyzed . Ventral views of the vulva expression were obtained from csuIs01 in the prl-1 roller background . The same 4 . 3 kb Ppa-obi-1 promoter was used to drive the full length 1 . 6 kb cDNA fused to GFP by injecting into PS312 and crossing into the Ppa-obi-1 ( tu404 ) background to generate csuEx27[Ppa-obi-1p::Ppa-obi-1 cDNA:gfp; Ppa-egl-20::rfp; Ppa-obi-1 ( tu404 ) gDNA] . For unknown reasons , the N-terminally fused gfp coding region did not express GFP . To create the C06G1 . 1p::gfp reporter gene construct , we PCR amplified a 3 . 2 kb upstream sequence from the C06G1 . 1 ATG using RHL169 and RHL187 and fused the promoter with GFP:Ppa-rpl-23 3′ UTR by fusion PCR , using RHL181 and RHL154 primers to obtain a 4183 bp amplicon . Independently amplified C06G1 . 1p::gfp fragments were co-injected with either pRF4 rol-6 ( su1006 ) or myo-2p::DsRED vectors along with the pBluescript vector as carrier DNA ( 50 ng/μl each ) . This reporter likely encompasses C06G1 . 1's maximum regulatory region because the 3 . 2 kb promoter fragment extends ∼300 bp upstream of the next gene on the opposite strand ( C06G1 . 14 ) , and C06G1 . 1 does not contain any intron larger than 1 kb that may act as potential intronic enhancers . Selection for DsRED animals in the reporter strain csuEx02 [C06G1 . 1p::gfp; Cel-myo-2::DsRED] results in 90% GFP+/DsRED+ progeny and equal DsRED+ expression in pharyngeal muscle cells , suggesting stable mitotic transmission of the transgene . All five F1 transgenic lines showed similar expression profile and intensity in post-hatching stages , but strong body epidermal expression was detected only in the csuEx03 [C06G1 . 1p::gfp; pRF4; pBST] line . Images were obtained using a Leica DM6000 fluorescent microscope and Leica SP5 confocal microscope . To create the T02B11 . 3p::Ppa-obi-1:gfp amphid sheath expressing translational fusion construct , an ∼2 . 1 kb T02B11 . 3 promoter fragment was amplified from C . elegans gDNA using RHL476/RHL490 ( primary reaction ) . The Cel-T02B11 . 3 promoter is then fused to the same 1 . 6 kb Ppa-obi-1 cDNA rescue construct by PCR using RHL477/RHL483 ( secondary ) to generate a 3 . 7 kb fusion . This construct and the GFP-unc-54-3′UTR fragment , amplified from the pPD95 . 75 vector using RHL480/RHL481 ( primary ) was fused using RHL501/RHL482 ( tertiary ) to create the complete construct . A similar scheme was used to generate F16F9 . 3p::Ppa-obi-1:gfp using RHL498/RHL500 ( promoter ) . C . elegans N2 were injected with a mixture of purified T02B11 . 3p::Ppa-obi-1:gfp ( 25 ng/μl ) or F16F9 . 3p::Ppa-obi-1:gfp , myo-2::DsRED ( 25 ng/μl ) and pBluescript carrier DNA ( 100 ng/μl ) . One stable line from each construct was characterized for ZTDO chemotaxis and paralysis . The GenBank accession for Ppa-obi-1 mRNA is JQ946290 . Primer sequences are listed in Table 2 . 10 . 7554/eLife . 03229 . 016Table 2 . PrimersDOI: http://dx . doi . org/10 . 7554/eLife . 03229 . 016PrimerPrimer sequence ( 5′→3′ ) Template targetRHL101TGAACGGGCTTTTAATCTGGPpa-obi-1 gDNARHL102GTGTCGAGATAGTCGGGCATPpa-obi-1 gDNARHL104ATGACTGCTCCAAAGAAGPpa-obi-1 cDNARHL107ATTTGCCCTTCGTTCCACPpa-obi-1 for RNAiRHL108CTGTCTCAGTATGCCGAATGPpa-obi-1 for RNAiRHL109ATTCAGCCGTGTTACAAATCPpa-obi-1 cDNARHL110CTGCATGTGTTCCGGTCTCGPpa-obi-1 cDNARHL113AAACCTCTGACACATGCAGCPpa-obi-1 cDNARHL121AGTTTCAAAGTTGCATGGPpa-obi-1 promoterRHL124CTTCTTTGGAGCAGTCATGTAGCGATAATCAGGAGTgfp; Ppa-obi-1 promoterRHL132ATGACTGCTCCAAAGAAGgfpRHL136CAACCGTGTGCAGTAGAACCPpa-obi-1 5′ RACERHL149ACTATCCTATCATCGGAAGCPpa-obi-1 promoterRHL151AGGGCATTGACAAATCATCGPpa-obi-1 cDNARHL154CACGACGTTGTAAAACGACGPpa-obi-1 cDNARHL169CGCGCTAAACCAATTCCGCCC06G1 . 1 promoterRHL181GTTGAGAGAAGAGGTGGAGTCCC06G1 . 1 promoterRHL187CTTCTTTGGAGCAGTCATCAGATCTGAAAATTTGACACATTGgfp; C06G1 . 1 promoterRHL267GGAACTAACATTAATGTCACC06G1 . 1 RNAiRHL268GAGAAGTGCATAGTTGATGC06G1 . 1 RNAiRHL477GCGCTACAGGGGATTTTGTCT02B11 . 3 promoterRHL480TGAAAGATGCAAGTAAAGGAGAAGAACTTTTCp95 . 75 GFPRHL481AAGGGCCCGTACGGCCGACTAp95 . 75 GFPRHL482GTAGGAAACAGTTATGTTTGGp95 . 75 GFPRHL484CTTCAGAAAATGATCAGGAGTCTGGTTCobi-1 cDNARHL487AGTGAAAAGTTCTTCTCCTTTACTTGCATCTTTCACGGAGTCobi-1 cDNA; gfpRHL490AACGAGAACCAGACTCCTGATCATTTTCTGAAGAAAGTTGAAAAACT02B11 . 3 promoter; obi-1 cDNARHL498GCGATAAGATCGGTCAATCTGAGF16F9 . 3 promoterRHL499GCCAGTAAGGGCTAGTAAGTGF16F9 . 3 promoterRHL500AACGAGAACCAGACTCCTGATCATATTTTGTTTCTTACTGTCTTGF16F9 . 3 promoterRHL501ATGGGTACGTTTCTGGGTATAGT02B11 . 3 promoter; obi-1 cDNAVR86GACTCCTGATTATCGCTACGPpa-obi-1 cDNAVR87AGATAGTCAAACTATGCATCPpa-obi-1 cDNAVR88CTAATAATCTACACTAAATGPpa-obi-1 cDNAAG11112CTCGGAGGAGGAACTGGATCPpa-beta-tubulin RT-qPCRAG11113GACCGTGTCAGAGACCTTAGPpa-beta-tubulin RT-qPCRRH12548GCAGGAACATATATTCGCAGPpa-egl-4 RT-qPCRRH12549TGTCACGGAACGTTTTGTACPpa-egl-4 RT-qPCRSL1GGTTTAATTACCCAAGTTTGAGSL1 for 5′ RACESSLP21FACTGGTGCTCATCGCAAGAMapping markerSSLP21RCCTTCTGTTTTCCTACCCCCMapping markerSSLP7FTCTTGCATAACACCGAACAAAMapping markerSSLP7RAGGGCGTTACGTGAATAAGCMapping markerSSLP17FCAATGCAGTAGCCGAATCATMapping markerSSLP17RCAATATTGGCCTTCACCCTGMapping markerS447FTGGGGATAGCGAAAATCATCMapping markerS447RCGCATCGTTATTCACGAAATTCMapping markerLetters in bold are based on different templates for fusion PCR . Body measurements were obtained as previously described ( Hong et al . , 2008a ) . In brief , we used a Leica DM6000 and LAS application to measure body dimensions at 24-hr intervals starting at mid-J4/L4 stage . To determine hatching rate , we picked single J4-stage PS312 hermaphrodites onto fresh OP50 plates and transferred the nematodes every 24 hr three times at 20°C . The total number of eggs present on the plates following each transfer and the resulting number of 3-day old animals for each worm during the 3-day period were used to calculate the percentage of hatching and brood size . The bacterial feeding strain containing the C06G1 . 1 fragment 1 was previously described in a large-scale C . elegans RNAi screen ( Simmer et al . , 2003 ) . C06G1 . 1 fragment 2 was made using RHL267 and RHL268 primers to amplify an ∼400 bp fragment of C06G1 . 1 , digesting this product with BglII and PstI , and ligating the 190 bp fragment into the BglII/PstI sites of pL4440 . In vitro synthesis of dsRNA from 1 μg of T7 amplified PCR template from the RNAi vector or the pL4440 negative control was accomplished using the Ambion in vitro synthesis kit ( Grand Island , NY ) . RNAi by bacterial feeding of C06G1 . 1 was performed by scoring the progeny of L4 stage rrf-3 ( pk1426 ) animals raised on RNAi or L4440 control plates containing 1 mM IPTG and 50 μg/ml carbenicillin . Means , SEM ( error bars ) , and p values for two-tailed t test were performed by Microsoft Excel . One-way ANOVA with multiple comparisons post-hoc testing were performed using Prism statistical software . Sample sizes represent total technical replicates from two or more biological replicates .

The nematode worm Pristionchus pacificus can live as a parasite inside the oriental beetle , where it waits for the beetle to die so it can feed off the bacteria that live on the beetle's decomposing carcass . This ecologically important interaction is called necromeny . P . pacificus is attracted to a new host by a sex pheromone produced by the beetle , but the genes and biological mechanisms that enable this interaction to occur are not understood in much detail . To identify the genetic basis of this interaction , Cinkornpumin et al . identified and examined a mutant form of P . pacificus that cannot sense the beetle sex pheromone . This revealed that although this pheromone attracts the adult nematodes , it stops P . pacificus embryos developing and can paralyze larvae . Cinkornpumin et al . suggest that the pheromone has likely evolved this ability in order to counteract the spread of the nematodes . This result implies that being invaded by P . pacificus makes life more difficult for the beetles than was previously thought . Further investigation of the gene damaged in the P . pacificus mutants revealed that it encodes a protein that may bind to molecules called lipids , which are needed to form cell membranes and are used in cell signaling . As well as helping the nematodes to detect the sex pheromone , the lipid-binding protein also appears to help protect the worms from the pheromone's detrimental effects . Cinkornpumin et al . observed that the gene for the lipid-binding protein is activated in several tissues , including the cells that form a sheath around some of the nerves that detect chemical signals . Whether this tissue is responsible for the chemical-sensing abilities of the lipid-binding protein , and whether these same tissues are responsible for protecting the nematodes from the damaging effects of the pheromone , remains to be discovered .

2014

A host beetle pheromone regulates development and behavior in the nematode Pristionchus pacificus

MreB is essential for rod shape in many bacteria . Membrane-associated MreB filaments move around the rod circumference , helping to insert cell wall in the radial direction to reinforce rod shape . To understand how oriented MreB motion arises , we altered the shape of Bacillus subtilis . MreB motion is isotropic in round cells , and orientation is restored when rod shape is externally imposed . Stationary filaments orient within protoplasts , and purified MreB tubulates liposomes in vitro , orienting within tubes . Together , this demonstrates MreB orients along the greatest principal membrane curvature , a conclusion supported with biophysical modeling . We observed that spherical cells regenerate into rods in a local , self-reinforcing manner: rapidly propagating rods emerge from small bulges , exhibiting oriented MreB motion . We propose that the coupling of MreB filament alignment to shape-reinforcing peptidoglycan synthesis creates a locally-acting , self-organizing mechanism allowing the rapid establishment and stable maintenance of emergent rod shape . Although many bacteria are rod shaped , the cellular mechanisms that construct and replicate this geometry have remained largely unknown . Bacterial shape is determined by the cell wall sacculus , a giant , encapsulating macromolecule that serves to resist internal osmotic pressure . One of the primary components of the cell wall is peptidoglycan ( PG ) , which is created by the polymerization of single glycan strands linked by peptide crossbridges . Studies of isolated cell walls from rod-shaped bacteria suggest material is generally oriented circumferentially around the rod , perpendicular to the long axis of the cell ( Gan et al . , 2008; Hayhurst et al . , 2008; Verwer et al . , 1980 ) , or in thick cables in others ( Hayhurst et al . , 2008 ) . This mostly circumferential , hoop-like organization of cell wall material allows the cell wall to better resist the internal osmotic pressure , as this pressure causes a stress twice as large in the circumferential direction ( on the rod sidewalls ) than in the axial direction ( on the poles ) ( Amir and Nelson , 2012; Chang and Huang , 2014 ) . This organization of material confers a mechanical anisotropy to the cell wall , causing it to stretch more along its length than across its width for a given stress; this anisotropy may assist rod-shaped cells in preferentially elongating along their length ( Baskin , 2005; Chang and Huang , 2014 ) . Concordantly , atomic force microscopy ( AFM ) has shown that Escherichia coli sacculi are 2–3 times more elastic along their length than across their width ( Yao et al . , 1999 ) . This rod-reinforcing circumferential organization is also observed in the cell walls of plants; hypocotyl and root axis cells rapidly elongate as rods by depositing cellulose fibrils in circumferential bands around their width , resulting not only in a similar dispersive rod-like growth , but also a similar anisotropic response to stress ( Baskin , 2005 ) . The organized deposition of cellulose arises from cortical microtubules self-organizing into a radial array oriented around the rod width , and this orients the directional motions of the cellulose synthases to insert material in circumferential bands ( Paredez et al . , 2006 ) . In contrast to our understanding of the self-organization underlying rod-shaped growth in plants , how bacteria construct a circumferential organization of glycan strands is not known . This organization may arise via the actions of a small number of genes essential for the formation and maintenance of rod shape . Collectively termed the Rod complex , ( or elongasome ) these include MreB , MreC , MreD ( encoded by the mreBCD operon ) ( Wachi et al . , 1989 ) , RodZ ( Alyahya et al . , 2009; Bendezú et al . , 2009 ) , and the glycosyltransferase/transpeptidase enzyme pair RodA/Pbp2 ( Cho et al . , 2016 ) . These components are conserved across a wide range of rod shaped bacteria , and mostly absent in cocci ( Alyahya et al . , 2009; Chastanet and Carballido-Lopez , 2012 ) , leading many to speculate they function as the central determinants of rod shape ( Carballido-Lopez , 2006; Jones et al . , 2001 ) . The spatial coordination of RodA/Pbp2-mediated PG synthesis is conferred by MreB , an actin homolog ( Jones et al . , 2001; van den Ent et al . , 2001 ) . MreB polymerizes onto membranes as antiparallel double filaments , which have been observed to bend liposome membranes inward ( Figure 1A ) ( Salje et al . , 2011; van den Ent et al . , 2014 ) . Loss or depolymerization of MreB causes rod-shaped cells to grow as spheres ( Gitai et al . , 2005; Jones et al . , 2001; Bendezú et al . , 2009 ) . B . subtilis contains 3 MreB paralogs ( MreB , Mbl , and MreBH ) that have been shown to co-polymerize into mixed filaments in vitro , and always colocalize in vivo ( Defeu Soufo and Graumann , 2004; Soufo and Graumann , 2010; Dempwolff et al . , 2011 ) . MreB filaments move circumferentially around the width of rod-shaped cells ( Domínguez-Escobar et al . , 2011; Garner et al . , 2011; van Teeffelen et al . , 2011 ) . Super-resolution imaging has demonstrated that MreB filaments always translocate along their length , moving in the direction of their orientation ( Olshausen et al . , 2013 ) . MreB filaments move in concert with MreC , MreD , and RodA/Pbp2 ( Domínguez-Escobar et al . , 2011; Garner et al . , 2011 ) , and loss of any one component stops the motion of the others . The directional motion of MreB filaments and associated Rod complexes depends on , and thus likely reflects , the insertion of new cell wall , as this motion halts upon the addition of cell wall synthesis-inhibiting antibiotics ( Domínguez-Escobar et al . , 2011; Garner et al . , 2011; van Teeffelen et al . , 2011 ) , or specific inactivation or depletion of Pbp2 ( Garner et al . , 2011; van Teeffelen et al . , 2011 ) or RodA ( Cho et al . , 2016 ) . It is not known how MreB and the rest of the Rod complex construct rod-shaped cells . As the motions of the Rod complexes reflect the insertion of new cell wall , their circumferential motions could deposit glycans in the hoop-like organization required to both build and reinforce rod shape . Therefore , we worked to understand the origin of this circumferential organization , seeking to determine what orients the motions of MreB and associated enzymes around the rod width in Bacillus subtilis . The mechanism by which MreB filaments and associated PG synthases orient their motion around the rod circumference is not known . Each filament-synthase complex is disconnected from the others , moving independently of proximal neighbors ( Garner et al . , 2011 ) . The organized , circumferential motion of these independent filament-synthase complexes could arise in two ways: ( 1 ) A templated organization , where cell wall synthetic complexes move along an existing pattern of ordered glycan strands in the cell wall as they insert new material into it ( Höltje , 1998 ) , or ( 2 ) A template-independent organization , where each synthetic complex has an intrinsic mechanism that orients its motion and resultant PG synthesis around the rod circumference . To explore the extent of order within the motions of the Rod complex , we analyzed the trajectories of GFP-Mbl and GFP-MreB with respect to the cell body using total internal reflection fluorescence microscopy ( TIRFM ) ( Figure 1B ) . Overall , these motions are circumferentially oriented , but not perfectly aligned , a characteristic reflected by the broad distribution of angles that GFP-MreB , its homologs , and the other components of the Rod complex move relative to the long axis of the cell ( Domínguez-Escobar et al . , 2011; Garner et al . , 2011 ) . However , examination of TIRFM time lapse movies revealed that both MreB and Mbl trajectories close in time ( within the period of one revolution ) frequently cross ( Figure 1C , Figure 1—video 1 ) , making it unlikely that MreB filaments move along a perfectly ordered template . As MreB movement reflects the insertion of new glycan strands , these motions indicate that the siacculus is built from somewhat disorganized , yet predominantly circumferential strands . This conclusion is in agreement with X-ray diffraction ( Balyuzi et al . , 1972; Labischinski et al . , 1979 ) and cryoelectron microscopy studies of E . colisacculi ( Gan et al . , 2008 ) which found that , while glycans are oriented circumferentially around the rod width on average , they are not ordered , running at variable angles to each other . Similarly , atomic force microscopy of B . subtilis sacculi has observed a generally oriented , but unaligned arrangement of 50 nm thick cables oriented roughly perpendicular to the long axis ( Hayhurst et al . , 2008 ) . Thus , both the motions of MreB and ultrastructural studies indicate the sacculus is not highly ordered , making it unlikely that it can serve as a self-propagating spatial template for rod shape . Furthermore , given that preexisting cell wall is not necessary for the regeneration of rod shape from wall-less B . subtilis L-forms ( Kawai et al . , 2014 ) , it is likely that both oriented MreB motion and rod shape can arise without an ordered template . As it appeared that organized MreB motion does not arise from patterns in the cell wall , we hypothesized there was an intrinsic mechanism orienting the motion of each MreB filament-cell wall synthetic complex . To test this hypothesis , we examined MreB motions as we changed the shape of cells from rods to spheres . As the internal osmotic pressure and stiffness of B . subtilis resists external mechanical perturbations to its shape ( Renner et al . , 2013 ) , we first altered the shape of cells by controlling the level of wall teichoic acids ( WTAs ) . WTAs are negatively charged cell wall polymers believed to increase the rigidity of the sacculus ( Matias and Beveridge , 2005 ) , a process that could occur via their coordination of extracellular Mg2+ ( Thomas and Rice , 2014; Kern et al . , 2010 ) , or modulation of hydrolase activity ( Atilano et al . , 2010 ) . Knockouts of tagO , the first gene in the WTA synthesis pathway , create large , slow-growing , round cells that still synthesize PG , building extremely thick and irregular cell walls ( D'Elia et al . , 2006 ) . We placed tagO under xylose-inducible control and grew cells at different induction levels . As expected , at high TagO inductions , cells displayed normal widths . As we reduced TagO levels , rods became gradually wider ( Figure 1D–E ) until , beneath a given induction , cells were no longer able to maintain rod shape , growing as spheres ( or clumps of spheres ) with no identifiable long axis . At intermediate induction levels , we observed a transition region between the two states , with cells growing as steady state populations of interconnected rods and spheres ( Figure 1D ) . In agreement with models that ( A ) WTAs work with PG to bind Mg2+ ( Thomas and Rice , 2014; Kern et al . , 2010 ) , and ( B ) are required for cell wall rigidity ( Matias and Beveridge , 2005 ) , both the cell width and the amount of TagO induction determining the rod/sphere transition could be modulated by Mg2+ levels ( Figure 1F , Figure 1—figure supplement 1B ) . Likewise , decreasing extracellular Mg2+ or tagO induction resulted in increasingly curved cell contours ( Figure 1—figure supplement 1B ) , suggesting the wall was becoming more flexible . By tracking the motion of GFP-MreB filaments in these differing cell shapes , we found that motion is always oriented in rods , moving predominantly circumferentially at all induction levels above the rod/sphere transition . However , in round cells ( those induced beneath the rod/sphere transition point or in tagO knockouts ) MreB filaments continued to move directionally , but their motions were isotropic , moving in all directions ( Figure 2A , Movie Figure 2—video 1 ) . To quantify the relative alignment of MreB under each condition , we calculated the angle between trajectory pairs less than 1 μm apart ( Figure 2B , Figure 2—figure supplement 1A ) . This analysis revealed that MreB motions are more aligned when cells are rods: above the rod/sphere transition , trajectories have a median angle difference of 26°; while at low TagO inductions , where cells are round , the angle difference increases to 42° , close to that of randomly oriented trajectories ( 45° ) . To verify that the loss of oriented MreB motion was due to the changes in cell shape , and not from some other effect of reduced WTA levels , we created round cells by alternate means . Depletion of both elongation PG transpeptidases ( Pbp2a and PbpH ) causes rod-shaped cells to become wider over time as they convert to spheres ( Garner et al . , 2011 ) . We used this gradual transition of rods into spheres to examine both the width and overall shape dependence of MreB motion . At initial points of depletion ( 1–2 hr ) the rods widened but maintained circumferential MreB motion . At 2 . 5 hr of PbpA depletion , cells were a mix of spheres and rods of differing widths . These cells displayed the same pattern of MreB orientation observed with tagO depletions: round cells contained unoriented MreB , while nearby rod-shaped cells showed circumferential motion . Identical behavior was observed for GFP-Mbl during PbpA depletions ( Figure 2C , Movie Figure 2—video 2 ) . Quantitation of trajectories from all cells ( both rods and spheres ) at each time point of depletion indicated an increase in the median angle between trajectories as the population grew wider and rounder over time ( Figure 2D , Figure 2—figure supplement 1B ) . In E . coli , the angle of mutant MreB filaments relative to the long axis has been reported to increase with cell width ( Ouzounov et al . , 2016 ) . To test if the angle of MreB movement changes with respect to cell width in B . subtilis , we calculated the angle of each trajectory to the long axis for all cells in our data with an identifiable width axis . At the same time , we also measured the curvature of each cell to determine how the overall shape of the cell affected the orientation of motion . This revealed that MreB motion in rods remained equivalently oriented over a wide range of rod widths , up to ~2 μm ( Figure 2E , Figure 2—figure supplement 1C–E ) . Beyond a 2 μm width , cells began to lose their rod shape as they became more spherical , and the predominantly circumferential orientation of MreB motion was lost ( Figure 2E , Figure 2—figure supplement 1E ) . This suggested that oriented MreB motion does not sense or rely on a specific cell radius; rather the orientation relies on differences between the two principal curvatures of the membrane . It appears that the motion of MreB filaments is oriented along the direction of greatest principal curvature: In rods , there is zero curvature along the rod length , and high curvature around the rod circumference , along which filaments orient . In contrast , in round cells where MreB motion is isotropic , the two principal curvatures are equal ( Figure 2F ) . To further verify that MreB filaments orient in response to overall cell shape , we externally imposed rod shape on cells with unoriented MreB motion . We loaded TagO-induced cells into long 1 . 5 ×1 . 5 μm microfluidic chambers , then reduced TagO expression to levels insufficient to produce rods in liquid culture ( Figure 3A , Figure 3—figure supplement 1A ) . After TagO depletion , cells expanded to fill the chamber indicating that WTA-depletion caused shape changes just as in bulk culture ( Figure 3A , Figure 3—figure supplement 1A ) . Within these chambers , cells grew as rods , but at a wider width ( 1 . 5 μm ) than wild-type cells , set by the chamber . When cells grew out of the chamber they swelled just as in bulk culture , showing confinement was required for rod shape at this induction level ( Figure 3B , Figure 3—figure supplement 1A ) . In the TagO-depleted cells confined into rod shapes , MreB moved circumferentially ( Figure 3C , Figure 3—video 1 ) , confirming that MreB orients in response to the cells having rod shape . This experiment demonstrates that the isotropic MreB motion observed in round cells arises from the lack of rod shape , and not from some other effect of our genetic perturbations . This experiment also showed another unexpected result: the doubling time of free ( unconstrained ) cells induced at similar TagO levels is long ( 53 ± 10 min ) , but confining them into rod shape restored their doubling time ( 44 ± 4 min ) toward wild-type times ( 39 ± 9 min ) ( Figure 3—figure supplement 1B ) . We next attempted to minimize any contribution to MreB filament orientation from ( A ) the directional motion of filaments , and ( B ) any pre-existing order within the sacculus . To accomplish this , we examined filament orientation in protoplasts ( cells that had their cell wall enzymatically removed ) confined into different shapes , using highly expressed GFP-MreB to assay long filaments , and GFP-Mbl to assay short filaments . We protoplasted cells in osmotically stabilized media ( Wyrick and Rogers , 1973 ) , then grew them under agar pads containing micro-patterned cross shapes . Cells in the center of these crosses ( ~5 μm diameter ) were forced to grow as spheres , whereas cells in the arms were constrained to grow into rods of various widths ranging from 2 to 5 μm ( Figure 3D ) . As reported previously ( Domínguez-Escobar et al . , 2011 ) , MreB filaments within protoplasts did not move directionally ( Figure 4—video 1 ) , likely because the cell wall provides the fixed surface along which the PG synthesis enzymes move . Within the protoplasts confined into the smallest rod shapes ( 2 μm ) , filaments oriented at a distribution of angles predominantly perpendicular to the cell length ( Figure 3E–F ) . The angular distributions of short GFP-Mbl filaments and longer GFP-MreB filaments were similar to each other , and also similar to the distribution of 1 ) filament trajectories observed in intact , wild-type cells and 2 ) filament trajectories of TagO depleted cells confined into the mother machine ( Figure 3Fi ) . As we increased the width of the imposed rod shape from 2 to 5 μm , filaments remained predominantly oriented in all cases ( Figure 3Fii ) , but their mean deviation from 90° increased as the rod width increased ( 34° at 2 μm , 35° at 4 μm , and 41° at 5 μm ) . In contrast to confinement in rods , both short and long filaments in spherically confined protoplasts remained unoriented ( Figure 3E ) . Together , these data demonstrate that MreB filaments orient to point around the rod width even in the absence of pre-existing cell wall or directional motion , as long as the cell has a rod shape . These experiments also demonstrate that MreB filaments will align even in wider rods , where the difference in principal curvatures is smaller than in wild-type cells , but that , as the difference in principal curvatures decreases , filament alignment becomes more disordered . To test if MreB filaments are themselves sufficient to align along the predominant direction of membrane curvature , we assembled purified T . maritima MreB within liposomes and visualized it using cryoelectron microscopy and tomography . While controlling the final concentration of protein encapsulated within liposomes ≤1 μm is difficult , we were able to assemble MreB inside liposomes at high concentrations . At these concentrations , MreB filaments tubulated liposomes , creating rod-like shapes ( Figure 4A , Figure 4—figure supplement 1A–B , Figure 4—video 1 ) . In tubulated regions , MreB filaments could be traced around the circumference of the liposome tube , while filaments in spherical regions were found in all possible orientations ( Figure 4A ) . At the highest concentrations , tubulated liposomes contained closely packed filament bundles , allowing us to observe a regular patterning of the canonical double filaments of MreB ( Figure 4B ) . Purified wild-type MreB did not bind to the outside surface of small liposomes contained within larger ones ( Figure 4A ) , indicating that MreB filaments preferentially polymerize on inward ( negative ) curvatures , akin to the inner leaflet of the bacterial membrane . In the absence of MreB , liposomes are spherical , with no deformations ( Figure 4—figure supplement 1C ) . Together , this data suggests that MreB filaments themselves are sufficient to align along the predominant direction of membrane curvature , as observed here with laterally associated filaments . We note that the experimental limitations of the liposomal system , combined with the tendency of MreB filaments to self-associate make it difficult for us to acquire and study the alignment of single filaments in vitro . Also , it remains to be determined if membrane-associated MreB filaments exist as bundles or isolated filaments in vivo . The above observations demonstrate that MreB filaments sense and align along the direction of greatest principal curvature , that is , the more curved inner surface of the rod circumference . The ultrastructure of MreB filaments provides a possible mechanism: MreB filaments are bent ( Salje et al . , 2011 ) , with the membrane-interacting surface on the outer face of the bend ( Figure 4C ) . This bent conformation could cause filaments to preferentially orient along the curved rod circumference , rather than the flat rod length , to maximize the burial of hydrophobic moieties into the membrane , a mechanism suggested by previous theory ( Wang and Wingreen , 2013 ) . As the curvature of MreB filaments bound to liposomes is much greater ( ~200 nm diameter [van den Ent et al . , 2014] ) than that of B . subtilis cells ( ~900 nm diameter ) , we performed analytical calculations to model how highly curved MreB filaments would align within a cell with a less curved surface ( Figure 4D–G , Appendix 1 ) . As many of the biochemical and physical parameters of MreB are still unknown , we first assumed a fixed set of parameters , and later verified that our results were robust over a large parameter range . We initially assumed a membrane interaction energy of 10 kT per monomer ( calculated from residues involved in membrane associations [Salje et al . , 2011] ) , and a similar Young’s modulus to actin ( 2 GPa ) . We modeled filaments as elastic beams made of two protofilaments . In addition , we used the Helfrich free energy to model the energetics of membrane deformation , and accounted for the work done against osmotic pressure due to changes in volume ( Appendix 1 ) . These calculations indicate that the total energy is minimized when filaments orient along the direction of maximal curvature ( Figure 4F ) and that , importantly , the energy penalty for incorrectly-oriented filaments is much greater than the energy of thermal fluctuations . Interestingly , this modeling indicates a decrease in energetic preference for the preferred filament orientation as the radius of the cell is increased ( Figure 4F ) , a prediction in qualitative agreement with our observations of alignment in protoplasts . Furthermore , our calculations indicate that orientation is robust over a large , biologically relevant range of parameters , including the membrane binding energy , filament length , and filament Young’s modulus ( Figure 4G ) . These calculations predict that filaments should orient circumferentially both if the membrane deforms to the filament ( at low osmotic pressures or if filaments are stiff ) ( Salje et al . , 2011 ) , or if filaments deform to the membrane ( at high osmotic pressures or if filaments are flexible ) ( Figure 4E ) . Our experimental data demonstrates MreB filament alignment across a range of pressures: high within cells , low to none within liposomes , and a pressure between the two within osmotically-stabilized protoplasts . In the absence of osmotic pressure , MreB filaments deform liposomes since it is energetically more favorable to bend the membranes than to bend the filaments , as observed in our in vitro data ( Figure 4A , Figure 4—figure supplement 1 ) . However , in live cells , our modeling predicts that MreB filaments cannot deform the inner membrane due to the large osmotic pressure , and instead deform to match the greatest principal membrane curvature . Hence filaments create curvature in liposomes and sense it in cells . Together , the above data demonstrate that MreB filaments are sufficient to preferentially orient along the direction of greatest principal membrane curvature . In rod-shaped cells , this direction is along the rod circumference . As filaments move along their length , their orientation constrains the spatial activity of the PG synthetic enzymes such that new cell wall is inserted in a mostly circumferential direction ( Hayhurst et al . , 2008 ) to reinforce rod shape ( Chang and Huang , 2014; Yao et al . , 1999 ) . While the ability of MreB filaments to orient in pre-existing rods can help explain how rod shape is maintained , we also wanted to understand how MreB filaments facilitate the de novo formation of rod shape . To explore this , we observed how cells interconvert between spheres and rods . We first examined how rod shape fails , by growing our TagO-inducible strain at induction levels that produced rods and then reducing the Mg2+ concentration to induce them to convert to spheres . This transition revealed that rods convert into round cells by continuous swelling: once a rod begins to widen , it continues to do so until reaching a fully spherical state with no reversion during the process ( Figure 5A ) . Similar rod to sphere transitions could be attained by holding Mg2+ constant while reducing TagO expression . Likewise , cells grown at intermediate TagO induction levels ( 8–12 mM ) grew as steady state populations of interconnected rods and spheres , indicating that cells underwent repeated cycles of rod shape formation followed by reversion to spheres ( Figure 1D , F ) . These results indicate that rod shape can be maintained only as long as the cell wall is sufficiently rigid to resist the internal osmotic pressure . We next examined how rod shape forms from round cells . As the recovery of protoplasted B . subtilis is so infrequent that it has never been directly visualized ( Mercier et al . , 2013 ) , we assayed how round cells with preexisting cell walls convert back into rods , using three systems: ( 1 ) re-inducing WTA expression within TagO-depleted , spherical cells , ( 2 ) holding TagO expression beneath the rod/sphere transition and increasing Mg2+ levels , and ( 3 ) re-inducing Pbp2a expression in spherical , Pbp2a-depleted cells . In all three cases , rods reformed in a discrete , local manner; spheres did not form into rods by progressively shrinking along one axis , but rather , rods abruptly emerged from one point on the cell , growing more rapidly than the parent sphere ( Figure 5B , Figure 5—video 1 and Figure 5—video 2 ) . This morphology is similar to the initial outgrowth of germinating B . subtilis spores ( Pandey et al . , 2013 ) . We occasionally observed another mode of recovery , occurring when round cells were constrained , or divided into , ovoid or near-rod shapes . Once these near-rod shaped cells formed , they immediately began rapid , rod-like elongation along their long axis ( Figure 5—figure supplement 1A ) . We focused on two salient features of the rod shape recoveries: ( 1 ) rod shape forms locally , most often at one point on the cell surface , and ( 2 ) once a rod-like region is formed , it appears self-reinforcing , both propagating rod shape and growing faster than adjacent or attached non-rod shaped cells . We first wanted to understand how rod shape initiates de novo from spherical cell surfaces . By examining the initial time points of recoveries , we found that rods begin as small outward bulges: local regions of outward ( positive Gaussian ) curvature flanked by regions of inward ( negative Gaussian ) curvature ( Figure 5C ) . These initial outward bulges showed a width distribution similar to that of the later emerging rods ( Figure 5D ) . Once these bulges formed , they immediately began rapid elongation into nascent rods , which would then thin down to wild type width over time . Bulge formation and rod recovery were independent of cell division , as cells depleted of FtsZ still recovered rod shape ( Figure 5—figure supplement 1B ) . Rather , these bulges appeared to arise randomly , evidenced by the fact that different cells produced rods at different times during WTA or Pbp2a repletion . We conclude that the appearance of a local outward bulge can act as the nucleating event of rod shape formation . As emerging rods appeared to grow faster than adjacent round cells , we tested if the doubling times of rod-shaped cells were faster than those of non-rods by measuring the doubling times in our inducible TagO strain at different induction levels using both OD600 measurements and single cell microscopy under steady state conditions ( Figure 3—figure supplement 1B ) . This revealed a sharp transition in doubling time that matched the conditions of the rod/sphere transition: growth is slow when cells are spheres , yet greatly increases when cells are rods ( Figure 3—figure supplement 1B , Figure 5—figure supplement 1C ) . Furthermore , the doubling times of recovering rods was similar to that of rods at steady state ( Figure 3—figure supplement 1B ) . We believe the lower doubling time of rods is likely due to cell shape and not another effect , such as the lack of WTAs , as ( 1 ) the doubling time of TagO-depleted cells confined in the microfluidic chambers matched that of wild type cells; and ( 2 ) both the doubling times and the boundary of the rod/sphere transition could be equivalently shifted by changing the Mg2+ concentration ( Figure 1F , Figure 1—figure supplement 1 , Figure 5—figure supplement 1C , Figure 5—video 2 ) . Combined , these results indicate that rod shape creates local , self-reinforcing regions that are poised for more rapid growth; once any small region of the cell approximates a rod shape , growth of the rod-like region is amplified , growing faster than other regions , and thereby outcompeting non-rod growth at the population level . We next sought to determine what features distinguished rods from round cells . As the elongation of rod-shaped cells requires a sufficiently rigid cell wall ( Figures 1D–E and 5A ) , the localized , self-reinforcing formation of rods in our system could arise from either of the two major cell wall components , PG and WTAs: ( 1 ) the PG strands could be arranged such that they better reinforce rod shape ( Amir and Nelson , 2012; Chang and Huang , 2014 ) , or ( 2 ) WTAs could be preferentially incorporated into the emerging rods to stabilize them . To assay the orientation of newly inserted cell wall , we imaged the motions of MreB as we induced TagO-depleted cells to recover into rods . This revealed that oriented MreB motion correlates with local shape: emerging rods displayed oriented MreB motion even at the initial points of their formation , while attached round parent cells displayed unaligned motion ( Figure 6A , Figure 6—video 1 ) . This demonstrates that oriented MreB motion correlates with local geometry and does not arise from a global , cell spanning change . We next examined the overall cellular distribution of MreB in recovering cells with confocal microscopy . This revealed that , immediately prior to rod emergence , MreB transiently accumulated in a bright ring oriented perpendicular to the direction of rod emergence , most often occurring at the interface of the bulge and the round cells ( Figure 6B , Figure 5—figure supplement 1D ) . To observe whether both spheres and rods inserted new PG during the process of rod shape recovery , we used fluorescent D-amino-acids ( FDAAs ) , which crosslink into newly inserted cell wall . We grew TagO-depleted cells in a microfluidic device in the presence of HADA , then switched the media to contain Cy3B-ADA as we re-induced TagO expression . During rod emergence , the old cell wall signal ( HADA ) remained in the sphere , while the emerging rod was almost entirely composed of new ( Cy3B-ADA ) material , confirming the discrete nature of rod shape recovery . However , the attached spheres also incorporated Cy3B-ADA , indicating PG synthesis occurs in both rods and spheres during recovery ( Figure 5—figure supplement 1E ) . The local reinforcement of rod shape in recovering cells could arise from preferential incorporation of the cell wall rigidifying WTAs . As the WTA ligases have been reported to interact with MreB ( Kawai et al . , 2011 ) , we tested if rod shape correlated with increased WTA accumulation in emerging rods . To test this , we labeled recovering cells with fluorescently labeled lectins that specifically bind to WTAs ( Figure 6—figure supplement 1A ) . Following TagO reinduction , WTAs in recovering cells were ( Figure 6—figure supplement 1B ) , equally present in the cell walls of both rods and spheres ( Figure 6C ) . To test if the WTA ligases move with MreB , we created GFP fusions to these proteins at their native locus and examined their dynamics with TIRFM . We were unable to observe any of the circumferential motions expected if the WTA ligases moved with MreB; instead they appeared to be rapidly diffusing on the membrane ( Figure 6—figure supplement 1C , Figure 6—video 2 , Appendix 2 ) . For further confirmation , we tracked the single molecule motions of these ligases , using JF549 labeled HaloTag fusions expressed at the native locus and promoter . We did not observe any directional motions of the ligases that would indicate they move or associate with MreB and Pbp2A ( Figure 6—figure supplement 2D–E , Figure 6—video 3 ) . In summary , these data gives new insights into what properties of the cell wall can be modulated to create and stabilize rod shape: rod shape is not formed by preferential localization of teichoic acids to rods , and both spheres and rods incorporate PG before and during rod shape recovery , in line with reports that PG synthesis is unchanged by the inhibition of WTA synthesis ( Pooley et al . , 1993 ) . Rather , the only differences we detected between rod shaped and round cells were increased growth rates and oriented MreB motion . We note that as WTAs have been shown to affect hydrolase activity ( Kasahara et al . , 2016 ) , their depletion may cause other rod-shape inhibiting PG abnormalities that we cannot observe . To construct regular , micron-spanning shapes made of covalently crosslinked material , nature must devise strategies for coordinating the activities of disperse , nanometer-scale protein complexes . This work reveals that the role of MreB in creating rod shape is to locally sense and subsequently reinforce differences in principal curvatures . The local , short-range feedback between differences in curvature , MreB orientation , and shape-reinforcing cell wall synthesis could provide a robust , self-organizing mechanism for the stable maintenance and rapid reestablishment of rod shape , allowing the local activity of short MreB filaments to guide the emergence of a shape many times their size . All B . subtilis strains were prepared for experimentation as follows: strains were streaked from −80°C freezer stocks onto lysogeny broth ( LB ) agar plates . Following >12 hr of growth at 37°C , single colonies were transferred to serially diluted overnight bulk liquid cultures in LB supplemented with 20 mM magnesium chloride , placed on a roller drum agitating at 60 rpm , and grown at 25°C . After >12 hr growth to OD600 <0 . 6 , these starter cultures were transferred to or inoculated into subsequent growth conditions . All strains with tagO under inducible control were grown overnight in the presence of 30 mM xylose unless otherwise noted . For the experiments in Figure 6C and Figure 3—figure supplement 1B , BEG300 cells were inoculated in the indicated medium ( LB with 20 mM MgCl2 unless otherwise stated ) from logarithmic phase overnights; ‘rods’ were grown from a low dilution with 30 mM xylose , and ‘spheres’ were grown with 0 mM xylose . For bulk culture doubling time measurements , doubling times were calculated from the slope of a graph of time vs . dilution for a succession of serial dilutions of a given strain . Time , the dependent variable , was taken as the time for a given dilution to pass the OD cutoff of OD600 = 0 . 20 . In place of technical replicates , a large number of replicates were performed on a continuous gradient of xylose induction , showing a consistent trend between the extreme values depicted in the figure . Single cell measurements were made in three ways . Phase contrast images were collected on a Nikon ( Tokyo , Japan ) Ti microscope equipped with a 6 . 5 μm-pixel Hamamatsu ( Hamamatsu City , Japan ) CMOS camera and a Nikon 100x NA 1 . 45 objective . Cells were collected by centrifugation at 6000 x g for 2 min and re-suspended in the original growth medium . Unless otherwise specified , cells were then placed on No . 1 . 5 cover glass , 24 × 60 mm , under a 1 mm thick agar pad ( 2–3% agar ) containing LB supplemented with 20 mM magnesium chloride . Unless otherwise noted , all cells were imaged at 37°C on a heated stage . Images were collected on a Nikon TI microscope with a 6 . 5 μm-pixel CMOS camera and a Nikon 100x NA 1 . 45 objective . Cells of strain BEG300 were grown overnight in LB supplemented with 30 mM xylose , 20 mM magnesium chloride , 1 μg/mL erythromycin , and 25 μg/mL lincomycin at 25°C at the specified xylose concentrations . 11 μM isopropyl β-D-1-thiogalactopyranoside ( IPTG ) was added to induce GFP-MreB and the cells were shifted to 37°C and allowed to grow for 2 hr before imaging . Cells of strain BEG202 ( ΔtagO ) with GFP-Mbl under a xylose-inducible promoter were grown overnight at 25°C in LB supplemented with 20 mM magnesium chloride and 0 . 125 mM xylose , and shifted to 37°C for 2 hr before imaging . Cells were placed on cleaned glass coverslips thickness No . 1 . 5 , as described in the next section . 3–6% agar pads were prepared in LB supplemented with 20 mM magnesium chloride , 11 μM IPTG and the desired concentration of xylose . Images were collected for 3 min at 1 or 2 s intervals , as specified . 30–50 cells were imaged in a day , and the experiments were repeated on at least one other day to test for technical variation . Coverslips were sonicated in 1 M KOH for 15 min , followed by five washes with water . Coverslips were washed twice with 100% ethanol , and then sonicated in 100% ethanol , followed by one more wash in 100% ethanol . They were stored in ethanol and dried for 10 min before use . Images were collected on a Nikon TI microscope with a Hamamatsu ImagEM ( EM-CCD ) camera ( effective pixel size 160 nm ) and Nikon 100x NA 1 . 45 TIRF objective . Z stacks were obtained at 0 . 2 μm slices . Total image depth was 3 μm . Only the top 3 slices of the cell were used in maximum intensity projections in Figure 3E . Cells were prepared as in ‘Imaging – MreB particle tracking’ above . Cells were placed under an agar pad in a MatTek ( Ashland , Massachusetts ) dish for imaging . Images were collected on an DeltaVision OMX SR Blaze system in SIM TIRF mode , using an Edge 5 . 5 sCMOS camera ( PCO ) and a 60x objective . 75msec exposures from a 488 nm diode laser were used for each rotation . Spherical aberration was minimized using immersion oil matching . Raw images were reconstructed using Softworx ( Applied Precision ) . All image processing unless otherwise specified was performed in FIJI ( Schindelin et al . , 2012 ) . Images used for particle tracking were unaltered , except for trimming five pixels from the edges of some videos to remove edge artifacts detected by the tracking software . Phase contrast images and fluorescent images of protoplasts were adjusted for contrast . Phase contrast images presented in the manuscript collected from cells in the custom microfluidic device , which did not undergo quantitative processing , were gamma-adjusted ( γ = 1 . 5 ) to compensate for changes in brightness occurring at the device’s feature borders; such processing was not used for growth quantification . Images were background-subtracted for viewing purposes; unaltered images were used for quantitative processing in all cases . The custom microfluidic setup used to confine cells in Figure 3A–C , Figure 3—figure supplement 1A , and Figure 3—video 1 was previously described in Norman et al . ( 2013 ) . Briefly , a polydimethylsiloxane slab with surface features was bonded to a 22 × 60 mm glass coverslip by oxygen plasma treatment followed by heating to 65°C for >1 hr . The features in our setup differed from those described in Norman et al . ( 2013 ) , particularly in the omission of a second , wider layer in the cell chambers , which enhanced growth at timescales beyond that of our experiments . Syringes containing growth medium were connected to the microfluidic features using Tygon tubing stainless steel dispensing needles ( McMaster Carr Supply Company , Elmhurst , Illinois ) . Medium was supplied to cells at a constant rate of 2–5 μL/min using automatic syringe pumps . Imaging was carried out using phase contrast microscopy as described above . For the microfluidics experiments in Figures 5 and 6 and Figure 5—video 1 ( top ) , Figure 5—video 2 , and Figure 6—video 1 , the CellASIC platform from Merck Millipore ( Billerica , Massachusetts ) was used with B04A plates . The cell confinement experiment in Figure 3A–C was conducted by first loading cells into the chamber: BEG300 cells were grown to stationary phase ( OD6003 . 0–5 . 0 ) in LB supplemented with 20 mM magnesium chloride , passed through a 5 μm filter , and concentrated 100-fold before loading in the custom-made microfluidic device . Both phase contrast and fluorescent imaging were performed as described in the ‘Imaging’ section above . For observing MreB movement , MreB-GFP expression was induced with 50 μM IPTG upon loading into the microfluidic chamber , and cells were imaged every 2 s with a camera exposure time of 300 ms . The biological phenomena depicted in Figure 3A–C , Figure 3—figure supplement 1A , and Figure 3—video 1 . are representative data drawn from three experiments for phase contrast data and two experiments for fluorescent data . Biological replicates , providing information on cell-to-cell clonal variation , for the purposes of this experiment should be considered individual channels , which are seeded by one or a few cells at the start of the experiment . Cells of strains bJS18 ( GFP-Mbl ) and bEG300 ( GFP-MreB ) were grown overnight at 25°C in the osmoprotective SMM media ( LB supplemented with 20 mM magnesium chloride , 17 mM maleic acid , 500 mM sucrose , brought to a pH of 7 . 0 ) with maximum xylose induction ( 30 mM ) ; cells were shifted to 37°C in the morning . For strain bEG300 , the SMM media was supplemented with 8 mM xylose ( for intermediate TagO induction ) . Following 2 hr of growth , 10 mg/mL of freshly suspended lysozyme was added to the cultures with OD600 >0 . 2 . After growing for 1–2 hr in lysozyme , the cells were spun and concentrated . 6% agar pads made in LB-SMM were made using a polydimethylsiloxane ( PDMS ) mold with crosses ( 2 , 4 and 5 μm arms and 5 μm center ) . The cells were placed on the agar pad for 2 min , allowing the cells to settle in the crosses . The pad was then placed in a MatTek ( Ashland , Massachusetts ) dish for imaging . 10–20 biological replicates were imaged per day and the experiment was replicated on 3 days . TagO depletions in Figure 2A were conducted using strain BEG300 in liquid culture . Cells were prepared as overnights , as described above , then grown at the specified xylose concentration at 37°C in LB with 20 mM magnesium chloride for 4 hr . The cells were then imaged as described above in the ‘Imaging – MreB particle tracking’ section . Pbp2A depletions shown in Figure 2 were conducted in liquid culture using strain BRB785 and BRB786 with an IPTG‐inducible Pbp2A fusion at the native locus with the redundant transpeptidase PbpH deleted . This strain was grown overnight in the presence of 2 mM IPTG , and then inoculated into CH media containing 2 mM IPTG , 0 . 015% xylose , and 20 mM magnesium chloride to stabilize the cells against lysis . At an OD600 of 0 . 6 , cells were spun down in a tabletop centrifuge and washed three times in CH media lacking IPTG . Cells were placed under agar pads containing 20 mM magnesium chloride , and spinning disk confocal images were taken every 5 s on a Nikon Ti microscope with a 100 × 1 . 49 TIRF objective and a Hamamatsu ImagEM C9100-13 EM-CCD camera ( effective pixel size of 160 nm ) . Depletions shown in Figure 5A were conducted using strain BEG300 . Cells were prepared as overnights in LB with 1 mM magnesium chloride and 12 mM xylose . In the morning , they were washed in LB with 12 mM xylose and no magnesium and placed under a 3% agar pad with the same medium . Phase contrast images were collected every 5 min using a Photometrics ( Tucson , Arizona ) CoolSNAP HQ2 CCD camera . The experiment was replicated on 2 days . Repletions of TagO or Pbp2a on pads , as shown in Figure 5B and Figure 5—video 1 ( bottom ) , were performed with strains BEG300 and BRB785 respectively . Cells were grown as overnights , as described above , then depleted at 37°C for >4 hr in LB with 20 mM magnesium chloride and collected by centrifugation at 6000 x g for 2 min . The cells were re-suspended in LB supplemented with 20 mM magnesium chloride and 1 mM IPTG ( BRB785 ) and 30 mM xylose ( BEG300 ) , placed under 5% agarose pads on coverslips with thickness No . 1 . 5 for imaging . Phase contrast images were collected every 5 min using a Photometrics CoolSNAP HQ2 CCD camera . For the repletions shown in Figures 5B–C and 6A , Figure 5—figure supplement 1 , and Figure 5—video 1 ( top ) and 2 , performed in the CellASIC microfluidic device in a B04A plate , BCW82 and BEG300 cells were grown to OD6001 . 2–1 . 5 in LB supplemented with 20 mM magnesium chloride , centrifuged to pellet large clumps for 3 min at <500 x g , and the supernatant loaded into the plate . Growth medium was supplied at 5–6 PSI . Cells were grown for at least an additional 30 min before the addition of inducer to the growth medium . Phase contrast images were collected every 10 min . Fluorescent images were collected on the imaging setup described in the ‘Imaging – MreB Particle Tracking’ section above: GFP-MreB was induced upon loading into the microfluidic chamber with 1 mM IPTG , and MreB dynamics were observed for 3 min after every 10 min , using 300 ms camera exposures taken every 2 s . The experiment was repeated six times , with the number of biological replicates per experiment limited by chance variation in the loading procedure , but on the order of one to four per experiment . For the repletions shown in Figure 6B and Figure 5—figure supplement 1D , the same procedure was used , but with imaging performed on the spinning disk confocal microscope described in ‘Imaging – Spinning Disk Confocal’ . Z-stacks were collected with a range of 3 μm around the focal plane and 0 . 2 μm steps . The MreB localization experiments were done using strain bEG300 with full induction of GFP-MreB ( 1 mM IPTG ) and recovering cells were imaged using the spinning disk microscope , collecting Z-stacks as described before . Where indicated , instead of visualizing MreB dynamics , fluorescent D-amino acids ( Kuru et al . , 2012 ) ( 7 μM ) were added to the growth medium in the CellASIC device: HADA during depletions of TagO ( 0 mM xylose ) and Cy3B-ADA during repletion of TagO ( 30 mM xylose ) . Cells were washed with LB supplemented with 20 mM magnesium chloride containing no D-amino acids for 1–2 min before imaging . To test if rod shape recovery occurs in the absence of cell division , three strains were tested ( BAB327 , BAB343 and BAB388 ) . Cells of BAB327 and BAB388 were grown in CH media with 25 mM magnesium chloride in the absence of xylose at 37°C until OD600 ~0 . 5 and diluted 10-fold in fresh media . After 2 hr of growth , IPTG was added to a final concentration of 1 mM ( MinCD and FtsA , respectively ) and cells were incubated for an extra 1 hr . Cells were imaged on a spinning disk confocal under pads with 1 mM IPTG and 60 mM xylose ( for TagO repletion ) . Phase-contrast and fluorescent images were acquired at 10 min intervals for a total of 8 hr . Cells of BAB343 were grown in LB supplemented with 20 mM , magnesium chloride in the absence of xylose at 25°C overnight . The next day , after 2 hr of growth in the same media at 37°C , IPTG was added to a final concentration of 1 mM ( MciZ ) and cells were incubated for an extra 1 hr . Cells were imaged on a spinning disk confocal under pads with 1 mM IPTG and 30 mM xylose ( for TagO repletion ) . Phase-contrast and fluorescent images were acquired at 10 min intervals for a total of 4 hr . For Figure 5—video 2 , cells of BCW51 were grown overnight at 25°C in LB supplemented with 8 mM xylose , 20 mM magnesium chloride , 1 μg/ml erythromycin and 25 μg/ml lincomycin ( MLS ) . Cells were shifted to 37°C for 2 hr and loaded into the CellASIC B04A plate at OD600 ~ 0 . 6 . At the start of imaging , magnesium was depleted by flowing in LB supplemented only with 8 mM xylose and MLS at 3 psi . Images were collected every 20 min over a 4 hr period . Magnesium was resupplied to the cells by changing to LB supplemented with 8 mM xylose , 20 mM magnesium chloride , and MLS . Imaging was continued every 20 min for an additional 4 hr . 5–10 cells were imaged and the experiment was done once . Cells were grown overnight at 25°C in LB supplemented with 30 mM xylose , 20 mM magnesium chloride , 1 μg/mL erythromycin and 25 μg/mL lincomycin . In the morning they were collected at OD600 ~ 0 . 2 , spun in a tabletop centrifuge at 9000 rpm for 3 min and washed in LB supplemented with various xylose ( 0–30 mM ) and magnesium chloride ( 0–20 mM ) levels . 25-fold serial dilutions into LB supplemented with the same xylose and magnesium chloride concentrations were made and allowed to grow at 37°C for 4 hr . Cells at OD600 ~ 0 . 2 were concentrated by spinning in a tabletop centrifuge at 9000 rpm for 3 min . They were placed on a coverslip thickness No . 1 . 5 under 3% agarose pads made in LB supplemented with the same concentrations of xylose and magnesium chloride . Images were collected using the imaging setup described in the ‘Imaging – phase contrast microscopy’ section above , as well as with a Photometrics CoolSNAP HQ2 CCD camera . The magnification and pixel size were the same in both setups . The experiment was replicated on two days and the data pooled together . The MATLAB based software uTrack was used for particle tracking ( Jaqaman et al . , 2008 ) . We used the comet detection algorithm to detect filaments ( difference of Gaussian: one pixel low-pass to 4–6 pixels high pass , watershed segmentation parameters: minimum threshold 3–5 standard deviations with a step size of 1 pixel ) which , at our MreB induction levels gave better localization of the resultant asymmetric particles over algorithms that search for symmetric Gaussians . Visual inspection of detected particles confirmed that most of the particles and none of the noise were being detected . A minimum Brownian search radius of 0 . 1–0 . 2 pixels and a maximum of 1–2 pixels was applied to link particles with at least five successive frames . Directed motion propagation was applied , with no joins between gaps allowed . Tracks were visualized using the FIJI plug-in TrackMate ( Tinevez et al . , 2017 ) . For sphere to rod transitions and cells confined in microfluidic channels , movies were processed by subtracting every 8th frame from each frame to remove stationary spots using the FIJI plugin StackDifference before tracking . The tracking was done as described earlier in this section . Strains containing fluorescent fusions to TagT , TagU , and TagV were grown as described in the ‘Overnight culture growth’ section but in CH medium instead of LB . Cells were grown for 3 hr at 37°C before imaging , then collected by centrifugation at 6000 x g for 2 min and re-suspended in CH . Cells were then placed on a glass coverslip thickness No . 1 . 5 under an agar pad thickness 1 mm made from CH and 1 . 5% agarose . Timelapse images were collected with TIRF illumination , using continuous 100 ms 488 nm exposures . Epifluorescent illuminated images were collected from a single exposure , while maximal intensity projections were formed from a series of continuous 100 ms TIRF exposures . Technical replicates were not collected , but each strain was imaged under a variety of appropriate imaging conditions to establish that the phenomena observed were not an artifact of the experimental setup . For single-molecule experiments imaging HaloTag-15aa-TagU ( bAB196 ) , HaloTag-15aa-TagV ( bAB197 ) , HaloTag-15aa-TagT ( bAB198 ) , Pbp2A-30aa-HaloTag ( BSY201 ) and MreB-30aa-HaloTag ( BYS40 ) , cells were grown at 37°C in CH medium from fresh colonies until reaching mid-exponential phase ( OD600 = 0 . 5 ) , ten-fold diluted back in fresh CH medium and grown one more round until OD600 = 0 . 5 . Cultures were then incubated for 15 min with 25 nM ( HaloTag-15aa-TagU/V/T and Pbp2A-30aa-HaloTag ) and 25 pM ( MreB-30aa-HaloTag ) of HaloTag-JF549 ligand . Cells were spun for 1 min at 4000 xg , 10-fold concentrated and imaged under 2% agarose pads . Images were collected on a Nikon TI microscope equipped with an EMCCD camera , together with a Nikon 100x NA 1 . 45 objective . Exposure times were 0 . 25 s , and illumination was accomplished using a 561 nm laser . BEG300 cells were grown from overnights as described in ‘Overnight culture growth’ but in CH medium instead of LB . Cells were then grown at 37°C for 4 hr without xylose to deplete WTAs , then induced with 30 mM xylose for 1 . 5 hr to re-induce WTA expression . Cells were then moved to 25°C for at least 30 min and incubated with 25 μg/mL Concanavalin A conjugated to Alexa Fluor 647 . Cells were collected by centrifugation at 6000 x g for 2 min , washed with CH medium , then re-suspended in fresh CH medium . Cells were then placed on a glass coverslip thickness No . 1 . 5 under an agar pad thickness 1 mm made from CH medium and 1 . 5% agarose . For PY79 and BCW61 controls , lectin-Alexa Fluor conjugate concentration was 200 μg/mL . Separate technical replicates were collected with the microscopes described in ‘Imaging – MreB Particle Tracking’ and ‘Imaging – Spinning Disk Confocal’ . For quantitative analysis , the latter setup was used . Quantification was performed in FIJI . Five individual cells were selected arbitrarily and pooled for quantitative analysis . The contours of each cell were manually traced , and intensity along these contours measured , then corrected for the mean fluorescent background . The per cell average was calculated from the mean of each pixel in the contour less background ( calculated from an empty field ) . The per strain average was taken as the mean of the per cell average . The output of uTrack is the position coordinates of tracks over frames . We fit a line through these coordinates using orthogonal least squares regression to minimize the perpendicular distance of the points from the line of best fit . We used principal component analysis for orthogonal regression using custom written MATLAB code . The R2 values we obtain range from 0 . 5 to 1 . We calculated mean track positions , angles and displacement using the line of best fit for all tracks . We also calculated the mean square displacement versus time of individual tracks and fit these curves to the quadratic equation MSD ( t ) = 4Dt+ ( Vt ) 2 , using nonlinear least squares fitting . As later times have fewer points and are noisier , we fit the first 80% of the data for each track . We determined α by fitting a straight line to the log⁡MSDt vs . log⁡ ( t ) curve . The goodness of fit was evaluated by determining the R2 value . We selected tracks for linearity and directional motion , based on the following cutoffs: R2 >0 . 9 , displacement >0 . 2 μm , , velocity >1e−9 μm/s , and R2 of the linear fit of log ( MSD ( t ) ) vs . log ( t ) >0 . 6 . The MATLAB-based software Morphometrics ( Ursell et al . , 2017 ) was used to segment phase contrast images of cells . We used the phase contrast setting for rod-shaped and intermediate states and the peripheral fluorescence setting for spherical states , because in this latter condition , peripheral fluorescence empirically did a better job of fitting cell outlines . The cell contours obtained were visually inspected and any erroneous contours were removed by custom written MATLAB code . Track angles were calculated with respect to the cell midline as defined by the Morphometrics ‘Calculate Pill Mesh’ feature , which identifies the midline based on a unique discretization of the cell shape determined from its Voronoi diagram . The difference between the track angle and midline angle was then calculated . Since the track angles θt and midline angles θm both ranged from −90° to 90° , the range of angle differences ∆θ= θt-θm was −180° to 180° . We changed the range to 0 to 180° by the transformation: Δθ=180+ Δθ if Δθ<0 , and 0 to 90° by the transformation: Δθ=180− Δθ if Δθ>90 . The mean deviation from 90° ( σ90 ) for each distribution was calculated using the following formula , where xi is each angle in the distribution and N is the total number of angles:σ90= ∑i=1N ( xi-90 ) 2N Custom written MATLAB code was used to calculate the normalized dot product ( DP ) of track pairs along with the distance ( d ) between their mean positions x- and y- as follows: DPij= cos ( θi −θj ) , dij= ( x−i−x−j ) 2+ ( y−i−yj− ) 2 To eliminate out-of-cell tracks we only considered those that had three other tracks within a 5 μm radius of their mean position . The dot product of track pairs ( DP ) and distance ( d ) between them was stored in data files , along with all the previous information for each individual track ( R2 , velocities , angles , mean positions , displacement etc ) . The files were then parsed using the cutoffs described in the ‘Data analysis – selecting directional tracks’ section . The tracks were binned based on the distance and the mean dot product calculated for each distance range as follows: DP-= 1N∑i>jNcos⁡ ( θi -θj ) A cutoff of 3 μm was chosen as the maximum binning distance , which is the average length of a cell . A data file containing simulated tracks was created by a custom written MATLAB script , which generates random angles distributed randomly on a 100 × 100 μm area . Each track has R2 = 0 . 95 , velocity = 25 nm/s and displacement = 1 μm . The same analysis code was run on these simulated tracks to generate track pairs with dot product and distance stored in a new data file . The data file was parsed using the same cutoffs as the real data and the mean dot product for each distance range calculated . The total numbers of trajectories within the simulation were much higher than the actual data ( 2–10 times higher ) . Pill meshes were created using Morphometrics ( Ursell et al . , 2017 ) , which calculates the coordinates of line segments perpendicular to the cell long axis . For cell widths at various steady state TagO and Mg2+ levels , the distance of these line segments was calculated using a custom written MATLAB script ( Hussain , 2017; copy archived at https://github . com/elifesciences-publications/hussain-2017-elife ) and the maximum width along the length of the cell was taken as the cell width . When measuring cell width nearest to a track ( for calculating track angle as a function of cell width ) , the mean width of the 10 nearest contour points from the track was calculated using a custom written MATLAB script . Cell widths of emerging bulges and rods from round cells were measured manually in FIJI . Our ability to segment individual spherical cells was limited by their nonuniform contrast , perhaps arising from the nonuniform thickness of these cells in the Z dimension; consequently , Morphometrics-based width measurements in these cells was limited , especially in cells exceeding 2 μm in diameter . Sidewall curvature of cells was extracted from the pill mesh obtained from Morphometrics . The curvature values are calculated from three successive contour points and smoothed over two pixels . The mean curvature of 3 nearest points to each track were calculated from both sides of the cell contour and called the mean curvature . Principal curvature ratio was calculated by dividing the sidewall curvature with the curvature in the radial direction ( calculated from cell width assuming the cell is radially symmetric ) . For radial curvature we used the following expression , where rcell is half the cell width:κ2=1rcell A value close to one indicates the two principal curvatures are similar and the cells are round . Phase contrast images were used to show rod shape emergence from local bulges . Edges were enhanced in FIJI and contrast adjusted to give bright cell outlines in the images . The stack was then colored in time using temporal color code function in FIJI . To create the curvature plot , the phase contrast images were run through Morphometrics which calculates the curvature at each contour point along the cell outline . The contour points of interest were selected and plotted using a custom written MATLAB script , which colored each point according to its local curvature as calculated by Morphometrics . To provide a good resolution for positive curvatures , we rescaled the color map such that negative curvatures were colored blue and positive curvatures were scaled by their curvature value . Data from agar pads experiments was analyzed using custom written MATLAB code . Data from cellASIC experiments was analyzed in Morphometrics to get areas for each cell . For doubling times during sphere to rod transitions , the data was collected by manually measuring the areas of the sphere and rod regions of the same cell in FIJI . In all cases , the area of each cell per frame was calculated and the log plot of area vs time was fit to a line . The doubling time was calculated using the slope of this line . N = 15 cells ( transitioning from spheres to rods ) . Cell contours ( Figure 1—figure supplement 1 ) were used to calculate tangent angles using the equation: θi=tan-1⁡yi+1-yixi+1-xi . The correlation between angles was calculated using the cosine of the angle difference binned as a function of number of points ( n ) between the angles: Gn= 1N∑i=1Ncos⁡ ( θi+n-θi ) . The number of points was converted to contour length using the pixel size of the camera to get the final correlation function: Gl= 1N∑i=1Ncos⁡ ( θi+l-θi ) . Tracks were generated using the TrackMate plugin in Fiji ( 2017 ) . Particles were detected with the Laplacian of Gaussians ( LoG ) detector , with a 0 . 4 μm spot diameter . Tracks were generated using the Simple LAP Tracker , with a 0 . 1 μm linking max distance and no frame gaps allowed . Tracks were exported into MATLAB for further processing . Mean squared displacement ( MSD ) was calculated for all tracks as a function of time delay ( t ) . For visualization of HaloTag tracks overlaid on phase images , we analyzed all tracks 10 frames or longer . For each track , the scaling exponent ( α ) was calculated by fitting MSD ( t ) =C ( t∧ α ) +4 ( σ∧2 ) using nonlinear least-squares fitting with constant C and where σ is the localization error ( Monnier et al . , 2012 ) . For analysis of alpha value and velocity frequencies tracks were first filtered using masks generated from phase images acquired after single-molecule imaging . We then included all tracks between 10 and 120 frames in length . α values were calculated by linear fitting log ( MSD ) versus log ( t ) . Velocity ( v ) was calculated by fitting MSD ( t ) =4D ( t ) + ( v*t ) ∧2 + 4 ( σ∧2 ) using nonlinear least-squares fitting , where D is the diffusion constant ( Monnier et al . , 2012 ) . For the graphs in Figure 6—figure supplement 2 , only plotted are molecules that moved in a consistent manner during their lifetime [>0 . 95 r2 fit to log ( MSD ) versus log ( t ) ] . Analysis of alpha value and velocity frequencies for MreB filaments was the same as for HaloTag , except we included all tracks between 10 and 60 frames in length and filtration by phase image masks was not necessary . Full length , un-tagged Thermotoga maritima MreB was purified as described previously ( Salje et al . , 2011 ) . The protein was encapsulated inside unilamellar liposomes following a previously published protocol ( Szwedziak et al . , 2014 ) . For this , 50 μL of E . coli total lipid extract , dissolved in chloroform at 10 mg/mL , was dried in a glass vial under a stream of nitrogen gas and left overnight under vacuum to remove traces of the solvent . The resulting thin lipid film was hydrated with 50 μL of TEN100 8 . 0 ( 50 mM Tris/HCl , 100 mM NaCl , 1 mM EDTA , 1 mM NaN3 , pH 8 . 0 ) , supplemented with 20 mM CHAPS ( Anatrace , Maumee , Ohio ) , and shaken vigorously at 800 rpm using a benchtop micro centrifuge tube shaker for 2 hr . The lipid-detergent solution was then sonicated for 1 min in a water bath sonicator . Subsequently , 50 μL of MreB protein solution at 30 μM , supplemented with 0 . 5 mM magnesium ATP ( Jena Bioscience , Germany ) was added and left for 30 min at room temperature . Next , the mixture was gradually diluted within 10–20 min to 600 μL with TEN100 8 . 0 plus 0 . 5 mM magnesium ATP ( without detergent ) to trigger spontaneous liposome formation . 2 . 5 μL of the solution was mixed with 0 . 2 μL 10 nm IgG immunogold conjugate ( TAAB , UK ) and plunge-frozen onto Quantifoil R2/2 carbon grid , using a Vitrobot automated freeze plunger ( FEI Company , Hillsboro , Oregon ) into liquid ethane . 2D electron cryomicroscopy images were taken on an FEI Polara TEM ( FEI Company ) operating at 300 kV with a 4k × 4 k Falcon II direct electron detector ( FEI Company ) at a pixel size of 1 . 8 Å . For electron cryotomography , samples were imaged using an FEI Titan Krios TEM ( FEI Company ) operating at 300 kV , equipped with a Gatan imaging filter set at zero-loss peak with a slit-width of 20 eV . A 4k × 4 k post-GIF K2 Summit direct electron detector ( Gatan , a subsidiary of Roper Technologies , Lakewood Ranch , Florida ) was used for data acquisition with SerialEM software ( 2005 ) at a pixel size of 3 . 8 Å at the specimen level . Specimens were tilted from −60˚ to +60˚ with uniform 1˚ increments . The defocus was set to between 8 and 10 μm , and the total dose for each tilt series was around 120–150 e/Å2 . Tomographic reconstructions from tilt series were calculated using RAPTOR ( 2008 ) and the IMOD tomography reconstruction package followed by SIRT reconstruction with the TOMO3D package ( 1996; 2011a ) . Movies showing liposomes were prepared with Chimera and PyMOL ( 2002; 2004 ) . Imaging of E . coli strain RM478 was conducted as in ‘Overnight culture growth’ , save the cells were grown in LB or M63-Glucose . When indicated , cells were grown and imaged under 2% agar pads made in the same media . For the media down shift ( LB at 37°C to M63-Glucose at 25°C ) , cells were first grown in LB at 37°C in a drum roller , then placed under agar pads made of M63-Glucose at 25°C and allowed to equilibrate 15 min prior to imaging . Imaging was conducted as in ‘Imaging – MreB particle tracking’ using a Nikon TI-E with TIRF illumination . All strains used in this study are available in Supplementary file 1 . All primers used in this study are available in Supplementary file 2 . BCW51 [ycgO::Pxyl-tagO , tagO::erm , amyE::sfGFP-mreB , sinR::phleo] was generated by transforming BEG300 with a Gibson assembly consisting of three fragments: ( 1 ) PCR with primers Sinr_up_F and Sinr_up_R and template PY79 genomic DNA; ( 2 ) PCR with primers oJM028 and oJM029 and template plasmid pWX478a ( containing phleo ) ; ( 3 ) PCR with primers Sinr_DOWN_R and Sinr_DOWN_F and template genomic DNA . BCW61 [tagE::erm] was generated by transforming PY79 with a Gibson assembly consisting of three fragments: ( 1 ) PCR with primers oCW054 and oCW055 and template PY79 genomic DNA; ( 2 ) PCR with primers oJM028 and oCW057 and template plasmid pWX467a containing cat; ( 3 ) PCR with primers oCW058 and oCW059 and template PY79 genomic DNA . BCW72 [yvhJ::PxylA-mazF ( cat ) ] was generated by transforming PY79 with a Gibson assembly consisting of three fragments: ( 1 ) PCR with primers oCW139 and oCW141 and template PY79 genomic DNA; ( 2 ) PCR with primers oJM029 and oMK047 and template DNA consisting of a fusion of cat and the mazF counterselectable marker from pGDREF ( Yu et al . , 2010 ) ; ( 3 ) PCR with primers oCW142 and oCW143 and template PY79 genomic DNA . BCW77 [ywtF::PxylA-mazF ( cat ) ] was generated by transforming PY79 with a Gibson assembly consisting of three fragments: ( 1 ) PCR with primers oCW159 and oCW161 and template PY79 genomic DNA; ( 2 ) PCR with primers oJM029 and oMK047 and template DNA consisting of a fusion of cat and the mazF counterselectable marker from pGDREF ( Yu et al . , 2010 ) ; ( 3 ) PCR with primers oCW164 and oCW165 and template PY79 genomic DNA . BCW78 [ywtF::msfGFP-ywtF] was generated by transforming BCW77 with a Gibson assembly consisting of three fragments: ( 1 ) PCR with primers oCW160 and oCW161 and template PY79 genomic DNA; ( 2 ) PCR with primers oCW072 and oCW073 and BMD61 genomic DNA; ( 3 ) PCR with primers oCW163 and oCW165 and template PY79 genomic DNA . BCW79 [yvhJ::msfGFP-yvhJ] was generated by transforming BCW72 with a Gibson assembly consisting of three fragments: ( 1 ) PCR with primers oCW139 and oCW146 and template PY79 genomic DNA; ( 2 ) PCR with primers oCW072 and oCW073 and BMD61 genomic DNA; ( 3 ) PCR with primers oCW143 and oCW145 and template PY79 genomic DNA . BCW80 [lytR::PxylA-mazF ( cat ) ] was generated by transforming PY79 with a Gibson assembly consisting of three fragments: ( 1 ) PCR with primers oCW101 and oCW109 and template PY79 genomic DNA; ( 2 ) PCR with primers oJM029 and oMK047 and template DNA consisting of a fusion of cat and the mazF counterselectable marker from pGDREF ( Yu et al . , 2010 ) ; ( 3 ) PCR with primers oCW100 and oCW125 and template PY79 genomic DNA . BCW81 [lytR::msfGFP-lytR] was generated by transforming BCW72 with a Gibson assembly consisting of three fragments: ( 1 ) PCR with primers oCW101 and oCW137 and template PY79 genomic DNA; ( 2 ) PCR with primers oCW072 and oCW073 and BMD61 genomic DNA; ( 3 ) PCR with primers oCW100 and oCW138 and template PY79 genomic DNA . BCW82 [tagO::erm , ycgO::PxylA-tagO , amyE::Pspac-gfp-mreB ( spec ) , dacA::kan] was generated by transforming BEG300 with genomic DNA from BGL19 . BEG202 [tagO::erm amyE::Pxyl-gfp-mbl ( spec ) ] was generated by transforming BEB1451 with genomic DNA from BJS18 . BEG281 [ycgO::PxylA-tagO] was generated by transforming with a plasmid created via ligating a Gibson assembly into pKM077 . pKM77 was digested with EcoRI and XhoI . The assembly was created with two fragments: ( 1 ) PCR with primers oEG85 and oEG86 and template py79 genomic DNA; ( 2 ) PCR with primers oEG87 and oEG88 . BEG291 [tagO::erm , ycgO::PxylA-tagO] was generated by transforming BEG281 with genomic DNA from BRB4282 . BEG300 [tagO::erm , ycgO::PxylA-tagO , amyE::Pspac-gfp-mreB ( spec ) ] was generated by transforming BEG291 with genomic DNA from BEG275 . BMD61 [mbl::mbl-msfGFP ( spec ) ] was generated by transforming py79 with a Gibson assembly consisting of four fragments: ( 1 ) PCR with primers oMD44 and oMD90 and template PY79 genomic DNA; ( 2 ) PCR with primers oMD47 and oMD56 and template synthetic , codon-optimized msfGFP; ( 3 ) PCR with primers oJM028 and oJM029 and template plasmid pWX466a ( containing spec ) ; ( 4 ) PCR with primers oMD48 and oMD50 and template genomic DNA . bSW99 [amyE::spc-Pspac-mciZ] was generated by transforming PY79 with a Gibson assembly consisting of five fragments: ( 1 ) PCR with primers oMD191 and oMD108 and template PY79 genomic DNA ( containing upstream region of amyE ) ; ( 2 ) PCR with primers oJM29 and oJM28 and template plasmid pWX466a ( containing spec ) ; ( 3 ) PCR with primers oMD234 and oSW76 and template plasmid pBOSE1400 ( a gift from Dr . Briana Burton , containing spec ) ; ( 4 ) PCR with primers oAB307 and oAB291 and template PY79 genomic DNA ( containing mciZ ) ; ( 5 ) PCR with primers oMD196 and oMD197 and template PY79 genomic DNA ( containing downstream region of amyE ) . bAB343 [tagO::erm , ycgO::cat-PxylA-tagO , amyE::spc-Pspac-mciZ , ftsAZ::ftsA-mNeonGreen-ftsZ] was generated by transforming bAB185 ( Bisson-Filho et al . , 2017 ) with genomic DNA from bSW99 . The resultant strain was then transformed with the genomic DNA from BEG291 and selected for Cm resistance . Subsequently , the resultant strain was transformed again with genomic DNA from BEG291 , but colonies were selected for MLS resistance in the presence of 30 mM of xylose and 25 mM MgCl2 . bAB327 [tagO::erm , ycgO::cat-PxylA-tagO , amyE::spc-Physpank-minCD , ftsAZ::ftsA-mNeonGreen-ftsZ] was generated by transforming bAB185 ( Bisson-Filho et al . , 2017 ) with genomic DNA from JB60 ( a gift from Dr . Frederico Gueiros-Filho ) . The resultant strain was then transformed with the genomic DNA from BEG291 and selected for Cm resistance . Subsequently , the resultant strain was transformed again with genomic DNA from BEG291 , but colonies were selected for MLS resistance in the presence of 60 mM xylose and 25 mM MgCl2 . bAB388 [tagO::erm , ycgO::cat-PxylA-tagO , amyE::spc-Physpank-ftsA , ftsAZ::ftsA-mNeonGreen-ftsZ] was generated by transforming bAB199 ( Bisson-Filho et al . , 2017 ) with genomic DNA from BEG291 and selected for Cm resistance . Subsequently , the resultant strain was transformed again with genomic DNA from BEG291 , but colonies were selected for MLS resistance in the presence of 60 mM xylose and 25 mM MgCl2 . bAB196 [tagU::erm-Pxyl-HaloTag-15aa-tagU] was generated by transforming PY79 with a Gibson assembly consisting of three fragments: ( 1 ) PCR amplifying the upstream region from tagU with primers oCW155 and oCW109 and template PY79 genomic DNA; ( 2 ) PCR amplifying the erm-Pxyl-HaloTag-15aa fragment with primers oJM029 and oCW73 and template bGS62 genomic; ( 3 ) PCR amplifying the downstream region from tagU with primers oCW138 and oCW156 and PY79 template genomic DNA . bAB197 [tagV::erm-Pxyl-HaloTag-15aa-tagV] was generated by transforming PY79 with a Gibson assembly consisting of three fragments: ( 1 ) PCR amplifying the upstream region from tagV with primers oCW140 and oCW141 and template PY79 genomic DNA; ( 2 ) PCR amplifying the erm-Pxyl-HaloTag-15aa fragment with primers oJM029 and oCW73 and template bGS62 genomic; ( 3 ) PCR amplifying the downstream region from tagV with primers oCW144 and oCW145 and PY79 template genomic DNA . bAB198 [tagT::erm-Pxyl-HaloTag-15aa-tagT] was generated by transforming PY79 with a Gibson assembly consisting of three fragments: ( 1 ) PCR amplifying the upstream region from tagT with primers oCW162 and oCW159 and template PY79 genomic DNA; ( 2 ) PCR amplifying the erm-Pxyl-HaloTag-15aa fragment with primers oJM029 and oCW73 and template bGS62 genomic; ( 3 ) PCR amplifying the downstream region from tagT with primers oCW163 and oCW166 and PY79 template genomic DNA . bYS09 [mreB::mreB-mNeonGreen] was generated by transforming BMD88 ( Schirner et al . , 2015 ) with a Gibson assembly consisting of three fragments: ( 1 ) PCR with primers oMD134 and oMD262 and bMD135 template genomic DNA ( containing the upstream of mreB and 40aa linker ) ; ( 2 ) PCR with primers oYS007 and oYS008 and gBlocks gene fragment containing mNeonGreen; ( 3 ) PCR with primers oMD92 and oMD116 and template PY79 genomic DNA . bYS40 [mreB::mreB-HaloTag] was generated by transforming BMD88 with a Gibson assembly consisting of three fragments: ( 1 ) PCR with primers oMD134 and oYS602 and bMD135 template genomic DNA; ( 2 ) PCR with primers oYS603 and oYS604 and template plasmid cdr1086 ( containing HALO ) ; ( 3 ) PCR with primers oMD92 and oMD116 and template PY79 genomic DNA . bYS201 [HaloTag-Pbp2A::cat] was generated by transforming PY79 with a Gibson assembly consisting of three fragments: ( 1 ) PCR with primers oMD083 and oYS136 and bMD98 sfGFP-Pbp2A::cat template genomic DNA ( 2 ) PCR with primers oYS599 and oYS598 and template plasmid cdr1086 ( containing HALO ) ; ( 3 ) PCR with primers oMD069 and oMD082 and template PY79 genomic DNA .

Many bacteria are surrounded by both a cell membrane and a cell wall – a rigid outer covering made of sugars and short protein chains . The cell wall often determines which of a variety of shapes – such as rods or spheres – the bacteria grow into . One protein required to form the rod shape is called MreB . This protein forms filaments that bind to the bacteria’s cell membrane and associate with the enzymes that build the cell wall . Together , these filament-enzyme complexes rotate around the cell to build and reinforce the cell wall in a hoop-like manner . But how do the MreB filaments know how to move around the circumference of the rod , instead of moving in any other direction ? Using a technique called total internal reflection microscopy to study how MreB filaments move across bacteria cells , Hussain , Wivagg et al . show that the filaments sense the shape of a bacterium by orienting along the direction of greatest curvature . As a result , the filaments in rod-shaped cells orient and move around the rod , while in spherical bacteria they move in all directions . However , spherical bacteria can regenerate into rods from small surface ‘bulges’ . The MreB filaments in the bulges move in an oriented way , helping them to generate the rod shape . Hussain , Wivagg et al . also found that forcing cells that lack a cell wall into a rod shape caused the MreB filaments bound to the cell membrane to orient and circle around the rod . This shows that the organization of the filaments is sufficient to shape the cell wall . In the future , determining what factors control the activity of the MreB filaments and the enzymes they associate with might reveal new targets for antibiotics that disrupt the cell wall and so kill the bacteria . This will require higher resolution microscopes to be used to examine the cell wall in more detail . The activity of all the proteins involved in building cell walls will also need to be extensively characterized .

2018

MreB filaments align along greatest principal membrane curvature to orient cell wall synthesis

Deep learning in in vitro fertilization is currently being evaluated in the development of assistive tools for the determination of transfer order and implantation potential using time-lapse data collected through expensive imaging hardware . Assistive tools and algorithms that can work with static images , however , can help in improving the access to care by enabling their use with images acquired from traditional microscopes that are available to virtually all fertility centers . Here , we evaluated the use of a deep convolutional neural network ( CNN ) , trained using single timepoint images of embryos collected at 113 hr post-insemination , in embryo selection amongst 97 clinical patient cohorts ( 742 embryos ) and observed an accuracy of 90% in choosing the highest quality embryo available . Furthermore , a CNN trained to assess an embryo’s implantation potential directly using a set of 97 euploid embryos capable of implantation outperformed 15 trained embryologists ( 75 . 26% vs . 67 . 35% , p<0 . 0001 ) from five different fertility centers . Assisted reproductive technologies ( ART ) such as in vitro fertilization ( IVF ) , while a solution to many infertile couples have been inefficient with an average success rate of approximately 30% reported in 2015 in the US ( CDC , 2015 ) . IVF is also an expensive solution costing patients well over $10 , 000 out-of-pocket for each ART cycle in the US with many patients requiring multiple cycles to achieve successful pregnancy ( CDC , 2015; Birenbaum-Carmeli , 2004; Toner , 2002 ) . Although multiple factors such as maternal age , medical diagnosis , gamete and embryo quality , and endometrium receptivity determine the success of ART cycles , the challenge of non-invasive selection of the highest available quality from a patient’s cohort of embryos ( top-quality embryo ) for transfer remains as one of the most important factors in achieving successful ART outcomes ( Vaegter et al . , 2017; Barash et al . , 2017; Conaghan et al . , 2013; Wong et al . , 2013; Racowsky et al . , 2015; Filho et al . , 2010; Machtinger and Racowsky , 2013; Demko et al . , 2016; Einarsson et al . , 2017; Hill et al . , 1989; Erenus et al . , 1991; Paulson et al . , 1990; Osman et al . , 2015 ) . Traditional methods of embryo selection rely on visual embryo morphological assessment and are highly practice-dependent and subjective ( Storr et al . , 2017; Baxter Bendus et al . , 2006; Paternot et al . , 2009 ) . Fully automated assessments of embryos are challenging owing to the complexity of embryo morphologies . Emulating the skill of highly trained embryologists in efficient embryo assessment in a fully automated system is a major challenge in all of the previous work done in computer-aided assessments of embryos due to focus on measuring specific expert-defined parameters such as zona pellucida thickness variation , number of blastomeres , degree of cell symmetry and cytoplasmic fragmentation , etc . ( Rocha et al . , 2017a; Rocha et al . , 2017b ) . Machine learning is loosely defined as a computer program that learns a given task over time through experience and improves itself to achieve the best possible task performance . In the past decade , advances in hardware compute performance and machine learning techniques have significantly improved their applicability in real-world medical and non-medical problems . Recently , machine learning has been proposed as a solution for automated analysis of embryo morphologies ( Rocha et al . , 2017b; Bormann et al . , 2020; Dimitriadis et al . , 2019; Thirumalaraju et al . , 2019; Khosravi et al . , 2019; Kanakasabapathy et al . , 2019a ) . This work makes use of a deep convolutional neural network ( CNN ) , a representation learning technique , that has been proven to be effective in image classification tasks . Unlike most prior computer-aided algorithms , including some techniques of machine learning used for embryo assessment , the reported CNN architecture allows automated embryo feature selection and analysis at the pixel level without any interference by an embryologist ( Rocha et al . , 2017a; Rocha et al . , 2017b ) . Such networks do not depend on human-specified features and can develop an ability to evaluate embryos categorically through iterative learning from thousands of examples . The use of deep-learning in IVF has also been explored; however , these recent neural network-based approaches have focused on either classifying embryos based on morphological quality and were not evaluated for transfer outcomes , or were developed with the use of time-lapse series of images toward the evaluation of implantation ( Khosravi et al . , 2019; Tran et al . , 2019 ) . It is important to emphasize here that most fertility centers do not possess time-lapse imaging hardware even in the United States of America ( Dolinko et al . , 2017 ) . The lack of availability of such hardware limits an otherwise promising technology mostly to resource-rich settings and fail to improve quality of and access to care in resource-constrained settings where such advances are sorely needed ( Wahl et al . , 2018; Hosny and Aerts , 2019 ) . Furthermore , in current clinical practice , embryos with the highest morphological grades ( top-quality ) are the first to be transferred , however , clinically these decisions are performed manually , even with time-lapse imaging systems . The development of networks that can measure an embryo’s potential for implantation and help in rank ordering embryos in a patient embryo cohort for transfer have utility in virtually all fertility centers . Conventionally , embryo transfers are performed at the cleavage or the blastocyst stage of development . Embryos are at the cleavage stage 2–3 days after fertilization and develop further in suitable culture conditions to reach the blastocyst stage 5–7 days after fertilization . Blastocyst embryo transfers , in particular , have been associated with better implantation rates and have helped lower the number of embryos transferred at a time ( De Croo et al . , 2019 ) . Therefore , in this study , we have investigated the use of a CNN pre-trained with 1 . 4 million ImageNet images and transfer-learned using 2440 static human embryo images recorded at a single time-point of 113 hr post insemination ( hpi ) for the development of neural networks that can help identify embryos capable of implantation and for identifying the top quality embryos ( Figure 1 ) . The top-quality embryos were identified by combining a previously developed network ( Xception architecture ) trained to classify embryos based on its blastocyst quality with a genetic algorithm scheme ( Figure 1; Thirumalaraju et al . , 2020 ) . The original neural network was trained on a hierarchical system of categorization , derived from a clinical Gardener-based grading system , to minimize data sparsity and improve overall network learning ( Kanakasabapathy et al . , 2019a; Thirumalaraju et al . , 2020; Kanakasabapathy et al . , 2019b; Esteva et al . , 2017 ) . The two major categories of non-blastocysts and blastocysts made up the inference classes , which included the training classes 1 , 2 , and 3 , 4 , 5 , respectively ( Figure 1 ) . Pre-training with a large dataset of images from ImageNet honed the ability of the developed CNN to identify the shape , structure , and texture variations between morphologically complex embryos with minimal data requirements while the genetic algorithm helped in rank ordering embryos by generating unified scores ( Figure 1 ) . The developed network was evaluated using an independent test set comprising of 97 patient-embryo cohorts . Embryos of the highest quality that were selected from the patient cohorts were evaluated using known implantation outcomes . Additionally , we also investigated if the neural network can be trained to directly differentiate between embryos based on their potential for implantation ( Figure 1 ) . Our tests with patient cohorts using the algorithm does not account for the ploidy status of the embryos . Since pre-implantation genetic screened ( PGS ) euploid embryos are associated with higher implantation chance , we also designed a neural network to evaluate the network performance in refining the screened embryos based on their implantation potential . The evaluations using the patient cohorts tend to yield embryo selections with unknown outcomes or ploidy status , therefore , for this section of the study , we utilized a test set of 97 euploid embryos with known implantation outcomes . The CNN was trained and evaluated in identifying euploid embryos capable of implantation and the performance was compared against those of 15 embryologists from five different fertility centers across the United States of America . In our evaluations of the CNN in categorizing embryos imaged at 113 hpi based on their morphology , the network performed with an accuracy of 90 . 97% ( area under the curve: 0 . 96 ) in differentiating between blastocysts and non-blastocysts ( n = 742 ) ( Kanakasabapathy et al . , 2019a; Thirumalaraju et al . , 2020; Figure 2—figure supplement 1 ) . The high accuracy indicated that the trained network was concordant with embryologists in categorizing embryos . These categorization scores ( five values per embryo ) need to be used by taking into account the scores of other embryos in the cohort to establish a rank order . In order to use the five probability values effectively for calculating the embryo score , we utilized a genetic algorithm , which is well-suited for optimization problems with multiple existing solutions . Here , the genetic algorithm empowered the developed CNN to make selections of the top-quality embryos within a patient’s embryo cohort at 113 hpi . Therefore , once we established that the network was capable of categorizing embryos based on their morphologies with high accuracy , we used a genetic algorithm and the network defined probability values of the embryos , belonging to each of the five training classes , to rank order the embryos for transfer . The 5 × 1 vector weights generated by the genetic algorithm during its training phase were used in evaluating retrospectively collected embryo cohorts from 97 patients . The final weights utilized in this study were −10 . 01226347 , –3 . 63697951 , −3 . 32090987 , 2 . 15367795 , and 2 . 8715555 for classes 1 through 5 , respectively . Embryos were ranked by the algorithm from highest to the lowest . According to the American Society for Reproductive Medicine guidelines on the limits to the number of embryos per transfer , one embryo is transferred for high prognosis patients with <37 years of age and two or more embryos are transferred for patients with >37 years of age as well as younger patients with low prognosis ( Practice Committee of the American Society for Reproductive Medicine . Electronic address: ASRM@asrm . org and Practice Committee of the Society for Assisted Reproductive Technology , 2017 ) . Therefore , in this study , the selection accuracy was assessed for scenarios of single embryo transfers ( SET ) and double embryo transfers ( DET ) . Using embryo cohort images ( n = 732 ) from the 97 patients , the accuracy of 5 well-trained embryologists’ selections were evaluated in comparison to selections made by the CNN + genetic algorithm ( CNNg ) . The rank-ordering performed by the algorithm may not utilize the same features used by embryologists in identifying the top embryos for transfer . Therefore , we initially evaluated the ability of both groups to effectively select ( i ) blastocyst ( s ) for transfer and ( ii ) the highest quality of blastocyst ( s ) ( HQB ) available for transfer . High-quality blastocysts are defined as embryos that met the freezing criteria ( >3 CC blastocyst grade; see Materials and methods ) of the Massachusetts General Hospital ( MGH ) fertility clinic . For blastocyst selections at 113 hpi , the CNNg algorithm performed with an accuracy of 98 . 96% for SET , which was similar ( p>0 . 05 ) to the average accuracy of the embryologists ( 96 . 91% , CI: 94 . 69% to 99 . 12% ) ( n = 5 ) ( Figure 2A ) . However , when two embryo selections for DET were allowed based on blastocyst and non-blastocyst classification , the CNNg algorithm performed with an accuracy of 100 . 00% , which was better ( p<0 . 05 ) than embryologists ( n = 5 ) who performed with an average accuracy of 98 . 76% ( CI: 97 . 69% to 99 . 83% ) ( Figure 2B ) . Toward the selection of HQB at 113 hpi , the accuracy of the CNNg algorithm for SET was 89 . 69% similar ( p>0 . 05 ) to the embryologists ( n = 5 ) who performed with an average accuracy of 90 . 31% ( CI: 87 . 50% to 93 . 11% ) ( Figure 2C ) . When two embryo selections for DET at 113 hpi were allowed , the system performed with a better ( p<0 . 05 ) accuracy of 97 . 94% in comparison to the embryologists who performed with an average accuracy of 96 . 91% ( CI: 96 . 00% to 97 . 81% ) ( Figure 2D ) . The evaluations indicated that the two groups made selections that were of similar quality or marginally different quality . Since the network was trained on the MGH classification criteria , the comparable performance of the CNNg algorithm and embryologists indicated that the neural network has trained itself sufficiently and made selections that were of clinically acceptable quality . In our evaluations , the selections made by each group , while were of similar quality , were observed to not necessarily be the same embryos from each cohort , and thus their transfer outcomes may be different . It is critical to evaluate the system performance in selecting the patient embryos based on pregnancy ( implantation ) outcome . Typically , in a clinical IVF cycle , the top-quality embryo is selected from the cohort of available embryos and is transferred to the patient . Embryos , which are similarly of a high-quality , are often frozen based on the freezing criteria used by the fertility center , for transfers in subsequent procedures for the same patient if needed . Frozen cycle transfers are not performed for all patients . Hence , the CNNg algorithm was evaluated in embryo selection for SET at 113 hpi using patient embryo cohorts based on actual implantation outcomes of the selected embryos and associated cycle characteristics ( n = 97 ) are provided in Supplementary file 1-table 1 . The test dataset was retrospectively collected based on pre-defined selection criteria and evaluations of transfer outcomes were performed using fresh embryo transfer cycles . The system selected 97 embryos in 97 patient embryo cohorts ( 742 embryos in total ) , out of which 44 embryos had known implantation outcomes . The accuracy of the system in SET through embryo selection at 113 hpi based on its implantation outcome was 59 . 1% while the implantation success rate for the 102 transferred embryos at the MGH fertility center was 44 . 1% for blastocyst transfers ( Supplementary file 1-table 2 ) . Furthermore , prior reports suggest that in general practice , the average implantation rates for manual-based embryo selection and transfers at blastocyst stages can be as low as 34% ( Martins et al . , 2017 ) . A limitation of a retrospective study is that not all embryos are transferred . Implantation outcomes of all embryos selected by the CNNg algorithm cannot be evaluated . Therefore , although the dataset was prepared not taking into consideration the availability of subsequent frozen cycle transfers , we investigated with the fertility center if the patients of the test set had any subsequent embryo transfers using the frozen embryos from the test set . When we consider subsequent frozen embryo transfers , five embryos originally selected by the CNNg algorithm at 113 hpi had known implantation outcomes of which four led to successful implantations ( Supplementary file 1-table 2 ) . The accuracy of the CNNg algorithm in SET , when both fresh and frozen embryo transfers were considered , was 61 . 2% . In such a scenario , for this specific dataset , the implantation success rate at MGH fertility center was 48 . 5% for blastocyst transfers when including both frozen and fresh transfers . The results suggest that the CNNg algorithm has the potential to improve clinical transfer outcomes . It should , however , be emphasized that in this particular analysis the performance of the system was evaluated by only using the embryos selected by the network and the embryologists . Furthermore , to evaluate if a CNN can potentially measure implantation potential through morphology alone , a pooled set of 29 embryo images with known transfer outcomes in a pilot study was used by the network to evaluate embryos based on their potential for implantation . The network was trained as a binary classifier and the SoftMax probability values outputted by the network was used as the embryo’s implantation potential . The CNN was retrained using 281 embryo images with known implantation outcomes that did not overlap with the test set and the final classification layer was replaced with the two classes- negative implantation and positive for implantation . The ability to differentiate embryo was measured through a receiver operating characteristic curve ( ROC ) analysis , establishing area under the curve ( AUC ) of 0 . 771 ( CI: 0 . 579 to 0 . 906 ) ( p<0 . 05 ) and the CNN performed with an accuracy of 82 . 76% ( CI: 64 . 23% to 94 . 15% ) ( Figure 3A ) . Ten out of 11 embryos had implanted with an implantation potential of over 0 . 47 and similarly , for embryos that scored less than 0 . 47 , 12 out of 18 embryos did not implant according to the patient cycle history . After we observed high performance in the artificial intelligence ( AI ) -based implantation potential prediction when compared with historical clinical data , we further conducted a multi-center AI system evaluation by comparing the implantation potential prediction accuracies obtained from the AI system and the embryo selections of 15 embryologists from five different fertility clinics . Here , we used 97 genetically screened euploid embryos transferred at 113 hpi to remove the effect of chromosomal abnormalities as a confounder , which existed in the pilot study ( 29 patient embryo ) . The IVF cycle characteristics in which these embryos were used are provided in Supplementary file 1-table 3 . The system performed with an accuracy of 75 . 25% while the embryologists performed with an average accuracy of 67 . 35% ( CI: 64 . 52% to 70 . 19% ) in differentiating euploid embryos based on their implantation outcome ( Figure 3B ) . A one-sample t-test revealed that the CNN significantly outperformed ( p<0 . 05 ) the embryologists in predicting embryo implantation by measuring the implantation potential of euploid embryos using a static image obtained at a single time-point of 113 hpi . The average implantation score of euploid embryos misclassified based on their implantation outcome using the CNN was 0 . 57 . 95% of the misclassified euploid embryos possessed scores ranging between 0 . 51 and 0 . 63 . Implantation scores closer to 0 . 5 indicate lower confidence in system predictions while implantation scores closer to 0 or 1 indicate higher confidence in system predictions ( Figure 3—figure supplement 1 ) . These results indicate that the majority of system errors in misclassifying the euploids occur among the embryos with the lowest confidence . Approximately 91% of euploid embryos with implantation potential scores of 0 . 80 or higher , and nearly 81% of embryos with implantation potential scores above 0 . 66 successfully implanted when transferred ( Figure 3—figure supplement 1 ) . Similarly , around 78% of euploid embryos with an implantation potential <0 . 33 , failed to successfully implant when transferred ( Figure 3—figure supplement 1 ) . These results suggest that the network’s implantation scores agree well with transfer outcomes even in high-quality euploid embryos . Deep neural networks hold value in aiding clinical decision making and have received significant attention from the IVF community . The deep-neural network-based approach showcased here is an objective approach to one of the more subjective but important parts of a clinical IVF process-embryo selections for transfer ( Bormann et al . , 2020 ) . Since over 80% of fertility clinics rely on non-time lapse imaging systems as part of their clinical processes , such neural network-based algorithms that rely purely on static single timepoint images can effectively assist in decision making ( Dolinko et al . , 2017 ) . In our study , we have evaluated two neural network-based approaches for improving embryo selection . Firstly , we have demonstrated that a deep-neural network in combination with a genetic algorithm ( CNNg ) can yield a continuous score that represents the quality of the embryo and that objective orders of transfer can be determined for a given set of embryos using such scores . The ranking algorithm studied here was able to consistently select embryos of the highest available morphological quality . Although the network was trained to classify embryos based on their quality , it performed well even in differentiating between embryos of the same class when combined with a genetic algorithm . The benefit of such systems is particularly evident in cases where selections made by the clinic/embryologist , although of similar grade , resulted in lower overall transfer success rates . Our networks only focused on the morphological features for embryo quality assessments due to data scarcity . The network’s learning can be compounded with data from additional timepoints , morphokinetics , and patient and cycle-specific information for more personalized IVF predictions and outcomes . Recently , Tran et al . studied the use of a deep-learning model ( IVY ) that can analyze whole time-lapse videos instead of specific time points for fetal heartbeat prediction ( Tran et al . , 2019 ) . However , the study was flawed since embryos with unknown outcomes ( non-transferred embryos ) were considered as negative outcome cases , which made up most of their dataset ( ~90% ) . The heavy class bias in their dataset and improper study design severely limits any conclusions that can be drawn from the work . A major hurdle for the development of networks capable of analyzing multi-timepoint images and with additional patient-specific information is the limited availability of diversified data with known clinical outcomes . During training , the lack of availability of such data prevents the networks from effectively learning relevant outcome-associated patterns in data . The need for data scales with the complexity of the task and the number of variables introduced . While this work focuses primarily on the utility of deep-learning algorithm for embryo evaluations at 113 hpi , it is also possible to develop similar networks for embryo evaluations at different timepoints , provided that sufficient data with matched outcomes/annotations are available . We have evaluated a similar network for use with cleavage-stage embryos ( 70 hpi ) and showed that deep-learning approaches can outperform trained embryologists in certain tasks such as embryo selection ( Thirumalaraju et al . , 2019; Kanakasabapathy et al . , 2020 ) . A major concern in any clinical practice , however , is the loss of viable embryos due to system errors . Therefore , the AI-based embryo selection algorithm reported here does not make any suggestion on discarding embryos . All embryos assessed by the CNNg in the selection process may be cryopreserved as per clinical practice . Thereby , our approach will not negatively affect the cumulative pregnancy rate since viable embryos will not be lost . However , it may improve the pregnancy rate as the system may be able to improve the chance of achieving a pregnancy faster with fewer embryos transferred . Furthermore , it is important to note that in its current stage this system is intended to act only as an assistive tool for embryologists . The embryologists can include the system’s prediction to make better judgments during embryo selection . The scores provided by the algorithm are continuous , but it can also be easily modified to present its scoring results in both binary and a more categorical format . Clinically , besides morphological features , various other important metrics and parameters are considered by embryologists at the time of decision making such as taking into account the ploidy status of the transferable embryos . PGS verified euploid embryos have been shown to possess a higher probability of successful outcome but cost a hefty premium on top of the cycle costs at most fertility centers in the United States ( Drazba et al . , 2014 ) . Furthermore , for patients with two or more euploid embryos , additional assessments of embryo morphology are required to select the best embryo based on their morphology for transfer , since euploids do not inherently guarantee implantation . Thus far , to the best of our knowledge , no system , deep-learning-based or otherwise , has been shown to be capable of differentiating between euploid blastocysts based on their capacity for implantation . Euploid embryos are usually of the highest available quality and differentiating between them objectively and reliably through manual analysis can be extremely challenging . The CNN-based approach , through direct estimations of implantation potential from 113 hpi embryo morphology , outperformed trained embryologists in identifying implanting embryos from a set of PGS euploid embryos . This accomplishment exhibits the potential of artificial intelligence-based approaches to improve success rates in the IVF lab . Our observations indicated that the system performed with a significantly better agreement with the actual implantation outcome for embryos with implantation scores closer to 1 or 0 ( Higher confidence ) . Furthermore , the comparison between the decisions made by 15 embryologists from different fertility centers in the US and the deep-neural network showcased that neural networks can outperform embryologists in identifying embryos capable of implantation . Hence , by applying the suggestions of a CNN , a trained embryologist can improve their selection of the embryo with the highest implantation potential . Advances in artificial intelligence have fostered numerous applications that have the potential to improve standard-of-care in the different fields of medicine . While other groups have also evaluated different use cases for machine learning in assisted reproductive medicine , this approach is novel in how it used a CNN trained on a large dataset to make predictions based on static images . The approach has shown the potential of CNNs to be used in aiding embryologists to select the embryo with the highest implantation potential , especially amongst high-quality euploid embryos . Although the current retrospective study shows that these systems can perform better than highly-trained embryologists , randomized control trials are required before routine use in clinical practice is adopted . Data were collected at the Massachusetts General Hospital ( MGH ) fertility center in Boston , Massachusetts . We used 3469 recorded videos of embryos collected from 543 patients with informed consent for research and publication , under an institutional review board approval for secondary research use . Videos were collected for research after institutional review board approval by the Massachusetts General Hospital Institutional Review Board ( IRB#2017P001339 and IRB#2019P002392 ) . All the experiments were performed in compliance with the relevant laws and institutional guidelines of the Massachusetts General Hospital , Brigham and Women's Hospital , and Partners Healthcare . The videos were collected using a commercial time-lapse imaging system ( Vitrolife Embryoscope ) . The imaging system used a Leica 20x objective that collected images at 10-min intervals under illumination from a single 635 nm LED . Each patient’s set of embryos were exported as videos ( . avi ) using the imaging system software . The videos of individual embryos were broken down into their respective frames to extract images from all timepoints post insemination . The images were identified by their timestamps and only images collected at 113 ± 0 . 05 hr post insemination were processed and used in this study . The extracted images were 250 × 250 pixels and they were cropped to 210 × 210 pixels . The cropping removed both the timestamps and identifiers present in the frame . All embryos used in the study were annotated using images from the fixed time-points ( 113 hpi ) by senior-level embryologists with a minimum of 5 years of human IVF training . Annotations for embryo implantation were assigned based on clinical outcomes . Out-of-focus images were included in the datasets and used for both testing and training . Only images of embryos that were completely non-discernable were removed from the study as part of the data cleaning procedure . The two networks in this study used two categorization systems . The network focused on the rank ordering of embryos used a hierarchical categorization system . The embryo images at 113 hpi time point were categorized between training classes 1 through five as described in detail elsewhere ( Thirumalaraju et al . , 2020 ) . Briefly , degenerated embryos , which did not begin compaction formed Class 1 while Class 2 embryos were those that reached the morula stage by 113 hpi . Classes 1 and 2 together formed ‘non-blastocysts’ inference class . Class three embryos exhibit features of an early blastocyst which is highlighted by the presence of blastocoel cavity and thick zona pellucida but lack expansion . Class four embryos were blastocysts with blastocoel cavities occupying over half of the embryo volume but either their inner cell mass ( ICM ) or trophectoderm ( TE ) was of poor quality . They are non-freezable quality embryos ( <3 CC ) , where three represents the degree of expansion ( range 1–6 ) and C represents the quality of ICM and TE ( range A-D ) , respectively . Class 5 embryos , however , met cryopreservation criteria ( >3 CC ) and included full blastocysts to hatched blastocysts . Classes 3 , 4 , and 5 together formed ‘blastocysts’ inference class . The two inference classes are used since the differentiation of blastocysts and non-blastocysts is a universally accepted categorization that is relevant to embryologists , while the five class categorization is specific to the neural network training , performance and evaluation ( Thirumalaraju et al . , 2020 ) . Networks that were focused on estimating an embryo’s implantation potential used a two-class training and inference system- positive for implantation and negative for implantation . The 113 hpi evaluation dataset included images of 2440 embryos categorized across five classes post-cleaning based on their clinical annotations made at 113 hpi . Our training set for this classification task used 1188 images with a validation dataset of 510 images obtained at 113 hpi . With the availability of unskewed validation sets prior to augmentation , we used a data generator during training , which performed random rotations and flips across all classes on the fly . The system performing with an accuracy of 90 . 97% was used in this study in combination with our genetic algorithm . The genetic algorithm was trained and tested with the training data prior to testing it with our independent test data . No human interaction was required/performed once the images were provided to the system during testing , as the entire process was fully automated . The independent non-overlapping test set consisted of 742 images of embryos originating from 97 patients . The selections were compared with embryologist selections . The network was also trained to classify embryos with successful and unsuccessful implantation . 281 embryo images with known implantation outcomes were used for training . Implantation signifies the attachment of a blastocyst into the endometrium . The status of implantation was clinically verified by ultrasound ~6 weeks after embryo transfer . Ninety-seven euploid embryos were evaluated by 15 embryologists , including director level embryologists from five different fertility centers . A genetic algorithm was designed to perform selections in combination with the neural network . The genetic algorithm component utilizes the probability scores of every embryo belonging to each of the five different classes to generate a transfer score that can be used to effectively identify the best embryo available in a cohort . For system evaluations , we used an independent set of embryos ( 100 patients; 2–12 embryos per patient ) , with no overlap with the training data set used for any prior exercise . The patient cohorts were chosen under the following criteria: ( i ) each patient embryo cohort had to possess at least two 2PN embryos , and ( ii ) at least one embryo of the patient embryo cohort developed to blastocyst stage by 113 hpi . We trained a genetic algorithm to select the morphologically highest quality embryo from a given cohort . There are four phases namely initialization , selection , crossover , and mutation . The classified embryos for each patient were sorted according to their identifier numbers allotted by the deep neural network . A population of weights was generated at random during initialization . A population size of 100 was generated with a 5 × 1 matrix representing each weight . Each weight defined a possible solution for the rank-ordering of embryos based on their quality using the five training classes . The dot product of the weights with the output logits provided by the CNN was used in the calculation of the fitness . The algorithm runs multiple cycles to select the optimal set of weight towards achieving the appropriately suitable rank order of embryos based on their qualities . At each cycle , all the weight sets obtained using the given population were used rank-ordering embryos within the training set . The best 20 wt sets were selected in each cycle . These selected weights ( specimens ) were then bred with each other with a probability set to 20% . It randomly selected two specimens from the selected top pool and created a random binary 5 × 1 matrix , where one represents that the given element should be switched in cohort and 0 represents that given element should not be switched within the cohort . The fitness function checks if the selected embryo belongs to the highest class available within the tested cohort . It checks if the selected solution ( specimen ) picked the embryo belonging to the top class in a given cohort of patient embryos . If the selected embryo belonged to the top class , the score was increased and if it did not , the score was not modified . After iterating for all patients’ cohorts , the total scores were used to select the best 20 weights of the given population and were taken for crossover and mutation to repeat the process . The new specimens replaced their parents in the top selected group of embryos . Otherwise , the matrix remained the same . After breeding , each specimen from the top selected group was mutated to give five mutations by adding a random float 5 × 1 matrix with a probability of 20% . These mutations were then added to the new population and the selection step was repeated with the new population of 100 . The genetic algorithm ran until the entire population converged to the same score after which a random weight was selected from the population as the final weight . Thus , final generated weights were used to further test the embryo cohorts within our test set .

Around one in seven couples have trouble conceiving , which means there is a high demand for solutions such as in vitro fertilization , also known as IVF . This process involves fertilizing and developing embryos in the laboratory and then selecting a few to implant into the womb of the patient . IVF , however , only has a 30% success rate , is expensive and can be both mentally and physically taxing for patients . Selecting the right embryos to implant is therefore extremely important , as this increases the chance of success , minimizes complications and ensures the baby will be healthy . Currently the tools available for making this decision are limited , highly subjective , time-consuming , and often extremely expensive . As a result , embryologists often rely on their experience and observational skills when choosing which embryos to implant , which can lead to a lot of variability . An automated system based on artificial intelligence ( AI ) could therefore improve IVF success rates by assisting embryologists with this decision and ensuring more consistent results . The AI system could learn how embryos develop over time and then uses this information to select the best embryos to implant from just a single image . This would offer a cheaper alternative to current analysis tools that are only available at the most expensive IVF clinics . Now , Bormann , Kanakasabapathy , Thirumalaraj et al . have developed an AI system for IVF based on thousands of images of embryos . Using individual images , the system selected embryos of a comparable quality to those selected by a human specialist . It also showed a greater ability to identify embryos that will lead to successful implantation . Indeed , the software outperformed 15 embryologists from five different centers across the United States in detecting which embryos were most likely to implant out of a group of high-quality embryos with few visible differences . Artificial intelligence has many potential applications to support expert clinical decision-making . Systems like these could improve success , reduce errors and lead to faster , cheaper and more accessible results . Beyond immediate IVF applications , this system could also be used in research and industry to help understand differences in embryo quality .

2020

Performance of a deep learning based neural network in the selection of human blastocysts for implantation

Smc–ScpAB forms elongated , annular structures that promote chromosome segregation , presumably by compacting and resolving sister DNA molecules . The mechanistic basis for its action , however , is only poorly understood . Here , we have established a physical assay to determine whether the binding of condensin to native chromosomes in Bacillus subtilis involves entrapment of DNA by the Smc–ScpAB ring . To do so , we have chemically cross-linked the three ring interfaces in Smc–ScpAB and thereafter isolated intact chromosomes under protein denaturing conditions . Exclusively species of Smc–ScpA , which were previously cross-linked into covalent rings , remained associated with chromosomal DNA . DNA entrapment is abolished by mutations that interfere with the Smc ATPase cycle and strongly reduced when the recruitment factor ParB is deleted , implying that most Smc–ScpAB is loaded onto the chromosome at parS sites near the replication origin . We furthermore report a physical interaction between native Smc–ScpAB and chromosomal DNA fragments . Compaction and individualization of sister DNA molecules is a prerequisite for efficient segregation of the genetic material to daughter cells during cell division . Multi-subunit Structural Maintenance of Chromosomes ( SMC ) protein complexes—such as cohesin and condensin—are major determinants of chromosome structure and dynamics during the cell cycle in eukaryotes as well as in prokaryotes ( Hirano , 2006; Thadani et al . , 2012; Gruber , 2014 ) . Condensin subunits were initially identified as abundant , non-histone components of mitotic chromosomes in metazoans ( Hirano and Mitchison , 1994 ) . In mitosis , condensin localizes together with topoisomerase II in punctate structures to the longitudinal core of chromatids , called the chromosome axis ( Coelho et al . , 2003; Maeshima and Laemmli , 2003; Ono et al . , 2004 ) . Inactivation of condensin subunits by mutation or depletion results in severe morphological aberrations and mechanical sensitivity of metaphase chromosomes , and subsequently to defects in their segregation during anaphase ( Hirano and Mitchison , 1994; Ono et al . , 2003; Gerlich et al . , 2006 ) . In bacteria , Smc–ScpAB is the prevalent version of SMC protein complexes . Its distant relatives MksBEF and MukBEF can be found scarcely scattered over most of the bacterial phylogenetic tree and in isolated branches of proteobacteria , respectively ( Gruber , 2011 ) . In Bacillus subtilis and Streptococcus pneumoniae , Smc–ScpAB is recruited to a region around the replication origin by ParB/Spo0J protein bound to parS sites , thereby forming a discrete focus—also called condensation center—on each nascent copy of the chromosome ( Gruber and Errington , 2009; Sullivan et al . , 2009; Minnen et al . , 2011 ) . Inactivation of Smc–ScpAB in B . subtilis under nutrient rich growth conditions blocks separation of sister replication origins and consequentially leads to lethal defects in chromosome partitioning ( Gruber et al . , 2014; Wang et al . , 2014 ) . Smc–ScpAB thus promotes the initial stages of chromosome segregation in B . subtilis , likely by condensing and individualizing the emerging copies of the chromosome in preparation for their segregation to opposite halves of the cell . The canonical SMC complex in bacteria comprises five subunits: ( 1 ) two Smc proteins , which each form a 45 nm long antiparallel coiled coil that connects an ABC-type ATPase ‘head’ domain at one end of the coiled coil with a ‘hinge’ homodimerization domain at the other end ( Hirano et al . , 2001 ) , ( 2 ) a single ScpA subunit , which belongs to the kleisin family of proteins and associates via its C-terminal winged-helix domain ( WHD ) with the bottom ‘cap’ surface of one Smc head and via its N-terminal helical domain with the ‘neck’ coiled coil region of the other Smc protein ( Bürmann et al . , 2013 ) , and ( 3 ) a dimer of ScpB protein , which binds to the central region of ScpA ( Bürmann et al . , 2013; Kamada et al . , 2013 ) . Overall , the pentameric Smc–ScpAB complex displays a highly extended conformation harboring a central channel , which is surrounded by a closed tripartite ring formed by the Smc dimer and the ScpAB2 sub-complex . The B . subtilis Smc coiled coils associate with one another to form rod-shaped Smc dimers ( Soh et al . , 2015 ) . Furthermore , the Smc head domains can interact directly with one another—via a composite interface that includes two molecules of ATP . Binding to ATP , head engagement and ATP hydrolysis likely control and drive the biochemical action of Smc–ScpAB . Models for SMC condensation activity have been proposed based on observations made with isolated SMC dimers , SMC fragments or holo-complexes . Such protein preparations support the bridging of given DNA molecules in vitro as indicated by the re-annealing of single stranded DNA , intermolecular DNA ligation , DNA catenation and the co-purification of labeled and unlabeled DNA molecules ( Sutani and Yanagida , 1997; Losada and Hirano , 2001; Cui et al . , 2008 ) . Many SMC complexes bound to different segments of DNA might thus come together and anchor DNA in condensation centers or at the chromosome axis . Oligomeric assemblies of bacterial Smc proteins have indeed been observed by Atomic Force Microscopy and Electron Microscopy ( Mascarenhas et al . , 2005; Fuentes-Perez et al . , 2012 ) . This model provides a straightforward explanation for the compaction activity of SMC . However , it is unclear how such apparently indiscriminate DNA aggregation would promote rather than block the individualization of sister chromosomes ( Gruber , 2014 ) . Local wrapping of DNA around the SMC complex could result in well-defined lengthwise condensation of DNA . However , too little SMC protein appears to be present in chromosomes to yield decent levels of compaction by simple wrapping . A different hypothesis is based on the finding that the structurally related cohesin complex holds sister chromatids in eukaryotes together by entrapping sister DNA fibers within its ring ( Gruber et al . , 2003; Gligoris et al . , 2014 ) . Accordingly , individual SMC complexes might entrap and expand loops of DNA , thereby driving lengthwise condensation of chromosomes with little limitations in the attainable levels of compaction ( Nasmyth , 2001; Alipour and Marko , 2012 ) . Here , we investigate how the prokaryotic SMC–kleisin complex binds to chromosomes in vivo using a novel whole-chromosome assay . We initially attempted to detect topological interactions between B . subtilis Smc–ScpAB and plasmid DNA using pull-down assays as previously described ( Ivanov and Nasmyth , 2005; Ghosh et al . , 2009; Cuylen et al . , 2011 ) . However , several attempts failed to provide clear evidence for entrapment of small circular DNA by prokaryotic condensin . Conceivably , Smc–ScpAB does not interact with these artificial substrates in a physiological manner . To circumvent this possibility , we established an inverse assay by immobilizing whole chromosomes of B . subtilis in agarose plugs and monitoring their association with covalently closed rings of Smc–ScpA under harsh protein denaturing conditions ( Figure 1A ) . To develop the chromosome entrapment assay we first performed experiments with the replicative sliding clamp , DnaN , in B . subtilis , which is known to entrap DNA in a topological manner . Furthermore , most of cellular DnaN protein is maintained in the vicinity of active replication forks in B . subtilis , presumably by its topological association with leading and lagging strand DNA ( Su'etsugu and Errington , 2011 ) . 10 . 7554/eLife . 06659 . 003Figure 1 . Development of the chromosome entrapment assay using DnaN . ( A ) Scheme for the chromosome entrapment assay . Cells are incubated with the cysteine cross-linker BMOE , lysed in agarose plugs and subjected to an electric field in the presence of SDS buffer . Proteins stably bound to chromosomal DNA are re-isolated from nuclease treated agarose plugs , concentrated and analyzed by SDS-PAGE . ( B ) Crystal structure of S . pneumoniae DnaN ( PDB: 3D1F ) in surface representation . The monomers of DnaN are shown in dark and light blue colours , respectively . The positions of an engineered pair of cysteine residues ( N114C and V313C ) at the monomer–monomer interface of B . subtilis DnaN are indicated by arrows . ( C ) Chromosome entrapment by DnaN . Cells of strains BSG1449 ( dnaN-HT ) and BSG1459 ( dnaN ( N114C , V313C ) -HT ) were cross-linked with BMOE and subjected to the chromosome entrapment assay . Input and eluate fractions were analysed by in-gel detection of fluorescently labeled HT fused to DnaN ( top panel ) . Eluate fractions of samples treated with or without nuclease during cell lysis are indicated as nuclease ‘+’ or ‘−’ , respectively . Eluate fractions were further analyzed by silver staining revealing that another protein was consistently co-isolated during the chromosome entrapment assay ( bottom panel ) . This protein—identified as flagellin by mass spectrometry—was retained independently of the integrity of the chromosome . The following figure supplement is available: Figure 1—figure supplement 1: DNA entrapment by DnaN . DOI: http://dx . doi . org/10 . 7554/eLife . 06659 . 00310 . 7554/eLife . 06659 . 004Figure 1—figure supplement 1 . DNA entrapment by DnaN . ( A ) Cross-linking time course of DnaN-HT . Cells of strain BSG1459 were incubated with the cross-linker BMOE and quenched at the indicated times . Extracts were analysed by in-gel fluorescence detection of DnaN-HT . At later time points the double cross-linked DnaN species ( ‘XX-DnaN-HT’ ) is enriched over non-cross-linked and single cross-linked DnaN-HT . ( B ) Cartoon represenation of the DnaN-HT construct based on the crystal structure of S . pneumoniae DnaN ( PDB: 3D1F ) and Rhodococcus Haloalkane dehalogenase ( PDB: 1BN6 ) . The structure of the DnaN dimer ( monomers in green and blue colours , respectively ) was juxtaposed to the structure of the Haloalkane dehalogenase linked by a flexible peptide . The binding pockets for HT ligand binding are indicated by orange arrows . The positions of the engineered pair of cysteine residues ( N114C and V313C ) at the DnaN–DnaN interface are indicated by blue and green arrows . ( C ) Chromosome entrapment assay of strain BSG1459 ( dnaN ( N114C , V313C ) -HT ) in presence or absence of the cross-linker BMOE . ( D ) Effect of benzonase ( ‘nuclease’ ) on protein extracts of strains BSG1001 , BSG1449 and BSG1459 . Cell extracts were analysed by in-gel fluorescence detection ( upper panel ) of DnaN-HT after long-time ( ‘80 min’ ) or short-time ( ‘10 min’ ) incubation with benzonase . Lower panel shows Coomassie staining of the same SDS-PAGE gel to control for equal protein extraction efficiency . DOI: http://dx . doi . org/10 . 7554/eLife . 06659 . 004 Based on the crystal structure of S . pneumoniae DnaN we engineered a pair of cysteine residues ( N114C , V313C ) into the B . subtilis protein so that DnaN can be cross-linked into covalent rings in the presence of a cysteine-specific cross-linker such as BMOE ( Figure 1B ) . For detection a cys-less variant of the HaloTag ( ‘HT’ ) was fused to the C-terminus of DnaN ( Figure 1—figure supplement 1B ) and the construct was integrated into the genome of B . subtilis via allelic replacement at the endogenous locus . The dnaN-ht genes with and without cysteine mutations supported normal growth of B . subtilis , implying that they encoded functional DnaN proteins ( data not shown ) . In vivo cross-linking of DnaN-HT resulted in two additional , slow migrating bands in SDS-PAGE gels ( Figure 1C ) , corresponding to single and double cross-linked species of DnaN dimers , designated as X-DnaN-HT and XX-DnaN-HT , respectively ( Figure 1—figure supplement 1A ) . We next embedded cells in agarose plugs and disrupted their cell walls by lysozyme digestion . Agarose plugs were then subjected to an electric field in the presence of SDS to denature and remove any unattached proteins from chromosomes . Plugs were finally treated with benzonase to digest genomic DNA and to release any stably entrapped protein . DnaN-HT protein was then analysed by in-gel fluorescence . Non-crosslinkable DnaN-HT was efficiently depleted from agarose plugs during the entrapment assay ( Figure 1C ) . In contrast , the double cross-linked , circular form of DnaN ( N114C , V313C ) -HT ( XX-DnaN-HT ) was retained in the agarose plug during electrophoresis with high efficiency ( ∼50% of input ) . A minor fraction of single cross-linked DnaN dimer ( X-DnaN-HT ) was also observed . This is likely generated from XX-DnaN-HT by spontaneous hydrolysis of thiol-malemide adducts during protein isolation ( Kalia and Raines , 2007; Baldwin and Kiick , 2013 ) . Importantly , the presence of benzonase during cell lysis eliminated all DnaN from the plug , indicating that circular DnaN is retained in plugs via its interaction with cellular DNA . Furthermore , in the absence of the cross-linker BMOE , no DnaN-HT was detected in the eluate fraction ( Figure 1—figure supplement 1C ) . The chromosome entrapment assay thus specifically detects a topological association of intact chromosomes with DNA sliding clamps and confirms that a major fraction ( at least 50% ) of DnaN is loaded onto DNA in rapidly growing cells . Next , we used the newly developed chromosome entrapment assay to test for an association between native chromosomes and Smc–ScpAB complexes . Cysteine pairs were introduced at the Smc–Smc and at both Smc–ScpA interfaces and a HT was fused at the C-terminus of Smc to allow in-gel fluorescence detection ( Figure 2A ) ( Bürmann et al . , 2013 ) . Strains bearing the cysteine mutations and the Smc-HaloTag fusion supported normal growth on nutrient rich medium demonstrating the functionality of the modified Smc complex ( Figure 2—figure supplement 1A ) . Cells were treated with BMOE and extracts were analysed by SDS-PAGE . As internal control for the chromosome entrapment assay we employed the DnaN ( N114C , V313C ) protein , whose double cross-linked form was detected in input and eluate samples by immunoblotting ( Figure 2—figure supplement 1B ) . Various species of Smc–ScpAB were identified in extracts of BMOE cross-linked cells by in-gel fluorescence . These correspond to fully cross-linked Smc–ScpA–Smc rings and several intermediate cross-linking species as reported previously ( Figure 2B , C ) ( Bürmann et al . , 2013 ) . To reveal the identity of all species , strains lacking one of six engineered cysteines were used as controls that collectively form several intermediate cross-linked species but no fully cross-linked rings of Smc–ScpAB ( Figure 2—figure supplement 1C ) . In these control samples little or no Smc-HT protein was retained in agarose plugs under denaturing conditions as expected for any non-circular protein ( Figure 2B ) . In the presence of all pairs of cysteine , however , a set of two closely migrating species was consistently detected at significant levels after the chromosome entrapment assay ( ∼10–20% of input material ) ( Figure 2B ) . We argued that the two closely migrating species might correspond to Smc–ScpAB with a single or a double cross-link at the Smc hinge . Consistent with this notion we find that only a single species of Smc2–ScpA accumulated during the chromosome entrapment assay when a single cysteine residue ( R643C ) was used to cross-link the Smc hinge domains ( Figure 2—figure supplement 2A ) . These findings strongly suggest that Smc–ScpAB is bound to chromosomes via entrapment of chromosomal DNA . If this were indeed the case , then its retention in agarose plugs should depend on the integrity of Smc–ScpAB rings and of chromosomal DNA . Incubation of agarose plugs with the nuclease benzonase during cell lysis eliminated the Smc-HT and DnaN signal in the sample ( Figure 2B , Figure 2—figure supplement 1B ) . To disrupt covalent Smc–ScpAB rings , we inserted cleavage sites for TEV protease into the linker region preceding the C-terminal WHD of ScpA and incubated cells during lysozyme treatment with recombinant TEV protease to open any circular Smc2–ScpA species . As expected , little or no Smc-HT signal was detected in agarose plugs after TEV cleavage of ScpA ( Figure 2D ) . To exclude any artefacts due to the presence of the HT on Smc we have repeated the chromosome entrapment assay with an untagged allele of Smc using immunoblotting with anti-Smc antibodies for the detection of cross-linked species , which yielded very similar results ( Figure 2—figure supplement 2C ) . Furthermore , we found that Smc2-ScpA rings are stably trapped in agarose plugs over extended periods of time in constant or alternating electric fields ( data not shown ) . Thus , our chromosome entrapment assay specifically detects the association between intact chromosomal DNA and rings of Smc–ScpAB in B . subtilis , demonstrating that DNA fibers pass through the Smc ring . 10 . 7554/eLife . 06659 . 005Figure 2 . Prokaryotic condensin entraps the chromosome . ( A ) Scheme for the cross-linking of Smc-HaloTag ( ‘HT’ ) and ScpA into a covalent Smc–ScpA–Smc ring . ( B ) Chromosome entrapment of covalent Smc2–ScpA rings . Cells of strains BSG1782 , BSG1809-1813 and BSG1831 were cross-linked and subjected to the chromosome entrapment assay . Cross-linked Smc-HT species were visualized by in-gel fluorescence detection . The presence or absence of cysteine pairs at each of the three ring interfaces are indicated by ‘+’ and ‘−’ , respectively . An aliquot of cells of strains BSG1782 was incubated with benzonase during cell lysis ( nuclease ‘+’ ) . The positions of uncross-linked Smc-HT and fully cross-linked , circular Smc–ScpA–Smc species are indicated by ‘Smc-HT’ and ‘Smc-HT circ . ’; all species are labelled by colour-coded arrowheads ( see panel C for legend ) . Circular species ( ‘h’ ) are labeled by a double pointed arrowhead . ( C ) Schematic depiction of the structure of cross-linked Smc–ScpA species ( ‘a’–‘i’ ) . ( D ) TEV cleavage of ScpA prevents entrapment of Smc–ScpAB in agarose plugs . In-gel fluorescence detection of Smc-HT derived from strains BSG1807 and BSG1832 . The presence or absence of TEV sites in ScpA and of TEV protease during cell lysis is indicated by ‘+’ and ‘−’ , respectively . Cleavage of ScpA ( TEVs ) by TEV protease creates new species of cross-linked Smc-HT ( see ‘input’ samples ) and prevents entrapment of Smc-HT in agarose plugs ( see ‘eluate fraction’ ) ( top panel ) . ‘XX-DnaN’ serves as internal assay control visualized by immunoblotting of cross-linked species of DnaN protein ( bottom panel ) . The following figure supplement is available: Figure 2—figure supplement 1: DNA entrapment by wild-type Smc–ScpAB ( I ) and Figure 2—figure supplement 2: DNA entrapment by wild-type Smc–ScpAB ( II ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06659 . 00510 . 7554/eLife . 06659 . 006Figure 2—figure supplement 1 . DNA entrapment by wild-type Smc–ScpAB ( I ) . ( A ) Colony formation assays of strains BSG1002 , BSG1007 , BSG1782 , 1809–1813 and BSG1831 were performed on minimal medium agar ( SMG ) or nutrient agar ( NA ) . Cultures were diluted 9^2-fold ( top row ) and 9^5-fold ( bottom row ) and grown at 37°C for 24 hr ( SMG ) and 14 hr ( NA ) , respectively . The presence and absence of cysteine pairs for cross-linking of the hinge , cap and neck interface are indicated by ‘+’ and ‘−’ , respectively . ( B ) Immunoblot of SDS-PAGE gels shown in Figure 2B using anti-DnaN antibody . All strains carry cysteine mutations for DnaN cross-linking . Doubly cross-linked DnaN ( ‘XX-DnaN’ ) serves as internal assay control as visualized by immunoblotting . ( C ) Identification of cross-linked species of Smc-HT and ScpA . The gel image is identical to Figure 2A with higher contrast settings to visualize low-intensity species . All cross-linked species are labelled ( with a colour-coded letter ) ( ‘a’–‘i’ ) . Circular species ( ‘h’ ) are labeled by a double pointed arrowhead . ( D ) Same as Figure 2 , panel C . Schematic depiction of the structure of cross-linked Smc–ScpA species ( ‘a’–‘i’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06659 . 00610 . 7554/eLife . 06659 . 007Figure 2—figure supplement 2 . DNA entrapment by wild-type Smc–ScpAB ( II ) . ( A ) Cross-linking of the Smc hinge interface with different cysteine residues . Cross-linking of the Smc hinge using cysteine residue ( ‘R643C’ ) instead of the ‘R558C’ and ‘N643C’ cysteine pair results in different migration pattern of cross-linked Smc-ScpA species ( see input fractions ) . In this case species ‘g’ and ‘i’ have the same migration behavior as the circular form ‘h’ . Only a single species of Smc-ScpA is retained in the agarose plug after the entrapment assay ( see eluate fractions ) . ( B ) Colony formation assay of strains BSG1002 , BSG1007 , BSG1807 and BSG1832 were performed as described before . The smc and scpA loci lacking mutations ( ‘WT’ ) , harbouring cysteine mutations for cross-linking alone ( ‘Cys’ ) or in combination with a TEV cleavage site ( ‘Cys-TEV site’ ) are denoted . ( C ) Chromosome entrapment of Smc–ScpAB lacking fusion tags ( see scheme on the left ) . The right panel shows immunoblotting of input and eluate fractions using an anti-Smc antibody . All strains harbour cysteine residues for crosslinking of Smc-ScpA in an otherwise wild-type background ( ‘WT’ ) in BSG680 , combined with a deletion of the parB gene ( ‘ΔparB’ ) in BSG1991 or with the E1118Q mutation at the smc locus in BSG1995 . The position of fully cross-linked , circular Smc2-ScpA species is indicated by double pointed arrowhead labelled ‘h’ ( see Figure 2—figure supplement 1C ) , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 06659 . 007 Next , we established the requirements for the formation of interconnections between Smc–ScpAB rings and chromosomes . The intrinsic ATPase activity of cohesin has previously been implicated in stable association with chromosomes ( Arumugam et al . , 2003; Weitzer et al . , 2003 ) . More specifically , ATP hydrolysis has been hypothesized to transiently open an entry gate for DNA in the cohesin ring during its loading onto chromosomes ( Gruber et al . , 2006; Hu et al . , 2011 ) . To test what steps of the ATP hydrolysis cycle in Smc–ScpAB are involved in the entrapment of chromosomal DNA , we made use of smc alleles harboring mutations that specifically prevent ATP binding ( K37I ) , engagement of Smc head domains ( S1090R ) or ATP hydrolysis ( E1118Q ) ( Figure 3A ) ( Hirano and Hirano , 2004 ) . The three mutant proteins are expressed at normal levels in B . subtilis being indicative of proper protein folding ( Figure 3—figure supplement 1A ) . However , they do not support growth on nutrient rich medium similar to smc null mutants , implying that all steps of the ATPase cycle are essential for Smc functionality ( Figure 3—figure supplement 1B ) ( Gruber et al . , 2014 ) . For the chromosome entrapment assay , these Smc ATPase mutations were combined with cysteine mutations for BMOE cross-linking . To support their viability , the resulting strains as well as the wild-type controls were grown in minimal medium . The three mutant Smc proteins assembled into normal Smc–ScpAB complexes as judged by Smc–ScpA cross-linking , albeit there is a slight decrease in the fraction of ScpA proteins bridging Smc dimers and a concomitant minor increase in ScpA subunits bound to single Smc proteins ( Figure 3—figure supplement 1C , species ‘e’ and ‘d’ , respectively ) ( Bürmann et al . , 2013 ) . Intriguingly , the ATP binding and engagement mutants abolished the fraction of covalent ring species retained in the agarose plug during the chromosome entrapment assay ( Figure 3B ) . In case of the ATP hydrolysis mutant Smc ( E1118Q ) only minute amounts of cross-linked rings were recovered from SDS treated plugs . This small fraction of stably bound condensin conceivably arises as a consequence of residual levels of ATP hydrolysis activity in Smc ( E1118Q ) ( Hirano and Hirano , 2004 ) . Thus , ATP binding and ATP dependent Smc head engagement—and most probably also ATP hydrolysis—are essential for entrapment of chromosomal DNA by condensin in bacteria , as has been supposed for cohesin in yeast . Furthermore , the strict requirement of several steps of the ATPase cycle strongly suggests that entrapment of DNA corresponds to the physiological form of association with the bacterial chromosome . 10 . 7554/eLife . 06659 . 008Figure 3 . The Smc ATPase is required for loading of DNA into Smc–ScpAB . ( A ) A scheme for the ATP hydrolysis cycle of Smc . Schematic positions for Walker A , Walker B and ABC-signature motifs on the Smc head domain are shown ( top row ) . ATP binding to the Walker A domain is blocked in Smc ( K37I ) ‘ ( 1 ) ’ . ATP-dependent engagement of two Smc heads is abolished in the Smc ( S1090R ) mutant ‘ ( 2 ) ’ . The E1118Q mutation strongly reduces ATP hydrolysis ‘ ( 3 ) ’ . ( B ) Smc ATPase mutations abolish chromosomal loading of Smc–ScpAB . In-gel fluorescence detection of Smc-HT of input and eluate fractions from a representative chromosome entrapment assay performed with strains BSG1782 and BSG1784-6 . Protein extracts ( 10% of input ) were loaded next to samples subjected to the entrapment assay . Selected cross-linked species of Smc-HT are labeled ( top panel ) . Detection of cross-linked species of DnaN by immunoblotting was used as internal assay control ( bottom panel ) . The following figure supplement is available: Figure 3—figure supplement 1: ATPase mutants of Smc–ScpAB . DOI: http://dx . doi . org/10 . 7554/eLife . 06659 . 00810 . 7554/eLife . 06659 . 009Figure 3—figure supplement 1 . ATPase mutants of Smc-ScpAB . ( A ) Upper panel shows Smc expression levels of strains with mutant smc alleles ( BSG1002 , 1007 , 1008 , 1045 , 1046 , 1047 and 1074 ) determined by immunoblotting using anti-Smc antibodies . Lower panel shows Coomassie staining of a SDS-PAGE gel loaded with the same whole cell extract samples to control for equal protein extraction efficiency . ( B ) Overnight cultures of strains ( BSG1002 , 1007 , 1008 , 1045 , 1046 , 1047 ) were spotted on SMG and NA as described before ( see Figure 2—figure supplement 1 ) . Smc locus was without mutation ( ‘WT’ ) , a smc deletion ( ‘Δsmc’ ) or mutations in the Smc ATPase domain . ( C ) Smc ATPase mutations show slightly decreased levels of cross-linking species ‘e’ and increased levels of ‘d’ . Image is identical to Figure 3B , with higher contrast . DOI: http://dx . doi . org/10 . 7554/eLife . 06659 . 009 What other factors might be required for the loading of condensin onto DNA ? The ScpB subunit forms homodimers that bind in an asymmetric manner to the central region of a single ScpA monomer . It thus is in close proximity of the Smc ATPase domains . Together with ScpA it putatively plays a role in the regulation of the Smc ATPase activity ( Kamada et al . , 2013 ) . Its precise molecular function , however , is not clear yet . To test whether ScpB is involved in the association of Smc–ScpA rings with chromosomes we combined the cysteine mutations in Smc and ScpA with an scpB in-frame deletion ( Figure 4—figure supplement 1 ) . Ring formation was only mildly affected by the absence of ScpB as judged by BMOE cross-linking and in-gel fluorescence detection ( Figure 4A ) ( Bürmann et al . , 2013 ) . However , Smc complexes lacking ScpB subunits failed to entrap chromosomes altogether demonstrating that ScpB is absolutely required for loading of prokaryotic condensin onto chromosomal DNA . 10 . 7554/eLife . 06659 . 010Figure 4 . ScpB and ParB are essential for efficient DNA entrapment by Smc complexes . ( A ) Deletion of scpB eliminates loading of chromosomal DNA into Smc complexes . In-gel fluorescence detection of Smc-HT in input and eluate fractions is shown from chromosome entrapment assays performed with strains BSG1782 ( ‘WT’ ) and BSG1850 ( ‘ΔscpB’ ) ( top panel ) . DnaN was used as internal control ( bottom panel ) . ( B ) Several parB mutations interfere with efficient chromosomal loading of Smc–ScpAB . Input and eluate fractions from chromosome entrapment assays with strains BSG1782 , BSG1783 and BSG1960-3 were analysed by in-gel fluorescence detection of Smc-HT ( top panel ) . DnaN was used as internal control ( middle panel ) . Immunoblotting using polyclonal rabbit anti-ParB antiserum confirms near-normal expression of mutant ParB proteins ( bottom panel ) . The following figure supplement is available: Figure 4—figure supplement 1: Growth of smc , parB double mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 06659 . 01010 . 7554/eLife . 06659 . 011Figure 4—figure supplement 1 . Growth of smc ( Cys ) mutants . Overnight cultures of strains ( BSG1002 , BSG1007 , BSG1782 , BSG1850 and BSG1783 ) were serially diluted and spotted as described before ( Figure 2—figure supplement 1 ) . Strains were wild-type ( ‘WT’ ) , harboured the smc deletion ( ‘Δ’ ) or cysteine mutations for cross-linking the Smc-ScpA ring ( ‘Cys’ ) . In addition , scpB ( ‘ΔscpB’ ) or parB ( ‘ΔparB’ ) were deleted where indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 06659 . 011 ParB proteins—bound to parS sites—are crucial for efficient targeting of Smc–ScpAB to a large region of the chromosome near the replication origin ( Gruber and Errington , 2009; Sullivan et al . , 2009; Minnen et al . , 2011 ) . ParB might act by simply increasing the local concentration of Smc–ScpAB around oriC either before or after its loading onto the chromosome . Alternatively , ParB bound to parS sites might be more directly involved in the loading reaction itself , for example , as catalytic factor , and its absence might thus affect levels of chromosomal condensin . To test this , we performed the chromosome entrapment assay with cells lacking the parB gene . Intriguingly , the levels of Smc–ScpAB entrapping chromosomal DNA were strongly reduced in the parB null mutant as judged by the limited retention of Smc–ScpA species in agarose plugs ( Figure 4B ) . Thus , ParB protein likely promotes the entrapment of chromosomal DNA by Smc–ScpAB . This strongly suggests that most condensin is loaded onto the chromosome at parS sites , where ParB protein is bound . In all other parts of the chromosome entrapment of DNA fibers by Smc–ScpAB might be very inefficient . The cysteine bearing smc allele causes growth defects when combined with ΔparB ( Figure 4—figure supplement 1 ) . Therefore , we cannot formally exclude the possibility that the decreased loading of Smc observed in ΔparB are due to the cysteine modifications in Smc and that chromosomal loading of wild-type Smc is not or much less affected by parB deletion . Previously , two parB point mutations ( N112S and R149G ) , which prevent the formation of Smc-GFP foci , have been isolated in B . subtilis ( Gruber and Errington , 2009 ) . We found that these mutations strongly impair loading of Smc onto the chromosome in the entrapment assay similar to ΔparB ( Figure 4B ) . The R149G mutation is positioned on the helix-turn-helix motif of ParB and might thus directly affect binding to parS sites ( Leonard et al . , 2004 ) . The N112S mutation , however , is located in another highly conserved region , which has been implicated in the ‘spreading’ of ParB protein from parS sequences into adjacent DNA ( Leonard et al . , 2004; Graham et al . , 2014 ) . The spreading of ParB along several kb of DNA is a feature conserved in plasmid and chromosome derived ParB proteins , however , the underlying mechanism is only poorly understood ( Rodionov et al . , 1999 ) . It might possibly involve the formation of a large nucleoprotein complex ( Broedersz et al . , 2014 ) . Several other mutants of ParB ( including B . subtilis ParB G77S and R80A ) have been reported to be defective in spreading from parS sites ( Breier and Grossman , 2007; Graham et al . , 2014 ) . Intriguingly , also these mutations resulted in largely reduced levels of Smc on the chromosome in our entrapment assay , being comparable to the levels found in a parB deletion mutant ( Figure 4B ) . This implies that ParB spreading from parS sites or formation of large nucleoprotein complexes might be essential for loading of DNA into the Smc ring by ParB . These findings are consistent with the observation that formation of Smc-GFP foci near the origin of replication are affected by the G77S mutation ( Sullivan et al . , 2009 ) . In summary , these results demonstrate that several factors—including ScpB protein , a ParB/parS nucleoprotein complex and the Smc ATPase cycle—are required to promote efficient loading of condensin rings onto the chromosome . Smc proteins and fragments thereof exhibit affinity for single- and double-stranded DNA in vitro ( Chiu et al . , 2004; Hirano and Hirano , 2006; Soh et al . , 2015 ) . The physical contacts with DNA might occur once condensin has been successfully loaded onto chromosomes and thus be a permanent feature of chromosomal Smc–ScpAB . Alternatively , the direct association with DNA might be restricted to certain intermediates in the chromosomal loading reaction . To test for interactions between Smc–ScpAB and specific chromosomal DNA fragments , we have affinity-purified endogenous Smc–ScpAB from B . subtilis cell lysates using a short Avitag peptide fused to the C-terminus of the Smc protein , which gets biotinylated when the biotin ligase gene birA is co-expressed ( ‘Smc-Avitag’ ) . We then examined fractions for the co-purification of fragments of chromosomal DNA—generated by restriction digest with XbaI—using quantitative PCR with primer pairs specific for different parts of the chromosome . Since we worried that Smc–ScpAB might not be sufficiently stable in diluted cell extracts , we cross-linked the three ring interfaces in Smc–ScpAB using BMOE cross-linking of engineered pairs of cysteines . A small fraction of chromosomal DNA was reproducibly co-purified with wild-type Smc-Avitag , whereas the yield of co-purified DNA was significantly improved by the presence of cross-linkable cysteine residues in Smc–ScpAB ( Figure 5 ) . In both cases origin-proximal regions ( yyaD , parS-359 , dnaA and dnaN ) of the chromosome were more efficiently enriched than distal regions ( amyE , trnS and ter ) by the co-purification with Smc implying that the observed Smc-DNA contacts are dependent on chromosomal loading of Smc–ScpAB by ParB protein at parS sites and are thus physiologically relevant ( Figure 5 , Figure 5—figure supplement 1 ) ( Gruber and Errington , 2009 ) . The association of DNA with wild-type and BMOE cross-linked Smc–ScpAB was highly sensitive to washes with a salt solution ( 2M NaCl ) , suggesting that it was dependent on electrostatic contacts between DNA and protein . These DNA contacts are presumably formed by the Smc–ScpAB complex itself . Alternatively , albeit less likely , other chromosomal proteins physically bound to DNA could prevent the release of condensin from DNA by blocking its sliding towards DNA ends . 10 . 7554/eLife . 06659 . 012Figure 5 . A physical interaction of Smc–ScpAB rings and chromosomal DNA . Co-purification of chromosomal DNA fragments with native Smc–ScpAB . Cells of strains BSG1104-5 and BSG1107-8 were treated with the cross-linker BMOE prior to cell lysis . Strains carrying ( ‘+’ ) or lacking ( ‘−’ ) cysteine mutations ( ‘6xCys’ ) in the Smc-AviTag construct were expressed in presence ( ‘+’ ) or absence ( ‘−’ ) of the biotin ligase ( ‘BirA’ ) . Beads were washed in the presence of either a 150 mM ammonium acetate buffer ( ‘low salt’ ) or a 2 M sodium chloride buffer ( ‘high salt’ ) . The co-purification of DNA fragments with Smc-biotin on streptavidin beads was measured by quantitative PCR using primer pairs specific for genomic positions indicated on a representation of the circular B . subtilis genome . Mean values and standard deviations were calculated from two independent biological replicates . The following figure supplement is available: Figure 5—figure supplement 1: Chromatin immuno-precipitation of Smc . DOI: http://dx . doi . org/10 . 7554/eLife . 06659 . 01210 . 7554/eLife . 06659 . 013Figure 5—figure supplement 1 . Chromatin immuno-precipitation of Smc . All Smc variants used in Figure 5 behave similarly in standard ChIP experiments using anti-Smc antibodies . Cells of strains BSG1104-5 and 1107-8 were grown to mid-exponential phase and subjected to chromatin immunoprecipitation ( ChIP ) using polyclonal rabbit anti-Smc serum . Input and eluate DNA samples were analysed by quantitative PCR using specific primer pairs for the genomic positions indicated on a circular representation of the B . subtilis genome . Pull-down efficiency ( ChIP-DNA/input-DNA *100% ) was plotted for each primer pair . The lack and presence of the BirA biotin ligase or cysteines for cross-linking of Smc and ScpA proteins are denoted as ‘BirA−’ and ‘BirA+’ or ‘6xCys−’ and ‘6xCys+’ , respectively . It is noted that the highly transcribed tRNA locus , trnS , is highly enriched by Smc-Avitag using standard ChIP ( but not native ChIP , see Figure 5 ) , being consistent with the notion that enrichment of highly transcribed genes by ChIP is prone to artefacts due to non-uniform formaldehyde cross-linking . DOI: http://dx . doi . org/10 . 7554/eLife . 06659 . 013 In many cases , it is challenging to measure the activity and outcome of biochemical processes in the living cell . Here , we report the establishment of a straight-forward method to determine the physical association of ring-shaped protein complexes with whole bacterial chromosomes . Two examples , the SMC condensin complex and the sliding clamp DnaN , document the significant potential of our simple entrapment approach . In principle , similar assays should also be possible with eukaryotic cells and for many other chromosomal proteins such as for example hexameric helicases and certain transcription factors . Furthermore , analogous procedures might be useful to address biological questions related to other denaturation-resistant cellular structures such as cell wall polymers ( e . g . , made up of peptidoglycan , chitin or cellulose ) . SMC–kleisin complexes are major governors of chromosome superstructure in most branches of the phylogenetic tree . The eukaryotic variants cohesin and condensin have been suggested to work as concatenases , which hold selected stretches of DNA together by simple embracement in their ring ( Haering et al . , 2008; Cuylen et al . , 2011 ) . Whether DNA entrapment is an ancestral and thus fundamentally conserved function of SMC–kleisin complexes , however , remained elusive so far . Furthermore , interaction studies developed for cohesin and condensin are based on small , artificial DNA substrates and might thus not necessarily reflect the mode of binding to native chromosomes . These assays also fall short of providing an estimate for the fraction of SMC complexes involved in interlocked associations with DNA and thus leave open the possibility that DNA entrapment might be an insignificant side reaction . Finally , it has not been tested under physiological conditions , whether the ATPase cycle is required for proper loading of DNA into any SMC–kleisin complex . To provide answers to these questions , we have established the chromosome entrapment assay to determine the association of prokaryotic condensin with native chromosomes . Our results clearly demonstrate that chromosomal DNA is loaded into condensin complexes in B . subtilis—in a manner that depends on the non-SMC subunit ScpB and at least one full cycle of Smc ATPase activity . The chromosome entrapment assay recovers about 10–20% of the fully cross-linked input material . This number probably understates the real proportion of chromosomally entrapped Smc complexes due to the loss of material during protein re-isolation from agarose plugs , due to possible adverse effects of cysteine mutations on Smc–ScpAB loading and because cysteine-maleimide linkages are vulnerable to hydrolytic reversal during and after the entrapment assay ( Kalia and Raines , 2007; Baldwin and Kiick , 2013 ) . Interestingly , a recent single-molecule tracking study in B . subtilis revealed two major populations of Smc: 80% of Smc-YFP proteins are displaying highly dynamic behavior on the nucleoid , whereas the other 20% ( and most ScpA-YFP protein ) are immobile and constrained within a small volume of the cell ( Kleine Borgmann et al . , 2013 ) . This immobile fraction possibly represents Smc–ScpAB complexes embracing origin proximal DNA after loading at parS sites as observed in the entrapment assay . Our results show that embracement of chromosomal DNA is a predominant feature of Smc–ScpAB , which has been evolutionarily retained in cohesin and likely all other SMC–kleisin complexes as well . Since the chromosome entrapment assay is based on the immobilization of intact replicating chromosomes , which possibly represent internally knotted and branched DNA structures , it is conceivable that Smc–ScpAB rings are linked to chromosomal DNA by non-topological capture of DNA loops , which themselves might be interlinked ( i . e . , knotted ) with other parts of the chromosome . Therefore , it remains to be determined whether DNA entrapment by Smc–ScpAB is of topological ( Figure 6A ) and/or non-topological ( Figure 6B ) nature . 10 . 7554/eLife . 06659 . 014Figure 6 . Models for entrapment of chromosomal DNA by Smc–ScpAB . ( A ) Loop capture model . DNA loops might be pre-formed within ParB/parS nucleoprotein assemblies . Driven by ATP dependent engagement of Smc head domains Smc–ScpAB adopts a ring-like configuration . Occasional opening of the Smc hinge then allows capture of ParB-DNA loops within Smc–ScpAB . Subsequent ATP hydrolysis by Smc locks the hinge in a closed state and stabilizes the structure . ( B ) Loop formation model . ParB/parS might serve as a landing platform for Smc–ScpAB allowing Smc–ScpAB in its ring-like conformation to guide DNA into its central cavity . Continuous extrusion of DNA through Smc–ScpAB then drives lengthwise condensation of chromosomes . Ring opening is not required in this model and DNA entrapment by Smc–ScpAB is thus strictly non-topological here . The following figure supplement is available: Figure 6—figure supplement 1: Quantitative Blotting of Smc protein and parB DNA . DOI: http://dx . doi . org/10 . 7554/eLife . 06659 . 01410 . 7554/eLife . 06659 . 015Figure 6—figure supplement 1 . Quantitative Blotting of Smc protein and parB DNA . To estimate Smc protein and parS-359 DNA abundance in B . subtilis cells , levels of Smc and parS-359 in extracts of wild-type cells were compared to defined amounts of purified Smc protein and parS-359 DNA spiked into equivalent extracts from cell harboring a tagged version of Smc and a modified parS-359 locus ( see panel C ) . The ratio of Smc protein ( see panel B ) to parS-359 DNA ( see panel A ) was calculated as average from three independent experiments ( biological replicates ) to be 61 ( ±12 SEM , standard error of mean ) Smc monomers per parS-359 DNA . Due to potential loss of chromosomal DNA during genomic DNA preparation this might be a slight over-estimation of the real number . ( A ) Quantitative Southern blotting to estimate the number of molecules ( in fmol ) of the parS359 locus in a given number of cells . Genomic DNA from 52× cell culture equivalents ( 1 equivalent equals 0 . 025 ml*OD600 ) of wild type ( ‘WT’ , BSG1001 ) has been digested with the PstI restriction enzyme ( see panel C ) and loaded into the left lane . DNA loaded into the second and third lane is derived from mixtures of wild-type and tagged ( ‘parB-GFP’ , BSG2058 ) cells . Defined amounts of a purified PCR product covering the wild-type parS-359 locus have been spiked into cell extracts of the tagged strain . Digested DNA was detected using a Southern probe specific for the parS-359 locus . ( B ) Quantitative Western blot to estimate the number of molecules of the Smc protein ( in fmol ) in a given number of cells . Cultures were identical to the ones used in ( A ) . Whole cell protein extract corresponding to 4 culture equivalents of wild-type cells ( ‘WT’ , BSG1001 ) was loaded into the left lane . Other lanes were loaded with samples derived from wild-type and tagged cells or mixtures thereof with or without spiked-in purified Smc protein as indicated in the figure panel . Smc protein was detected by immunoblotting using a polyclonal rabbit anti-Smc antibody . ( C ) Protein and DNA spike-in was performed in extracts derived from cells ( BSG2058 ) with modifications at the parS-359 locus ( blue filled circle ) and the smc gene ( purple filled circle ) . PstI restriction sites ( in red ) in the parB gene ( light blue box ) and the neighbouring yyaC gene ( grey box ) generate a fragment of 1270 bp in size in case of the wild-type locus . From parB-gfp cells ( BSG2058 ) , however , a bigger DNA fragment of 3147 bp in size is generated . Strain BSG2058 also harboured a modified smc gene , which is fused to a halotag cassette ( pink box ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06659 . 015 Smc–ScpAB plays a crucial role in the segregation of replication origins in B . subtilis cells ( Gruber , 2014; Wang et al . , 2014 ) , presumably by organizing nascent sister chromosomes so that their spatial overlap and entanglement is minimized . It is tempting to speculate that ParB/parS not only enriches Smc in the vicinity of the replication origin but also sets up lengthwise compaction of chromosomes by presenting DNA of a certain topology to the Smc–ScpAB complex . Consistent with this notion , we found that several parB mutants , which are defective in the ability to form large nucleoprotein complexes and to spread from parS sites , fail to promote loading of Smc–ScpAB onto the chromosome . Thus , Smc–ScpAB might capture , stabilize and expand structures–such as DNA loops or coils–that are pre-formed within larger ParB/parS nucleoprotein assemblies ( Figure 6A ) . Alternatively , ParB/parS might serve as an elaborate landing platform on the chromosome , where Smc–ScpAB initiates with the help of ParB the lengthwise compaction of chromosomes by forming and extruding loops of DNA ( Figure 6B ) . Extrusion of DNA might involve the translocation of Smc–ScpAB along DNA fibers made up of naked or ParB coated DNA . These models are not mutually exclusive . A reasonable first step towards understanding the architecture of Smc/ParB/parS assemblies might be the investigation of physical interactions between ParB/parS and Smc–ScpAB and their functional interconnection with the Smc ATPase cycle . In the future , Smc/ParB/parS structures could serve as a relatively simple paradigm for chromosome organization by the more intricately regulated cohesin and condensin complexes in eukaryotes . ParB serves a supplemental—albeit important—role in the Smc loading process . In contrast , Smc functionality and its loading onto DNA in vivo is critically dependent on the ScpB subunit and the ATPase cycle . It remains unclear though what the exact role of ScpB and ATP hydrolysis in the entrapment of DNA might be . If loading indeed depends on transient opening of a DNA entry gate , the open state would likely represent an energetically unfavorable reaction intermediate ( Figure 6A ) . Timely opening would require energy input as well as tight regulation . We have recently demonstrated that the ATP dependent engagement of Smc head domains—together with DNA binding to the Smc hinge domain—can transform the configuration of the Smc coiled coil from a rod to a more open ring-like conformation ( Soh et al . , 2015 ) . Hydrolysis of ATP and/or ScpB binding could drive a subsequent conformational change that might open the SMC–kleisin ring ( Figure 6A ) . Alternatively , ScpB and/or ATP hydrolysis might stabilize Smc–ScpAB once loaded onto DNA or promote Smc's sliding along DNA to allow efficient extrusion of DNA loops by the Smc–ScpAB ring ( Figure 6B ) . Deletion or mutation of the parB gene results in a clear drop in the levels of chromosomally bound condensin ( ∼5–10 fold less ) in our chromosome entrapment assay ( Figure 4B , Figure 2—figure supplement 2C ) . In addition to this loading defect , also the specific recruitment of Smc–ScpAB towards the replication origin is lost in the absence of ParB ( Gruber and Errington , 2009; Sullivan et al . , 2009 ) . Thus , in parB mutants only a very small proportion of cellular Smc–ScpAB is bound to chromosomes within the replication origin region , where it presumably performs its essential function by promoting the separation of nascent sister chromosomes ( Gruber et al . , 2014; Wang et al . , 2014 ) . Nevertheless , defects in chromosome segregation are rather mild in parB mutants when compared to mutants of smc . Using quantitative blotting of Smc protein and replication origin DNA from cell extracts , we have estimated the average number of Smc protein to be around 30 dimers per replication origin in a fast growing population of cells ( Figure 6—figure supplement 1 ) . Assuming that all Smc complexes entrap chromosomal DNA in wild-type cells , only three to six Smc dimers ( 10–20% of total ) are loaded onto the chromosome in parB mutants according to our measurements . Thus , a handful of Smc–ScpAB complexes , which are presumably randomly distributed over the chromosome , appears to be capable of supporting near-normal chromosome segregation under these conditions—when chromosome segregation is already compromised by the loss of the parABS system . Few Smc–ScpAB therefore seem to be able to provide enough organization to the replication origin region and the remainder of the chromosome to prevent lethal accumulation of inter-linked sister chromosomes . It is conceivable that individual Smc–ScpAB complexes are able to organize large chunks of a bacterial chromosome , possible by forming giant loops of DNA . Alternatively , Smc activity might be needed only at a limited number of defined locations on the B . subtilis chromosome and/or for very short periods of time . However , when levels of functional Smc dimers are in addition reduced for example by hypomorphic mutations in the smc gene itself , the loss of ParB protein becomes lethal ( Gruber and Errington , 2009 ) . This work reveals the mode of association of Smc–ScpAB with bacterial chromosomes , highlights its striking evolutionary conservation and demonstrates the involvement of the SMC ATPase cycle in chromosomal loading . Future work must address the underlying biochemical mechanisms to get basic insight into the architectural role of SMC in chromosome biology . Genetic modifications at smc , scpAB , parB and dnaN loci were generated via double cross-over recombination in strains derived from B . subtilis 1A700 or B . subtilis 168ED . Genotypes of strains used in this study are listed in Supplementary file 1 . Cells were transformed with plasmids or B . subtilis genomic DNA using a 2-step starvation protocol as previously described ( Hamoen et al . , 2002; Bürmann et al . , 2013 ) . Transformants were selected by growth on nutrient agar ( NA ) plates ( Oxoid , UK ) supplemented with antibiotics as required: 5 µg ml−1 kanamycin , 80 µg ml−1 spectinomycin , 10 µg ml−1 tetracycline , 5 µg ml−1 chloramphenicol , 1 µg ml−1 erythromycin and 25 µg ml−1 lincomycin . Strains displaying a condensin null phenotype were selected on SMG medium instead: SMM salt solution ( 2 g l−1 ammonium sulphate , 14 g l−1 dipotassium hydrogen phosphate , 6 g l−1 potassium dihydrogen phosphate , 1 g l−1 trisodium citrate , 0 . 2 g l−1 magnesium sulphate , 6 g l−1 potassium hydrogen phosphate ) supplemented with 5 g l−1 glucose , 20 mg l−1 tryptophan and 1 g l−1 glutamate with the respective antibiotics . Strains were single-colony purified and grown in the absence of antibiotics for experiments . Cells were pre-grown in a 96-well plate in SMG medium for 24 hr at 37°C . Overnight cultures were diluted 9^2-fold ( high density spots ) or 9^5-fold ( low density spots ) and spotted onto NA or SMG agar plates . Plates were incubated at 37°C for 12 hr on NA or 24 hr on SMG agar . Cells were grown in either LB Miller medium ( 10 g l−1 tryptone , 5 g l−1 yeast extract , 10 g l−1 sodium chloride ) or SMG medium to mid-exponential phase at 37°C ( in LB Miller medium , OD600 of 0 . 4; in SMG medium , OD600 of 0 . 03 ) . Cells were harvested by centrifugation or vacuum filtration and washed in ice-cold PBS supplemented with 0 . 1% glycerol ( ‘PBSG’ ) . Cell aliquots ( corresponding to 1 ml at an OD600 of 1 . 25 ) were re-suspended in PBSG and incubated with the cross-linker BMOE ( bis-maleimidoethane , Applichem , Germany ) at a concentration of 1 mM ( diluted from a 20 mM stock solution in DMSO ) . After a 10 min incubation on ice the reaction was quenched by addition of 2-mercaptoethanol ( ‘2-ME’ ) to a final concentration of 28 mM . For the preparation of protein extracts ( ‘input’ ) a mixture of following components was added to an aliquot of cells: 400 units ready-lyse lysozyme ( Epicentre , Madison , WI ) , 12 . 5 units benzonase ( Sigma-Aldrich , St . Louis , MO ) and a protease-inhibitor cocktail ( Sigma ) . In addition , 1 µM HT Oregon Green substrate ( Promega , Madison , WI ) was added to cell suspensions with HaloTag bearing alleles . Samples were then incubated for 20 min at 37°C protected from light . Finally , the samples were heated to 70°C for 5 min in LDS Sample Buffer ( NuPage , Thermo Scientific , Waltham , MA ) containing 200 mM DTT and loaded onto a SDS-PAGE gel ( see below ) . Gels with Oregon Green labeled samples were scanned on a Typhoon scanner ( GE Healthcare , UK ) with Cy3-DIGE filter setup . Cells were grown , cross-linked and quenched as described above . Lysozyme stock solution , protease inhibitor and HT substrate were added to an aliquot of cells at concentrations given above . The cell suspension was mixed immediately in a 1:1 ratio with a 2% solution of Megabase agarose ( BioRad , Hercules , CA ) or low-melt agarose ( BioRad ) and casted into 100 µl agarose plugs using plug molds ( BioRad ) . Agarose plugs were incubated for 20 min at 37°C , protected from light , and then loaded into the wells of a 6% SDS-PAGE Tris-glycine gel . The polyacrylamide mini-gel was run for 60 min at 25 mA protected from light . Agarose plugs were then re-extracted from the gel and transferred into 1 . 5 ml Eppendorf tubes . 1 ml of Wash Buffer ( ‘WB’: 0 . 01 mM EDTA , 0 . 5 mM Tris , 0 . 5 mM MgCl2 , 0 . 01% SDS ) was added per agarose plug . Plugs were incubated for 10 min with gentle agitation protected from light . This step was repeated once . Wash buffer was then discarded and replaced by 100 µl fresh WB supplemented with 50 units of benzonase ( Sigma ) . Plugs were incubated at 37°C for 30 min . Plugs were melted at 85°C for 2 min under vigorous agitation . The samples were frozen at −80°C and stored overnight . Samples were then thawed , centrifuged for 10 min at 4°C and 14 , 000×g and transferred to a 0 . 45 µm Cellulose acetate spin column ( Costar , Tewksbury , MA ) and spun for 1 min at 10 , 000×g . The flow-through was concentrated in a Speed Vac ( Thermo Scientific , no heating , 2 . 5 hr running time ) . The concentrated sample was resuspended in LDS Sample Buffer ( NuPage ) containing 200 mM DTT and heated for 3 min at 70°C . Samples were loaded onto Tris-acetate gels ( 3–8% Novex , Thermo Scientific ) and run for 2 . 5 hr at 35 mA per gel at 4°C . For DnaN detection Bis-Tris gels ( 8–12% Novex ) were run for 1 hr at 200 V at room temperature . Gels were either scanned on a Typhoon scanner ( FLA 9000 , GE Healthcare ) with Cy3-DIGE filter setup or immuno-blotted using antibodies against DnaN or Smc ( see below ) . For cleavage of ScpA ( TEVs ) or degradation of chromosomal DNA 15 units of His-TEV protease or 12 . 5 units of benzonase , respectively , was added before casting agarose plugs . B . subtilis strains containing smc-tev-avitag alleles were grown to OD600 of 0 . 4 in 100 ml LB Miller at 37°C . Part of the culture was fixed with formaldehyde and subjected to chromatin immunoprecipitation ( ChIP ) as described by ( Gruber and Errington , 2009 ) using a rabbit anti-Smc antiserum . In parallel , 10 ml of the culture were mixed with ice , harvested by centrifugation and washed in cold PBSG . Cells were resuspended in 200 µl PBSG and treated with 0 . 5 mM BMOE for 10 min on ice . The reaction was quenched with 14 mM 2-ME and cells were washed once in CutSmart buffer ( New England Biolabs , Ipswich , MA ) . Cells were resuspended in 200 µl CutSmart containing 10 kU Ready-Lyse lysozyme ( Epicentre ) , 40 U XbaI ( New England Biolabs ) and a protease inhibitor cocktail ( Sigma ) . The suspension was incubated for 15 min at 37°C before addition of 1800 µl buffer LS ( 10 mM Tris/HCl , 150 mM NH4OAc , 1 mM EDTA , 6 mM 2-ME , 0 . 05% Tween-20 , 0 . 01% NaN3 , final pH 7 . 9 at 23°C ) . Lysates were centrifuged for 5 min at 20 , 000×g . Subsequently , 1400 µl of the extract were incubated with 100 µl Dynabeads Streptavidin C1 for 30 min at room temperature . Beads were washed once in buffer LS , then split , resuspended either in buffer LS or in buffer HS ( 10 mM Tris/HCl , 2 M NaCl , 1 mM EDTA , 6 mM 2-ME , 0 . 05% Tween-20 , 0 . 01% NaN3 , final pH 7 . 9 at 23°C ) and incubated for 15 min at room temperature . Beads were washed twice with buffer LS , and protein/DNA complexes were eluted for 1 hr at 22°C by incubation with 350 µl LS containing TEV protease and 1 mM DTT . DNA from input and eluate fractions was purified by treatment with 0 . 5 mg/ml Proteinase K for 1 hr at 55°C followed by phenol/chloroform extraction . Samples were analysed by quantitative PCR using the second derivative maximum of a four parameter logistic model similar to the method described by ( Zhao and Fernald , 2005 ) . After gels were scanned for in-gel fluorescence detection , they were immediately transferred onto a PVDF membrane ( Immobilon-P , Merck Millipore , Germany ) using semi-dry transfer . Membranes were blocked with 3 . 5% ( wt/vol ) milk powder in PBS with 0 . 1% Tween 20 . Rabbit polyclonal sera against B . subtilis DnaN ( Lenhart et al . , 2013 ) , B . subtilis Smc ( this paper ) and B . subtilis ParB ( this paper ) were used as primary antibodies for immunoblotting at dilutions of 1:5000 each . The membrane was developed with HRP-coupled secondary antibodies and chemiluminescence ( SuperSignal West Femto , Thermo Scientific ) and visualized on a LAS-3000 scanner ( FujiFilm , Germany ) . To estimate DnaN cross-linking kinetics ( Figure 1—figure supplement 1A ) samples were grown as described above . An aliquot of cells was incubated with the cross-linker BMOE ( 1 mM ) for the indicated length of time before the reaction was quenched with 2-ME ( 28 mM ) .

The genome of any living organism holds all the genetic information that the organism needs to live and grow . This information is written in the sequence of the organism's DNA , and is often divided into sub-structures called chromosomes . Different species have different sized genomes , but even bacteria with some of the smallest genomes still contain DNA molecules that are thousand times longer than the length of their cells . DNA molecules must thus be highly compacted in order to fit inside the cells . DNA compaction is particularly important during cell division , when the DNA is being equally distributed to the newly formed cells . In plants , animals and all other eukaryotes , large protein complexes known as condensin and cohesin play a major role in compacting , and then separating , the cell's chromosomes . Many bacteria also have condensin-like complexes . At the core of all these complexes are pairs of so-called SMC proteins . However , it is not clear how these SMC proteins direct chromosomes to become highly compacted when cells are dividing . Wilhelm et al . have now developed two new approaches to investigate how SMC proteins associate with bacterial DNA . These approaches were then used to study how SMC proteins coordinate the compaction of chromosomes in a bacterium called Bacillus subtilis . The experiments revealed that SMC proteins are in direct physical contact with the bacterial chromosome , and that bacterial DNA fibers are physically captured within a ring structure formed by the SMC proteins . Wilhelm et al . suggest that these new findings , and recent technological advances , have now set the stage for future studies to gain mechanistic insight into these protein complexes that organize and segregate chromosomes .

2015

SMC condensin entraps chromosomal DNA by an ATP hydrolysis dependent loading mechanism in Bacillus subtilis

Differentiation of cellular lineages is facilitated by asymmetric segregation of fate determinants between dividing cells . In budding yeast , various aging factors segregate to the aging ( mother ) -lineage , with poorly understood consequences . In this study , we show that yeast mother cells form a protein aggregate during early replicative aging that is maintained as a single , asymmetrically inherited deposit over the remaining lifespan . Surprisingly , deposit formation was not associated with stress or general decline in proteostasis . Rather , the deposit-containing cells displayed enhanced degradation of cytosolic proteasome substrates and unimpaired clearance of stress-induced protein aggregates . Deposit formation was dependent on Hsp42 , which collected non-random client proteins of the Hsp104/Hsp70-refolding machinery , including the prion Sup35 . Importantly , loss of Hsp42 resulted in symmetric inheritance of its constituents and prolonged the lifespan of the mother cell . Together , these data suggest that protein aggregation is an early aging-associated differentiation event in yeast , having a two-faceted role in organismal fitness . Aging results in an increasing decline of the organism's fitness over time ( Lopez-Otin et al . , 2013 ) . Remarkably , this process segregates asymmetrically during budding yeast division: the mother cell forms an aging lineage , whereas the daughters generated by these mothers rejuvenate to form eternal lineages , similar to the segregation of soma and germ lineages in metazoans . Such lineage separation requires that the inheritance of factors that promote aging , such as defective/deleterious organelles , proteins , and DNA , is asymmetric during cell division ( Sinclair and Guarente , 1997; Aguilaniu et al . , 2003; Erjavec et al . , 2007; Henderson and Gottschling , 2008; Shcheprova et al . , 2008; Liu et al . , 2010; Zhou et al . , 2011; Hughes and Gottschling , 2012; Higuchi et al . , 2013; Clay et al . , 2014; Denoth Lippuner et al . , 2014; Henderson et al . , 2014; Higuchi-Sanabria et al . , 2014; Thayer et al . , 2014; Katajisto et al . , 2015 ) . Therefore , how cells are able to recognize , sort , and coordinate the asymmetric segregation of aging factors and other fate determinants is an outstanding question in biology ( Neumuller and Knoblich , 2009 ) . Protein aggregates and/or damaged proteins are a hallmark in the etiology of many human disorders associated with aging ( Hartl et al . , 2011; Wolff et al . , 2014 ) , and their presence correlates with aging of mitotically active yeast and drosophila stem cells ( Aguilaniu et al . , 2003; Erjavec et al . , 2007; Bufalino et al . , 2013; Coelho et al . , 2013 ) . Studies on budding yeast have shown a correlation between the accumulation of protein aggregates and replicative aging by demonstrating that Hsp104-mediated protein disaggregation is required for full replicative life span ( Erjavec et al . , 2007 ) , and that over-expression of Mca1 , which counteracts the formation of stress- and age-associated protein aggregates ( Lee et al . , 2010; Hill et al . , 2014 ) , extends the life span of yeast mother cells ( Hill et al . , 2014 ) . How cells respond to protein aggregation that occurs specifically during aging has remained elusive since most studies investigating the cellular responses to protein aggregation have relied on over-expression of non-native , aggregation prone proteins , proteostasis inhibitors , or other stressors , such as heat ( Kaganovich et al . , 2008; Liu et al . , 2010; Specht et al . , 2011; Zhou et al . , 2011; Malinovska et al . , 2012; Spokoini et al . , 2012; Winkler et al . , 2012; Escusa-Toret et al . , 2013; Zhou et al . , 2014 ) . These studies have uncovered specific modes of cytosolic compartmentalization that take place when cells encounter proteotoxic stress . For example , cells stressed with heat respond by forming multiple protein aggregates ( referred to as peripheral aggregates , stress foci , Q-bodies , or CytoQ ) at the surface of the ER ( Specht et al . , 2011; Spokoini et al . , 2012; Escusa-Toret et al . , 2013; Miller et al . , 2015; Zhou et al . , 2014; Wallace et al . , 2015 ) . These structures , hereafter referred as Q-bodies , contain acutely misfolded proteins that are sorted between the nuclear and cytoplasmic degradation/deposit sites by the Hook family proteins Btn2 and Cur1 ( Malinovska et al . , 2012 ) , and coalesce together by the aid of small heat shock proteins; Hsp42 in budding yeast ( Specht et al . , 2011; Escusa-Toret et al . , 2013 ) , and Hsp16 in fission yeast ( Coelho et al . , 2014 ) . Simultaneously , Q-bodies are being rapidly resolved by the protein disaggregase Hsp104 ( Parsell et al . , 1994; Specht et al . , 2011; Zhou et al . , 2011; Spokoini et al . , 2012; Escusa-Toret et al . , 2013 ) , together with other heat shock responsive chaperones such as Hsp70 and Hsp82 ( Escusa-Toret et al . , 2013 ) . The formation of Q-bodies seems to aid stress survival , as the deletion of HSP42 resulted in defective tolerance of prolonged heat stress ( Escusa-Toret et al . , 2013 ) . The asymmetric inheritance of Q-bodies by the mother cells is promoted by the geometry of the bud neck ( Zhou et al . , 2011 ) , tethering to mitochondria ( Zhou et al . , 2014 ) , and by actin cable-mediated retrograde transport , which is dependent of Hsp104 and Sir2 ( Liu et al . , 2010; Song et al . , 2014 ) . Notably , Sir2 is also a key player in processes that underlie the asymmetric segregation of damaged mitochondria ( Higuchi et al . , 2013 ) and the accumulation of extrachromosomal DNA circles ( Sinclair and Guarente , 1997; Kaeberlein et al . , 1999 ) to the aging mother cell . Prolonged Q-body-inducing stress ( heat or over-expression of thermolabile proteins ) combined with proteasome inhibition can lead to the formation of a dynamically exchanging deposit of ubiquitylated proteins named the juxtanuclear quality compartment , JUNQ ( Kaganovich et al . , 2008; Escusa-Toret et al . , 2013 ) . This structure is regulated by the Upb3 deubiqutinase ( Oling et al . , 2014 ) , by proteosomal activity ( Andersson et al . , 2013 ) and by lipid droplets ( Moldavski et al . , 2015 ) , and it was also shown to appear during replicative aging ( Oling et al . , 2014 ) . The faithful inheritance of this structure by the mother cell is dependent on its association with the nucleus ( Spokoini et al . , 2012 ) . More recently , it was shown that the ‘JUNQ’ might actually reside inside the nucleus , and it was thus renamed as intranuclear quality control compartment , INQ ( Miller et al . , 2015 ) . The JUNQ/INQ assembly is dependent on Btn2-aggregase ( Miller et al . , 2015 ) , a protein also found to be involved in prion curing ( Kryndushkin et al . , 2008 , 2012 ) . Apart from the JUNQ/INQ structure , terminally aggregating proteins , such as the amyloidogenic prions Rnq1 and Ure2 , were shown to partition to an non-dynamic , vacuole-associated deposit called the insoluble protein deposit IPOD ( Kaganovich et al . , 2008; Tyedmers et al . , 2010b ) , which has remained less well characterized . Despite this wealth of data , it remains unclear how these exogenous/stress-induced aggregation models relate to protein aggregation that takes place during physiological ‘healthy’ aging . Particularly , it is unclear why/how protein aggregates arise during aging , how are they segregated during cell division and , importantly , what is their consequence to the protein quality control of the aging cell , as well as to the aging process itself . To illuminate these aspects , we probed the role of protein aggregation during unperturbed replicative aging . Our findings indicate that protein aggregation is a prevalent and highly coordinated event of early aging and is not solely associated with proteostasis deterioration . Instead , we provide evidence that age-associated protein aggregation may initially benefit the cytosolic protein quality control , but eventually becomes involved with age-associated loss of fitness . To address the role of protein aggregation in unperturbed , physiological aging , we analyzed microscopically the replicative age-associated protein aggregation landscape in budding yeast by visualizing different chaperone proteins that mark aberrantly folded and aggregated proteins . By employing the Mother Enrichment Program ( MEP ) ( Lindstrom and Gottschling , 2009 ) ( Figure 1—figure supplement 1A ) , we harvested cells of different age and first analyzed the localization of endogenous GFP-tagged protein-disaggregase Hsp104 ( Parsell et al . , 1994; Glover and Lindquist , 1998 ) , a broad sensor for protein aggregates ( Figure 1A , Haslberger et al . , 2010 ) . Interestingly , we found many cells displaying an aggregate ( typically a single bright Hsp104-labeled focus ) and this portion increased in a progressive , age-dependent manner such that >80% of cells that had undergone more than 6 divisions displayed such a structure ( Figure 1A , B ) , as previously reported ( Aguilaniu et al . , 2003; Erjavec et al . , 2007 ) . Co-localization analysis with Hsp104 demonstrated that the Hsp70 proteins Ssa1 and Ssa2 , the small heat shock protein Hsp42 , and the Hsp40 protein Ydj1 readily localized to these aggregates , the Hsp26 was found to be enriched in only 15% of Hsp104-labeled foci , while no accumulation of Hsp40 protein Sis1 or the Hsp90 protein Hsp82 was detected ( Figure 1C , Figure 1—figure supplement 2A–C ) . Importantly , age-dependent appearance of these structures was also detected in diploid cells , in other strain backgrounds ( W303 ) , independently of the MEP procedure , and when different fluorophores where used for tagging Hsp104 ( Figure 1—figure supplement 2D–H ) , indicating that their formation represents a general , age-dependent phenomenon of budding yeast cells . 10 . 7554/eLife . 06197 . 003Figure 1 . Replicative aging leads to the formation of age-associated protein deposit . ( A ) Representative images of cells expressing endogenous Hsp104 tagged with GFP . The C-terminal tagging does not hamper Hsp104 disaggregation activity ( Specht et al . , 2011 ) . Cells of different age were harvested using the Mother Enrichment Program ( MEP ) ( Lindstrom and Gottschling , 2009 ) and stained with calcofluor . ( B ) Percentage of cells of different age groups containing at least one Hsp104-focus , ( N = 135 to 472 cells per age group ) . ( C ) Fraction of Hsp104-mCherry foci that are enriched with the indicated chaperones ( N = 16 to 92 Hsp104-focus containing cells per strain , n . d . = not detected ) . ( D ) Representative frames of a movie of dividing cells expressing Hsp104-mCherry ( black ) . Red arrowhead indicates an aggregate that is retained in the mother cell ( M ) . This cell divided five times giving rise to four daughter cells that start to form aggregates after dividing 2–3 times ( blue arrows at 390 min ) . ( E ) Integrated density at newly forming age-associated protein deposits ( deposit/cytoplasm ) over time , ( N = 26 ) . ( F ) Fraction of divisions during which the age-associated protein deposit is asymmetrically inherited by the mother cell as a function of the age of the mother cell ( N = 66 to 306 divisions per age group from 2-3 independent experiments . The approximate mother cell age was estimated from separate bud scar analysis ) . ( G ) Representative micrographs of a dividing mother cell expressing Hsp104-GFP followed for 66 hr in a microfluidic chip ( Lee et al . , 2012 ) . ( H ) Proportion of cells having the indicated number of Hsp104 foci . Hsp104 foci containing cells that could be followed >10 consecutive divisions were quantified for the number of Hsp104-foci/cell at each time point , ( N = 44 ) . ( I ) Total time spent with and without an Hsp104-focus for cells undergoing >10 consecutive divisions , starting with a focus , ( N = 44 ) . ( J ) Fluorescent recovery after photobleaching ( FRAP ) analysis of Hsp104-mCherry turnover at age-associated protein deposit . The mCherry signal at the age-associated protein deposit was photobleached in HSP104-GFP/HSP104-mCherry diploid cells and the kinetics of recovery were monitored , using the GFP signal to localize the age-associated protein deposit over time . ( K ) Fitting of nine recovery curves showed that on average a large ( 59% ) fraction was immobile and the half-time recovery for the mobile fraction was 8 . 9 s , ( N = 9 ) . Scale bars 5 μm . Graphs display mean ± SEM , *p < 0 . 05 , ***p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 06197 . 00310 . 7554/eLife . 06197 . 004Figure 1—figure supplement 1 . Schematic representation of the strategies used here to study aged cells . ( A ) Strategy to obtain old cells: cells were biotinylated and grown for ( x ) amount of time in YPD media containing estradiol , which activates the MEP ( Lindstrom and Gottschling , 2009 ) and results in daughter specific cell cycle arrest due to excision of two essential genes UBC9 and CDC20 . After desired time , aged mother cells containing the cell wall-linked biotin coat were coupled to streptavidin microbeads , purified with affinity chromatography , and imaged under the microscope . ( B ) The microfluidic system ( Lee et al . , 2012 ) allows microscopic examination of trapped mother cells as they undergo repetitive divisions . Cells are captured between a PDMS pad and a glass cover slip and subjected to constant media flow , which washes away the newly born daughter cells that are not captured between the pad due to their smaller size . DOI: http://dx . doi . org/10 . 7554/eLife . 06197 . 00410 . 7554/eLife . 06197 . 005Figure 1—figure supplement 2 . Age-associated protein deposit formation is a general age-dependent phenomenon marked by a subset of chaperones . ( A–C ) Representative images of Hsp104-mCherry expressing cells co-tagged with ( A ) Ssa1-GFP ( B ) Hsp42-GFP , or ( C ) Hsp82 . Age-associated protein deposit are depicted with a red arrowhead . ( D ) W303 cells expressing Hsp104-YFP . ( E ) S288C cells expressing Hsp104-mCherry or ( F ) Hsp104-GFP . ( G ) Diploid S288C cells expressing Hsp104-GFP . ( H ) Comparison of the relative amount of age-associated protein deposit containing cells in calcofluor treated MEP and non-MEP background cells vs non-calcofluor treated MEP strain background , ( N = 177–243 cells analyzed per genotype ) . Scale bars: ( A–C ) 2 μm , ( D–G ) 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06197 . 005 To characterize the nature of these protein deposits further , we performed live-cell imaging by acquiring images every 15 min over several hours and covering the entire depth of the cell . This showed that most mother cells started to form this aggregate already after budding three to four times ( Figure 1D , 390 min ) . The fastest growth phase of the aggregate took place during the first half an hour following its detection , after which the Hsp104 signal intensity increased only marginally ( Figure 1D , E ) . While the aggregates initially underwent occasional dissolution , they became stable within a few hours from nucleation ( Figure 1E ) . Pedigree analysis of aggregate inheritance in cells of different age showed that they faithfully segregated to the aging mother cell ( 98% of divisions ) , irrespective of its age ( Figure 1D , F , Video 1 ) . To further analyze the persistence and behavior of the deposits over the entire replicative life span of cells , we used a microscope-coupled microfluidic dissection platform ( Lee et al . , 2012 ) ( Figure 1—figure supplement 1B ) . This showed that the aggregate was efficiently maintained as one compartment , and whenever new foci emerged , they typically merged soon after with the pre-existing deposit . We quantified the number of Hsp104 foci ( 1 or >1 ) and its post-nucleation stability in cells that were tracked for at least 10 divisions after the aggregate had appeared . These cells preferentially ( >80% of time ) displayed only a single Hsp104 focus ( Figure 1G–H ) , which was very stable , typically persisting until the last divisions of the cell ( Figure 1G , I ) . Moreover , this structure was largely non-dynamic , displaying limited exchange of Hsp104 with the cytoplasm , as determined by fluorescent recovery after photobleaching ( FRAP ) analysis in diploid cells that expressed one GFP- and one mCherry-tagged copy of HSP104 ( Figure 1J ) . We bleached and measured the recovery of the mCherry signal ( Figure 1J ) and by fitting nine recovery curves , found that on average a large fraction ( 59% ) of Hsp104 was immobile , while the mobile fraction displayed a half-time recovery rate of 8 . 9 s ( Figure 1K ) . Together , these data show that a large majority of unstressed yeast cells develop a protein deposit early during replicative aging . This deposit displays limited exchange with the cytoplasm ( assessed by Hsp104 ) , is efficiently maintained as a single compartment , and is faithfully inherited by the aging mother cell . 10 . 7554/eLife . 06197 . 006Video 1 . Age-associated protein deposit is a stable structure that is faithfully inherited by the aging mother cell lineage . Aged yeast cell expressing Hsp104-mCherry ( in green ) to mark protein aggregates and Cdc10-GFP ( in red ) to mark the mother bud interface undergoing four consecutive divisions . DOI: http://dx . doi . org/10 . 7554/eLife . 06197 . 006 To better understand how the age-associated protein deposit forms and what are its physiological constituents , we selected two proteins containing glutamine- and/or asparagine-rich domains , Sup35 and Dcp2 , which are known to undergo conformational switches and to aggregate , and we monitored their localization in respect to the age-associated protein deposit . Sup35 ( eRF3 ) is a translation termination factor that can undergo stable amyloid-like prion conversion from non-prion [psi–] to prion state [PSI+] ( Chernoff et al . , 1993; Ter-Avanesyan et al . , 1994; Wickner , 1994; Patino et al . , 1996; Paushkin et al . , 1996 ) . Thus , we tested whether Sup35-GFP is targeted to the age-dependent aggregate and if the prion status plays a role in the targeting to and/or in the formation of the age-associated protein deposit . Importantly , Sup35-GFP clearly accumulated into Hsp104-mCherry-labeled age-associated protein deposit in 33% of [PSI+] cells that contained such a deposit , whereas no accumulation was detected in the non-prion [psi−] cells ( Figure 2A–C ) . As expected , the formation of this bright Sup35 focus was age-dependent ( Figure 2D ) , in line with earlier observations ( Derdowski et al . , 2010 ) , but interestingly the [PSI+] state did not have an overall effect on Hsp104-labeled deposit formation ( Figure 2E ) . Similarly , the elimination of the [RNQ+] prion by deleting RNQ1 did not negatively influence the formation of the age-associated deposit ( data not shown ) . Notably , time-lapse imaging showed that Sup35 was recruited into the pre-existing age-associated protein deposit ( marked by Hsp104-mCherry ) . These Sup35-enriched age-associated protein deposits segregated asymmetrically towards the aging mother cell in 98% of mitoses ( Figure 2F , G ) . These data provide evidence that the Hsp104-labeled focus is a bona fide deposit site for aggregating proteins , and that prion conversion of Sup35 promotes its gradual storage into the age-associated protein deposit , but does not potentiate its formation per se . 10 . 7554/eLife . 06197 . 007Figure 2 . Prion form of Sup35 is deposited to the age-associated protein deposit . ( A , B ) Co-localization of Sup35-GFP and Hsp104-mCherry in diploid [psi−] ( A ) and [PSI+] ( B ) cells , where a single locus of the respective gene was tagged with the fluorescent marker indicated . White arrowhead indicates the age-associated protein deposit ( Hsp104 positive ) and the red arrowhead Sup35-aggregates not associated with the age-associated protein deposit . ( C ) Percentage of age-associated protein deposits in [psi–] and [PSI+] cells with enriched Sup35 , ( N = 130 ) . ( D ) Percentage of cells with large Sup35-foci in cells of indicated age groups ( N = 157–184 cells pre age group , average age young [psi–] 1 . 1 , young [PSI+] 1 . 1 , aged [psi–] 6 . 9 , aged [PSI+] 5 . 9 generations ) . ( E ) Percentage of [psi–] and [PSI+] cells of the indicated age group ( see D ) containing an age-associated protein deposit , ( N = 157–184 cells pre age group ) . ( F ) Time-lapse images of a [PSI+] cell co-expressing Sup35-GFP ( green ) Hsp104-mCherry ( red ) at the indicated time points . Arrowheads point at the age-associated protein deposit as observed in the different channels . The newborn daughters are indicated in the bottom row . ( G ) Percentage of divisions where the Sup35-GFP-labeled age-associated protein deposit is retained in the aging mother cell lineage , ( N = 204 divisions ) . ( H ) Fluorescent images of a cell co-expressing the P-body protein Dcp2 tagged with GFP and Hsp104 tagged with mCherry . ( I ) Percentages of cells of indicated age groups that contain the indicated fluorescent foci , ( N = 18–71 per group ) . ( J ) Time-lapse , fluorescent images of Dcp2-GFP and Hsp104-mCherry expressing cells at the indicated time points after switching the cells to 0 . 1% glucose . Scale bars ( A , B , H , J ) 5 μm , ( F ) 2 μm . Graphs display mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 06197 . 007 In contrast , Dcp2 , a Q/N-rich component of the reversible P-body mRNP aggregates ( Reijns et al . , 2008 ) , did not accumulate into the age-associated deposit ( Figure 2H , I ) , even after induction of P-body formation by reducing the glucose level in the medium to 0 . 1% ( Decker et al . , 2007 ) ( 0/57 Hsp104 foci with Dcp2-GFP , Figure 2J ) . Furthermore , we never observed age-associated protein deposits and P-bodies fusing with each other ( Figure 2J ) . This suggests that P-bodies and age-associated protein deposits have different physicochemical properties , representing two distinct modes of cytosolic sub-compartmentalization ( Hyman et al . , 2014; Kroschwald et al . , 2015 ) . Altogether , these data show that the substrates of the age-associated deposit are non-random and possibly amyloid-like , which might explain their irreversible nature . During recent years , numerous distinct protein deposits , including Q-bodies , JUNQ/INQ , and the IPOD , have been discovered and thus we wanted to explore whether the age-associated protein deposit matches the identity of any of these structures . We first looked at the behavior of Q-bodies , which assemble in response to heat-stress . In stark contrast to the age-associated aggregates ( Figure 1G , I ) , Q-bodies induced upon heat stress ( 42°C , 30 min ) were transient and readily solubilized after the removal of the stress factor ( Figure 3A; Liu et al . , 2010; Escusa-Toret et al . , 2013; Wallace et al . , 2015 ) . Comparative analysis of the localization of different chaperone proteins between Q-bodies and the age-associated protein deposit showed that all markers of the age-associated protein deposit were also found to accumulate in Q-bodies ( Figure 3B ) . Conversely , we found chaperones such as Hsp82 and Btn2 localizing to Q-bodies ( Figure 3B ) albeit being excluded from the age-dependent protein deposit ( Figure 1C ) . Together , the reversible nature and the difference in associated chaperones demonstrate that the Q-bodies and age-associated protein deposits can be distinguished . 10 . 7554/eLife . 06197 . 008Figure 3 . The age-associated protein deposit can be distinguished from Q-bodies , JUNQ/INQ , and IPOD . ( A ) Cells were heat shocked ( 42°C for 30 min ) to induce the formation of Q-bodies ( stress induced Hsp104-labeled aggregates ) and the kinetics of Q-bodies dissolution was followed with time-lapse microscopy of Hsp104-GFP . ( B ) Cells were imaged after heat shock ( 42°C for 30 min ) . Arrowhead in 3D-projected images depicts co-localization of Hsp104 with age-associated deposit-resident ( Ssa1 , Hsp42 ) and non-resident ( Hsp82 , Btn2 ) markers in Q-bodies ( compare with Figure 1C ) . ( C ) The expression of JUNQ/INQ marker VHL-GFP was induced at the start of the imaging to follow its recognition by Hsp104-mCherry . ( D ) Quantification of VHL-GFP foci recognized by Hsp104-GFP within 30 min of their appearance ( N = 29 cells ) . ( E ) 3D-projected images of RPN4 deleted cells expressing Hsp82-GFP and Hsp104-mCherry . Hsp82-GFP , which localizes to Q-bodies ( Figure 3A ) but not to age-associated protein deposits ( Figure 1C ) , co-localizes with one of the two Hsp104-mCherry foci in aged cell ( see fluorescence intensity line-scan over the two foci ) . ( F ) 3D-projected image of Btn2-GFP and Hsp104-mCherry expressing cell . Btn2 is typically very low abundant and does not accumulate ( 98 . 6% of cases , N = 72 ) to the age-associated deposit ( white arrowhead ) . ( G ) Cell with a Btn2 focus , which typically ( 96 . 4% of cases , N = 30 ) did not display accumulation of Hsp104-mCherry ) . ( H ) Rnq1-overexpression was induced at the onset of imaging . The panel shows the z-sections displaying the age-associated deposit ( white arrowhead in ‘merge’ ) . ( I ) Quantification of newly formed Rnq1-GFP foci that accumulate Hsp104-mCherry within 30 min of their appearance ( N = 53 cells ) . ( J ) Representative image of [PIN+ , PSI+] cell harboring GFP-tagged endogenous Rnq1 and mCherry-tagged Hsp104 . Arrowhead indicates the age-associated protein deposit . ( K ) Example of a [PIN+ , PSI+] cell that displays an Rnq1-aggregate . Arrowhead indicates Hsp104-labeled foci , of which only one has accumulated Rnq1 . Scale bars: A-B , E , J-K 5 μm , C , G 2 μm . Graphs display mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 06197 . 008 We then examined whether the age-associated protein deposit would bear the characteristics of the JUNQ/INQ compartment , which is formed during prolonged stress and/or in response to proteasome inhibition . To this end , we induced the GAL-promoter-driven expression of the JUNQ/INQ marker , the unstable proteasome substrate human von Hippel-Lindau tumor suppressor protein ( VHL ) ( Kaganovich et al . , 2008; Miller et al . , 2015 ) , and monitored the accumulation of Hsp104-mCherry into the newly forming VHL-foci . VHL typically formed a single focus , which in 87% of the cases did not accumulate Hsp104-mCherry within 30 min after its appearance ( Figure 3C , D , N = 29 ) . However , the VHL expression was often associated with overall increased Hsp104-mCherry expression over time , suggesting that its expression activates a stress response . Hence , we visualized the JUNQ/INQ without over-expression of exogenous substrates . To accomplish this , we used an Hsp82-GFP , Hsp104-mCherry expressing strain deleted of the RPN4 gene , leading to increased burden of proteasomal substrates , due to decreased amount of functional proteasomes ( Xie and Varshavsky , 2001 ) . Indeed , many of rpn4Δ cells displayed numerous Hsp82-Hsp104 double-positive Q-body-like puncta ( data not shown ) . In addition , we found cells that displayed two puncta: a Hsp104-Hsp42 double-positive JUNQ-like deposit , and a Hsp104-positive , Hsp82-negative deposit that fills the criteria of the age-associated deposit ( Figure 3E ) . To consolidate this further , we investigated the co-localization of the endogenous JUNQ/INQ marker Btn2 ( Miller et al . , 2015 ) together with the age-associated protein deposit ( Hsp104 ) . This analysis showed that in 98 . 6% of the cases , Btn2 did not localize to the age-associated Hsp104-foci ( Figure 1C and Figure 3F ) . Moreover , from all thirty identified cells displaying endogenous Btn2 foci , we found only one case in which Hsp104 was enriched at this site ( Figure 3G , N = 1268 cells ) . Together , these data suggest that the age-associated protein deposit and the JUNQ/INQ are discrete structures that may exist in parallel . Finally , we monitored the resemblance between the age-associated protein deposit and IPOD . To visualize cells during IPOD appearance , we imaged Hsp104-mCherry-expressing cells together with the canonical IPOD marker , the galactose-inducible Rnq1-GFP ( Kaganovich et al . , 2008 ) , immediately after placing cells to galactose-containing media . This showed that the IPOD typically appeared as a single focus to which Hsp104 rapidly accumulated ( Figure 3H , I ) ( >98% of Rnq1-foci displayed accumulation of Hsp104 within 30 min after their appearance , N = 53 ) . However , by dissecting the Rnq1-GFP appearance dynamics in cell with a pre-existing age-associated protein deposit ( marked by Hsp104-mCherry ) , we found that the aggregating Rnq1-GFP did not accumulate to the age-associated deposit ( see white arrowhead in Figure 3H ) , but rather it formed a new aggregate to which Hsp104 then strongly accumulated , while its intensity at the age-associated protein deposit declined ( Figure 3H ) . To rule out possible artifacts induced by Rnq1-GFP over-expression , we also looked at the co-localization between the age-associated protein deposit and GFP-tagged endogenous Rnq1 in its prion [RNQ+]-state . We analyzed altogether 117 Hsp104 foci-containing [PIN+ , PSI+] cells and found that in 98 . 3% of the cases , Rnq1-GFP did not accumulate to this deposit ( Figure 3J ) . Within this cohort , we found two cells with an Rnq1-GFP-aggregate , which was in both cases enriched with Hsp104-mCherry ( Figure 3K ) . However , both of the Rnq1-aggregate containing cells also displayed an additional Hsp104 focus to which Rnq1 had not accumulated ( Figure 3K ) . Together with the results from the Rnq1 over-expression , these data suggest that Rnq1 aggregate is likely to represent a structure that is different from the majority of the age-associated protein deposits . In sum , these data point out marked differences between the age-dependent protein deposit and the previously characterized Q-bodies , IPOD , and the JUNQ/INQ that could derive from context dependent ( stress vs aging ) differences in cellular responses to protein aggregation . Appearance of protein aggregates has commonly been associated with defects in protein quality control ( Tyedmers et al . , 2010a ) . Therefore , we wanted to elucidate if the formation of the age-associated protein deposit is a sign of defective protein quality control . First , we examined whether aged cells display a decline in dealing with proteotoxic stress . As shown before , young cells clear heat-induced protein aggregates ( Q-bodies ) rapidly after stress removal ( Figure 3A ) . Hence , we first wanted to test if this recovery period is affected by the age of the cell ( Figure 4A ) . We measured the clearance time of Q-bodies ( defined as two or more Hsp104 foci ) between young ( average age 0 . 4 generations , between 0 and 1 generations , N = 70 ) and aged cells ( average age 9 . 2 generations , between 6 and 19 generations , N = 50 ) following acute heat stress ( Figure 4B ) . Surprisingly , no difference in the mean time of protein aggregate clearance was detected between these populations: young: 71 ±5 min old: 72 ±4 min , n . s . ) ( Figure 4C ) , suggesting that aged cells , despite having a protein deposit , are fit to cope with proteotoxic stress . 10 . 7554/eLife . 06197 . 009Figure 4 . Aged cells are not impaired in handling proteotoxic stress . ( A ) Experimental scheme . ( B ) Representative image of displaying the dynamics of recovery from heat shock induced proteotoxic stress ( Q-bodies ) between aged ( red star ) and young cell ( blue star ) . ( C ) Q-body clearance time of individual cells of the indicated age groups , ( old: average 9 . 2 generations , age between 6 and 19 , N = 50; young: 0 . 4 generations , age between 0 and 1 , N = 70 ) . ( D ) Experimental scheme: Hsp104-mCherry-expressing cells were captured on a temperature-controlled microfluidic device and imaged prior ( −50 min ) , during ( 30 min ) and after ( up to 300 min ) mild heat shock . It is important to note , that the strength of the heat stress induced on the microfluidic platform is not comparable with Figure 3A , B and Figure 4A–C . ( E ) Single cell analysis of Q-body formation ( cells with >1 Hsp104-foci ) in cells with a pre-existing age-associated protein deposit ( red line , N = 53 ) , cells without a pre-existing deposit ( blue line , N = 97 ) , and cells that did not experience stress ( N = 82 ) . Data from two independent experiments . ( F ) Transitions in the aggregate state were recorded 50 min before the heat stress and 300 min after the heat stress . ( G ) Quantification of transitions ( N = 32–77 per group , from two independent experiments ) . White bars indicate cells that did not experience heat stress and red bars denote transitions in aggregate state in heat-stress experienced cells . Scale bar 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06197 . 009 We then used a temperature-controlled microfluidic device that enabled us to categorize cells depending on their pre-stress age-associated protein deposit-status , and monitor them under the microscope prior to , during and after undergoing acute proteotoxic stress conditions ( Figure 4D ) . The duration of the Q-body response ( state in which cells display two or more Hsp104-foci ) was plotted over time and demonstrated that cells with and without a pre-existing age-associated deposit showed a comparable Q-body response when heat was applied , while cells that were not exposed to heat did not show similar Q-body response during this time period ( Figure 4E , N = 53–97 cells ) . Interestingly , the deposit-containing cells responded slightly faster and to a lesser extend when compared to their clonal counterparts without a pre-existing protein deposit ( Figure 4E ) . We then asked if exposure to proteotoxic stress ( Q-body state ) would promote the formation of age-associated protein deposits . To this end , we compared the aggregate status of single cells ( Hsp104-mCherry: no/one/several puncta ) at time points 0 and 380 min , between stress-experienced ( between 50 and 80 min ) and non-stressed ( constantly at 30 °C ) cells ( Figure 4F , N = 32–77 per group ) . This analysis showed that the majority of non-stressed cells ( 62 . 5% ) without a deposit at the beginning of the experiment displayed a single aggregate at the end of the experiment ( Figure 4G ) , in accordance with the appearance kinetics of age-associated deposit ( Figure 1B , D ) . Surprisingly , when the cells without a prior aggregate encountered heat stress ( i . e . conditions that induce proteotoxic stress ) , a substantially smaller portion of them ( 36 . 4% ) displayed an aggregate at the end of the experiment ( Figure 4G ) . Similarly , there was a slight increase of cells without an aggregate in the stress-encountered cohort ( 17 . 7% vs 6 . 8% ) among cells that displayed an age-associated deposit at the beginning of the experiment ( Figure 4G ) . However , irrespectively of stress , cells with an age-associated deposit at the beginning of the experiment preferentially displayed a single deposit at the end of the experiment ( non-stressed 82% , stress-experienced 79% ) ( Figure 4G ) . These results indicate that mild exposure to proteotoxic stress conditions counteract the formation of age-associated protein deposits and demonstrate that cells with an age-associated deposit prior to encountering the stress predominantly reverted back to this state . Altogether , these data indicate that despite forming a protein deposit , aged cells are still fit to coping with acute proteotoxic stress and suggest that the formation of the age-dependent protein deposit is not due to the overload of the general quality control machinery . The notion that aged cells with a protein deposit handled proteotoxic stress comparably to their young counterparts prompted us to test the effect of the age-associated deposit on the function of the ubiquitin–proteasome system ( UPS ) , which has an important role in yeast replicative aging ( Kruegel et al . , 2011 ) . Pathological polyQ protein fragments have been shown to impair the UPS function both in yeast and mammalian cells ( Bennett et al . , 2005; Park et al . , 2013 ) , and UPS substrates have been shown to accumulate in aged cells ( Andersson et al . , 2013 ) . On the other hand , stress-induced protein aggregates ( Q-bodies ) do not result from failed protein degradation , but may actually benefit the proteostasis of cells undergoing stress ( Escusa-Toret et al . , 2013 ) . We therefore wanted to analyze the contribution of the age-associated deposit on the functionality of the UPS . In order to selectively measure the effect of the age-associated protein deposit on UPS function , we made use of the auxin-inducible degron ( AID ) system ( Nishimura et al . , 2009 ) , which enabled us to measure the degradation rate of cytosolic ( AID-GFP ) and nuclear ( AID-GFP-NLS ) UPS substrates in vivo ( Figure 5A , B ) . Cells co-expressing Hsp104-mCherry and AID-GFP were switched to auxin-containing media at the onset of imaging , and the GFP intensities were measured over time from neighboring cells containing or not an Hsp104-puncta . The normalized values were pooled and fitted with a non-linear one-phase decay function . Importantly , this showed that the decay rate of AID-GFP was significantly accelerated in cells with an age-associated deposit when compared to neighboring cells that did not contain an Hsp104-focus ( Figure 5C , D ) ( rate constant ( K ) : 0 . 155 ± 0 . 007 ( with ) vs 0 . 129 ± 0 . 006 ( without ) , p < 0 . 01 , N = 91 cells/category ) . However , no difference was detected in the decay rate of nuclear GFP between cells with or without an age-associated protein deposit ( Figure 5E , F ) ( K: 0 . 075 ± 0 . 004 ( with ) vs 0 . 079 ± 0 . 005 ( without ) , n . s . ) , demonstrating that the effect of the age-associated protein deposit on the UPS function is specific to the cytosolic compartment . Collectively , these results suggest that the age-associated deposit promotes cytosolic quality control by enabling more efficient clearance of degradation substrates . 10 . 7554/eLife . 06197 . 010Figure 5 . Presence of age-associated protein deposit promotes the function of cytosolic ubiquitin–proteasome system in vivo . ( A ) Schematic representation of the auxin-induced degron ( AID ) -system ( Nishimura et al . , 2009 ) . Addition of auxin facilitates the recognition of the degron-motif in the target protein ( GFP ) by the exogenously expressed Arabidobsis thaliana E3 ligase SCF-Tir1 and subsequent ubiquitination and degradation by the endogenous ubiquitin–proteaome system ( UPS ) . ( B ) Representative images of cells expressing Hsp104-mCherry and AID-GFP taken immediately ( 0 min ) and 48 min after auxin-addition . ( C ) Examples of time frames GFP degradation in the absence of auxin ( upper panel ) in the presence of auxin in cells with ( middle ) or without ( below ) an age-associated protein deposit . ( D ) The decay rates of GFP in the indicated groups . The error bars depict the normalized intensity values ( average ± SEM ) derived from 53 ( ctrl ) and 91 ( auxin added ) cells from three pooled replicates . The solid lines indicated non-linear one-phase decay fit . ( E ) GFP-NLS decay in the representative groups . ( F ) The decay rates and the graph fitting of NLS-GFP in the indicated groups as in ( D ) . N = 15 ( ctrl ) , 65–66 ( auxin added ) cells from three pooled replicates . The error bars depict the normalized intensity values ( average ± SEM ) , while the solid lines indicate the non-linear one-phase decay fit . DOI: http://dx . doi . org/10 . 7554/eLife . 06197 . 010 Next , we wanted to understand what are the assembly principles of the age-associated deposit . Molecular chaperones generally recognize , refold , or sort aberrantly folded proteins , making them prime candidates to regulate the age-associated protein deposit assembly . Thus , we addressed the contribution of chaperones localizing to the age-associated protein deposit to its formation . Time-lapse microscopy showed that in 24 out of 25 cells , Hsp42 was at the age-associated protein deposit before Hsp104 and was the first protein we find to mark this structure ( Figure 6A ) . Ssa1 and Hsp104 were recruited subsequently , appearing concurrently up to 120 min after appearance of the Hsp42-focus ( Figure 6A and data not shown ) . To identify the role of these chaperones in the assembly of the age-associated protein deposit , we investigated the consequences of their deletion . Interestingly , loss of the Hsp70 proteins Ssa1 and Ssa2 , which act synergistically with the disaggregase Hsp104 in protein refolding ( Glover and Lindquist , 1998; Kim et al . , 2013 ) , resulted in rapid formation of the Hsp104 focus ( Figure 6B , C ) and a similar , but milder effect was detected upon deletion of HSP26 . In stark contrast , deletion of HSP42 abolished age-associated protein deposit formation ( Figure 6B , C ) , which parallels to its function in Q-body assembly ( Specht et al . , 2011; Escusa-Toret et al . , 2013 ) . To probe the role of Hsp104 and Hsp42 further , we analyzed the consequences of their over-expression on age-associated protein deposit formation ( monitored by endogenous Hsp104-GFP ) . Over-expression of Hsp42 caused precocious nucleation of age-associated protein deposit and promoted its growth ( Figure 6D–F ) . On the contrary , a significantly smaller fraction of cells over-expressing Hsp104 contained an age-associated protein deposit and while age-associated deposits still occasionally formed , they were rapidly destabilized ( Figure 6D–F ) . Collectively , these data suggest that these chaperones functionally oppose each other ( Figure 6G ) : the Hsp104/Hsp70 proteins prevent age-associated protein deposit formation possibly by refolding cytoplasmic substrates , whereas Hsp42 coordinates the collection of these substrates from the cytoplasm into the age-associated protein deposit ( Figure 6G ) . 10 . 7554/eLife . 06197 . 011Figure 6 . Identification of the roles of chaperones in age-associated protein deposit assembly . ( A ) Representative time-lapse images of cells expressing endogenously tagged Hsp42-GFP ( upper panel ) and Hsp104-mCherry ( middle panel ) . Arrowhead depicts the age-associated protein deposit structure . ( B ) Representative images of aged Hsp104-GFP expressing wild-type , ssa1Δ , and hsp42Δ mutant cells . ( C ) Quantification of Hsp104-foci containing cells in indicated age groups , ( N = 317–1020 cells per genotype ) . ( D ) Time-lapse images of cells expressing endogenous Hsp104-GFP in wild-type , Hsp104 over-expressing , and Hsp42 over-expressing cells ( age-associated protein deposit depicted by a red arrowhead ) . ( E ) Quantification of percentage of cells with age-associated protein deposit in wild-type , Hsp104 over-expressing , and Hsp42 over-expressing cells of indicated age groups ( N = 432–1020 cells per genotype ) . ( F ) Quantification of endogenous Hsp104-GFP integrated density ( age-associated protein deposit/cytoplasm ) following its appearance in wild-type ( black line ) and Hsp42 over expressing cells ( red line ) , ( N = 19–23 ) . ( G ) A summarizing model of the pathway underlying age-associated protein deposit formation . Scale bars 5 μm . Graphs display mean ± SEM , n . s not significant , *p < 0 . 05 , **p < 0 . 01 , ***p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 06197 . 011 To consolidate this model , we further investigated the interplay between the refolding machinery ( Hsp70/Hsp104 ) and the nucleation/growth promoting Hsp42 . We used cells expressing Ssa1-GFP to visualize the age-associated protein deposit in the absence of Hsp104 ( Figure 7A ) . Similarly to what we found for the deletion of SSA1 and SSA2 ( Figure 6B , C ) , the hsp104Δ mutant cells formed the age-associated protein deposit more rapidly during aging . Again , the deletion of HSP42 had an opposite effect , hampering formation of the Ssa1-GFP focus ( Figure 7A , B ) . Remarkably , age-associated protein deposit formation was severely impaired in the hsp104∆ hsp42∆ double-mutant cells . These cells were crowded with punctate Ssa1-GFP ( speckles ) , which failed to coalesce into a single regular age-associated protein deposit ( Figure 7A , B ) . The portion of cells that contained multiple Ssa1-GFP foci was significantly higher in hsp104Δ hsp42Δ double-mutant cells than in hsp104Δ single-mutant cells . Interestingly , whereas >80% of wild-type and hsp104∆ mutant cells that contained an aggregate displayed a single Ssa1-GFP focus , the majority of hsp42∆ and hsp104∆ hsp42∆ mutant cells harbored several foci ( Figure 7C ) . 10 . 7554/eLife . 06197 . 012Figure 7 . The assembly of age-associated protein deposit is promoted by Hsp42 and counteracted by Hsp104/Hsp70 . ( A ) Representative aged wild-type , hsp104Δ , hsp42Δ and hsp104Δ and hsp42Δ double-mutant cells . ( B ) Quantification of the fraction of cells in indicated age groups that display Ssa1-GFP foci ( age-associated protein deposit ) , ( N = 363–493 cells per genotype ) . ( C ) Quantification of the portion of cells that display >1 Ssa1 foci from all Ssa1-foci containing cells , ( N = 363–493 cells per genotype ) . ( D ) Illustration of the experimental scheme of the reconstitution assay . Cells that lack both HSP104 and HSP42 and display fragmented aggregate phenotype were mated with cells of the opposite mating type to reintroduce either Hsp104 and/or Hsp42 . The resulting zygote was imaged with time-lapse microscopy and analyzed 5 min after fusion to score for aggregation phenotype ( analyzed by Ssa1-GFP ) . ( E ) Quantification of the Ssa1-GFP phenotype ( no foci , one focus , multiple foci ) 5 min post fusion , ( N = 73–100 fusion events per genotype ) . Scale bars 5 μm . Graphs display mean ± SEM , n . s not significant , **p < 0 . 01 , ***p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 06197 . 012 To decisively dissect the individual contributions of Hsp104 and Hsp42 on age-associated protein deposit formation , we conceived an in vivo reconstitution assay where we could rapidly re-introduce either Hsp104 or Hsp42 to the aggregate-enriched hsp104Δ hsp42Δ double-mutant cells . To accomplish this , we mated hsp104Δ hsp42Δ double-mutant cells with wild-type ( re-introducing both Hsp104 and Hsp42 ) , hsp104∆ single-mutant ( to reintroduce only Hsp42 ) , hsp42Δ single-mutant ( to reintroduce only Hsp104 ) , or with the hsp104 hsp42 double-mutant cells ( reintroducing none of the chaperones ) ( Figure 7D ) . We monitored the deposit by imaging the fusing cells at 1-min intervals , using Ssa1-GFP as a reporter ( Figure 7D ) . Importantly , simultaneous reintroduction of Hsp104 and Hsp42 was sufficient to clear the cytoplasm of Ssa1-GFP speckles already within 5 min ( Figure 7E , no foci ) . In contrast , introducing Hsp42 alone resulted in rapid disappearance of cytoplasmic speckles with concurrent formation of a single Ssa1-GFP focus ( Figure 7E , one focus ) . In contrary , reintroduction of only Hsp104 led to the slow and incomplete clearance of the Ssa-1 speckles , which was a reminiscent , but a milder phenotype as that observed upon conjugating the cell with hsp104Δ hsp42Δ double-mutant partner ( Figure 6E , multiple foci ) . Collectively , this analysis supports the conclusion that Hsp42 functions to collect aggregates into the age-associated protein deposit structure , whereas Hsp104 functions to disaggregate/refold age-associated protein deposit destined substrates . In the context of the age-associated protein deposit cargo Sup35 ( Figure 2 ) , it is interesting to note that Hsp42 over-expression promotes [PSI+] curing , while its deletion results in enhanced [PSI+] induction , further supporting its role as the depositor for the age-associated protein deposit ( Duennwald et al . , 2012 ) . We then asked whether age-associated protein deposit formation ensures the asymmetric inheritance of protein aggregates by the aging lineage . Examining its mitotic segregation ( monitored with Ssa1-GFP ) in 96 wild-type and 186 hsp104Δ mutant cell divisions demonstrated that the age-associated protein deposit was in both cases inherited by the aging mother cell with high fidelity ( in 98% of divisions; data not shown ) . In strong contrast , Ssa1-GFP foci were frequently inherited by the daughter cells when both HSP104 and HSP42 were deleted ( Figure 8A , Video 2 ) . By examining 61 mitotic events by time-lapse microscopy in the double-deleted cells , we observed Ssa1-GFP-foci relocating from the mother to the bud in 53 cases , but did not find any biased retrograde movement from the bud to the mother ( Fig . 8A , Video 2 and data not shown ) . Quantification of the percentage of buds with protein deposits demonstrated that more than 65% of all hsp104Δ hsp42Δ double-mutant buds displayed at least one aggregate , compared to less than 4% in hsp104Δ and hsp42Δ single-mutant and in wild-type cells ( Figure 8B ) . Altogether , these data provide evidence that collection of protein aggregates into a single deposit by Hsp42 promotes their asymmetric retention in the mother cell during cell division . 10 . 7554/eLife . 06197 . 013Figure 8 . Age-associated protein deposit formation establishes asymmetric inheritance of protein aggregates and correlates with replicative age . ( A ) Time-lapse imaging of Ssa1-GFP in hsp42Δ hsp104Δ double-mutant cells ( arrowheads denote Ssa1-GFP aggregates relocating from the mother cell to the bud ) . ( B ) Quantification of Ssa1-GFP foci found in the buds of mitotic yeast cells ( N = 216–434 buds per genotype ) . ( C ) Replicative aging experiments of wild-type ( black line , 27 generations ) , hsp104Δ mutant ( blue line , 21 . 5 generations ) , hsp42Δ mutant ( red line , 39 generations ) , and hsp104Δ and hsp42Δ double mutant ( purple , 23 generations ) single cells , ( N = 101–140 cells per genotype ) . ( D ) A schematic model for the age-associated protein deposit pathway: Hsp42 acts as a collector of protein aggregate seeds and promotes their deposition at the ER membrane ensuring their asymmetric inheritance by the aging lineage during mitosis . These cytoplasmic seeds are subjected to Hsp104/Hsp70-dependent refolding , which is inactive ( blue rectangle ) at the site of the age-associated protein deposit assembly . Scale bar 5 μm . Graph displays mean ± SEM , n . s not significant , ***p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 06197 . 01310 . 7554/eLife . 06197 . 014Video 2 . The lack of the age-associated protein deposit assembly pathway results in the inheritance of the aggregates by the daughter cells . Expression of Ssa1-GFP was followed in hsp104Δ hsp42Δ double-mutant cell as it underwent cell division . DOI: http://dx . doi . org/10 . 7554/eLife . 06197 . 014 Finally , we tested the significance of the age-associated protein deposit pathway and asymmetric protein aggregate segregation on replicative aging ( Figure 8C ) . In accordance to previous work ( Erjavec et al . , 2007 ) , mutant hsp104∆ cells were short-lived compared to wild-type cells ( 21 . 5 vs 27 generations , p < 0 . 001 ) , supporting the idea that accelerated aggregate accumulation promotes aging . Further supporting this notion , deletion of HSP42 extended the lifespan of the cells by more than 40% ( 39 generations , p < 0 . 001 ) . The lifespan of the hsp104Δ hsp42Δ double-mutant cells was similar to that of the hsp104Δ single-mutant cells ( 23 vs 21 . 5 generations , respectively , ns . ) and significantly shorter than that of the wild-type ( p < 0 . 01 ) and hsp42Δ mutant cells ( p < 0 . 001 ) . The over-expression of HSP42 had no effect on the lifespan ( data not shown ) , suggesting that the wild-type levels of Hsp42 are sufficient to provide its maximal effects on life span , which is consistent with the appearance of the deposit early in life span in the wild-type cells . Together , these data imply that the Hsp104/Hsp70 system , which counteracts protein aggregation , is essential for longevity , while the Hsp42-dependent circuit promotes aging in the mother lineage , presumably by building the age-associated protein deposit and thereby establishing the asymmetric inheritance of age-associated protein aggregates . Protein aggregation has been frequently associated with aging and aging-associated diseases . Here , by using the budding yeast as a model for replicative aging , we identify an Hsp42-promoted and Hsp104/Hsp70-counteracted pathway that deposits age-associated protein aggregates and thereby ensures their biased segregation to the aging mother-lineage upon cell division ( see model in Figure 7D ) . This might represent a common mechanism to drive asymmetric segregation of aberrant proteins in dividing cells , as stressed fission yeast cells utilize a similar Hsp16-dependent mechanism for asymmetric partitioning of misfolded proteins ( Coelho et al . , 2014 ) . Intriguingly , our data suggest that depositing age-associated protein aggregates to the mother cells might be a mixed blessing . On one hand , the age-associated deposit can promote spatial protein quality control ( Figures 4 , 5 ) and promote cell diversity as an asymmetrically dividing structure that may harbor protein conformation encoded epigenetic information , such as the prion fold of Sup35 ( Figure 2 ) . At the same time , the long-term consequences of this pathway appear negative for the aging lineage ( Figure 8C ) . Why the deposit-forming pathway ultimately becomes associated with age-dependent loss of fitness remains to be determined . In this regard , it will be of great importance to elucidate how age-associated deposit relates to other aging pathways , such as those influencing the function of mitochondria ( McMurray and Gottschling , 2003; McFaline-Figueroa et al . , 2011 ) , vacuoles ( Hughes and Gottschling , 2012 ) and nuclear pores ( Shcheprova et al . , 2008; Denoth-Lippuner et al . , 2014; Webster et al . , 2014 ) , and how the deposit localizes relative to these organelles and their components ( Chong et al . , 2015 ) . Protein deposit-containing cells efficiently coped with acute proteotoxic stress ( Figure 4 ) and displayed improved degradation of cytosolic UPS substrate during early- to mid-life span ( Figure 5 ) . It is important to note , however , that our assays were conducted with relatively young cells that had recently acquired the protein deposit , and hence it is possible that UPS function starts to decline later during the aging process , as shown previously ( Andersson et al . , 2013 ) . Curiously , the decline in proteostasis has been associated with protein aggregation pathologies such as Alzheimer's , Huntington's , and Parkinson's disease ( Vilchez et al . , 2014 ) . The decreased degradation of cytosolic UPS substrates in a polyQ disease model was found to be due to the sequestration of the Hsp40 protein Sis1 ( Park et al . , 2013 ) —a protein that was not associated with the age-associated deposit ( Figure 1C ) . It is thus plausible that the presence of the deposit might , for example , lead to increased levels of available ( substrate unbound ) Sis1 , thereby allowing more efficient nuclear import and degradation of cytosolic UPS substrates . Altogether , these data suggest that there is a need to better differentiate between pathophysiological protein aggregation and ‘homeostatic’ protein aggregation that takes place during physiological aging , of which the latter might initially help to coordinate chaperones and UPS factors involved in protein quality control . Related to this , the age-associated protein deposit did not clearly fill out the set criterion for any of the previously characterized cytosolic deposits , including Q-bodies , JUNQ/INQ , or IPOD , but rather seems to exist in parallel with the JUNQ and IPOD structures ( Figure 1 , Figure 3 ) —although , for example , the chaperones regulating these deposits were often shared between these structures . This favors the notion that pathological aggregation processes ( mimicked e . g . , by VHL and Rnq1 over-expression ) can hijack endogenous proteostasis regulatory mechanisms , possibly underlying their harmful effects in cells ( Park et al . , 2013 ) . Altogether , these apparent differences between the stress-induced , pathological , and aging-associated protein aggregates ( Figure 2 , Figure 3 ) demonstrates that protein aggregates are not all equal in their composition or in the way they are being recognized and dealt by the cell and reinforces the importance to discriminate between different aggregation models . In the future , it will be important to identify the cargo deposited to the age-associated protein deposits . In this aspect , it is interesting to note that many of the long-lived asymmetrically retained proteins that accumulate to mother cells as they age ( Thayer et al . , 2014 ) were among those localizing to the age-associated protein deposit ( including Hsp104 , Ssa1 , Ssa2 , and Hsp26 ) . One example of an age-associated deposit resident protein was the prion-like Sup35 . This storage rendered Sup35 to be inherited by the mother cells ( unlike a prion , analogous to a mnemon ( Caudron and Barral , 2013 ) ) ( Figure 2 ) , which is consistent with the size-dependent transmission model of Sup35-aggregates ( Derdowski et al . , 2010 ) . However , aging does not enhance Sup35-dependent [PSI+] prion generation ( Shewmaker and Wickner , 2006 ) . Together , these notions suggest that Sup35 cannot escape the age-associated deposit under normal growth conditions , perhaps owing to the rigid amyloid-like packing ( Saibil et al . , 2012 ) , being consistent with the durable and non-dynamic nature of these deposits . However , this notion may come with a caveat: we found that the age-associated deposit could be , at least temporarily , disassembled by the heat shock response ( Figure 4 ) . This suggests the captivating possibility that changes in the environment might trigger the spread from the age-associated deposit . For cells facing fluctuating environments in the wild , the strategy to stochastically store prion-like proteins in aged cells might prove beneficial , as it could allow stress-dependent spread of the prion conformers from a subpopulation of aged-cells to their daughters , promoting faster adaptation to the changing environment ( Newby and Lindquist , 2013 ) . Since asymmetric inheritance of damaged proteins and protein aggregates seems to be conserved in metazoan stem cells ( Rujano et al . , 2006; Bufalino et al . , 2013 ) that are also subjected to both replicative decline and segregation of lineages , we hypothesize that similar pathways that selectively segregate protein-aggregate based fate determinants asymmetrically during cell division are likely to be conserved and may contribute to cellular lineage specification across species . Yeast strains used in this study are listed in Supplementary file 1 . Strains were generated either manually ( Janke et al . , 2004 ) or derived from collections: ( http://web . uni-frankfurt . de/fb15/mikro/euroscarf/col_index . html or http://clones . lifetechnologies . com/cloneinfo . php ? clone=yeastgfp ) . Plasmids ( pAG415-GDP ) over-expressing Hsp104 ( O-1360 ) and Hsp42 ( O-2252 ) , and the [PIN+][PSI+] ( yJW508 ) and [PIN+][psi–] ( yJW584 ) parental strains ( Osherovich and Weissman , 2001 ) were a generous gift from Simon Alberti ( Max Planck Institute of Molecular Cell Biology and Genetics Dresden , Germany ) . The Mother Enrichment parental strains ( Lindstrom and Gottschling , 2009 ) were a kind gift from Dan Gottschling ( Fred Hutchinson Cancer Research Center , Seattle USA ) and the plasmids encoding VHL-GFP and Rnq1-GFP were provided by Judith Frydman ( via Addgene ) . The integrative plasmids for the auxin-mediated degradation assays ( pADH1-OsTIR1-9myc , pADH1-eGFP-IAA17-NLS , pADH1-eGFP-IAA17 ) have been described in ( Nishimura et al . ) and were kindly provided by Matthias Peter ( ETH Zurich , Switzerland ) . Cells subjected to imaging were re-inoculated from overnight cultures to O . D . 600 nm 0 . 05 in YPD and grown to O . D . 0 . 5–0 . 8 at 30°C , centrifuged 500 g for 5 min , resuspended in synthetic complete ( SC ) media ( -his ) , and mounted between a coverslip and an agar pad ( SC-his ) . The over-expression of Rnq1-GFP and VHL-GFP was induced by switching cells to 2% galactose at the onset of imaging . The auxin-mediated degradation assays were performed similarly as above by placing cells to 0 . 5 mM 3-Indoleacetic acid ( Sigma–Aldrich , #12886 ) at the onset of imaging . The MEP was performed as described in ( Lindstrom and Gottschling , 2009 ) . Briefly , cells were grown in log phase for 15 hr before coupling the cell wall with Sulfo-NHS-LC-Biotin ( Pierce ) . Following 2 hour recovery period , the MEP was engaged by addition of estradiol to a 1 μM final concentration . After desired incubation period , cells were coupled to uMACS Streptavidin MicroBeads ( Miltenyi Biotec ) and affinity purified with MACS separation columns ( Miltenyi Biotec ) . Calcofluor staining was done by incubating cells with 5 μg/ml Fluorescent Brightener 28 ( Sigma–Aldrich ) for 1 min prior to centrifugation and resuspension to SC media . The Q-bodies were induced by incubating cells in a water bath at +42 °C for 30 min , or by using a temperature-controlled microfluidic chamber ( see microfluidics ) . For the mating experiments , cells ( 1 . 85×107 ) of the opposite mating type were centrifuged 500 g for 5 min and resuspended to 40 μl SC media , mixed and immediately imaged with 1-min intervals ( see microscopy for details ) . Wide-field microscopy was performed at 30 °C with a DeltaVision microscope coupled to a coolSNAP CCD HQ2 camera ( Roper ) , 250W Xenon lamps and 100X/1 . 40 and 60x/1 . 42 NA Olympus oil immersion objectives . Z-sectioning was performed with 500-nm spacing ( unless otherwise indicated ) , obtaining 9 or 11 ( live-cell imaging ) or 15 ( calcofluor stained cells ) stacks . Images were deconvolved with Softworx software ( Applied Precision ) . Microfluidics on the aging chip were imaged with a DeltaVision microscope every 45 min for total of 66 hr by obtaining seven z-sections with 0 . 6-μm spacing . FRAP was done with temperature-controlled Zeiss LSM 510 microscope controlled with AIM LSM4 . 0 software , 63x 1 . 4NA Oil DIC Plan-Apochromat objective at 30°C , using DPSS and Argon lasers . Diploid cells expressing HSp104-GFP/Hsp104mCherry were imaged every 5 s at three z-planes with 800-nm spacing five times before bleaching the mCherry signal at the age-associated protein deposit with 100% DPSS laser power , after which the recovery of mCherry at the age-associated protein deposit was monitored over the period of two minutes by orienting with the GFP-signal . The temperature ramp experiments ( Figure 4 ) were carried out with ONIX microfluidic perfusion system equipped with a micro-incubator temperature controller CellASIC ) , using Y04C microfluidic plates ( CellASIC ) . The temperature setup was the following: 30°C ( 50 min ) , 42°C ( 30 min ) 30°C ( 300 min ) and the cells were imaged every 10 min as described in ( microscopy ) . The experiments were carried out in synthetic full medium with even flow rate of 2 psi . Microfluidics on the aging chip was performed with 1 . 5 μl/min flow rate as described in ( Denoth-Lippuner et al . , 2014 ) . All image analyses were performed with Image J software ( http://imagej . nih . gov/ij/ ) . The aggregation foci were scored by eye from maximum intensity projected images ( spanning the entire volume of the cell ) and were defined as puncta that display high-intensity over the surrounding cytoplasmic background signal . For age-associated protein deposit intensity measurements , the integrated density was measured at the site over time at the age-associated protein deposit from stacked , non-processed sum-projected videos and was then normalized to the cytoplasmic Hsp104-GFP intensity . For the auxin-mediated degradation experiments , the 11-plane z-stack covering 5 . 5 μm was sum projected and an integrated density of a defined area ( 18 μm2 ( GFP ) or 16 μm2 ( GFP-NLS ) was measured over time from the region of interest ( ROI ) and the neighboring background region ( BG ) . To distinguish between cells that contained an aggregate from those that did not contain an aggregate , cells were categorized based on the first frame Hsp104-mCherry signal into the two respective groups . The ROI values were background subtracted [ROI ( t ) -BG ( t ) ] and normalized to the first value [ROI ( tx ) /ROI ( t1 ) ] . The average of the pooled values was fitted with Prism5 software using non-linear one-phase decay . For the FRAP analysis , raw data were background subtracted , corrected for acquisition-induced bleaching and normalized , and the curve was fitted from pooled values with Prism5 software using one-phase association non-linear fitting . The FRAP data where the aggregate structure was lost from the focal plane during recovery period image acquisition were discarded . Lifespan of virgin daughters was analyzed on YPD plates using Zeiss Axioscope 40 microdissection microscope as previously described in ( Denoth-Lippuner et al . , 2014 ) . All statistical analyses and graph preparations were done with Prism5 software . The error bars represent ±SEM from experimental triplicates with independent clones and statistical analyses were conducted with one-way ANOVA using Newman–Keuls post test , t-test , or Gehan-Breslow-Wilcoxon test .

Aging is a complex process . Studies involving a single-celled organism called budding yeast are commonly used to investigate the factors that contribute to aging . When these yeast cells divide , a small daughter cell buds out from a large mother cell . A mother cell has a limited lifespan and produces a finite number of daughter cells and then dies ( a phenomenon referred to ‘replicative aging’ ) . However , when a daughter cell forms , the daughter's age is reset to zero , giving it the full potential to produce new offspring . Previous research on budding yeast has shown that damaged or aggregated proteins accumulate in the mother cells but not the daughter cells , and that this accumulation of proteins contributes to shortening the lifespan of the mother cell . Furthermore , protein aggregation has also been associated with a number of age-related diseases in humans , including neurodegenerative disorders such as Alzheimer's and Parkinson's disease . However , it remains unclear how cells respond to protein aggregation that occurs during aging . Many studies that have previously investigated this question have relied on exposing cells to stressful conditions , such as high temperatures , in order to trigger proteins to aggregate . But now , Saarikangas and Barral have studied how proteins aggregate under normal , unstressed conditions in budding yeast as they age . The experiments revealed that most unstressed yeast cells develop a single deposit of aggregated proteins already during early aging . This age-associated structure proved to be a different type of response than the protein aggregation that occurs during stress . Furthermore , the deposit did not form as a consequence of the cell having obvious problems with folding its proteins , nor did the deposit hinder cells from coping with stressful conditions that trigger protein misfolding . Rather , this deposit supported the ability of the cell to degrade defective proteins . This suggests that the deposit represents an early adaptive response to aging , which might consequently provide aged cells some advantage over their younger counterparts . Saarikangas and Barral also found that this protein deposit was always retained in the mother cell and not passed onto the daughters at cell division . Further experiments showed that an enzyme called heat shock protein 42 was responsible for collecting target proteins and bring them together to form the single deposit . Reducing the levels of this enzyme prevented the deposit from forming and extended the lifespan of the mother cells . Thus , these findings suggest that mother cells collect harmful protein aggregates into a single deposit and prevent them from entering the daughter cells . Further work is needed to understand how the deposit is preferentially retained within the mother cell .

2015

Protein aggregates are associated with replicative aging without compromising protein quality control

Autophagy is required for the homeostasis of cellular material and is proposed to be involved in many aspects of health . Defects in the autophagy pathway have been observed in neurodegenerative disorders; however , no genetically-inherited pathogenic mutations in any of the core autophagy-related ( ATG ) genes have been reported in human patients to date . We identified a homozygous missense mutation , changing a conserved amino acid , in ATG5 in two siblings with congenital ataxia , mental retardation , and developmental delay . The subjects' cells display a decrease in autophagy flux and defects in conjugation of ATG12 to ATG5 . The homologous mutation in yeast demonstrates a 30-50% reduction of induced autophagy . Flies in which Atg5 is substituted with the mutant human ATG5 exhibit severe movement disorder , in contrast to flies expressing the wild-type human protein . Our results demonstrate the critical role of autophagy in preventing neurological diseases and maintaining neuronal health . Macroautophagy , referred to hereafter as autophagy , is a cellular process by which proteins and organelles are degraded and recycled through sequestration within autophagosomes and delivery to lysosomes ( Levine and Klionsky , 2004 ) . The autophagy pathway is highly conserved and required for organismal development and function . Defects in autophagy are associated with diseases including cancer , metabolic disruption , and neurodegenerative disorder ( Choi et al . , 2013; Cuervo and Wong , 2014; Frake et al . , 2015 ) . Patients with mutations in any of the non-redundant core autophagy-related ( ATG ) genes have not previously been reported . Ataxia is a neurodegenerative disease caused by disruption of the cerebellum and Purkinje cells , which results in the lack of coordinated muscle movements . Large phenotype diversity is present in individuals with ataxia , including age of onset , rate of progression , and other accompanying neurological and non-neurological features ( Jayadev and Bird , 2013 ) , with corresponding genotypic heterogeneity ( Sandford and Burmeister , 2014 ) . Even within the more defined phenotype of childhood ataxia with developmental delay , there are a large number of associated genes , such that similar phenotypic features alone are often insufficient information for an accurate diagnosis ( Burns et al . , 2014; De Michele and Filla , 2012; Jayadev and Bird , 2013 ) . Identification of genetic causes of childhood ataxia is important for understanding disease pathogenesis and for possible future treatment development . Whole exome sequencing has been successfully utilized to identify known and novel genetic mutations responsible for ataxia ( Burns et al . , 2014; Fogel et al . , 2014 ) . Identification of candidate genes can be further verified through additional molecular analysis and utilization of specific and general animal models . Here we identified a novel mutation in a core autophagy gene , ATG5 , in two children with ataxia , and demonstrate a reduction in autophagic response , also reproducing the phenotype in yeast and fly models . Two Turkish siblings presented with ataxia and developmental delay in childhood , as previously described ( Yapici and Eraksoy , 2005 ) . We performed linkage analysis on both affected siblings , their unaffected siblings , and their unaffected mother , using a model of remote parental consanguinity and identified a single broad ( >14 Mbp ) peak with LOD score 3 . 16 on chromosome 6q21 , between 102 and 116 Mb ( Figure 1 ) . Whole exome sequencing identified a homozygous missense mutation , hg19 chr6:106 , 727 , 648 T>A , corresponding to E122D in ATG5 ( Figure 2A ) as the only damaging mutation within the genetically identified chromosomal linkage interval . The mutation was Sanger verified and found absent from variant databases and from Turkish controls . 10 . 7554/eLife . 12245 . 003Figure 1 . Linkage analysis in consanguineous family with two siblings with ataxia , mental retardation and developmental delay maps defect to chromosomal interval containing ATG5 . Remote consanguinity was detected between parents of two previously described siblings having ataxia ( Yapici and Eraksoy , 2005 ) , illustrated here as third cousins . SNP and linkage results for chromosome 6 ( B ) are illustrated below the pedigree ( A ) . The shared homozygous region lies between rs4334996 and rs1204817 , encompassing ATG5 at 106 . 6 Mb . Father ( 501 ) ’s alleles were inferred , 0 denotes unknown alleles . Affected siblings , 601 and 602 , are denoted by black squares and unaffected family members by open symbols . The proximal boundary is defined by a recombination event between rs1547384 and rs4334996 in affected individual 602 , while the distal boundary is defined as an ancestral recombination event ( lack of homozygosity , dark green ) between rs1204817 and rs648248 . Orange arrows indicate the position of ATG5 . DOI: http://dx . doi . org/10 . 7554/eLife . 12245 . 00310 . 7554/eLife . 12245 . 004Figure 2 . The primary sequence of ATG5 , including the mutant E122 residue , as well as the protein structure is highly conserved across eukaryotic species . ( A ) Amino acid sequence alignment between ATG5 orthologs from human ( HsaATG5 ) , mouse ( MmuAtg5 ) , Drosophila melanogaster ( DmeAtg5 ) and Saccharomyces cerevisiae ( SceAtg5 ) was constructed at GenomeNet ( Kyoto University Bioinformatics Center ) through CLUSTALW and rendered in Genedoc v . 2 . 7 using default settings . E122 in human ATG5 and E141 in yeast Atg5 , which are homologous residues , are indicated by red arrows . ( B ) Location of E122 residue is highlighted in yellow on the crystal structure of a human ATG12 ( residues 53–140 ) –ATG5 -ATG16L1 ( residues 11–43 ) complex ( PDB ID: 4NAW ) . ATG5 is shown in cyan , ATG16L1 in magenta , and ATG12 in green ( Otomo et al . , 2013 ) . ( C ) Location of the E141 residue in yeast Atg5 , which corresponds to the E122 in human ATG5 , is indicated in yellow on the crystal structure of a yeast Atg12 ( 100–186 ) –Atg5 -Atg16 ( 1–46 ) complex , colored as for the human counterparts as in panel B ( PDB ID: 3W1S ) ( Noda et al . , 2013 ) . ( D ) Superimposition of crystal structure of ATG5E122D-ATG16L1 with ATG5WT-ATG16L1 ( PDB: 4TQ0 ) ( Kim et al . , 2015a ) . Close-up view of ATG5 structure around WT ( E ) and E122D ( F ) mutation . DOI: http://dx . doi . org/10 . 7554/eLife . 12245 . 004 ATG5 plays a role in elongation of the phagophore and its subsequent maturation into the complete autophagosome . ATG12 is a ubiquitin-like protein that covalently binds ATG5 ( Mizushima et al . , 1998a ) , and this conjugate noncovalently binds ATG16L1 . Crystal structure of the resulting ATG12–ATG5-ATG16L1 complex indicated that E122 is located in the vicinity of the ATG12–ATG5 interaction surface ( Figure 2B ) ; hence , we predicted that the mutation in ATG5 could affect the conjugation of ATG12 . Comparison of protein isolated from control lymphoblastoid cell lines ( LCL ) and of affected subjects revealed a severe reduction of the ATG12–ATG5 conjugate in the mutant cells under basal conditions ( Figure 3A ) , suggesting that the E122D mutation may have impaired autophagy by inhibiting conjugation between ATG12 and ATG5 . 10 . 7554/eLife . 12245 . 005Figure 3 . Cells from ataxia patients with ATG5E122D/E122D mutation exhibit autophagy defects . ( A ) Decreased expression of ATG12–ATG5 conjugates in cells from ataxia patients with ATG5E122D/E122D mutation . ATG5 immunoblotting ( IB ) of ATG12–ATG5 conjugates of LCLs from individuals whose ATG5 genotype corresponds to wild type ( A to I ) or E122D ( J and K ) . ( B ) Decreased autophagic flux in ATG5E122D/E122D LCL cells . A subset of LCLs from ( A ) were treated with 0 . 1 μM bafilomycin A1 ( Baf ) for the indicated hours and analyzed by IB . LC3-II is an autophagosome marker , and LC3-I is a precursor for LC3-II . Baf inhibits lysosomal degradation of LC3-II . Actin is shown as a loading control . ( C ) Decreased autophagic flux and increased expression of SQSTM1 , an autophagy substrate , in ATG5E122D/E122D LCL cells . A subset of LCLs from ( A ) were treated with 250 nM Torin 1 or 0 . 1 μM Baf , for 2 hr and analyzed by IB . Torin 1 is an autophagic flux activator . DOI: http://dx . doi . org/10 . 7554/eLife . 12245 . 005 The ATG12–ATG5-ATG16L1 complex functions in part as an E3 ligase to facilitate the conjugation of LC3 to phosphatidylethanolamine , generating LC3-II ( Fujita et al . , 2008; Hanada et al . , 2007 ) . Compared to control cells , LCLs from patients with the E122D mutation exposed to bafilomycin A1 showed a substantial reduction in LC3-II accumulation under basal conditions ( Figure 3B ) , suggesting a possible decrease in E3 activity and subsequent attenuation of basal autophagic flux . The patient LCLs were also unable to upregulate their autophagic flux in response to Torin 1 ( Figure 3C ) , which is a strong inducer of autophagy ( Thoreen et al . , 2009 ) . ATG5E122D LCLs also showed elevated levels of SQSTM1/p62 , an autophagy receptor and substrate , further indicating disruption of basal autophagy ( Figure 3C ) . To examine the effect of the ATG5E122D mutation on formation of the ATG12–ATG5-ATG16L1 complex , we expressed the recombinant human proteins in insect Hi5 cells and analyzed the complexes by affinity isolation . We could detect the ATG12–ATG5 complex when both wild-type proteins were co-expressed , but we could only detect a minimal amount of the ATG12–ATG5E122D complex ( Figure 4A ) . Although overexpression of human ATG5WT in HEK293 cells or Drosophila tissues resulted in efficient covalent conjugation with overexpressed human ATG12 ( Figure 4B and C ) or endogenous Drosophila Atg12 ( Figure 4D ) , mutant ATG5E122D was dramatically impaired in this process ( Figure 4B , C and D ) . Interestingly , expression levels of ATG5WT and ATG5E122D monomers were comparable to each other , indicating that the mutation affects the conjugation process , rather than the stability of proteins . This was consistent with the structural location of ATG5 E122 adjacent to the surface that interacts with ATG12 ( Figure 2B ) . To confirm that the mutation does not overtly alter the structure of ATG5 or binding to ATG16L1 , we analyzed formation of the noncovalent ATG5-ATG16L1 complex using constructs containing a TEV protease site . Both wild-type and mutant ATG5 protein were efficiently co-precipitated with ATG16L1 ( Figure 4E ) . Indeed , the co-crystal structure of a human ATG5E122D-ATG16L1 complex ( Figure 2 and Table 1 ) superimposes well with the previously determined structure of the WT proteins ( Figure 2D ) , with the major obvious difference being replacement of the side-chain ( Figure 2E and F ) . Thus , it appears that the E122D mutation interferes with the ATG12–ATG5 conjugation process , but not with ATG5 folding or binding of ATG16L1 . 10 . 7554/eLife . 12245 . 006Figure 4 . E122D mutation interferes with formation of the ATG12–ATG5 conjugate . ( A ) Coomassie Blue-stained SDS-PAGE gel following glutathione affinity purification from lysates of Hi5 cells infected with baculoviruses expressing GST-ATG12 and either WT or E122D mutant ATG5 . ( B and C ) HEK293 cells expressing the indicated proteins were analyzed by IB . ( D ) Drosophila whole bodies expressing the indicated transgenes under the control of Tub-Gal4 were analyzed by IB . ( E ) Lysates from Hi5 cells expressing the indicated proteins were subjected to His/Ni-NTA purification and subsequent TEV protease treatment . Proteins were analyzed by Coomassie Blue staining . DOI: http://dx . doi . org/10 . 7554/eLife . 12245 . 00610 . 7554/eLife . 12245 . 007Table 1 . Crystallography data collection and refinement statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 12245 . 007Data collectionBeam lineAPS 24-ID-CSpace groupC2Unit cell parametersa , b , c ( Å ) 217 . 1 , 84 . 5 , 151 . 9α , β , γ ( ° ) 90 , 133 . 8 , 90Resolution ( Å ) ( highest shell ) 50–3 . 0 ( 3 . 12-3 . 0 ) Wavelength ( Å ) 0 . 9792Number of measured reflections179 , 310Number of unique refections39 , 496Overall Rsym0 . 057 ( 0 . 645 ) Completeness ( % ) 98 . 9 ( 99 . 3 ) Overall I/σI14 . 3 ( 2 . 3 ) Multiplicity4 . 5RefinementResolution ( Å ) 50–3 . 0Rwork/Rfree0 . 198/0 . 244rmsd bond lengths ( Å ) 0 . 008rmsd bond angles ( ° ) 0 . 994Number of protein atoms9403Ramachandran statisticsPreferred ( % ) 97 . 69Allowed ( % ) 2 . 22Disallowed ( % ) 0 . 09 ATG5 is a highly conserved protein , and sequence alignment demonstrated that E122 corresponds to yeast E141 ( Figure 2A and C ) . We extended our analysis of the effect of the mutation on autophagy activity , by taking advantage of the yeast system . To test whether autophagy was affected by the Atg5 mutation in yeast , we initially relied on the GFP-Atg8 processing assay ( Shintani and Klionsky , 2004 ) . During autophagy a population of Atg8 is continuously transported to the vacuole inside of autophagosomes . Tagging the N terminus of Atg8 with GFP makes it possible to monitor autophagy flux because Atg8 is rapidly degraded inside the vacuole whereas GFP is relatively resistant to vacuolar hydrolases; the generation of free GFP is an indication of autophagic activity . We observed a consistent decrease in autophagy activity with the Atg5E141D mutant relative to Atg5WT following autophagy induction by starvation ( Figure 5A ) . Atg8 , or GFP-Atg8 , does not measure autophagic cargo per se ( Klionsky , 2016 ) , and the amount of GFP-Atg8 processing only corresponds to the inner surface of the autophagosome . To corroborate the effects observed through the GFP-Atg8 processing assay we examined autophagy using the quantitative Pho8∆60 assay ( Noda and Klionsky , 2008 ) . Pho8∆60 is an altered form of a phosphatase that is only delivered to the vacuole via autophagy; subsequent proteolytic processing generates an active form of the hydrolase . After 4 and 6 hr of starvation , yeast cells expressing the plasmid-based Atg5E141D mutant showed a significant decrease in autophagy levels compared to cells expressing Atg5WT ( Figure 5B ) , and similar results were obtained when the WT and mutant ATG5 genes were integrated back into the chromosomal ATG5 locus ( Figure 5C ) . 10 . 7554/eLife . 12245 . 008Figure 5 . E141D mutation of yeast Atg5 attenuates autophagy . ( A-D ) Yeast cells were grown in SMD to mid-log phase and nitrogen starved for the indicated times . ( A ) WLY176 atg5∆ yeast cells expressed GFP-Atg8 through its endogenous promoter and plasmid-based Atg5WT-PA , Atg5E141D-PA or an empty vector . Protein extracts were analyzed for GFP-Atg8 processing by western blot . The ratio of free GFP to Dpm1 ( loading control ) is presented below the blots , and quantification is presented on the right ( Student’s t test , n=4; *p < 0 . 05 ) ; the value for Atg5WT at 6 hr was set to 1 . 0 and other values were normalized . ( B ) WLY176 atg5∆ yeast cells expressed either plasmid-based Atg5WT-PA , Atg5E141D-PA or an empty vector . Protein extracts were used to measure autophagy through the Pho8Δ60 assay ( Student’s t test , n=6; *p < 0 . 05 ) . ( C ) WLY176 cells with genomic integrated Atg5WT or Atg5E141D were used to generate protein extracts and autophagy was monitored through the Pho8Δ60 assay ( Student’s t test , n=3; *p < 0 . 05 ) . ( D ) WLY176 atg5∆ yeast cells expressing plasmid-based Atg5WT-PA , Atg5E141D-PA or an empty vector were used to generate protein extracts . The ratio of Atg8–PE to total Atg8 is presented below the blots based on western blot using antiserum to Atg8 . Dpm1 was used as a loading control . ( E ) MKO ATG3 ( YCY137 ) cells were co-transformed with pATG8∆R-ATG7-ATG10 ( 414 ) , and either pATG5WT-HA-ATG12 ( 416 ) , pATG5E141D-HA-ATG12 ( 416 ) , pATG5WT-HA-ATG12-ATG16 ( 416 ) , or pATG5E141D-HA-ATG12-ATG16 ( 416 ) . Overnight cultures were diluted to OD=0 . 02 in SMD -Ura -Trp . The cells were incubated at 30°C for 18 hr to mid-log phase before they were shifted to SD-N for nitrogen starvation . Samples at the corresponding time points were collected , TCA precipitated and subsequently analyzed by western blot . S . E . , short exposure; L . E . , long exposure . DOI: http://dx . doi . org/10 . 7554/eLife . 12245 . 008 To determine the reason for reduced autophagic activity we tested the effects of the Atg5E141D mutant on Atg8 lipidation . As shown by the ratio of Atg8–PE:total Atg8 , cells expressing Atg5E141D displayed a decrease in Atg8–PE conjugation at 30 and 60 min of starvation compared to cells expressing Atg5WT ( Figure 5D ) . We extended this analysis using the in vivo reconstitution of Atg8–PE conjugation as described previously ( Cao et al . , 2008 ) . In brief , we examined Atg8 lipidation in a multiple-knockout ( MKO ) strain in which 23 ATG genes are deleted , when expressing only the E1 , E2 and E3-like conjugation enzymes of the autophagy machinery . Atg8∆R that lacks the C-terminal arginine was used in the assay to bypass the initial activation step initiated by Atg4; due to the absence of Atg4 , there is no cleavage of Atg8–PE from the membrane , resulting in stabilization of this form of the protein . We found that the Atg5E141D mutant was significantly defective in Atg8–PE conjugation compared to the cells with Atg5WT when Atg16 was not present ( Figure 5E ) . Atg16 is not required mechanistically for Atg8 conjugation , but its presence increases the efficiency of this process and may dictate the site of conjugation ( Cao et al . , 2008; Hanada et al . , 2007 ) . Thus , the presence of Atg16 may partially mask the Atg8 lipidation defects of the Atg5E141D mutant , and this may explain why the E122D/E141D mutation induces a hypomorphic rather than a complete null phenotype . To further characterize the effect of the E122D mutation on the development of ataxia , we generated Drosophila melanogaster knockouts for Atg5 ( Figure 6A ) , and reconstituted the Atg5-null mutant flies with transgenes expressing wild-type ( WT ) or E122D human ATG5 ( Figure 6B-D ) . Unlike mouse models , Atg5-null flies are viable , although they exhibit severe mobility defects after adult eclosion as demonstrated by a negative geotaxis assay ( Figure 6E and I , and Video 1 ) , similar to Atg7 null mutant flies ( Juhasz et al . , 2007 ) . These mobility defects were substantially restored by expression of ATG5WT ( Figure 6F and I , and Video 2 ) , suggesting that the molecular function of ATG5 is conserved between human and Drosophila . However , Atg5-null mutant flies expressing ATG5E122D were still defective in mobility although slightly better than Atg5-null controls ( Figure 6G–I , and Videos 3 and 4 ) , demonstrating again that ATG5 activity is compromised but not eliminated by the E122D mutation . ATG5E122D was also inferior to ATG5WT in suppressing Ref ( 2 ) P ( fly p62/SQSTM1 ) accumulation ( Figure 6J and K ) and cell death ( Figure 6L and M ) in the brain of Atg5-null mutant flies . 10 . 7554/eLife . 12245 . 009Figure 6 . Ataxic phenotype of Atg5-null flies is suppressed by human ATG5WT but not by ATG5E122D . ( A ) Genomic organization of the Atg5 locus and the Atg5-null mutant ( Atg55cc5 ) . Atg55cc5 mutants have a CRISPR-Cas9-mediated deletion in approximately 1 . 5 kb residues that eliminate more than 85% of Atg5-coding sequences including the translation start site . Open boxes , untranslated exons; closed boxes , protein-coding exons . Scale bar , relative length of 1 kb genomic span . ( B ) Schematic representation of how ATG5 transgenic flies were made . Plasmid which can express wild-type or E122D-mutated human ATG5 was inserted into an identical genomic location ( the attP site ) through phiC31-mediated recombination ( Bateman et al . , 2006; Bischof et al . , 2007; Venken et al . , 2006 ) . The scheme was adapted from a previous publication ( Kim and Lee , 2015 ) . ( C ) Genetic scheme of how ATG5 transgenes were placed into the Atg5-null mutant flies . Atg5 , UAS-ATG5 and Tub-Gal4 loci are on the X-chromosome , second chromosome and third chromosome , respectively . ( D ) Whole flies of indicated genotypes were analyzed by IB . ( E to H ) Photographs of the vials containing 2-week-old adult male flies of indicated genotypes taken at 3 sec after negative geotaxis induction: ( E ) Atg5-null flies exhibit severely impaired mobility . ( F ) Ataxic phenotype of Atg5-null flies is complemented by human ATG5WT expression . ( G and H ) Human ATG5E122D is less capable than human ATG5WT in suppressing the fly ataxia phenotype . ( I ) Quantification of the climbing speeds of 2-week-old adult male flies ( n≥20 ) of the indicated genotype . Climbing speed is presented as mean ± standard deviation ( n=5 ) . P values were calculated using the Student’s t test ( ***p<0 . 001 ) . ( J ) Drosophila heads from two-weeks-old flies of the indicated genotypes were analyzed by IB . ( K ) Ref ( 2 ) P [p62] is an autophagy substrate . Relative protein expression was measured by densitometry and presented in a bar graph ( mean ± standard error; n=4 ) . ( L ) Terminal deoxynucleotidyl transferase dUTP nick end labeling ( TUNEL ) of Drosophila brain ( middle layer of the medial compartment ) . ( M ) TUNEL-positive cells per field were quantified and presented in a bar graph ( mean ± standard error; n≥5 ) . K and M: P values were calculated using the Student’s t test ( *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12245 . 00910 . 7554/eLife . 12245 . 010Video 1 . Climbing assay in 2 weeks-old wild-type flies ( left ) and Atg5-null flies ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12245 . 01010 . 7554/eLife . 12245 . 011Video 2 . Climbing assay in 2 weeks-old Atg5-null flies ( left ) and Atg5-null flies expressing ATG5WT ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12245 . 01110 . 7554/eLife . 12245 . 012Video 3 . Climbing assay in 2 weeks-old Atg5-null flies ( left ) and Atg5-null flies expressing ATG5E122D ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12245 . 01210 . 7554/eLife . 12245 . 013Video 4 . Climbing assay in 2 weeks-old Atg5-null flies expressing ATG5WT ( left ) or ATG5E122D ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12245 . 013 In summary , we demonstrate that the homozygous E122D mutation of ATG5 , a unique mutation found in two human subjects with ataxia , results in reduced conjugation to ATG12 and in an overall decrease in autophagy activity . The homologous mutation in yeast also interferes with autophagy , and the ataxia phenotype was replicated in a fly model . Based on these results we propose that this ATG5 mutation , and the consequent disruption in autophagy activity , is the cause of the ataxic phenotype and disturbance of the cerebellum in the affected siblings . This hypothesis is in agreement with previously characterized mouse models , in which neuron-specific knockout of Atg5 results in ataxia-like phenotypes ( Hara et al . , 2006; Nishiyama et al . , 2007 ) . By contrast , mice with complete knockout of Atg5 die shortly after birth , demonstrating that autophagy is essential for mammalian survival ( Kuma et al . , 2004 ) . Our results indicate that E122D is a partial loss-of-function allele that impairs but does not completely abolish ATG5 activity . Although the overall structure of the E122D mutant ATG5 superimposes well with the wild-type protein , the mutation causes a striking decrease in the level of ATG12–ATG5 conjugate that is formed when the C terminus of ATG12 is covalently linked to Lys130 of ATG5 . We speculate that the E122D mutation causes subtle changes in the conformational dynamics that propagate to Lys130 , which is less than 10 Å away , resulting in less ATG12–ATG5 , which in turn leads to reduced LC3/Atg8 conjugation . In neurons , which are among the cells most dependent on autophagy for tissue homeostasis ( Button et al . , 2015 ) , the residual function of the E122D allele is inadequate , resulting in predominantly neurological symptoms in the two patients . Since homozygous mutations with complete loss-of-function have not been reported , we predict that individuals carrying such mutations , similar to Atg5-null mice ( Kuma et al . , 2004 ) , might not be viable . Autophagy is quickly gaining importance for its roles in preventing neurodegeneration . WDR45 is a redundant , non-core autophagy gene , one of four mammalian homologs to Atg18 , and mutations in WDR45 cause SENDA , static encephalopathy of childhood with neurodegeneration in adulthood ( Haack et al . , 2012; Saitsu et al . , 2013 ) . Autophagy appears to be critical in ataxia , whether mutant proteins evade autophagy processes or normal autophagy is disrupted . Several ataxias are attributed to intranuclear or cytoplasmic aggregation of mutant proteins within the cell ( Matilla-Duenas et al . , 2014 ) . These protein aggregates , in humans and in mouse models , not only evade autophagic sequestration but may even inhibit autophagy ( Alves et al . , 2014 ) , or lead to reduction in autophagy available for other proteins due to saturation . Assessment of autophagy in patient cells may be used to refine and identify the genetic cause of a patient’s ataxia . Further discovery of the role of autophagy in neurodegenerative diseases should be used to investigate therapies targeted at the autophagy process . Many drugs enhance autophagy and their effects on a multitude of neurodegenerative diseases have been studied ( Sarkar et al . , 2009 ) . Recently more studies have been conducted assessing the value of autophagy enhancers in ataxia models and patients . Induction of autophagy through administration of Temsirolimus , a rapamycin ester , and lentiviral overexpression of BECN1 in SCA3 model mice increase autophagy and the clearance of mutant protein aggregates , and reduce the ataxic phenotype ( Menzies et al . , 2010; Nascimento-Ferreira et al . , 2013 ) . In a single patient trial , trehalose treatment of patient fibroblasts increased autophagy and alleviated cellular pathogenic features by improving mitochondrial morphology and reducing free radicals in the cell ( Casarejos et al . , 2014; Sarkar et al . , 2007 ) . Trehalose also showed success in trials involving models of SCA17 ( Chen et al . , 2015 ) . Lithium , another inducer of autophagy , improved symptoms in a SCA1 mouse model ( Watase et al . , 2007 ) , but did not slow or reduce symptoms in a treatment trial in SCA2 patients ( Sacca et al . , 2015 ) . An autophagy enhancer may be an appropriate treatment to test in the presented subjects , as autophagic flux is attenuated , but not completely abrogated , by the ATG5E122D/E122D mutation . This study’s finding of the pathogenic human E122D mutation in ATG5 , a gene encoding part of the autophagy-controlling core machinery , is important and novel , but consistent with reports of neurodegenerative disorders in other autophagy-related genes ( Frake et al . , 2015 ) . Our results suggest that other mutations in this and other ATG genes , which impair but do not completely abolish autophagy , may result in similar forms of ataxia , intellectual disability and developmental delay . This study exemplifies the utility of exome sequencing in the identification of rare disease-causing variants , and supports the role of impaired autophagy in neurodegenerative disease . In addition , we demonstrate the utility of a combined genetic , biochemical and cell biological analysis in multiple model systems to elucidate the underlying pathogenic mechanism of rare human diseases . Study protocols including written informed consents have been approved by the University of Michigan Institutional Review Board and the Boğaziçi University Institutional Review Board for Research with Human Participants . Two Turkish brothers , ages 5 and 7 in 2004 , presented with ataxia and developmental delay , as previously described ( Yapici and Eraksoy , 2005 ) . Parents were initially reported to be unrelated , but recently suggested they might be remotely related . Both patients were delayed in walking , had truncal ataxia and dysmetria , nystagmus , and lower IQ ( 68 and 70 ) . MRI revealed cerebellar hypoplasia . Follow-up examinations showed no progression of symptoms . DNA was isolated from peripheral whole blood using the Qiagen ( Germantown , MD ) Gentra Puregene isolation kit . Linkage analysis was performed using the genotype data generated with Illumina HumanOmniExpress-24 chip for the mother and the four sibs . The Allegro module ( Gudbjartsson et al . , 2000 ) of easyLINKAGE software was used , assuming autosomal recessive inheritance and parents as third cousins . No deletion or duplications common to just the two affected brothers were detected using cnvPartition plug-in in Illumina Genome Studio v . 1 . 02 software . Exome sequencing was performed independently twice on one subject . Capture for whole exome sequencing was performed with NimbleGen SeqCap EZ Exome Library v1 . 0 kit ( Roche , Indianapolis , IN ) . Captured regions were sequenced with Illumina HiSeq2000 instruments . Variants were filtered to remove common variants based on 1000 Genomes , Exome Sequencing Project , and Exome Aggregation Consortium databases , variants outside of identified linkage regions , variants not expected to change protein coding , and variants not following a recessive model of inheritance ( Exome Aggregation Consortium ( ExAC ) , 2015; Genomes Project Consortium et al . , 2012; NHLBI Go Exome Sequencing Project , 2015 ) . PCR followed by Sanger sequencing was performed to validate the variant identified through exome sequencing and test for segregation within the family . The variant of interest was further examined in two separate collections of a total of 500 Turkish samples , and found absent . Lymphoblastoid cell lines ( LCL ) of both subjects were generated from heparinized whole blood samples and cultured as described ( Doyle , 1990 ) . As they are made in house and cultured briefly , mycoplasma contamination risk is minimized . We used a baculovirus/insect cell expression system to examine formation of the human ATG12–ATG5 conjugate in a heterologous system described previously ( Qiu et al . , 2013 ) . Hi5 insect cells ( Invitrogen , Carlsbad , CA ) were infected with baculoviruses expressing human ATG7 , ATG10 , a GST-tagged version of ATG12 ( residues 53–140 , corresponding to the ubiquitin-like domain ) and a His-tagged version of either ATG5WT or ATG5E122D . Three days post infection , lysates were subjected to glutathione affinity chromatography , and the GST-ATG12–His-ATG5 conjugate was detected by SDS-PAGE followed by Coomassie Blue staining . To confirm that the baculoviruses produce protein , Hi5 cells were coinfected with baculoviruses expressing the His-tagged WT and mutant versions of ATG5 and the N-terminal domain of ATG16L1 ( residues 1–69 , as an MBP fusion ) . At three days post infection , lysates were subjected to nickel affinity purification . The ATG5-ATG16L1 complex formation was detected by SDS-PAGE and Coomassie Blue staining . The complex containing ATG5E122D and the N-terminal domain of ATG16L1 ( residues 1–69 ) was expressed in Hi5 insect cells , and purified by nickel affinity , ion exchange , and size exclusion chromatography into a final buffer of 20 mM Tris , pH 8 . 5 , 50 mM NaCl , 10 mM DTT . The complex was concentrated to 18 . 5 mg/ml , aliquoted , flash-frozen and stored at -80°C until further use . Crystals were grown by the hanging drop vapor diffusion method by mixing purified protein 1:1 with reservoir solutions of 37 . 5 mM MES , pH 5 . 2–5 . 8 , 0 . 2 M sodium tartrate , and 11–13% polyethylene glycol 3350 . Final crystals were obtained by micro-seeding with reservoir solution of 40 mM MES , pH 5 . 5 , 0 . 2 M sodium tartrate , 8 . 5% PEG3350 , 10 mM DTT . Crystals were cryoprotected in reservoir solution supplemented with 25% xylitol , and flash frozen in liquid nitrogen prior to data collection . Diffraction data were processed with XDS . The structure was determined by molecular replacement using Phaser ( McCoy et al . , 2007 ) with the structure of the WT ATG5-ATG16L1 ( 1–69 ) ( PDB: 4TQ0 ) complex as a search model ( Kim et al . , 2015a ) . Model construction and rebuilding were performed using Coot ( Emsley et al . , 2010 ) . The structure was refined using Phenix ( Adams et al . , 2010 ) . Diffraction data and refinement statistics are provided in Table 1 . Cells or tissues were lysed in cell lysis buffer ( 20 mM Tris-HCl , pH 7 . 5 , 150 mM NaCl , 1 mM EDTA , 1 mM EGTA , 2 . 5 mM sodium pyrophosphate , 1 mM beta-glycerophosphate , 1 mM Na3VO4 , 1% Triton X-100 ) or RIPA buffer ( 50 mM Tris-HCl , pH 7 . 4 , 150 mM NaCl , 1% sodium deoxycholate , 1% NP-40 , 0 . 1% SDS ) containing protease inhibitor cocktail ( Roche ) . After being clarified with centrifugation , lysates were boiled in SDS sample buffer , separated by SDS-PAGE , transferred to polyvinylidene difluoride membranes and probed with the indicated antibodies . ATG5 ( 12994 ) , LC3 ( 3868 ) and SQSTM1/p62 ( 5114 ) antibodies were purchased from Cell Signaling Technology . Hemagglutinin ( HA , 3F10 ) antibody was from Roche . Actin ( JLA20 ) and tubulin ( T5168 ) antibodies were from Developmental Studies Hybridoma Bank and Sigma , respectively . Ref ( 2 ) P antibody was previously described ( Pircs et al . , 2012 ) . Wild-type human ATG5-coding sequence was from Addgene #24922 ( deposited by Dr . Toren Finkel ) ( Lee et al . , 2008 ) . The E122D mutation was introduced into ATG5 by PCR-based site-directed mutagenesis . ATG5WT and ATG5E122D were cloned into the plasmid pLU-CMV-Flag . The HA-ATG12-expressing plasmid was from Addgene #22950 ( deposited by Dr . Noboru Mizushima ) ( Mizushima et al . , 1998b ) . HEK293 cells ( the 293 A substrain from Invitrogen , tested negative for mycoplasma by PCR ) were cultured in Dulbecco’s modified Eagle’s medium ( DMEM , Invitrogen ) containing 10% fetal bovine serum ( FBS ) and penicillin/streptomycin at 37°C in 5% CO2 . For transient expression of proteins , HEK293 cells were transfected with purified plasmid constructs and polyethylenimine ( PEI , Sigma ) as previously described ( Horbinski et al . , 2001 ) . Cells were harvested 24 hr after transfection for immunoblot analyses . Saccharomyces cerevisiae strain WLY176 was used to generate an ATG5 knockout strain ( atg5∆ ) as previously described ( Gueldener et al . , 2002 ) . The MKO strain YCY137 ( SEY6210 atg1∆ , 2∆ , 4∆ , 5∆ , 6∆ , 7∆ , 8∆ , 9∆ , 10∆ , 11∆ , 12∆ , 13∆ , 14∆ , 16∆ , 17∆ , 18∆ , 19∆ , 20∆ , 21∆ , 23∆ , 24∆ , 27∆ , 29∆ ) ( Cao et al . , 2008 ) , was used for in vivo reconstitution of Atg8 conjugation . Site-directed mutagenesis was performed to generate ATG5 amplicons with the E141D mutation as previously described ( Liu and Naismith , 2008 ) . A pRS406 empty plasmid was digested with Spel and SalI , and then ligated with a DNA fragment encoding either wild-type or mutant Atg5-PA . atg5∆ was transformed with an empty pRS406 vector , or plasmids encoding Atg5-PA WT or Atg5-PA E141D . Wild-type WLY176 colonies were transformed with empty pRS406 vector as a control . Colonies were grown on SMD-URA medium and starved in nitrogen-deficient medium . Pho8Δ60 and western blot analyses were performed as described previously ( Noda and Klionsky , 2008; Shintani and Klionsky , 2004 ) . Quantification was performed using ImageJ software . The pATG8∆R-ATG7-ATG10 ( 414 ) , pATG5 ( WT ) -HA-ATG12 ( 416 ) and pATG5 ( WT ) -HA-ATG12-ATG16 ( 416 ) plasmids were described previously ( Cao et al . , 2008 ) . The pATG5 ( E141D ) -HA-ATG12 ( 416 ) and pATG5 ( E141D ) -HA-ATG12-ATG16 ( 416 ) plasmids were made by site-directed mutagenesis based on the wild-type constructs . Atg5-null Drosophila flies ( Atg55cc5 ) were generated by CRISPR-Cas9-mediated genome editing , using a double gRNA approach , both targeting the same gene , as described ( Kondo and Ueda , 2013 ) . The Atg55cc5 mutant was recovered by screening viable candidate lines for accumulation of the specific autophagy cargo Ref ( 2 ) P using western blots , followed by PCR and sequencing . Atg55cc5 mutants have a deletion in X:7 , 322 , 242–7 , 323 , 717 residues ( Drosophila melanogaster R6 . 06 ) , which deletes five out of six exons of the Atg5 gene , eliminating more than 85% of protein-coding sequences including the translation start site ( Figure 6A ) . The PhiC31 integrase-mediated site-specific transformation method was used to express human ATG5WT and ATG5E122D from an identical genomic locus ( Bateman et al . , 2006; Bischof et al . , 2007; Venken et al . , 2006 ) . In brief , flag-tagged ATG5WT and ATG5E122D were cloned into a pUAST-attB vector ( Bischof et al . , 2007 ) and fully sequenced . pUAST-attB-ATG5WT and pUAST-attB-ATG5E122D were microinjected into y1 M{vas-int . Dm}ZH-2A w*; M{3xP3-RFP . attP}ZH-51D flies and stable transformants were isolated by the presence of the mini-white+ marker ( Figure 6B ) . The UAS-ATG5WT or UAS-ATG5E122D transgenes were crossed with a double balancer strain ( Bl/CyO; TM2/TM6B ) and then with +/CyO; Tub-Gal4/TM2 to be constructed as stable Tub>ATG5 lines ( UAS-ATG5/UAS-ATG5; Tub-Gal4/TM6B ) . The Tub>ATG5 male flies were crossed with Atg55cc5/FM7 female flies to generate Atg5-null flies expressing human ATG5 transgenes . Climbing assays and TUNEL staining were performed as previously described ( Kim et al . , 2015b ) .

Ataxia is a rare disease that affects balance and co-ordination , leading to difficulties in walking and other movements . The disease mostly affects adults , but some children are born with it and they often have additional cognitive and developmental problems . Mutations in at least 60 genes are known to be able to cause ataxia , but it is thought that there are still more to be found . Kim , Sandford et al . studied two siblings with the childhood form of ataxia and found that they both had a mutation in a gene called ATG5 . The protein produced by the mutant ATG5 gene was less able to interact with another protein called ATG12 . Furthermore , the cells of both children had defects in a process called autophagy – which destroys old and faulty proteins to prevent them accumulating and causing damage to the cell . Next , Kim , Sandford et al . examined the effect of this mutation in baker’s yeast cells . Cells with a mutation in the yeast equivalent of human ATG5 had lower levels of autophagy than normal cells . Further experiments used fruit flies that lacked fly Atg5 , which were unable to fly or walk properly . Inserting the normal form of human ATG5 into the flies restored normal movement , but the mutant form of the gene had less of an effect . These findings suggest that a mutation in ATG5 can be responsible for the symptoms of childhood ataxia . Kim , Sandford et al . think that other people with severe ataxia may have mutations in genes involved in autophagy . Therefore , the next step is to study autophagy in cells from many other ataxia patients .

2016

Mutation in ATG5 reduces autophagy and leads to ataxia with developmental delay

In humans , listening to speech evokes neural responses in the motor cortex . This has been controversially interpreted as evidence that speech sounds are processed as articulatory gestures . However , it is unclear what information is actually encoded by such neural activity . We used high-density direct human cortical recordings while participants spoke and listened to speech sounds . Motor cortex neural patterns during listening were substantially different than during articulation of the same sounds . During listening , we observed neural activity in the superior and inferior regions of ventral motor cortex . During speaking , responses were distributed throughout somatotopic representations of speech articulators in motor cortex . The structure of responses in motor cortex during listening was organized along acoustic features similar to auditory cortex , rather than along articulatory features as during speaking . Motor cortex does not contain articulatory representations of perceived actions in speech , but rather , represents auditory vocal information . Our motor and sensory cortices are traditionally thought to be functionally separate systems . However , an accumulating number of studies has revealed their roles in action and perception to be highly integrated ( Pulvermüller and Fadiga , 2010 ) . For example , a number of studies have demonstrated that both sensory and motor cortices are engaged during perception ( Gallese et al . , 1996; Wilson et al . , 2004; Tkach et al . , 2007; Cogan et al . , 2014 ) . In humans , this phenomenon has been observed in the context of speech , where listening to speech sounds evokes robust neural activity in the motor cortex ( Wilson et al . , 2004; Pulvermüller et al . , 2006; Edwards et al . , 2010; Cogan et al . , 2014 ) . This observation has re-ignited an intense scientific debate over the role of the motor system in speech perception over the past decade ( Lotto et al . , 2009; Scott et al . , 2009; Pulvermüller and Fadiga , 2010 ) . One interpretation of the observed motor activity during speech perception is that “the objects of speech perception are the intended phonetic gestures of the speaker”- as posited by Liberman’s motor theory of speech perception ( Liberman et al . , 1967; Liberman and Mattingly , 1985 ) . The motor theory is a venerable and well-differentiated exemplar of a set of speech perception theories that we could call 'production-referencing' theories . Unlike motor theory , more modern production referencing theories do not assume that sensorimotor circuits are necessarily referenced in order for speech to be recognized , but they allow for motor involvement in perception in certain phonetic modes . For example , Lindblom , 1996 suggested that a direct link between spectrotemporal analysis and word recognition is the normal mode of speech perception ( the 'what' mode of perception ) , but in some cases listeners do use a route through sensorimotor circuits ( the 'how' mode of perception ) if , for example , the listener is attempting to imitate a new sound ( Lindblom , 1996 ) . While demonstrations of evoked motor cortex activity by speech sounds strengthen production-referencing theories , it remains unclear what information is actually represented by such activity . Determining what phonetic properties are encoded in the motor cortex has significant implications for elucidating the role it may play in speech perception . To address this , we recorded direct neural activity from the peri-Sylvian speech cortex in nine human participants undergoing clinical monitoring for epilepsy surgery . This includes but is not limited to two relevant areas comprising the supra-Sylvian ventral half of the lateral sensorimotor cortex ( vSMC ) for the motor control of articulation ( Penfield and Boldrey , 1937 ) and the infra-Sylvian superior temporal gyrus ( STG ) for the auditory processing of speech sounds ( Ojemann et al . , 1989; Boatman et al . , 1995 ) . Since cortical processing of speech sounds is spatially discrete and temporally fast ( Formisano et al . , 2008; Chang et al . , 2011; Steinschneider et al . , 2011 ) , we used customized high-density electrode grids ( a four-fold increase over conventional recordings ) ( Bouchard et al . , 2013; Mesgarani et al . , 2014 ) . Importantly , these recordings have simultaneous high spatial and temporal resolution in order to study the detailed speech representations in the vSMC ( Crone et al . , 1998; Edwards et al . , 2009 ) . With this approach , we seek to address unanswered questions about the representation of speech sounds in motor cortex , including how the spatiotemporal patterns compare when speaking and listening and whether auditory representations in motor cortex are organized along articulatory or acoustic dimensions . Participants first listened passively to consonant-vowel ( CV ) syllables ( 8 consonants followed by the /a/ vowel ) . In a separate trial block , they spoke aloud these same CV syllables . We measured the average evoked cortical activity during these listening and speaking CV tasks . We focused our analysis on high gamma ( 70–150 Hz ) cortical surface local field potentials , which strongly correlate with extracellular multi-unit neuronal spiking ( Steinschneider et al . , 2008; Ray and Maunsell , 2011 ) . We aligned neural responses to the onset of speech acoustics ( t = 0 ) in listening and speaking tasks to provide a common reference point across speech sounds . We first determined which peri-Sylvian cortical areas were activated during passive listening to speech sounds . Figure 1a and b shows the locations of cortical areas that demonstrated cortical evoked responses in a single representative subject during listening and speaking respectively . During listening , evoked responses spanned middle and posterior STG as expected , with weaker responses in middle temporal gyrus ( MTG ) ( Figure 1a ) . In the vSMC , ( composed of the pre- and post- central gyri ) we found electrodes in the superior-most and inferior-most aspects ( Figure 1a , Figure 1—figure supplement 1 , 2 ) that demonstrated reliable and robust single-trial responses to speech sounds during passive listening ( Figure 1b ) . Neural responses were also found at a few sites scattered across supramarginal , inferior- , and middle- frontal gyri—though these were not consistent across subjects ( Figure 1—figure supplement 1 ) . By performing spatial clustering analysis on the electrode positions in each subject , we found that 3/5 subjects showed significant clustering of regions responsive to auditory stimuli ( Hartigan’s Dip statistic , p<0 . 05 ( see Materials and methods ) ; Figure 1—figure supplement 1 ) . Out of these 3 subjects , k-means clustering revealed two subjects with k=2 electrode clusters ( subjects 1 and 4 , clusters in inferior and superior vSMC ) , and one subject with k=5 clusters . When participants spoke the same CV syllables , in contrast , articulatory movement-related cortical activity was well distributed throughout vSMC ( Figure 1c ) , with auditory feedback cortical activity seen in the STG . 10 . 7554/eLife . 12577 . 003Figure 1 . Speech sounds evoke responses in the human motor cortex . ( a ) Magnetic resonance image surface reconstruction of one representative subject’s cerebrum ( subject 1: S1 ) . Individual electrodes are plotted as dots , and the average cortical response magnitude ( z-scored high gamma activity ) when listening to CV syllables is signified by the color opacity . CS denotes the central sulcus; SF denotes the Sylvian fissure . ( b ) Acoustic waveform , spectrogram , single-trial cortical activity ( raster ) , and mean cortical activity ( high gamma z-score , with standard error ) from two vSMC sites and one STG site when a subject is listening to /da/ . Time points significantly above a pre-stimulus silence period ( p<0 . 01 , bootstrap resampled , FDR corrected , alpha < 0 . 005 ) are marked along the horizontal axis . The vertical dashed line indicates the onset of the syllable acoustics ( t=0 ) . ( c ) Same subject as in ( a ) ; distributed vSMC cortical activity when speaking CV syllables ( mean high gamma z-score ) . ( d ) Total number of significantly active sites in all subjects during listening , speaking , and both conditions ( p<0 . 01 , t-test , responses compared to silence and speech ) . Electrode sites are broken down by their anatomical locations . S denotes superior vSMC sites; I denotes inferior vSMC sites . DOI: http://dx . doi . org/10 . 7554/eLife . 12577 . 00310 . 7554/eLife . 12577 . 004Figure 1—figure supplement 1 . Average cortical responses to speaking and listening in all subjects ( S2-S5 ) . ( a ) Average Z-scored responses in active electrodes when listening to ( left ) or speaking ( right ) CV syllables in all subjects , as shown in Figure 1a and c . ( b ) Results from spatial clustering of significantly active electrodes in each subject . S1 , S4 , and S5 showed spatially segregated clusters ( see Materials and methods ) and were included in k-means clustering analysis . The silhouette index shows that the best number of clusters for S1 ( from Figure 1 ) and S4 was k=2 , whereas the best number of clusters for S5 was k=5 . DOI: http://dx . doi . org/10 . 7554/eLife . 12577 . 00410 . 7554/eLife . 12577 . 005Figure 1—figure supplement 2 . Neural responses while listening to CV syllables in 4 additional subjects not included in MDS analyses ( S6 - S9 ) . Responses are plotted on each subject’s brain surface and are shown for each electrode as the average Z-scored high gamma activity across all syllables , aligned to the acoustic onset of the stimulus ( indicated by dashed line at t = 0 ) . As in Figure 1 and Figure 1—figure supplement 1 , strong responses are seen during listening in inferior and superior regions of the vSMC ( primarily precentral gyrus ) in addition to the superior temporal gyrus ( STG ) . For each subject , SF indicates the location of the Sylvian fissure , and CS indicates the central sulcus . DOI: http://dx . doi . org/10 . 7554/eLife . 12577 . 005 Across all participants , we identified 115 electrodes that demonstrated significant neural activity in vSMC during listening ( p<0 . 01 , t-test , compared to pre-stimulus silent rest period; Figure 1d ) . When speaking , in contrast , a total of 362 electrodes in vSMC were found to be significantly active ( Figure 1d , p<0 . 01 , t-test , compared to pre-stimulus silent rest period ) . We compared the relative proportions of electrodes that were found in different supra-Sylvian anatomical regions . Critically , only a subset of sites in vSMC ( 98 out of 362 , ~27% ) was active during both listening and speaking ( Figure 1d ) . These sites were primarily localized to the pre-central gyrus , whereas speaking evoked activity across both pre- and post-central gyri sites . Neural responses in the vSMC during listening were found in the superior ( S in Figure 1d ) pre-central gyrus and inferior , anterior aspect of the sub-central gyrus of the vSMC ( I in Figure 1d ) . We next compared the patterns of cortical activity to specific speech sounds during listening and speaking . During speaking , specific articulator representations have been identified in the somatotopically-organized vSMC ( Bouchard et al . , 2013 ) . For example , the plosive consonants /b/ , /d/ , and /g/ are produced by the closure of the vocal tract at the lips , front tongue , and back tongue , respectively ( Figure 2a , b , see Figure 2—figure supplement 1 for all syllable tokens ) ( Ladefoged and Johnson , 2010 ) . The cortical representations for these articulators are laid out along a superior-to-inferior ( medial-to-lateral ) sequence in the vSMC ( Penfield and Boldrey , 1937 ) . We first examined average cortical activity at single electrode sites distributed along the vSMC axis for articulating individual speech sounds . Figure 2c shows single electrode activity from a single representative subject ( the same from Figure 1 ) for speaking ( blue lines ) and listening ( red lines ) for three CV syllables , which have different place of articulation ( /ba/ , /da/ , and /ga/ ) . The exact location of these electrodes on the vSMC is shown in Figure 2d . The production of labial consonants ( /b/ ) is associated with activity in lip cortical representations as evidenced by strong responses to the bilabial /ba/ ( Figure 2c , electrodes 5–6 , blue lines ) . These are located superior to the tongue representations associated with the /d/ and /g/ consonants , as shown previously ( Bouchard et al . , 2013 ) . Those tongue sites were sub-specified by ‘coronal’ ( i . e . anterior-based ) tongue position for /d/ ( electrodes 8–10 , blue lines ) superiorly , and ‘dorsal’ ( i . e . posterior-based ) tongue position for /g/ inferiorly ( electrode 13 , blue line ) . Other sites ( electrodes 1–4 , 11–12 , blue lines ) showed the same neural activity across all three syllables . 10 . 7554/eLife . 12577 . 006Figure 2 . Site-by-site differences in vSMC neural activity when speaking and listening to CV syllables . ( a ) Top , vocal tract schematics for three syllables ( /ba/ , /da/ , /ga/ ) produced by occlusion at the lips , tongue tip , and tongue body , respectively ( arrow ) . ( b ) Acoustic waveforms and spectrograms of spoken syllables . ( c ) Average neural activity at electrodes along the vSMC for speaking ( blue ) and listening ( red ) to the three syllables ( high gamma z-score ) . Solid lines indicate activity was significantly different from pre-stimulus silence activity ( p<0 . 01 ) . Transparent lines indicate activity was not different from pre-stimulus silence activity ( p>0 . 01 ) . Vertical dashed line denotes the onset of the syllable acoustics ( t=0 ) . ( d ) Location of electrodes 1–13 in panel c , shown on whole brain and with inset detail . CS = central sulcus , SF = Sylvian fissure . DOI: http://dx . doi . org/10 . 7554/eLife . 12577 . 00610 . 7554/eLife . 12577 . 007Figure 2—figure supplement 1 . Syllable token set . ( a ) Consonants of all syllable tokens . Consonants were paired with the vowel /a/ , and are organized by place and manner of articulation . ( b ) Vocal tract schematics for three occlusions made with the lips , tongue tip , and tongue body , respectively ( arrow ) . ( c ) Acoustic waveform and spectrogram of the syllable tokens . DOI: http://dx . doi . org/10 . 7554/eLife . 12577 . 007 We next examined those same vSMC electrodes during listening , and found that the majority of those cortical vSMC electrodes were not active ( p>0 . 01 , t-test compared to silence , Figure 2c transparent red lines ) . The few that were active ( electrodes 1 , 2 , 4 , 11–12 , solid red lines ) were similar for all three CV syllables , with activity increasing approximately 100ms after the acoustic onset . Across the entire population of vSMC electrodes that were active during listening , onset latencies were generally shorter than those in STG sites , with significant increases in both inferior vSMC ( p<0 . 001 ) and superior vSMC ( p<0 . 05 ) compared to STG ( Figure 3a , Wilcoxon rank sum test , see Figure 3c for average responses to all syllables ) . The latency to the response peak was also significantly higher in superior vSMC compared to STG ( Figure 3b , p<0 . 01 , Wilcoxon rank sum test ) . A cross-correlation analysis between these vSMC electrodes and STG electrodes revealed a diverse array of relationships between these populations ( Figure 3d–f ) , including STG electrode activity leading vSMC electrode activity and vice versa . In contrast to speaking , we did not observe somatotopic organization of cortical responses when listening to speech . Therefore , the pattern of raw evoked responses during listening shows critical differences from those during speaking . 10 . 7554/eLife . 12577 . 008Figure 3 . Dynamics of responses during CV listening in STG , inferior vSMC , and superior vSMC . ( a ) STG onset latencies were significantly lower than both inferior vSMC ( p<0 . 001 , Z = −4 . 03 ) and superior vSMC ( p<0 . 05 , Z = −2 . 28 ) . ( b ) STG peak latencies were significantly lower than superior vSMC ( p<0 . 01 , Z = −2 . 93 ) , but not significantly different from peak latencies in inferior vSMC ( p>0 . 1 ) . In ( a ) and ( b ) , red bar indicates the median , boxes indicate 25th and 75th percentile , and error bars indicate the range . Response latencies were pooled across all subjects . All p-values in ( a ) and ( b ) are from the Wilcoxon rank sum test . ( c ) Average evoked responses to all syllable tokens across sites in superior vSMC ( n=32 ) , inferior vSMC ( n=37 ) , and STG . Responses were aligned to the syllable acoustic onset ( t=0 ) . A random subset of STG responses ( n=52 out of the 273 that were used in the latency analysis in ( a ) and ( b ) ) are shown here for ease of viewing . ( d ) Example cross-correlations between three vSMC electrodes and all STG electrodes in one patient , for a maximum lag of ± 0 . 75 s . More power in the negative lags indicates a faster response in the STG compared to the vSMC electrode , and more power in the positive lags indicates a faster response in vSMC compared to STG . We observe vSMC electrodes that tend to respond later than STG ( e248 , left panel ) , vSMC electrodes that tend to respond before STG ( e136 , middle panel ) , and vSMC electrodes that respond at similar times to some STG electrodes ( e169 , right panel ) . ( e ) Average evoked responses during CV listening for all STG electrodes from this patient and the three vSMC electrodes shown in panel ( d ) . Responses were aligned to the syllable acoustic onset ( t=0 ) , as in panel ( c ) . ( f ) Percentage of sites with STG leading , coactive , or vSMC leading as expressed by the asymmetry index ( see Materials and methods ) . Both inferior and superior vSMC show leading and lagging responses compared to STG , as well as populations of coactive pairs . DOI: http://dx . doi . org/10 . 7554/eLife . 12577 . 008 We next evaluated quantitatively whether the structure of distributed vSMC neural activity during listening was more similar to that of vSMC during speaking or the STG during listening . In previous studies , we demonstrated that the structure of evoked responses are primarily organized by different feature sensitivities: place of articulation in the vSMC ( Bouchard et al . , 2013 ) , and manner of articulation in the STG ( Mesgarani et al . , 2014 ) . We visualized the similarity of population activity evoked by different consonants using unsupervised multidimensional scaling ( MDS ) , where the 2-dimensional Euclidean distances between stimuli correspond to the similarity of their neural responses . Visual inspection of MDS plots shows that , during speaking , evoked activity in vSMC clustered into place of articulation features ( Figure 4a ) : labials ( /b/ , /p/ ) , alveolars ( /s/ , /sh/ , /t/ , /d/ ) , and velars ( /g/ , /k/ ) ( Figure 4b ) . In contrast , neural responses during listening did not cluster into the same features ( Figure 4c ) . To quantify the degree to which the evoked activity clustered into place of articulation features , we used unsupervised K-means clustering to assign the neural responses to clusters ( k=3 ) , and the adjusted Rand Index ( RIadj ) ( Rand , 1971; Hubert and Arabie , 1985 ) to measure the degree to which the neural clustering agreed with linguistically defined place of articulation consonant clusters . The RIadj quantifies the degree of agreement between two clustering patterns , where RIadj = 1 denotes identical clustering patterns and RIadj = 0 denotes independent clustering patterns . We found that while evoked activity during speaking clustered by place of articulation features , activity during listening did not ( Figure 4d; see Figure 4—figure supplement 1 for moving time window analysis ) . Even when the vSMC electrode subset was restricted to short-latency vSMC electrodes leading STG activity ( as evidenced by a positive asymmetry index in Figure 3f ) , activity during listening did not cluster according to place of articulation features ( Figure 4—figure supplement 2 ) . Thus , responses in motor areas during speech perception do not show a spatially distributed representation of speech motor articulator features . 10 . 7554/eLife . 12577 . 009Figure 4 . Organization of motor cortex activity patterns . ( a ) Consonants of all syllable tokens organized by place and manner of articulation . Where consonants appear in pairs , the right is a voiced consonant , and the left is a voiceless consonant . ( b ) Relational organization of vSMC patterns ( similarity ) using multidimensional scaling ( MDS ) during speaking . Neural pattern similarity is proportional to the Euclidean distance ( that is , similar response patterns are grouped closely together , whereas dissimilar patterns are positioned far apart ) . Tokens are colored by the main place of articulation of the consonants ( labial , velar , or alveolar ) . ( c ) Similarity of vSMC response patterns during listening . Same coloring by place of articulation . ( d ) Organization by motor articulators . K-means clustering was used to assign mean neural responses to 3 groups ( labial , alveolar , velar ) for both listening and speaking neural organizations ( b , c ) . The similarity of the grouping to known major articulators was measured by the adjusted Rand Index . An index of 1 indicates neural responses group by place of articulation features . ***p<0 . 001 , Wilcoxon rank-sum ( e ) Organization of mean STG responses using MDS when listening . In contrast to c and d , tokens are now colored by their main acoustic feature ( fricative , voiced plosive , or voiceless plosive ) . ( f ) Organization of mean vSMC responses using MDS when listening colored by their main acoustic feature . ( Identical to C , but recolored here by acoustic features ) . ( g ) Organization by manner of articulation acoustic features ( fricative , voiced plosive , voiceless plosive ) for both STG and vSMC organizations when listening ( e , f ) . The similarity of the grouping to known acoustic feature groupings was measured by the adjusted Rand Index . ***p<0 . 001 , Wilcoxon rank sum . ( h ) During listening , responses in vSMC show significantly greater organization by acoustic manner features compared to place features as assessed by the adjusted Rand Index , indicating an acoustic rather than articulatory representation ( ***p<0 . 001 , Wilcoxon rank-sum ) . Bars in this panel are the same as the red bars in ( d ) and ( g ) . In ( d ) , ( g ) , and ( h ) , bars indicate mean ± standard deviation , DOI: http://dx . doi . org/10 . 7554/eLife . 12577 . 00910 . 7554/eLife . 12577 . 010Figure 4—figure supplement 1 . Clustering trajectory analysis of neural responses to syllables . The clustering trajectory across time was derived from vSMC activity ( n=98 electrodes ) when speaking ( blue ) and listening ( red ) using the ΔRIadj metric . An index of 1 denotes the neural responses organized by acoustic features , and an index of -1 denotes an organization by place of articulation features . Time points indicated by the blue and red windows are significantly organized according to place or acoustics , respectively ( FDR-corrected p<0 . 05 , permutation test comparing random clustering at each time point to clustering by acoustics or place ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12577 . 01010 . 7554/eLife . 12577 . 011Figure 4—figure supplement 2 . Analysis of short latency responses in vSMC . We restricted the MDS analysis to vSMC responses with a positive asymmetry index ( see Figure 3d and f , n=50 electrodes ) and found significant organization by acoustic manner features during listening in these sites . ( a ) Similarity of responses to syllables during listening in the subset of vSMC electrodes leading activity in STG . Tokens are colored according to place of articulation features ( left ) or acoustic manner features ( right ) as in Figure 4c and f . ( b ) During listening , responses of short latency sites in the vSMC still show significantly greater organization by acoustic manner features compared to place features as assessed by the adjusted Rand Index , indicating an acoustic rather than articulatory representation ( ***p<0 . 001 , Wilcoxon rank-sum ) . Compare with Figure 4h . Bars indicate mean ± standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 12577 . 01110 . 7554/eLife . 12577 . 012Figure 4—figure supplement 3 . Organization of syllable tokens and auditory cortical activity patterns . ( a ) Organization of the acoustic spectrograms of the CV tokens using MDS . ( b ) Organization of mean STG responses using MDS when speaking . Tokens in ( a ) and ( b ) are colored by their main acoustic feature . ( c ) Organization of vSMC responses by place of articulation using all speech-responsive electrodes ( n=362 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12577 . 012 Finding no evidence that major articulator features are either locally or spatially distributed in the vSMC in response to speech sounds , we next compared vSMC responses to population responses in the STG . STG has an acoustic sensory representation of speech that best discriminates speech sounds by manner of articulation features with salient acoustic differences ( Mesgarani et al . , 2014 ) . Using multidimensional scaling , STG spatial patterns during listening showed clustering according to three high-order acoustic features ( Figure 4e ) : voiced plosives ( /b/ , /d/ , /g/ ) , unvoiced plosives ( /p/ , /t/ , /k/ ) , and fricatives ( /s/ , /sh/ ) ( Ladefoged and Johnson , 2010 ) . This is consistent with the relational organization derived by analysis of structure in the stimulus acoustics ( Figure 4—figure supplement 3a ) , and the structure of STG during speaking ( Figure 4—figure supplement 3b ) . With the same analyses , we observed that activity in motor cortex clustered into the same three acoustic features ( Figure 4f , note this panel is identical to Figure 4c simply re-colored ) . Unsupervised K-means clustering analysis confirmed that vSMC activity , during listening , organized into these linguistically defined acoustic feature groups , but was significantly weaker than the organization of STG ( p<0 . 001 , Wilcoxon rank-sum , Figure 4g ) . Importantly , however , clustering by acoustic manner features was significantly stronger than clustering by place features in vSMC electrodes during listening ( p<0 . 001 , Wilcoxon rank-sum , Figure 4h ) . This organization suggests that motor cortex activity during speech perception reflects an acoustic sensory representation of speech in the vSMC that mirrors acoustic representations of speech in auditory cortex . To further define the acoustic selectivity and tuning of vSMC motor electrodes , participants listened to natural , continuous speech samples from a corpus with a range of American English speakers ( Garofolo et al . , 1993 ) . We fit spectrotemporal receptive field ( STRF ) models for each vSMC electrode using normalized reverse correlation ( see Materials and methods ) , which describes the spectrotemporal properties of speech acoustics that predict the activity of a single site in motor cortex . To compute the STRF , we calculate the correlation between the neural response at an electrode and the stimulus spectrogram at multiple time lags . The result is then normalized by the auto-correlation in the stimulus . This results in a linear filter for each electrode ( the STRF ) , which , when convolved with the stimulus spectrogram , produces a predicted neural response to that stimulus . The prediction performance of each STRF was determined by calculating the correlation between the activity predicted by the STRF and the actual response on held out data . A fraction of vSMC sites ( 16/98 sites total ) were reasonably well-predicted with a linear STRF ( r>=0 . 10 and p<0 . 01 , permutation test ) ( Theunissen et al . , 2001 ) . STRFs with significant correlation coefficients were localized to superior and inferior vSMC ( primarily precentral gyrus ) in addition to STG ( Figure 5a ) . Still , the prediction performance of STRFs in vSMC was generally lower than that of the STG ( Figure 5b ) . Furthermore , the majority of STRFs in both regions showed strong low frequency tuning ( 100–200 Hz ) properties related to voicing ( Figure 5c ) , though some also showed high frequency tuning consistent with selectivity for fricatives and stop consonants by visual inspection ( Mesgarani et al . , 2014 ) . We also estimated the mean cortical response at each motor site to every phoneme in English and found a diverse set of responses ( Figure 5—figure supplement 1a ) that were notably weaker in magnitude compared to STG responses ( Figure 5—figure supplement 1b ) . Weak selectivity to phonetic features measured by the Phoneme Selectivity Index ( PSI ) was also observed ( Figure 5—figure supplement 1c ) ( Mesgarani et al . , 2014 ) . These findings reveal that individual sites in motor cortex reflect sensory responses to definable spectrotemporal features speech acoustics , including voicing attributes . Presumably , this tuning gives rise to the acoustic organization found in the previous analysis of distributed spatial patterns of neural activity . 10 . 7554/eLife . 12577 . 013Figure 5 . Acoustic spectrotemporal tuning in vSMC . ( a ) All STRF correlations and locations are plotted with opacity signifying the strength of the correlation . CS denotes the central sulcus; SF denotes the Sylvian fissure . ( b ) Distribution of STRF prediction correlations for significantly active vSMC and STG sites . Cut-off at r = 0 . 1 is shown as a dashed line . ( c ) Individual STRFs from all subjects ( S1-S5 , STRF correlation>0 . 1 ) plotted as a function of distance from the central sulcus and Sylvian fissure , with opacity signifying the strength of the STRF correlation . DOI: http://dx . doi . org/10 . 7554/eLife . 12577 . 01310 . 7554/eLife . 12577 . 014Figure 5—figure supplement 1 . Summary and comparison of vSMC and STG responses to syllables . ( a ) Average responses at sample electrodes to all English phonemes and their PSI vectors . ( b ) Peak high-gamma z-score distributions for significantly active vSMC and STG sites when listening ( p<0 . 01 , comparing silence to stimulus activity ) . ( c ) Average PSI distributions for significantly active vSMC and STG sites . DOI: http://dx . doi . org/10 . 7554/eLife . 12577 . 014 Our principal objective was to determine the vSMC motor cortex representation of auditory speech sounds . We used high-resolution cortical recordings and a wide array of speech sounds to determine how the vSMC structure of speech sounds compared to the structure of motor commands in vSMC and sensory processing in STG . We found evidence for both spatially local and distributed activity correlated to speech acoustics , which suggests an auditory representation of speech in motor cortex . The proposal that the motor cortex critically integrates observations with motor commands largely stems from the discovery of mirror neurons ( in area F5 of macaques ) that fire both when a monkey produced an action and observed a similar action ( di Pellegrino et al . , 1992; Rizzolatti and Craighero , 2004; Pulvermüller and Fadiga , 2010 ) . This 'integrative' view is reminiscent of linguistic production-referencing theories , including the motor theory of speech perception , which propose that motor circuits are involved in speech perception ( Liberman et al . , 1967; Liberman and Mattingly , 1985 ) . In line with these theories , human neuroimaging studies have showed mirror activity in ventral premotor cortex during listening ( Wilson et al . , 2004; Pulvermüller et al . , 2006; Edwards et al . , 2010 ) , and modulated premotor activity in phoneme categorization tasks ( Alho et al . , 2012; Chevillet et al . , 2013 ) . Our results extend these findings by detailing the representational selectivity and encoding of vSMC in perception . Consistent with previous findings , we demonstrated local ‘audiomotor’ responses to speech sounds in vSMC . When the responses were further examined for phonetic structure , we found major motor articulatory place features , such as labial , alveolar , and velar , were not represented with single site activity or distributed spatial activity . This observation is in direct contrast with structural predictions made by the original motor theory of speech perception ( Liberman et al . , 1967; Liberman and Mattingly , 1985 ) , while confirming that motor cortex plays a role in perception ( Lindblom , 1996; Hickok and Poeppel , 2007 ) . We localized activity during speech perception to regions of the vSMC that have been implicated in phonation and laryngeal control ( Penfield and Boldrey , 1937; Brown et al . , 2008 ) . When listening to speech , we observed these regions reflected acoustic sensory properties of speech , with individual sites tuned for spectrotemporal acoustic properties . The tuning properties of responsive sites in vSMC are similar to properties observed in STG during listening ( Mesgarani et al . , 2014 ) and appear to give rise to an acoustic sensory organization of speech sounds ( rather than purely motor organization ) in motor cortex during listening . There is an emerging consensus that frontal and motor regions are recruited during effortful listening ( Du et al . , 2014 ) . For example , previous studies have demonstrated that frontal areas come online to process degraded speech for the attentional enhancement of auditory processing ( Wild et al . , 2012 ) . Our results may complement this interpretation in that the audiomotor cortex enhancement is specific to an auditory representation , without transforming information to a motor articulatory representation . That being said , the auditory encoding that we observed in the motor cortex did not appear to be as strong as that as that observed in the STG , and exhibited comparatively weaker activity and weaker phoneme selectivity ( Figure 5—figure supplement 1b and c , and see ( Mesgarani et al . , 2014 ) . In addition to having implications for perceptual models , we speculate that these results have strong implications for speech production , as auditory feedback is potentially processed directly in the vSMC in addition to the canonical auditory cortex . Speech production models currently propose a complex role for sensory feedback , where pathways exist for the activation of auditory cortex from vSMC activation ( the forward prediction of production consequences ) , and the activation of vSMC from auditory and somatosensory input ( the error correction signal ) ( Guenther et al . , 2006; Houde and Nagarajan , 2011 ) . In the current study , it appears that the motor cortex contains both sensory and motor representations , where the sensory representations are active during passive listening , whereas motor representations dominate during speech production . Analysis of the time course of vSMC and STG responses revealed a heterogeneous population of both short- and longer-latencies in the inferior and superior vSMC that are generally slower than the STG ( Figure 3a–c ) . Early responses in vSMC may reflect bidirectional connections from STG ( Zatorre et al . , 2007 ) , primary auditory cortex ( Nelson et al . , 2013; Schneider et al . , 2014 ) or auditory thalamus ( Henschke et al . , 2014 ) , whereas later responses might reflect indirect connectivity in areas downstream from the STG ( Rauschecker and Scott , 2009 ) . Indeed , our cross-correlation analysis revealed bidirectional dynamical relationships between vSMC and STG responses , in which STG responses led vSMC responses and vice versa ( Figure 3d–f ) . Still , this analysis was independent of the diverse tuning properties in the vSMC and STG electrode sets , so longer latency responses likely reflect the later responses to vowels relative to consonants . Even so , we found a wide variety of tuning and dynamical profiles in the vSMC electrodes that responded during listening . Given these proposed functional connections , activity in vSMC from speech sounds may be a consequence of sounds activating the sensory feedback circuit ( Hickok et al . , 2011 ) . Alternatively , evoked responses in the motor cortex during passive listening may directly reflect auditory inputs arising from aggregated activity picked up by the electrode . We believe the latter scenario to be less likely , however , given that auditory responses were observed in dorsal vSMC on electrode contacts several centimeters away from auditory inputs in the STG . In addition , the spatial spread of neural signals in the high gamma range is substantially smaller than this difference – high gamma signal correlations at <2 mm spacing are only around r=0 . 5 , and at distances of 1 cm reach a noise floor ( Chang , 2015 ) ; Muller et al , unpublished findings ) . Given the observed acoustic rather than place selectivity observed during listening in the vSMC , our results suggest that motor theories of speech perception may need to be revised to incorporate a novel sensorimotor representation of sound in the vSMC . Nine human participants were implanted with high-density multi-electrode cortical surface arrays as part of their clinical evaluation for epilepsy surgery . The array contained 256 electrodes with 4 mm pitch . Arrays were implanted on the lateral left hemispheres over the peri-Sylvian cortex , but exact placement was determined entirely by clinical indications ( Figure 1—figure supplement 1 and Figure 1—figure supplement 2 ) . Using anatomic image fusion software from BrainLab ( Munich , Germany ) , electrode positions were extracted from the computed tomography ( CT ) scan , co-registered with the patient’s MRI and then superimposed on the participant’s 3D MRI surface reconstruction image . All participants were left hemisphere language dominant , as assessed by the Wada test . Participants had self-reported normal hearing . The study protocol was approved by the UC San Francisco Committee on Human Research , and all participants provided written informed consent . Participants completed three separate tasks that were designed to sample a range of phonetic features . First , participants listened to eight consonant-vowel ( CV ) syllables ( /ba/ , /da/ , /ga/ , /pa/ , /ta/ , /ka/ , /∫a/ , /sa/ ) produced by a male speaker unknown to the participant . Stimuli were presented randomly , with 4–21 repetitions of each CV syllable for 5 out of the 9 subjects included in all subsequent analyses , and one repetition of each CV syllable for 4 subjects shown only in Figure 1—figure supplement 2 . To remain alert , participants were asked to identify the syllable they heard by selecting from a multiple-choice question on a computer with their ipsilateral ( left ) hand . In the second task , participants spoke aloud the same CV syllables prompted by a visual cue on the laptop computer display . In the third task , participants passively listened to natural speech samples from a phonemically transcribed continuous speech corpus ( TIMIT ) . We chose 499 unique sentences from 400 different male and female speakers . Each sentence was repeated two times . For the phoneme selectivity analysis , we chose a subset of TIMIT phonemes that occurred more than 30 times . This resulted in an analysis of 33 phonemes . For spectrotemporal receptive field analysis ( see below ) , data from all sentences were used . Electrocorticographic ( ECoG ) signals were recorded with a multichannel PZ2 amplifier connected to an RZ2 digital signal acquisition system ( Tucker-Davis Technologies , Alachua , FL , USA ) sampling at 3 , 052 Hz . The produced speech was recorded with a microphone , digitized , and simultaneously recorded . The speech sound signals were presented monaurally from loudspeakers at a comfortable level , digitized , and also simultaneously recorded with the ECoG signals . Line noise ( 60 Hz and harmonics at 120 and 180 Hz ) was next removed from the signal with a notch filter . Each time series was visually and quantitatively inspected for excessive noise , and was subsequently removed from further analyses if its periodogram deviated more than two standard deviations away from the average periodogram of all other time series . The remaining time series were then common-average referenced ( CAR ) and used for analyses . The CAR was taken across 16 channel banks in order to remove non-neural electrical noise from shared inputs to the PZ2 . We find that this method of CAR significantly reduces movement-related and other non-neural artifacts while not adversely affecting our signals of interest . The analytic amplitude of each time series was extracted using eight bandpass filters ( Gaussian filters , logarithmically increasing center frequencies ( 70–150 Hz ) with semi-logarithmically increasing bandwidths ) with the Hilbert transform . The high-gamma power was calculated by averaging the analytic amplitude across these eight bands , and downsampling the signal to 100 Hz . The signal was finally z-scored relative to the mean and standard deviation of baseline rest data for each channel . Supra-Sylvian cortical sites with robust evoked responses to both speech sounds and speech production were selected for this analysis . To identify if a site was responsive to speech sounds , we implemented a bootstrap t-test comparing a site's responses randomly sampled over time during speech sound presentations to responses randomly sampled over time during pre-stimulus silent intervals ( p<0 . 01 ) . This resulted in 10 , 22 , 29 , 27 , and 27 sites for the five participants ( n=115 ) . Next we implemented a bootstrap t-test comparing neural responses during speech production and pre-stimulus silence ( p<0 . 01 ) , resulting in 25 , 74 , 87 , 92 , and 84 sites ( n=362 ) . Finally , we took the intersection of these two groups to arrive at our final supra-Sylvian sites set of 8 , 16 , 28 , 22 , and 24 sites active during listening and speaking ( n=98 ) . To analyze the responses of the auditory cortex , we restricted the infra-Sylvian cortical sites to those that were reliably evoked by speech sounds ( p<0 . 01 , t-test between silence and speech sounds neural responses ) . This resulted in 73 , 61 , 40 , 77 , and 89 infra-Sylvian temporal cortical sites ( n=340 ) responsive to speech sounds . To investigate the degree of spatial clustering in the vSMC electrodes responsive during listening , we used the Dip-means method ( Kalogeratos and Likas , 2012 ) , which allows us to test whether data shows any form of clustering . Importantly , unlike the silhouette index , this allows us to distinguish between k=1 and k>1 clusters . For each subject , the pairwise distances between the spatial locations of all electrodes in a single subject were computed . Using each electrode in turn as a 'viewer' ( Kalogeratos and Likas , 2012 ) , we tested to see whether the distribution of distances to that electrode significantly deviated from unimodality ( Hartigan and Hartigan , 1985 ) . If one or more electrodes showed a signficantly non-unimodal pairwise distance histogram , then the data were considered to be clustered . Following this procedure , k-means clustering was performed with k=2 through k=6 clusters , and the silhouette index was used to determine the best number of clusters for a given subject . The silhouette index for a given data point is defined ass ( i ) =b ( i ) −aimax{a ( i ) , b ( i ) } where b ( i ) is the lowest average distance of i to any other cluster of which i is not a member , and a ( i ) is the average distance between i and any other data point assigned to the same cluster . The silhouette index ranges from −1 to 1 , with higher positive values indicating good clustering . For the speaking and listening CV syllable tasks , the start of the syllable acoustics was used to align the responses of each electrode site . For the phoneme responses , the TIMIT phonetic transcriptions were used to align responses to the phoneme onset . Once responses were aligned to a stimulus , the average activity for each site to each stimulus was measured by taking the mean response over different trials of the same stimuli . The maximum of the mean responses to different stimuli were then used to measure the peak-high gamma distributions between different tasks and sites . We measured the onset latencies for responses to listening in STG and vSMC by calculating the average z-scored high gamma activity across all CV syllables , and then calculating the first time at which activity was significantly higher than the 500-ms pre-stimulus silent rest period ( one-tailed Wilcoxon rank sum test , p<0 . 001 ) . We also calculated the peak latency as the time at which the average z-scored response reached its maximum value . Differences in onset and peak latencies were compared across STG , inferior , and superior vSMC using the a two-tailed Wilcoxon rank sum test at a significance level of p<0 . 05 ( uncorrected ) . To measure the timing/dynamics between pairs of vSMC and STG sites during CV syllable listening , we performed a cross-correlation analysis between pairs of electrodes in these two regions . The cross-correlation measures the similarity of two time series at different time lags by taking pairs of electrode responses and calculating the correlation between one response and a time-shifted version of the second response . If the peak in the cross-correlation between an STG electrode and a vSMC electrode occurs at a negative lag , this indicates that the STG response leads ( occurs earlier than ) the vSMC response and that STG activity in the past is predictive of future activity in the vSMC . In contrast , if the peak in the cross-correlation between an STG electrode and a vSMC electrode occurs at a positive lag , this indicates that the vSMC response leads ( occurs earlier than ) the STG response . The cross-correlation at time lag τ is calculated between the response at an STG electrode ( denoted x ) and the response at a vSMC electrode ( y ) as follows:x*y[τ] = def∑t=-0 . 5 st=1 sx*[t]y[t+τ] Where the maximum lag τ was chosen to be 0 . 75 s . Cross-correlations were normalized by 1M−|τ| ( where M is the total number of time points in the response ) to obtain an unbiased estimate at each time lag τ . The cross-correlation between vSMC and STG electrodes was calculated separately for each CV syllable trial , and then averaged across trials ( see examples in Figure 3d ) . To determine the incidence of relationships within our electrode population where STG leads vSMC , vSMC leads STG , or both are coactive , we calculated an asymmetry index . This index ranges from −1 to 1 and describes the relative power in the positive versus negative lags for each vSMC electrode . It is calculated for each vSMC electrode by taking the sum of the positive cross-correlations in the negative lags and the sum of the positive cross-correlations in the positive lags , and then computing the ratio:asymmetry index = Ppos−PnegPpos+Pneg For a given vSMC electrode , an asymmetry index of −1 indicates that the cross-correlations lie fully in the negative lags ( indicating that STG responses lead the vSMC response in that electrode ) . In contrast , a value of 1 indicates that the cross-correlations are in the positive lags only , indicating that the vSMC electrode leads all STG electrodes . To examine the relational organization of the neural responses to syllables , we applied unsupervised multidimensional scaling ( MDS ) to the distance matrix of the mean neural responses at the sites of interest described in Materials and methods: Electrode selection . For analysis of speaking and listening responses , the vSMC sites used were those identified as significantly active during both speech production and speech perception ( n=98 , Figure 4b , c , f ) . However , clustering results for speaking were similar when all vSMC sites identified as significantly active during speech production were included ( n=362 , Figure 4—figure supplement 3c ) . The STG sites used were those identified as significantly active during speech perception ( n=340 , Figure 4e , Figure 4—figure supplement 3b ) . Syllables placed closer together in MDS space elicited similar neural response patterns , and those further apart from one another elicited more dissimilar patterns . To calculate the distance between a pair of mean neural responses , a mean neural response to one syllable was linearly correlated to another , and the resulting correlation coefficient was subtracted from 1 . We used unsupervised K-means clustering to examine the grouping of the mean neural activity to syllables of the electrodes of interest described in Methods: Electrode selection . We clustered the mean activity into 3 distinct clusters . This number of clusters was chosen because there are 3 major place of articulations and manner of articulations in the syllable stimuli set ( Figure 4a ) that have been shown to play a major role in the neural organization of motor cortex during speech production and auditory cortex during speech perception . After clustering the neural responses into three distinct groups , we measured the similarity of the grouping to the linguistically defined grouping of consonants by place of articulation and acoustic features ( Figure 4a and Figure 2—figure supplement 1 ) using the adjusted Rand Index ( RIadj ) . The RIadj is frequently used in statistics for cluster validation . It measures the amount of agreement between two clustering schemes: one by a given clustering process ( e . g . K-means ) , and the other by some external criteria , or gold-standard ( e . g . place of articulation linguistic features ) . The RIadj takes an intuitive approach to measuring cluster similarity by counting the number of pairs of objects classified in the same cluster under both clustering schemes , and controlling for chance ( hence , 'adjusted' RI ) . It has an expected value of 0 for independent clusterings , and a maximum value of 1 for identical clustering . It is defined as the following: Let S be a set of n objects , S = ( o1 , o2 , … , on ) . Partitioning the objects in two different ways such that U = ( U1 , … , Ur ) is a partition of S into r subsets , and V = ( V1 , … , Vt ) is a partition of S into t subsets , let: a = number of pair of objects that are in the same set in U and in the same set in V , b = number of pair of objects that are in the same set in U and in different sets in V , c = number of pair of objects that are in different sets in U and in the same set in V , d = number of pair of objects that are in different sets in U and in different sets in V . Without adjusting for chance , the RI is simply:RI = a+da+b+c+d=a+dn2 . Taking into account chance pairings , RIadj becomes:RIadj = n2 ( a+d ) - ( a+b ) ( a+c ) + ( c+d ) ( b+d ) n22- ( a+b ) ( a+c ) + ( c+d ) ( b+d ) . To localize an unbiased time window for analysis , the ΔRIadj metric was derived for all time windows by subtracting the RIadj measured with the place of articulation features gold-standard from the RIadj measured with the acoustic feature gold-standard ( Figure 4—figure supplement 1 ) . An ΔRIadj = 1 denotes organization by acoustic features , and an ΔRIadj = −1 denotes organization by place features . The significance of the ΔRIadj was computed by calculating the RIadj for a randomized labeling of neural responses compared to either acoustic feature or place feature clustering , taking the difference ( ∆RIadj ) , and repeating this procedure 1000 times with different randomized labelings to create a null distribution of ∆RIadj values . The p-value was calculated as the number of times this random ∆RIadj exceeded the observed ∆RIadj , and was thresholded at an FDR-corrected p<0 . 05 using the Benjamini-Hochberg procedure ( Benjamini and Hochberg , 1995 ) . To characterize the phoneme selectivity of each electrode site , we implemented the PSI calculation described by Mesgarani et al . , 2014 . In short , for a single site , we summed the number of responses that were statistically different ( Wilcoxon rank-sum test , p<0 . 01 , corrected for multiple comparisons ) from the response to a particular phoneme . This resulted in a PSI that ranges from 0 to 32 , where a PSI = 32 is an extremely selective electrode and a PSI = 0 is not selective . A PSI describes an electrode’s selectivity to one phoneme , and a vector of PSIs describes an electrode’s selectivity profile to all phonemes . The spectrotemporal representation of speech sounds was first estimated using a cochlear frequency model , consisting of a bank of logarithmically spaced constant Q asymmetric filters . The filter bank output was subjected to nonlinear compression , followed by a first order derivative along the spectral axis modeling a lateral inhibitory network , and an envelope estimation operation ( Wang and Shamma , 1994 ) . This resulted in a two dimensional spectrotemporal representation ( spectrogram ) of speech sounds simulating the pattern of activity on the auditory nerve . We then estimated the spectrotemporal receptive fields ( STRFs ) of the sites from passive listening to TIMIT using normalized reverse correlation ( Aertsen and Johannesma , 1981; Klein et al . , 2000; Theunissen et al . , 2001; Woolley et al . , 2006 ) between spectrotemporal representation of the sentences and the evoked neural activity ( STRFLab software package: http://strflab . berkeley . edu , DirectFit routine ) . The STRF is a linear filter that describes which combinations of spectrotemporal features will elicit a neural response in a given electrode . The relationship between the STRF , H , stimulus spectrogram , S ( as estimated above ) , and the predicted response , r^ ( t ) , of an electrode are given by the following equation:r^ ( t ) =∑i=0M-1 ∑τ=0N-1 H ( τ , f ) S ( t -τ , f ) where N is the number of delays of length τ after which the STRF will be estimated ( reflecting memory for the stimulus ) , and M is the number of frequency bands in the spectrogram . To estimate the STRF , we minimize the mean squared error between the predicted and observed responses . To prevent overfitting , we used an L2 regularization procedure in which a ridge hyperparameter and sparseness hyperparameter were calculated for each electrode's STRF ( details in [Woolley et al . , 2006] ) . The ridge hyperparameter acts as a smoothing factor on the STRF , whereas the sparseness hyperparameter controls the number of non-zero weights in the STRF . These hyperparameters were optimized with a systematic hyperparameter grid search maximizing for mutual information ( bits/s ) . With the optimized hyperparameters , we calculated the final STRF and correlation between the predicted and actual neural response using cross-validation . To do this , a STRF was derived using 9/10 of the stimuli-response pairs , and the Pearson correlation coefficient ( indicating the STRF goodness-of-fit ) was measured by predicting the remaining one-tenth responses . This was repeated 10 times with 10 non-overlapping stimuli-response pair sets . The final STRF and correlation number were derived by averaging the 10 STRFs and correlation coefficients . To assess statistical differences , we used independent sample t-tests when the data were found not to deviate significantly from normality ( KS test ) . When data were not normally distributed , we used the nonparametric Wilcoxon rank sum test . In some cases , a bootstrap t-test was used .

When we speak , we force air out of our lungs so that it passes over the vocal cords and causes them to vibrate . Movements of the jaw , lips and tongue can then shape the resulting sound wave into speech sounds . The brain’s outer layer , which is called the cortex , controls this process . More precisely , neighboring areas in the so-called motor cortex trigger the movements in a specific order to produce different sounds . Brain imaging experiments have also shown that the motor cortex is active when we listen to speech , as well as when we produce it . One theory is that when we hear a sound , such as the consonant ‘b’ , the sound activates the same areas of motor cortex as those involved in producing that sound . This could help us to recognize and understand the sounds we hear . To test this theory , Cheung , Hamilton et al . studied how speech sounds activate the motor cortex by recording electrical signals directly from the brain’s surface in nine human volunteers who were undergoing a clinical evaluation for epilepsy surgery . This revealed that speaking activates many different areas of motor cortex . However , listening to the same sounds activates only a small subset of these areas . Contrary to what was thought , brain activity patterns in motor cortex during listening do not match those during speaking . Instead , they depend on the properties of the sounds . Thus , sounds that have similar acoustic properties but which require different movements to produce them , such as ‘b’ and ‘d’ , activate the motor cortex in similar ways during listening , but not during speaking . Further research is now needed to work out why the motor cortex behaves differently when we hear as opposed to when we speak . Previous work has suggested that the region increases its activity during listening when the sounds heard are unclear , for example because of background noise . One testable idea therefore is that the motor cortex helps to enhance the processing of degraded sounds .

2016

The auditory representation of speech sounds in human motor cortex

Secondary metabolites have a great potential as pharmaceuticals , but there are only a few examples where regulation of gene cluster expression has been correlated with ecological and physiological relevance for the producer . Here , signals , mediators , and biological effects of terrein production were studied in the fungus Aspergillus terreus to elucidate the contribution of terrein to ecological competition . Terrein causes fruit surface lesions and inhibits plant seed germination . Additionally , terrein is moderately antifungal and reduces ferric iron , thereby supporting growth of A . terreus under iron starvation . In accordance , the lack of nitrogen or iron or elevated methionine levels induced terrein production and was dependent on either the nitrogen response regulators AreA and AtfA or the iron response regulator HapX . Independent signal transduction allows complex sensing of the environment and , combined with its broad spectrum of biological activities , terrein provides a prominent example of adapted secondary metabolite production in response to environmental competition . Fungal secondary metabolites ( SMs ) are often encoded by gene clusters and have been the subject of intensive analysis in recent years . This has resulted from the large number of completed fungal genomes as well as the intrinsic interest of SMs as environmental/pathogenic agents and their potential for pharmacological use . However , there are few examples where the regulation of the expression of these clusters has been shown to be ecologically/physiologically relevant ( Rohlfs and Churchill , 2011 ) . Except for penicillin and perhaps aflatoxin and gliotoxin , the potential real-life functions of most SMs are unknown ( Bhatnagar et al . , 2006; Vargas et al . , 2014 ) . Most clusters contain a pathway-specific transcription factor , and the common approach is to artificially over-express the cluster genes by constitutive expression of this transcription factor and chemically characterise the product ( Yin and Keller , 2011 ) . In most cases the relevance of the SM is not known . In this study the biological function of terrein from the filamentous ascomycete Aspergillus terreus was investigated . A . terreus is a filamentous ascomycete of biotechnological and medical importance since it produces itaconic acid ( Steiger et al . , 2013 ) and lovastatin ( Bizukojc and Ledakowicz , 2015 ) , but it is also a causative agent of life-threatening invasive aspergillosis in immunocompromised patients ( Baddley et al . , 2003; Slesiona et al . , 2012b ) and has recently been described as a pathogen of potato leaves ( Louis et al . , 2013 ) . While searching for a candidate protein required for pigment synthesis in A . terreus conidia , we serendipitously identified the gene cluster responsible for terrein production ( Zaehle et al . , 2014 ) , which is the major SM formed by A . terreus . Terrein production is highly pronounced on sugar-rich plant-derived media such as potato dextrose broth ( PDB ) ( Zaehle et al . , 2014 ) and shows phytotoxic activities such as inhibition of seed germination and lesion formation on fruit surfaces ( Kamata et al . , 1983; Zaehle et al . , 2014 ) . The terrein biosynthesis gene cluster consists of 11 genes ( terA–J , terR ) , whereby terA encodes the key enzyme , which is a non-reducing polyketide synthase ( Zaehle et al . , 2014 ) , and terR codes for the transcriptional activator TerR containing a GAL4-type Zn2Cys6 binuclear cluster DNA-binding domain ( Gressler et al . , 2015a ) , as frequently found in transcriptional activators of fungal SM gene clusters . Although activation of cluster genes is strictly TerR-dependent , no signals that lead to TerR activation under in vivo conditions have yet been identified . Since terrein is the major SM produced by A . terreus—gram scales are easily achieved—we assumed a benefit from its production in the natural habitat . To elucidate this question , a detailed knowledge on the inducing factors stimulating terrein production and analyses of its biological activities were required . Therefore , we aimed to investigate environmental signals that result in terR expression and , eventually , the production of terrein . From these analyses , the impact of different global transcription factors on cluster induction was deduced . Two global transcription factors sense the quality and availability of nitrogen sources and specifically respond to the plant and fruit environment . Additionally , the iron responsive regulon plays a vital role in cluster induction , which indicates a specific contribution of terrein in modulating iron homeostasis . In previous studies we showed that terrein is produced on plant-derived media such as PDB , which is in agreement with its phytotoxic biological activity ( Zaehle et al . , 2014 ) . To address the question of specific conditions that induce the gene cluster , an A . terreus reporter strain was generated expressing the β-galactosidase gene lacZ from Escherichia coli under control of the terrein synthase promoter PterA ( PterA:lacZ ) . Due to the dependence of PterA expression on TerR , this strain served as a direct indicator of terR expression and terrein production . In agreement with a lack of terrein production , β-galactosidase activity was near the detection limit when the A . terreus PterA:lacZ strain was grown on glucose minimal medium . In contrast , and in agreement with previous observations , a 200–500 fold induction was observed on PDB medium ( Figure 1A ) . Similarly , on Sabouraud and yeast extract-peptone-dextrose ( YPD ) medium , induction levels reached 10–30% compared with PDB . However , potato broth or casamino acids did not induce the cluster without the addition of glucose , indicating that glucose appears to be required for terrein production rather than repressing gene cluster induction as shown for other SM gene clusters ( Theilgaard et al . , 1997; Brakhage et al . , 2004; Gressler et al . , 2011 ) . Indeed , when glucose medium was supplemented with 1% casamino acids as the nitrogen source , a 20–30 fold activation was detected . Since terrein can cause lesions on fruit surfaces and inhibits plant seed germination ( Kamata et al . , 1983; Zaehle et al . , 2014 ) , we assumed that sugar-rich fruit and root juices might have a strong stimulatory capacity . Therefore , we cultivated the reporter strain on banana , carrot , peach , and apple juice . β-Galactosidase activities from these media exceeded the activity of the already strong inducing PDB medium ( Figure 1B ) by a factor of at least five . Additionally , in subsequent LC analyses of culture extracts , a distinct ultraviolet signal for terrein was detected ( Figure 1—figure supplement 1A–D ) . This led us to infect fresh bananas with the A . terreus ΔakuB strain ( the parental strain for gene deletions; Gressler et al . , 2011 ) , a ΔterR mutant lacking the transcriptional activator , and a ΔterA deletion mutant that lacks the key polyketide synthase from the cluster . 10 . 7554/eLife . 07861 . 003Figure 1 . Terrein production and expression of terA on plant-derived media . ( A ) Promoter activity of strain PterA:lacZ after 48 hr and 72 hr on minimal media supplemented with glucose ( Gluc ) , casamino acids ( CA ) , or glucose + casamino acids ( Gluc CA ) or the complex media Sabouraud ( Sab ) , yeast extract-peptone-dextrose ( YPD ) , potato dextrose broth ( PDB ) or potato extract ( PB ) . ( B ) Promoter activity of strain PterA:lacZ grown for 24 , 36 and 48 hr in carrot , banana , apple , and peach juice . ( C ) High performance liquid chromatography analysis of banana extracts infected with Aspergillus terreus SBUG844 strains ΔakuB , ΔakuBΔterA or ΔakuBΔterR . A mock-infected fruit served as control . 1—terrein . ( D ) Lesion formation on banana surfaces caused by extracts shown in ( C ) . Photographs were taken after 40 and 60 hr . Lesions only occur with extracts from the wild-type and the parental strain of the mutants ( ΔakuB ) . Numbers indicate the serial twofold dilution of the extracts starting from undiluted crude extracts down to 1:256 dilutions . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 00310 . 7554/eLife . 07861 . 004Figure 1—figure supplement 1 . High performance liquid chromatography ( HPLC ) analysis of extracts from Aspergillus terreus strains after cultivation in fruit juices and from infected nectarines and apples . ( A–D ) HPLC analyses of culture extracts from fruit juices after inoculation with SBUG844_PterA:lacZ ( blue line ) and mock-inoculated juices that served as negative controls ( green line ) . Culture supernatants were extracted 48 hr post inoculation: ( A ) carrot juice , ( B ) banana juice , ( C ) apple juice , ( D ) peach juice . 1 = terrein . ( E , F ) HPLC analyses of fruit extracts after infection with A . terreus wild-type SBUG844ΔakuB ( blue line ) and terrein biosynthesis gene cluster mutants ΔterA ( red ) and ΔterR ( violet ) . Mock-infected fruits served as negative controls ( green line ) : ( E ) nectarine , ( F ) apple . 1 = terrein . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 00410 . 7554/eLife . 07861 . 005Figure 1—figure supplement 2 . β-Galactosidase activity of PterA:lacZ grown in banana juice without ( control ) or with different supplementations . +Fe = 30 µM FeCl3; +N = 70 mM NH4Cl; +Met = 10 mM L-methionine . Tests were performed in biological triplicates that were analysed in technical triplicates . Promoter activity of terA is reduced by nitrogen supplementation and is completely repressed in the presence of nitrogen and iron in the medium . The addition of methionine partially overwrites this repression . Statistical significance was calculated in comparison to the control condition without supplementation using the Student's paired t-test with a two-tailed distribution: **p<0 . 01 , ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 005 Ethyl acetate extraction of bananas infected with ΔakuB strain revealed high amounts of terrein , whereas no terrein was detected after infection with the ΔterR or ΔterA mutant ( Figure 1C ) . Similar results were obtained when fresh apples or nectarines were infected with the three A . terreus strains ( Figure 1—figure supplement 1E , F ) . Furthermore , the extracts of bananas infected with A . terreus wild-type ( and ΔakuB ) caused strong lesion formation on fresh banana peels , which was hardly observed with extracts of bananas infected with the mutants ( Figure 1D ) . Taken together , these results confirm terrein production in a natural habitat of fruit infection , although the specific factors leading to terrein production still remain unclear . Since casamino acid-supplemented glucose medium revealed moderate gene cluster activation ( Figure 1A ) , we assumed that specific amino acids could act as inducers . Therefore , we tested the inducing effect of amino acids as sole nitrogen sources by preparing six distinct pools that covered all 20 canonical proteinogenic amino acids ( Figure 2—figure supplement 1 ) . While most of the pools did not activate terA expression , the pool with aliphatic amino acids caused moderate induction and media containing the sulphur-rich amino acids cysteine and methionine strongly induced gene expression , especially after 72 hr of cultivation ( Figure 2—figure supplement 1 ) . In the presence of the preferred nitrogen source glutamine , the induction by aliphatic amino acids was mainly lost . Also , the activation by cysteine was low and strongly concentration-dependent ( Figure 2A ) . In contrast , methionine provoked a significant induction even at low concentrations . Therefore , we also analysed the effect of other sulphur-containing compounds in two concentrations ( Figure 2B ) . However , no induction was observed when homocysteine , cystathionine , glutathione , dimethylsulfoxide , sulfate , or sulfide were tested . This indicates that methionine itself rather than a sulphur source causes the induction . 10 . 7554/eLife . 07861 . 006Figure 2 . Methionine-dependent terA expression . ( A ) β-Galactosidase activity of SBUG844_PterA:lacZ after 48 hr of cultivation in glutamine-containing minimal media in combination with 5 , 10 , or 20 mM of the aliphatic ( Ala , Ile , Leu , Val ) or sulphur-containing amino acid ( Cys , Met ) . Significance calculated against the glutamine control . ( B ) β-Galactosidase assay of SBUG844_PterA:lacZ in the presence of various sulphur sources . Glutamines containing minimal media were supplemented with low ( 5 mM ) or high ( 10 mM ) concentrations of Na2SO4 , Na2S , methionine ( Met ) , cysteine ( Cys ) , homocysteine ( Hcy ) , reduced gluthathione ( GSH ) , and dimethylsulfoxide ( DMSO ) . Cystathionine ( Cth ) was used in final concentrations of 1 and 3 mM . Activity was determined after 48 hr of growth . Significance calculated against medium supplemented with sodium sulfate . All cultivations were performed in biological triplicates and activity determinations were made in technical duplicates . Statistical significance was calculated by the Student's paired t-test with a two-tailed distribution: *p<0 . 05 , **p<0 . 01; ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 00610 . 7554/eLife . 07861 . 007Figure 2—figure supplement 1 . Amino acid-dependent terA expression . All cultivations were performed in biological triplicates and activity determinations were made in technical duplicates . Statistical significance was calculated by the Student's paired t-test with a two-tailed distribution: *p<0 . 05 , **p<0 . 01; ***p<0 . 001 . β-Galactosidase assay of SBUG844_PterA:lacZ grown in AMM ( -N ) G100 with 1% casamino acids ( CA ) or 50 mM of various amino acid pools as indicated in the table on the right . The promoter activity of terA was determined after 48 and 72 hr . Significance calculated against CA control . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 007 Methionine and PDB medium induced the terrein biosynthesis gene cluster to a similar extent , but remained approximately five times below that of banana juice . This suggested that additional or alternative inducing signals might exist . Fruit juices are rich in sugars and the C:N ratio in fruits is very high , which could result in severe nitrogen limitation at a later growth state . Therefore , another set of experiments was performed in which the total concentration of amino acids was either high ( 50 mM ) or low ( 10 mM ) and PterA induction from the lacZ reporter strain was determined at different time points ( Figure 3A ) . While high concentrations of amino acids or inorganic nitrogen sources such as nitrate or ammonia did not induce the cluster , low nitrogen contents resulted in a time-dependent 40–400 fold induction . To correlate gene expression with nitrogen exhaustion , the nitrogen consumption from a medium supplemented with 10 mM ammonium chloride was monitored and cluster induction was simultaneously analysed . Indeed , as soon as nitrogen levels reached the detection limit , cluster expression was strongly induced ( Figure 3B ) . Subsequently , a culture shift experiment was performed in which mycelium was pre-grown in the presence of 70 mM ammonium chloride and shifted to medium either with or without a nitrogen source ( Figure 3C ) . While no reporter activity and terrein formation was observed from cultures of the nitrogen-rich medium , high β-galactosidase activity accompanied by terrein accumulation was observed in the nitrogen starvation medium . 10 . 7554/eLife . 07861 . 008Figure 3 . Terrein biosynthesis gene cluster activation under nitrogen starvation . ( A ) β-Galactosidase activity of SBUG844_PterA:lacZ was cultivated for 24 , 36 , 48 , 72 , and 96 hr in glucose minimal medium supplemented with different concentrations of various nitrogen sources: 70 mM and 10 mM NaNO3 or NH4Cl , 1% or 0 . 1% of casamino acids ( CA ) , and 50 or 10 mM aspartate ( Asp ) , asparagine ( Asn ) , glutamate ( Glu ) , or glutamine ( Gln ) . ( B ) Correlation of nitrogen consumption and terA promoter activity determined by β-galactosidase activity and ammonia consumption of SBUG844_PterA:lacZ in glucose minimal medium with 10 mM NH4Cl . ( C ) β-Galactosidase activity of SBUG844_PterA:lacZ in nitrogen shift experiments . Cultures grown for 48 hr in glucose minimal medium with 70 mM NH4Cl ( PC ) were washed and transferred to medium with ( +N ) or without 70 mM NH4Cl ( −N ) and promoter activity was determined after 15 hr of cultivation . ( D ) Carbon source dependent terA promoter activation under nitrogen starved conditions . Strain SBUG844_PterA:lacZ was pre-cultured for 48 hr on casamino acids without sugar supplementation . The mycelium was washed and transferred to nitrogen-free media with different carbon sources . Reporter activity was determined 24 hr after the shift . ( E ) β-Galactosidase activity from bananas infected with conidia suspension of SBUG844 wild-type and PterA:lacZ . Sections ( red boxes ) were cut from bananas , ground to a fine powder and subjected to β-galactosidase activity determination . Activity was only detected from the reporter strain . ( F ) Quantification of terrein from wild-type infected bananas with or without ammonium supplementation . All tests were performed in biological triplicates and technical duplicates; p values were calculated by the Student's paired t-test with a two-tailed distribution: **p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 008 These results confirmed that nitrogen starvation acts as an inducing factor that might be responsible for the high terrein production levels on fruit juices . In agreement , when ammonium chloride was added to banana juice , β-galactosidase activity was strongly reduced ( Figure 1—figure supplement 2 ) . Similarly , the PterA:lacZ reporter strain displayed high β-galactosidase activity when bananas were directly infected ( Figure 3E ) . When bananas were supplemented with ammonium chloride prior to infection with the wild-type , ethyl acetate extracts of the bananas revealed a 50% reduction in terrein content compared with bananas without nitrogen supplementation ( Figure 3F ) . This finding clearly indicates that nitrogen limitation is a major inducer for terrein production under natural conditions . However , it should be mentioned that the presence of sugars was always required , since especially hexoses from mono- and disaccharides provoked strong expression under nitrogen-limited conditions ( Figure 3D ) , which is in direct contrast to the production of dihydroisoflavipucine in A . terreus that belongs to the class of fruit and root rot toxins . This metabolite is only produced in the strict absence of sugars and requires preferred nitrogen sources such as glutamine or asparagine for induction ( Gressler et al . , 2011 ) . Nitrogen starvation and methionine marked important signals for terrein cluster induction . To unveil the global regulators that could be involved in signal transduction , the A . terreus genome was analysed for the presence of transcription factors known to play a role in nitrogen sensing , cross pathway control of amino acid synthesis , and stress response . The global nitrogen regulator AreA ( ATEG_07264 ) ( Arst and Cove , 1973; Hynes , 1975; Davis et al . , 2005 ) , the cross pathway control regulator CpcA ( ATEG_03131 ) ( Hoffmann et al . , 2001; Krappmann et al . , 2004 ) , the stress response bZIP transcription factor AtfA ( ATEG_04664 ) ( Balazs et al . , 2010; Lara-Rojas et al . , 2011 ) , and the nitrogen starvation-induced ras-protein RhbA ( ATEG_09480 ) ( Panepinto et al . , 2002 ) were selected for gene deletions . All mutants were tested for their growth properties in the presence of proteinogenic amino acids or ornithine , citrulline , urea , and casamino acids , all of which were used as the sole nitrogen source . Additionally , several complex media were analysed ( Figure 4—figure supplement 1 ) . Only minor growth defects were observed with the cpcA and rhbA mutants , with some general growth reduction on selected amino acids ( Asp , His , Ser , Thr ) or complex media . However , among all the mutants tested , the areA mutant revealed the most severe growth defects . This mutant grew like the wild-type on glutamine and displayed some reduced growth on the nitrogen-rich amino acids asparagine , aspartate , and histidine as well as on urea and ammonium chloride . However , the ΔareA strain was unable to use any other proteinogenic amino acid . As described for other Aspergillus species ( Hunter et al . , 2014 ) , these results confirm an essential role of AreA in nitrogen sensing and utilisation in A . terreus . In contrast , the atfA mutant was only impaired in growth at high aspartate concentrations ( 50 mM ) . However , conidia of this mutant lacked the typical yellow-brown pigmentation . Additionally , the atfA mutant completely lacked the typical red colouration of the medium in the presence of methionine , which has previously been demonstrated to be associated with terrein production ( Zaehle et al . , 2014 ) . This implied a regulatory role of atfA on SM production in A . terreus . On the contrary , the atfA mutant did not show increased sensitivity against oxidative or osmotic stress as described for other Aspergillus species ( Balazs et al . , 2010; Lara-Rojas et al . , 2011 ) . Finally , a double deletion of the areA and atfA genes was generated ( ΔareAΔatfA ) resembling the growth phenotypes of both single mutants , since it was only able to grow on the media that supported growth of the areA mutant where it formed white conidia , as observed for the ΔatfA strain . To test the effect of transcription factor mutations on terrein biosynthesis gene cluster activation , all mutants were pre-grown on non-inducing glucose medium with glutamine and transferred to medium with or without nitrogen . Terrein was quantified after 24 hr ( Figure 4A ) . The wild-type , the ΔcpcA , and the ΔrhbA mutant revealed low terrein production in the presence of nitrogen , but high terrein titers when nitrogen was omitted . This indicates that cpcA and rhbA are dispensable for terrein production . In contrast , both the ΔareA and the ΔatfA mutant only produced marginal amounts of terrein under nitrogen-limited conditions ( Figure 4A ) , and this effect was completely cured in the complemented mutants areAc and atfAc . No terrein was detected in a ΔareAΔatfA double mutant . Therefore , both areA and atfA appear essential for terrein production under nitrogen limitation . To confirm this assumption , we expressed the cluster-specific transcription factor gene terR under control of the gpdA promoter in a ΔareAΔatfA mutant background . A constitutive terrein production was observed , indicating that AreA and AtfA regulate terR expression , but are not essential for expression of the structural genes which directly depend on TerR ( Figure 4A ) . Second , we analysed the effect of areA and atfA deletion on gene expression by qRT-PCR from strains shifted for 0 hr , 12 hr , and 24 hr to nitrogen starvation . qRT-PCR was performed on the cluster genes terA , terB , the specific activator terR , and the global transcription factors areA and atfA ( Figure 4B ) . 10 . 7554/eLife . 07861 . 009Figure 4 . Terrein quantification and expression of terrein cluster genes in nitrogen regulator mutants . ( A ) Terrein quantification from the parental strains SBUG844ΔakuB ( ΔakuB ) , regulator mutants ( ΔcpcA ) , ( ΔrhbA ) , ( ΔareA ) , ( ΔatfA ) , ( ΔareAΔatfA ) , complemented mutants ( areAC ) , ( atfAC ) , and strain SBUG844ΔakuBΔareAΔatfAΔ::AnPgpdA:terR with terR overexpression in the ΔareAΔatfA background . Mycelia were pre-grown in glutamine-supplemented media , washed and transferred to minimal medium with ( +N ) or without ( −N ) 50 mM glutamine . Terrein was quantified from supernatants 24 hr after the shift . ( B ) qRT-PCR of strains ΔakuB , ΔareA , areAC , ΔatfA , atfAC , and ΔareAΔatfA were pre-cultivated for 40 hr in glutamine-supplemented media and transferred to nitrogen starvation . RNA was isolated at 0 , 12 , and 24 hr of starvation . Transcript levels were normalised against the actin gene actA by fold expression = 2^ ( CTtarget − CTactA ) . ( C ) Terrein quantification from strains shown in ( B ) after 72 hr of cultivation in glutamine-containing minimal medium supplemented with 10 mM methionine . ( D ) qRT-PCR from RNA of strains shown in ( C ) isolated after 48 hr of cultivation . Transcript levels were normalized against the enolase gene enoA by fold expression = 2^ ( CTtarget − CTenoA ) . ( E ) Top and bottom view of colonies of Aspergillus terreus wild-type ( ΔakuB ) and mutants ( ΔterA , ΔatfA , ΔareA ) grown for 72 hr on solid minimal media supplemented with 25 mM methionine as sole nitrogen source . The red pigmentation of the wild-type ( bottom view ) is lost in the ΔterA and ΔatfA mutants that show some enhanced growth while unable to produce terrein . The ΔareA strain is unable to grow . In all experiments biological triplicates with technical duplicates were analysed . Statistical significances in comparison to the parental ΔakuB strain were calculated by the Student's paired t-test with a two-tailed distribution: *p<0 . 05 , **p<0 . 01; ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 00910 . 7554/eLife . 07861 . 010Figure 4—source data 1 . Genotypes of strains used in the study . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 01010 . 7554/eLife . 07861 . 011Figure 4—source data 2 . List of oligo nucleotides used in the study . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 01110 . 7554/eLife . 07861 . 012Figure 4—figure supplement 1 . Analysis of colony formation and growth phenotypes of nitrogen regulator mutants in the presence of different nitrogen sourcesAs basal medium AMM-G100 without nitrogen was used , which was supplemented with 70 mM inorganic nitrogen ( NaNO3 or NH4Cl ) , 50 mM of standard amino acids , ornithine ( Orn ) , citrulline ( Cit ) , urea , or 1% casamino acids ( CA ) . Additionally , AMM-CA1% ( without glucose ) and the complex media yeast extract-peptone-dextrose ( YPD ) , potato dextrose agar ( PDA ) , and malt extract agar ( MA ) were used . Photographs were generally taken after 96 hr of incubation at 37°C . Asterisks denote different incubation times: * = 72 hr; ** = 48 hr . Note the severely reduced colony formation of the ΔareA strain on virtually all single amino acids except for arginine , glutamine , and histidine . Additionally , note the loss of conidia colouration in the ΔatfA strain . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 01210 . 7554/eLife . 07861 . 013Figure 4—figure supplement 2 . Biochemical characterisation of the recombinant AreA DNA-binding domain and in vitro binding to HGATAR motifs identified in the terR promoter of the Aspergillus terreus terrein biosynthesis gene cluster . ( A ) Amino acid sequence alignment of the AreA DNA binding domains from Aspergillus nidulans and A . terreus . Underlined amino acids denote the 65-residue peptide used for solving the solution structure of the AreA:DNA complex ( Starich et al . , 1998 ) . Cysteine residues of the Cys2–Cys2 zinc finger module are highlighted in yellow . ( B ) SDS-PAGE analysis of purified AreA663-797 . ( C ) In solution oligomeric state analysis of AreA663-797 as determined via size exclusion chromatography and multiangle static light scattering . The light scattering signal ( LS ) is shown overlaid with the calculated molar mass ( Mw ) across the elution profile as monitored by the absorbance at 280 nm ( UV ) and changes of the refractive index ( dRI ) . ( D ) Schematic presentation of putative GATA AreA binding sites ( BS 1 and 2 ) in the terR promoter . ( E–G ) Real-time in vitro surface plasmon resonance ( SPR ) interaction analysis of AreA663-797 with DNA containing the identified GATA motifs from the terR promoter of the terrein biosynthesis gene cluster . Sequences of DNA duplexes used for SPR analysis are shown on top of the sensorgrams . Numbers represent the GATA motif positions relative to the start of the open reading frame . HGATAR sites are highlighted in red . ( H ) Dissociation constants and stoichiometry of analysed AreA663–797:DNA interactions analysed by SPR . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 01310 . 7554/eLife . 07861 . 014Figure 4—figure supplement 3 . High performance liquid chromatography analyses from culture filtrates of SBUG844 wild-type and two independent atfA overexpression mutants ( AnPgpdA:atfA; OE 1 and 2 ) . Strains were cultivated under non-inducing conditions for the wild-type . ( A ) AMM-G100; ( B ) AMM-G100Gln50 . 1 = terrein . Terrein and its side products are detected from the atfA overexpressing strains but not from the wild-type . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 01410 . 7554/eLife . 07861 . 015Figure 4—figure supplement 4 . Southern blot analyses for Aspergillus terreus strains generated in this study . Genomic DNA was isolated and restricted with the respective restriction enzymes indicated in each panel . Digoxygenin-labelled probes were used in all experiments and were amplified by the oligonucleotide couples as indicated . Signals were detected by the CDPstar chemiluminescent substrate . Expected fragment sizes are also shown . ( A ) General scheme for generation of gene deletions using the pyrithiamine ( ptrA ) resistance cassette ( ΔGOI ) and for complementation of deletion mutants using the phleomycine ( ble ) resistance cassette ( GOIC ) via homologous recombination . ( B ) Generation of the β-galactosidase reporter strain SBUG844_PterA:lacZ in A . terreus SBUG844 wild-type . ( C ) Overexpression of atfA in SBUG844 under control of the A . nidulans gpdA promoter . The native atfA locus ( signal at 7008 bp ) is still intact in strains that integrated the overexpression construct . Strains with ectopic single copy integration of the construct used for downstream experiments are denoted by ‘X’ . ( D ) Deletion of the cpcA locus ( ATEG_03131 ) in SBUG844ΔakuB . ( E ) Deletion of the rhbA locus ( ATEG_09480 ) in SBUG844ΔakuB . ( F ) Deletion of the areA locus ( ATEG_07264 ) in SBUG844ΔakuB and complementation with FGSC A1156 areA . ( G ) PCR and PstI digest of PCR products from the areA upstream fragments amplified from genomic DNA from FGSC A1156 , SBUG844 , and SBUG844/areAC . The complemented strain shows the identical PstI restriction pattern as the amplified product from FGSC A1156 , whereas no PstI restriction site is present in the SBUG844 product , confirming the complementation of the ΔareA strain by the areA gene from A1156 . ( H ) Deletion of the atfA locus ( ATEG_04664 ) in SBUG844ΔakuB and complementation with SBUG844 atfA . ( I ) Deletion of the atfA locus ( ATEG_04664 ) in the areA negative background of SBUG844ΔakuBΔareA ( Δ ) and replacement of atfA in SBUG844ΔakuBΔareA with the terR overexpression construct under control of the A . nidulans gpdA promoter ( Δ::AnPgpdA:terR ) . ( J ) Partial deletion of the hapX locus ( ATEG_08 , 014 ) in SBUG844ΔakuB and complementation with the Aspergillus nidulans FGSC A4 hapX ( AN08251 ) . ( K ) Deletion of the sreA locus ( ATEG_07714 ) in SBUG844ΔakuB and complementation with SBUG844 sreA . ( L ) Deletion of the sidA locus ( ATEG_06879 ) in SBUG844ΔakuB and complementation with SBUG844 sidA . ( M ) Deletion of sidA ( ATEG_06879 ) in the terA negative background of SBUG844ΔakuBΔterA . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 01510 . 7554/eLife . 07861 . 016Figure 4—figure supplement 5 . Terrein production and susceptibility to osmotic stress in a sakA mutant . In all experiments biological triplicates with technical duplicates were analysed . ( A ) Terrein quantification from strains SBUG844ΔakuB ( ΔakuB ) and two sakA deletion strains SBUG844ΔakuBΔsakA ( 1 ) and ( 2 ) ( ΔsakA ) cultivated in AMM ( -N ) G100 for 72 hr in the presence of either 70 mM inorganic ( NaNO3 , NH4Cl ) or 50 mM organic nitrogen ( Glu , Gln , Asp , Asn ) . ( B ) Terrein quantification from strains SBUG844ΔakuB ( ΔakuB ) , SBUG844ΔakuBΔsakA and the complemented strain SBUG844ΔakuBΔsakA/sakAC ( sakAC ) . Cultures were pre-grown in nitrogen-rich AMM-G100Gln50 , mycelia were washed and transferred to AMM ( -N ) G100 with ( +N ) or without ( −N ) 50 mM Gln . Terrein was quantified from supernatants 24 hr after the shift . ( C ) Susceptibility assay towards NaCl stress . 105 to 101 conidia of strain SBUG844ΔakuB ( ΔakuB ) , the deletion mutants SBUG844ΔakuBΔsakA ( ΔsakA ) and SBUG844ΔakuBΔatfA ( ΔatfA ) or their complemented strains SBUG844ΔakuBΔsakA/sakAC ( sakAC ) and SBUG844ΔakuBΔatfA/atfAC ( atfAC ) were point inoculated on AMM ( -N ) G100 plates supplemented with 10 mM Gln and 0 , 0 . 5 , 1 . 0 , and 2 . 0 M NaCl . Plates were incubated 72 hr or 5 d ( * ) at 37°C . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 016 In the wild-type the regulators areA and atfA showed a time-dependent increase in gene expression . While the regulator terR was most strongly upregulated after 12 hr , expression of the TerR-controlled terrein biosynthesis genes terA and terB continued to increase after 12 hr and reached 13 . 3 and 3 . 2 times the expression level of the actin control gene at 24 hr ( Figure 4B ) . On the contrary , while complemented mutants behaved like the wild-type , the areA and atfA deletions strongly reduced activation of terR and , in turn , the expression of terA and terB . Results from qRT-PCR perfectly coincided with the substantially reduced terrein production rates under nitrogen limitation in these mutants ( Figure 4A ) . AreA recognises the DNA-binding motif HGATAR , and two adjacent binding sites are generally required for transcriptional activation due to dimer formation of AreA monomers ( Ravagnani et al . , 1997 ) . In this respect , the terR promoter contains two putative AreA binding sites that match the HGATAR consensus ( BS1and BS2; positions −59 and −72 relative to the translational start point ) . Surface plasmon resonance ( SPR ) analyses with the Aspergillus nidulans AreA DNA-binding domain , which is 91% identical to the respective A . terreus AreA domain , showed that BS1 and BS2 are recognised with high affinity ( Figure 4—figure supplement 2 ) . This strengthens the model of a direct involvement of AreA in the activation of terR expression . However , the reduction of terR expression was less pronounced in the ΔareA than in the ΔatfA background ( Figure 4B ) , which indicates that AreA is not the only activator acting on the terR promoter . Therefore , we additionally searched for putative palindromic AtfA/Sko1 binding sites ( 5′-TKACGTMA-3′ ) in the promoter regions of the cluster ( Proft et al . , 2005 ) . Only one hit ( TGACGTCA ) was identified in the promoter of the structural gene terC . However , if one mismatch is allowed , there is a putative binding site at position −731 relative to the ATG start codon of terR ( 5´-TGGCGTCA-3´ ) , but it remains speculative whether this binding site is recognised by A . terreus AtfA . Nevertheless , it should be mentioned that even a single half site of the suggested motif could promote transcription factor binding and promoter induction ( Proft et al . , 2005 ) . We therefore conclude that , although direct evidence for AtfA binding at the terR promoter is lacking , both transcription factors seem to regulate terR expression . In agreement , terR , terA , and terB expression and terrein production showed the strongest decrease in the ΔareAΔatfA double knock-out mutant ( Figure 4A ) . In addition , a constitutive expression of atfA by the gpdA promoter led to increased terrein production ( Figure 4—figure supplement 3 ) . This supports our hypothesis of direct involvement of AtfA in terrein cluster regulation . Due to the requirement for atfA and areA in terrein biosynthesis gene cluster induction under nitrogen starvation , we also tested their contribution to methionine-dependent induction in nitrogen-supplemented medium ( Figure 4C ) . While the wild-type strain and complemented mutants showed high terrein production rates of up to 55 mg/g mycelium , the terrein levels in the ΔareA , ΔatfA , or ΔareAΔatfA double mutant remained near the detection limit and qRT-PCR was performed to confirm this result on the transcriptional level . Although areA and atfA were expressed in the wild-type only at low levels on glutamine/methionine medium ( Figure 4D ) , deletion of areA or atfA reduced terR and , consequentially , terA and terB expression . In the ΔatfA mutant terR transcription was completely abolished and , in agreement , transcription of terA and terB was no longer detected . An areA mutant is unable to use methionine as a nitrogen source ( Figure 4E ) , and its uptake may be limited leading to the loss of transcriptional activation . In contrast , the atfA mutant still uses methionine as a nitrogen source , but neither produces terrein nor pigmented conidia nor the red colouration of the medium which is associated with terrein production in the presence of methionine ( Figure 4E ) . This implies that atfA may be induced by a methionine-dependent signaling cascade that subsequently leads to terR expression . While nitrogen starvation stimulated terrein production , starvation of other macroelements such as carbon , sulphur , or phosphate did not result in terrein production ( data not shown ) . However , this did not exclude limitation of trace elements as inducing signals . Therefore , the β-galactosidase activity from the PterA:lacZ reporter strain was determined from cultures with reduced amounts of trace elements . Indeed , a decrease in trace elements was accompanied by increased terA promoter activity ( Figure 5A ) . To attribute this activation to a specific trace element , media with limited amounts of trace elements were supplemented with each of the single trace elements FeSO4 , ZnSO4 , CuSO4 , MnCl2 , Na2MoO4 , CoCl2 , and H3BO3 . Cluster induction was observed in all cultures except for that supplemented with FeSO4 ( Figure 5B ) , indicating cluster activation from iron limitation . Subsequently , a minimal medium was prepared that contained all trace elements but iron which led to strong cluster induction , and the addition of 40 µM FeCl3 completely repressed the induction ( Figure 5C ) . Thus , besides methionine and the limitation of nitrogen , a lack of iron—but no other trace element—induces the terrein biosynthesis gene cluster . 10 . 7554/eLife . 07861 . 017Figure 5 . Expression of terA , production of siderophores under iron limited conditions and iron reducing properties of terrein . All analyses were performed in biological triplicates and technical duplicates . Statistical significances were calculated by the Student's paired t-test with a two-tailed distribution . ( A ) β-Galactosidase activity from SBUG844_PterA:lacZ in AMM ( -N ) G100Gln50 medium with limited amounts of trace elements . Asterisks indicate p values vs 100% trace elements: **p<0 . 01; ***p<0 . 001 . ( B ) β -Galactosidase activity from SBUG844_PterA:lacZ in AMM ( -N ) G100Gln50 medium with 1% trace elements supplemented with 20 µM of specific trace elements: FeSO4 , ZnSO4 , CuSO4 , MnCl2 , Na2MoO4 , CoCl2 , or H3BO3 . Asterisks indicate p values vs activity without supplementation: ***p<0 . 001 . ( C ) β-Galactosidase activity from SBUG844_PterA:lacZ in AMM ( -N ) G100Gln50 medium with and without 40 µM FeCl3 . Asterisks indicate p values vs 40 µM FeCl3: ***p<0 . 001 . ( D ) High performance liquid chromatography ( HPLC ) profiles of lyophilised culture supernatants of ΔterA after 72 hr of cultivation in AMM ( -N ) G100Gln50 supplemented with 0 , 20 , or 200 µM FeCl3 . Peaks for the siderophores ferrichrysin ( 2 ) and coprogen ( 3 ) are indicated . ( E ) Fe ( III ) reduction by terrein determined by the TPTZ assay . Ascorbic acid served as control and maximum reduction by ascorbic acid was set as 100% . Analyses were made from duplicates . ( F ) HPLC profiles of terrein after incubation with different oxidised ( upper , red lines ) or reduced ( lower , blue lines ) metal ions . Left: Cu ( II ) SO4 and Cu ( I ) Ac; right: Fe ( III ) Cl3 and Fe ( II ) SO4 . 1 = terrein; 4 = propylene maleic acid . ( G ) Scheme of terrein oxidation during iron reduction leading to the formation of propylene maleic acid . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 01710 . 7554/eLife . 07861 . 018Figure 5—source data 1 . Analytical data of coprogen . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 01810 . 7554/eLife . 07861 . 019Figure 5—source data 2 . Analytical data of ferrichrysin . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 01910 . 7554/eLife . 07861 . 020Figure 5—figure supplement 1 . qRT-PCR expression analysis of genes from iron acquisition systems under iron-supplemented and limited conditions . All analyses were performed from biological triplicates and technical duplicates . Statistical significances were calculated by the Student's paired t-test with a two-tailed distribution . SBUG844 wild-type was grown in AMM-G100Gln50 under iron-supplemented ( 40 µM; +Fe ) and starved conditions ( −Fe ) . qRT-PCR was performed on genes assumed to be responsible for siderophore biosynthesis , siderophore transport , or reductive iron assimilation . Transcript quantities were normalised against enoA by fold expression = 2^ ( CTtarget − CTeno ) . Asterisks indicate p values vs 0 µM FeCl3: **p<0 . 01; ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 02010 . 7554/eLife . 07861 . 021Figure 5—figure supplement 2 . Chrome azol S ( CAS ) assay of coprogen and terrein . Both substances were dissolved in methanol and serial dilutions ( 125–15 . 63 µg ) were added to punched holes of CAS plates . Photographs were taken 48 hr after incubation at 30°C . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 02110 . 7554/eLife . 07861 . 022Figure 5—figure supplement 3 . pH- and time-dependent Fe ( III ) reduction by terrein assayed by TPTZ . 10 µM FeCl3 were incubated with 4 mM fixed concentration of terrein in acetate buffer in a pH range from 3 . 0 to 6 . 0 . The analyses were performed from duplicates . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 02210 . 7554/eLife . 07861 . 023Figure 5—figure supplement 4 . 1H NMR ( 500 MHz , MeOD; upper panel ) and 13C NMR ( 150 MHz , MeOD; lower panel ) of compound 4 , 2- ( ( E ) -prop-1-en-1-yl ) maleic acid . 1H NMR ( 500 MHz , MeOD ) : δ 6 . 22 ( 1H , d , 3J = 15 . 8 Hz ) , 6 . 15 ( 1H , dq , 3J = 15 . 8 Hz , 3J = 6 . 5 Hz , ) , 5 . 73 ( 1H , s ) , 1 . 85 ppm ( 3H , d , 3J = 6 . 5 Hz ) ; 13C NMR ( 600 MHz , MeOH ) : δ 172 . 6 , 169 . 5 , 151 . 7 , 136 . 4 , 130 . 2 , 118 . 5 , 18 . 8 ppm; HRMS: ( ESI+ ) : m/z calculated for C7H9O4: 157 . 0495 , found 157 . 0495 [M + H]+ . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 02310 . 7554/eLife . 07861 . 024Figure 5—figure supplement 5 . HSQC NMR ( 600 MHz , MeOD; upper panel ) and HMBC NMR ( 600 MHz , MeOD; lower panel ) of compound 4 , 2- ( ( E ) -prop-1-en-1-yl ) maleic acid . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 024 Fungal iron acquisition has been well investigated in A . nidulans ( Eisendle et al . , 2003 ) and Aspergillus fumigatus ( Haas , 2014 ) , but only limited information was available on the iron acquisition systems from A . terreus . A BLASTp search for orthologous genes in the genome of A . terreus NIH2624 ( Table 1 ) revealed that all genes required for a reductive iron assimilation ( RIA ) system ( fetC , freB , ftrA ) , siderophore biosynthesis ( sid genes , except sidG ) , siderophore transport ( mirA-D , sitA ) , and regulators of iron homeostasis ( srbA , sreA , and hapX ) are well conserved in the A . terreus genome . When tested under iron limitation , qRT-PCR analyses confirmed a strong induction of genes from siderophore biosynthesis , siderophore transport , and the RIA pathway ( Figure 5—figure supplement 1 ) . To confirm siderophore production by A . terreus , we cultivated the wild-type strain under iron-rich ( 200 µM FeCl3 ) and iron-limited conditions ( no or 20 µM FeCl3 ) and identified ferrichrysin ( 2 ) and coprogen ( 3 ) only under iron limitation ( Figure 5D and Figure 5—source data 1 , 2 ) ( Zähner et al . , 1963; Bertrand et al . , 2010 ) . Coprogen revealed a higher stability than ferrichrysin . Subsequent analyses involving siderophores were therefore based only on coprogen . When terrein was compared with purified coprogen for chelating iron , only coprogen showed the expected iron chelating activity in CAS agar plate assays ( Figure 5—figure supplement 2 ) . Thus , although terrein is produced under iron limitation , it does not depict a new kind of siderophore . 10 . 7554/eLife . 07861 . 025Table 1 . BLASTp analysis of homologous genes for iron uptake and regulation of iron homeostasis ( adapted by [Haas , 2012] ) DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 025Aspergillus fumigatus functionGene code‡ExpressionAspergillus terreus gene codeIdentity/similarityReductive iron assimilation ( RIA ) FetCFerroxidaseAFUA_5G03790−FeATEG_0803279%/89% FreBFerric reductaseAFUA_1G17270−FeATEG_1032253%/64% FtrAIron permeaseAFUA_5G03800−FeATEG_0803175%/84%Siderophore biosynthesis ( SB ) EstATAFC esteraseAFUA_3G03660−FeATEG_0407244%/58% NpgA/PptAPhosphopantetheinyl transferaseAFUA_2G08590−ATEG_0969556%/65% SidAOrnithine monooxygenaseAFUA_2G07680−FeATEG_0687978%/85% SidCFC NRPSAFUA_1G17200−FeATEG_0507360%/76% SidDFSC NRPSAFUA_3G03420−FeATEG_0748843%/59% SidFTransacylaseAFUA_3G03400−FeATEG_0507552%/67% SidGTransacetylaseAFUA_3G03650−Fenone– SidHMevalonyl hydrataseAFUA_3G03410−FeATEG_0150953%/67% SidIMevalonyl ligaseAFUA_1G17190−FeATEG_0507486%/91% SidLTransacetylaseAFUA_1G04450–ATEG_0377064%/76%Siderophore transporter ( SIT ) MirAEnterobactin transporterAN7800; -−FeATEG_0407168%/77% MirBTAFC transporterAN8540; -AFUA_3G03640−FeATEG_0271150%/68% MirC*AN7485; AFUA_2G05730−FeATEG_0676278%/87% MirD†Trichotecene efflux pumpAFUA_3G03440–ATEG_0748740%/58% SitA/SitT*AN5378; AFUA_7G06060−FeATEG_0632962%/73%Regulatory proteins HapXbZip-TFAFUA_5G03920−FeATEG_0801477%/83% ( hapX re-annotated ) SreAGATA TFAFUA_5G11260+FeATEG_0774167%/75% SrbA†HLH TFAFUA_2G01260−FeATEG_0815672%/82%Genes selected for qPCR analyses are highlighted in bold . *Genes annotated according to ( Schrettl et al . , 2008 ) . †Genes annotated according to ( Blatzer et al . , 2011 ) . ‡Gene codes ANxxxx refer to the A . nidulans genome . Since some antioxidative properties of terrein had previously been described ( Trabolsy et al . , 2014 ) , the ability of terrein in reducing ferric ( Fe3+ ) to ferrous ( Fe2+ ) iron was investigated by using the ferrous iron chelator 2 , 4 , 6-tripyridyl-S-triazine ( TPTZ ) . The strong antioxidant ascorbic acid ( Elmagirbi et al . , 2012 ) served as control ( Figure 5E ) . Although ascorbic acid showed a much higher reducing potential , terrein was also able to convert ferric to ferrous iron in a concentration-dependent ( Figure 5E ) and pH-dependent ( Figure 5—figure supplement 3 ) manner with a pH optimum of 3–4 . In this respect , we noticed a significant fall in the culture pH under iron limitation from 6 . 5 to about 3 . 5 , which agrees with the optimum pH for terrein-mediated iron reduction . Thus , since terrein is produced during growth under iron limitation , the combination with a fall in pH could indeed increase iron availability . To elucidate the structure of the terrein oxidation product , the terrein-mediated iron reduction assay was scaled up and extractions were subjected to high performance liquid chromatography ( HPLC ) analysis . Here , besides a major proportion of terrein that remained in its original structure ( Figure 5F ) , a new peak ( 4 ) was observed , which was identified by NMR analyses ( Figure 5—figure supplements 4 , 5 ) as 2- ( ( E ) -prop-1-en-1-yl ) maleic acid ( PMA , Figure 5G ) . The structure of the oxidation product implies that oxidation took place at the two hydroxylated carbon atoms of terrein , leading to a cleavage of the pentenone ring system as proposed by Trabolsy et al . ( 2014 ) . In contrast , conversion of terrein to PMA was not observed during incubation with Fe2+ , confirming that iron reduction was the cause of PMA formation . No conversion of terrein was observed with oxidised copper ions ( Cu2+ ) , indicating a limited reductive potential which is sufficient for iron but not for copper ( Figure 5F ) . In conclusion , while terrein cannot chelate iron to support the siderophore-mediated iron uptake , its reductive potential is sufficient to reduce ferric to ferrous iron , which may increase iron solubility and could ease the direct uptake of iron via the ferrous iron transport system . Due to the co-regulation with siderophore biosynthesis , we assumed that transcriptional regulators involved in iron homeostasis could also regulate terrein biosynthesis . In A . nidulans and A . fumigatus , siderophore biosynthesis is regulated by HapX , a transcriptional inducer under iron limitation , and SreA , a repressor in the presence of iron ( Haas , 2012 ) . To investigate the impact of SreA and HapX on siderophore ( coprogen ) and terrein synthesis , the sreA gene was completely and the hapX gene partially deleted ( the latter due to incomplete sequence information at the hapX locus tag ATEG_08014 ) . Additionally , while the sreA mutant was complemented with the A . terreus sreA gene , the hapX mutant was complemented with the hapX gene from A . nidulans FGSC A4 . All strains were cultivated in iron-supplemented and iron-limited media and terrein and coprogen were quantified from culture supernatants ( Figure 6A , B ) . All complemented mutants behaved similar to the wild-type with production of only marginal amounts of coprogen and terrein in the presence of iron . Under iron starvation the production rates for both metabolites strongly increased . 10 . 7554/eLife . 07861 . 026Figure 6 . Effect of hapX and sreA deletion on terrein and coprogen biosynthesis . ( A ) Coprogen or ( B ) terrein quantification from SBUG844ΔakuB ( ΔakuB ) , SBUG844ΔakuBΔhapX ( ΔhapX ) , SBUG844ΔakuBΔhapX/AnhapXC ( AnhapXC ) , SBUG844ΔakuBΔsreA ( ΔsreA ) , and SBUG844ΔakuBΔsreA/sreAC ( sreAC ) grown for 72 hr in AMM_G100Gln50 with ( +Fe ) or without 40 µM FeCl3 ( −Fe ) . Coprogen was quantified from lyophilised culture supernatants and terrein from culture extracts . ( C ) qRT-PCR from strains and media described in ( A ) and ( B ) . RNA was isolated after 40 hr of cultivation . Transcript levels were normalised against enoA by fold expression = 2^ ( CTtarget − CTenoA ) . ( $ ) denotes the lack of hapX transcripts from the complemented ΔhapX strain , since the Aspergillus nidulans hapX was used for complementation . qRT-PCR on the complemented mutant with oligonucleotides specific for AnhapX are shown in Figure 6—figure supplement 1 . All analyses were performed from biological triplicates and technical duplicates . Statistical significances were calculated in comparison to the parental ΔakuB strain by the Student's paired t-test with a two-tailed distribution: *p<0 . 05; **p<0 . 01; ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 02610 . 7554/eLife . 07861 . 027Figure 6—figure supplement 1 . qRT-PCR analysis of Aspergillus nidulans hapX expression in the Aspergillus terreus wild-type SBUG844ΔakuB , the hapX mutant SBUG844ΔakuBΔhapX and its complemented strain SBUG844ΔakuBΔhapX/AnhapXC . All analyses were performed from biological triplicates and in technical duplicates . All strains were grown for 40 hr in the presence ( +Fe ) or absence ( −Fe ) of 40 µM FeCl3 . Transcript levels of AnhapX were normalised against enoA gene from A . terreus by fold expression = 2^ ( CTtarget − CTeno ) . The heterologously expressed AnhapX gene behaves in a similar way to the native hapX gene from A . terreus ( compare with Figure 6 ) . Statistical significances were calculated by the Student's paired t-test with a two-tailed distribution: Asterisks indicate p values vs 0 µM FeCl3: ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 027 On the other hand , partial hapX deletion reduced coprogen production under iron limitation by approximately 50% and terrein concentrations by about 90% ( Figure 6A , B ) . This indicates that HapX directly activates both siderophore and terrein biosynthesis in A . terreus . In contrast , the ΔsreA mutant produced significantly higher amounts of coprogen in the presence of iron whereas terrein production did not significantly increase ( Figure 6A , B ) . This confirms SreA as a negative feedback regulator in siderophore biosynthesis , whereas it does not control terrein production . In contrast , HapX positively controls both pathways . These results were confirmed on the transcriptional level by qRT-PCR ( Figure 6C and Figure 6—figure supplement 1 ) . The inactivation of hapX prohibited terR transcription and resulted in an inability to induce terrein cluster genes terA , terB , and terC . In contrast , deletion of sreA influenced neither terR activation nor expression of other terrein cluster genes in the presence of iron . We therefore conclude that hapX is the major regulator for terrein cluster induction under iron starvation . To elucidate the complete sequence of the hapX gene we subsequently used degenerate primers to amplify the main proportion of hapX from cDNA . The complete sequence of the A . terreus hapX gene is found under accession number KP233834 ( Gressler et al . , 2015b ) . The sequence of the full-length HapX protein matches with that of HapX proteins from other Aspergillus species such as A . nidulans ( 73% identity ) , A . fumigatus ( 77% ) , Aspergillus niger ( 81% ) , and A . oryzae ( 81% ) . In general , siderophore-based iron acquisition is highly efficient and assumed to be more important than the reductive iron assimilation pathway . In A . fumigatus , growth and virulence defects caused by the interruption of the reductive iron assimilation pathway are only observed when the siderophore-based system is also inactivated ( Blatzer et al . , 2011 ) . Therefore , to elucidate a positive effect of terrein on iron acquisition , we deleted the sidA gene in A . terreus that encodes the L-ornithine-N5-monooxygenase , a key enzyme in hydroxamate siderophore biosynthesis . Coprogen production was confirmed in the wild-type and a complemented mutant , but was completely lacking from the ΔsidA mutant ( Figure 7—figure supplement 1 ) . When analysed for growth phenotypes on solid media , all complemented mutants and the ΔterA strain behaved like the wild-type ( Figure 7A and Figure 7—figure supplement 1A ) . The hapX mutant showed a reduced growth rate without iron supplementation ( Figure 7A ) , which is in agreement with reduced coprogen production as shown above ( Figure 6A ) . However , severe iron limitation from the addition of the iron chelator bathophenanthroline sulfonate ( BPS ) completely repressed growth of the hapX mutant ( Figure 7A ) . The ΔsidA mutant displayed the most severe phenotype: while ΔsidA showed normal growth with slightly reduced conidiation in the presence of high to moderate iron concentrations ( 100 or 20 µM FeCl3 ) ( Figure 7A ) , growth was strongly retarded when the iron concentration was reduced to 2 µM FeCl3 . No growth was observed even after prolonged incubation when iron was omitted ( Figure 7A ) . These phenotypes were cured when purified coprogen was externally added ( Figure 7—figure supplement 1B ) . Interestingly , growth of the ΔsidA strain was also partially restored in the presence of ascorbic acid and , although to a lesser extent , by the addition of terrein ( Figure 7A ) . This result is supported by previous studies on an A . nidulans ΔsidA mutant which was able to grow under iron limitation in the presence of ascorbic acid ( Eisendle et al . , 2003 ) . Unexpectedly , terrein supplementation inhibited growth of an A . fumigatus sidA mutant , which was also true for an A . fumigatus wild-type strain ( Figure 7—figure supplement 2 ) . Subsequent analyses showed that growth of the phytopathogen Fusarium graminearum was also inhibited by terrein-containing culture extracts from A . terreus ( Figure 7—figure supplement 2 ) , indicating some antifungal properties of terrein against environmental competitors . 10 . 7554/eLife . 07861 . 028Figure 7 . Siderophore production in sidA mutants and growth-supporting effect of terrein under iron limitation . ( A ) Growth of wild-type ΔakuB , ΔterA , ΔhapX , and ΔsidA on AMM ( −N ) G100Gln10 plates containing different iron concentrations . Photographs were taken after 4 d and , as indicated , after 17 d of incubation at 37°C . Iron-free medium was supplemented with 100 , 20 , 2 , or 0 µM FeCl3 . Additionally , plates without iron were supplemented with bathophenanthroline disulfonate ( BPS; 100 µM ) to simulate harsh starvation conditions or with either ascorbic acid ( 1 mM ) or terrein ( 5 or 10 mM ) . ( B , C ) Impact of terrein production on growth of ΔsidA mutants . AMM-G100Gln50 plates were supplemented with 0 , 2 , 20 , or 100 µM FeCl3 and either inoculated with ( B ) conidia or ( C ) mycelial pellets from cultures pre-grown for 40 hr in AMM-G100Gln50 with 200 µM FeCl3 . The parental Aspergillus terreus wild-type ΔakuB , the ΔakuBΔsidA strain , and the ΔakuBΔterAΔsidA are shown . Mycelia from a 40 hr culture in AMM-G100Gln50 with 200 µM FeCl3 was washed with iron-free medium and three pellets were applied to the plates . Plates were incubated at 37°C for 5 d . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 02810 . 7554/eLife . 07861 . 029Figure 7—figure supplement 1 . Dependence of colony formation of Aspergillus terreus mutants on iron availabilty , ascorbic acid , terrein , and coprogen . ( A ) As basic media AMM ( -N ) G100Gln10 ( -Fe ) agar plates were used that were supplemented either with different iron concentrations ( 100 , 20 , 2 , or 0 µM FeCl3 ) , the iron chelator bathophenanthroline disulfonate ( BPS; 100 µM ) , ascorbic acid ( Asc , 1 mM ) , or terrein ( 10 mM ) . Plates were incubated for 4 d at 37°C , except for the ΔterAΔsidA mutant that was further incubated for up to 17 d ( B ) Effect of purified coprogen on growth of A . terreus mutants and complemented strains under iron starvation . Conidia ( 1 × 105 ) of A . terreus SBUG844ΔakuB , SBUG844ΔakuBΔsidA , SBUG844ΔakuBΔsidA/sidAC , SBUG844ΔakuBΔterA , or SBUG844ΔakuBΔterAΔsidA were plated on AMM-G100Gln10 plates supplemented with ( +Fe ) or without 20 µM FeCl3 ( −Fe ) . To ( −Fe ) media filter disks soaked with either 20 µl methanol ( MeOH , negative control ) or 20 µl of 2 . 5 mg/ml coprogen ( in methanol ) were applied . Plates were incubated at 37°C for 72 hr until photographed . ( C ) Coprogen quantification from ΔakuB wild-type , ΔsidA mutant and complemented mutant sidAC . Strains were cultivated for 40 hr in AMM ( -N ) G100Gln50 supplemented with 200 µM FeCl3 . The mycelium was washed and transferred to iron-free AMM ( −N ) G100Gln50 medium . Coprogen was quantified after 30 hr of incubation . Analyses were performed from biological triplicates and technical triplicates . Statistical significance was calculated in comparison to the parental ΔakuB strain by the Student's paired t-test with a two-tailed distribution: *p<0 . 05; ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 02910 . 7554/eLife . 07861 . 030Figure 7—figure supplement 2 . Antifungal activity of terrein and potato dextrose broth ( PDB ) medium from Aspergillus terreus wild-type cultivations . ( A ) AMM-G100Gln10 containing no iron addition ( −Fe ) or 2 mM FeCl3 were supplemented with 0 , 1 , 10 , or 20 mM terrein and inoculated with conidia of Aspergillus fumigatus wild-type ATCC46645 . Plates were incubated for 84 hr at 37°C until photographed . Terrein inhibits growth of A . fumigatus independently of the available iron concentration . ( B ) Analysis of the effect of A . terreus inoculated PDB culture broth on growth of A . terreus , A . fumigatus , and Fusarium graminearum . The basal medium for A . terreus and A . fumigatus was AMM-G50 medium with nitrate as the nitrogen source , whereas the medium for F . graminearum was additionally supplemented with 0 . 2% potato broth ( AMM-G50 + 0 . 2%PB ) . All plates were supplemented either with 150 µl of PDB ethyl acetate extract from a mock-inoculated culture ( PDB extract ) or inoculated with A . terreus SBUG844 and cultivated for 4 d at 30°C ( PDB metabolite extract ) . Photographs were taken after 4 d of incubation at 30°C . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 03010 . 7554/eLife . 07861 . 031Figure 7—figure supplement 3 . Terrein determination in siderophor deletion mutants . High performance liquid chromatography profiles of agar plugs were recorded from plates shown in panel Figure 7C . Wild-type ( A ) and ΔsidA mutant ( B ) produced terrein ( 1 ) under iron-limited conditions , whereas no terrein was formed by the ΔterAΔsidA mutant ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 031 Due to the beneficial effect of terrein on the A . terreus ΔsidA mutant , we investigated the impact of intrinsic terrein production on growth under iron limitation . For this , we deleted the sidA gene in the ΔterA background and compared growth of wild-type , ΔsidA , and ΔterAΔsidA under iron-supplemented ( 20 and 200 µM FeCl3 ) and iron-limited conditions ( 0 and 2 µM FeCl3 ) ( Figure 7B ) . Both mutants ( ΔsidA and ΔterAΔsidA ) were unable to grow on media containing 2 µM or less FeCl3 . However , since terrein is only produced by vegetative mycelium , strains were pre-grown in the presence of 200 µM FeCl3 , washed , and transferred to the plates with different iron contents ( Figure 7C ) . While growth of mycelium from the ΔakuB strain looked identical to that from conidia , the ΔsidA mutant started to form small colonies even in the absence of iron , whereas the ΔterAΔsidA mutant showed some weak growth only at 2 µM FeCl3 but not in the absence of external iron supplementation . Terrein production was analysed from extracted agar plugs , which confirmed that the wild-type and ΔsidA mutant produced substantial amounts of terrein under iron-limited conditions whereas terrein production was fully abrogated in the ΔterAΔsidA mutant ( Figure 7—figure supplement 3 ) . These results indicate that terrein production supports iron acquisition , although the siderophore system is the dominant iron acquisition system . A . terreus is known as a human pathogen that causes severe invasive bronchopulmonary and disseminated aspergillosis ( Lass-Flörl et al . , 2005; Slesiona et al . , 2012a , 2012b ) . In addition , A . terreus has been described as a causative agent of foliar blight of potato leaves ( Louis et al . , 2013 , 2014a , 2014b ) . Besides its pathogenic potential , A . terreus can inhibit proliferation of other plant pathogens such as Fusarium udum ( Upadhyay and Rai , 1987 ) and acts as a mycoparasite on sclerotia of the plant pathogenic fungus Sclerotinia sclerotiorum ( Melo et al . , 2006 ) . SMs are assumed to support competition in the plant and soil environment . The natural biological activities of terrein include the inhibition of plant seed germination ( Kamata et al . , 1983 ) , the induction of fruit surface lesions ( Zaehle et al . , 2014 ) , and the newly discovered reduction of ferric to ferrous iron accompanied by some antifungal activity . These activities are all appropriate to supporting the competitiveness of A . terreus in the environment . However , the production of terrein at the correct timing requires coordinated sensing and transduction of environmental signals . The terrein biosynthesis gene cluster is directly controlled by the Zn2Cys6 transcription factor TerR ( Gressler et al . , 2015a ) . Therefore , terrein production requires transcriptional activation of terR . While sugars are indispensable for high terrein production rates ( Zaehle et al . , 2014 ) , three major signals were identified that resulted in terR transcription: methionine-dependent induction , nitrogen limitation , and iron starvation . These signals resemble the plant and rhizosphere environment . Plant roots are leaky for carbon compounds that derive from C-fixation during photosynthesis . Up to one third of plant-assimilated carbon may end up in the rhizosphere , and especially in the extramatrical mycelium of ectomycorrhiza ( Churchland and Grayston , 2014 ) . Here , at least a fraction of carbohydrates is subsequently exuded by ectomycorrhiza , making it available to other soil microorganisms ( Sun et al . , 1999 ) . Mycorrhizas mobilize nitrogen to feed their symbiotic plant partner , resulting in nitrogen limitation for surrounding microorganisms . These conditions stimulate terrein production in A . terreus , which is typically isolated from the rhizosphere ( Gao et al . , 2013; Rajalakshmi and Mahesh , 2014 ) and might affect existing plant–microbe interactions by its phytotoxic and antifungal activity ( Figure 8 ) . 10 . 7554/eLife . 07861 . 032Figure 8 . Scheme of the regulation of terrein biosynthesis gene cluster expression during interactions in the rhizosphere . Plants secrete methionine ( +Met ) with root exudates into the soil . Additionally , competing microorganisms reduce the available pool of nitrogen sources ( −N ) and iron ( −Fe ) . Signals from nitrogen limitation and methionine are sensed via AreA and AtfA , whereas iron limitation is sensed via HapX . All three transcription factors activate the promoter of the terrein biosynthesis gene cluster-specific transcription factor . TerR leads to transcription of the structural genes required for terrein biosynthesis and terrein is produced from acetyl- and malonyl-CoA units . The two-cluster specific major facilitator superfamily ( MFS ) transporters export terrein into the rhizosphere . Here terrein can counteract iron limitation by its ferric iron reducing activity , supports degradation of organic matter by its phytotoxic activities , and reduces growth of competitors by its antifungal activity . DOI: http://dx . doi . org/10 . 7554/eLife . 07861 . 032 Another inducing signal derives from methionine , which may be of particular importance when A . terreus acts as a potato plant pathogen ( Louis et al . , 2013 ) . Notably , PDB induces high terrein production rates ( Zaehle et al . , 2014 ) , which we assume to be methionine-dependent . In plants , methionine is required for ethylene production in environmental stress response ( Wang et al . , 2002 ) , fruit ripening ( Yang and Hoffman , 1984 ) , and polyamine production in plant defence reactions . Furthermore , methionine is a precursor for iron chelating siderophores such as nicotianamine and mugineic acid ( Roje , 2006 ) and has been identified in plant root exudates ( Dakora and Phillips , 2002 ) . Therefore , elevated levels of methionine depict a signal that indicates the presence of a plant environment and stimulates terrein production . However , although further studies are required to elucidate the specific contribution of terrein to plant infection , its ability to induce lesion formation on several fruit surfaces indicates that terrein interacts with plant defence mechanisms ( Zaehle et al . , 2014 ) . The third—at first sight unrelated—inducing factor is iron starvation . Although terrein can reduce ferric to ferrous iron , a direct positive growth-promoting effect of terrein during iron limitation in A . terreus monocultures was only observed in the absence of a functional siderophore system . However , extracellular siderophores show a high affinity for ferric iron but not for ferrous iron ( Hider and Kong , 2010 ) . Thus , the reduction of ferric iron could reduce the efficiency of iron chelation by xenosiderophores secreted from competing microorganisms . A similar strategy for iron acquisition has previously been described for the human pathogenic fungus Histoplasma capsulatum ( Timmerman and Woods , 1999 ) . H . capsulatum secretes glutathione accompanied by the γ-glutamyltransferase Ggt1 . This enzyme releases the dipeptide cysteinylglycine from gluthathione , which in turn reduces extracellular ferric iron ( Zarnowski et al . , 2008 ) . While H . capsulatum also secretes hydroxamate siderophores ( Howard et al . , 2000 ) , silencing of ggt1 mRNA reduced virulence in macrophages even in the presence of the siderophore system ( Zarnowski et al . , 2008 ) . Thus , terrein may interfere with competing siderophore systems to increase the competitiveness of A . terreus in the environment . In this study we showed that transduction of environmental signals to the promoter of the transcriptional activator TerR requires activation by the global transcription factors AreA , AtfA , and HapX . While AreA has previously been described to regulate several SM gene clusters in fungi ( Tudzynski et al . , 1999 ) , HapX activates iron acquisition systems ( Schrettl et al . , 2010 ) and AtfA plays a specific role in osmotic and oxidative stress responses ( Lara-Rojas et al . , 2011 ) . In Aspergillus species and maybe all saprophytic ascomycetes , the GATA transcription factor AreA plays a substantial role in regulation of nitrogen . Under nitrogen starvation , AreA is transported into the nucleus ( Todd et al . , 2005 ) where it binds a 5′-HGATAR-3′ sequence in its target promoters ( Ravagnani et al . , 1997 ) , as shown for the genes encoding nitrate reductase ( NiaD ) and nitrite reductase ( NiiD ) ( Chang et al . , 1996 ) that are upregulated in the absence of preferred nitrogen sources ( Johnstone et al . , 1990; Punt et al . , 1995 ) . In Aspergillus parasiticus AreA also binds to the promoter regions of the transcriptional regulator genes aflR and aflJ from the aflatoxin biosynthesis cluster ( Chang et al . , 2000 ) . In contrast to nitrate utilisation , this binding impedes aflatoxin production probably due to inhibition of transcriptional activation ( Yu et al . , 2004 ) . We have shown that terrein biosynthesis in A . terreus during nitrogen starvation is AreA-dependent , and at least two areA binding sites in the terR promoter were recognised by the AreA DNA-binding domain . Therefore , AreA directly binds to the terR promoter and supports the expression of the terrein biosynthesis genes . Similarly , biosynthesis of various mycotoxins and gibberellins is activated by AreA in phytopathogenic Fusarium species ( Mihlan et al . , 2003; Kim and Woloshuk , 2008; Giese et al . , 2013 ) . This implies that AreA plays a general role in regulation of SM production of plant pathogens , and the plant-derived environment may be characterised as a habitat with a high C:N ratio . In several ascomycetes the nitrogen starvation response is also mediated by the basic leucine zipper transcription factor AtfA ( Hong et al . , 2013 ) . In A . nidulans , nitrogen starvation , oxidative or osmotic stresses induce phosphorylation of the MAP kinase SakA that is subsequently transported into the nucleus to activate transcription of target genes by interaction with the constitutively expressed transcription factor AtfA ( Lara-Rojas et al . , 2011 ) . Interestingly , in A . terreus the interaction between AtfA and SakA has not been solved in detail . While a sakA deletion does not affect terrein production levels under nitrogen limitation , it allows terrein production in the presence of some nitrogen sources , indicating that it may possess a repressor function under some conditions . In addition , a ΔsakA strain still produces pigmented conidia ( Figure 4—figure supplement 5 ) and the sakA mutant—but not the ΔatfA strain—showed some increased osmosensitivity ( Figure 4—figure supplement 5 ) , which agrees with recent data on ΔatfA and ΔsakA mutants of Penicillium marneffei ( Nimmanee et al . , 2014 ) . Therefore , it is conceivable that terrein biosynthesis gene cluster activation via AtfA does not require SakA . In A . terreus AtfA seems to act as a general inducer of SM production since its deletion resulted in non-pigmented conidia and in the loss of terrein production under nitrogen starvation or methionine supplementation . In contrast , deletion of the atf1 gene in Botrytis cinerea resulted in accumulation rather than depletion of the polyketide botcinin A and the sesquiterpene botrydial ( Temme et al . , 2012 ) . This points to AtfA as a transcriptional inhibitor in B . cinerea whereas AtfA acts as an inducer in A . terreus . Interestingly , AreA and AtfA sense the same environmental signals to induce the terrein biosynthesis gene cluster , although their global role in cellular physiology differs significantly . This adaptation of different transcription factors towards terrein production underlines a special importance of terrein for A . terreus during environmental competition . Independent from AreA and AtfA , activation of the terrein biosynthesis gene cluster was also triggered by the iron response regulator HapX under iron limitation . To overcome severe iron starvation , microorganisms frequently produce siderophores ( Pollack and Neilands , 1970; Eisendle et al . , 2003; Kreutzer and Nett , 2012 ) or utilise siderophores from competing species ( xenosiderophores ) ( Boukhalfa and Crumbliss , 2002 ) . In A . fumigatus and A . nidulans siderophore biosynthesis and uptake is regulated by the opposing transcription factors HapX and SreA depending on the presence or absence of iron ( Schrettl et al . , 2007 , 2008 , 2010 ) . Similarly , A . terreus produces the HapX and SreA regulated siderophores ferrichrysin and coprogen that allow efficient acquisition of iron . However , in competition with other micoorganisms , terrein-mediated iron reduction could reduce binding of iron to high-affinity siderophores from competitors that cannot be used by A . terreus , as described for the cysteinylglycine iron reduction system from H . capsulatum . The control of terrein production by HapX depicts an unprecedented example of the activation of a SM gene cluster outside the siderophore biosynthesis system . Interestingly , terrein production is only under positive HapX control , but not under negative SreA feedback control , which is in agreement with a lack of SreA binding sites in the terR promoter ( Schrettl et al . , 2008 ) , whereas a CCAAT box required for binding of the CCAAT-binding complex ( CBC ) and its subsequent CBC interaction with HapX ( Hortschansky et al . , 2007; Gsaller et al . , 2014 ) is present . In the HapX target promoter regions of cycA , sreA , acoA , and lysF from Aspergillus species , a bipartite CBC–HapX-DNA binding motif with consensus 5′-C ( C/A ) AATCAN11-12GAT-3′ is present , of which CCAAT is recognised by the CBC complex while the GAT sequence is bound by HapX ( Hortschansky et al . , 2015 ) . Interestingly , these are all types of promoters that are repressed under iron limitation . In contrast , the cccA promoter , which encodes for a vacuolar iron transporter , is the only HapX-induced promoter known so far with a confirmed in vitro HapX binding site ( Gsaller et al . , 2014 ) . HapX binds to the GAT motif of a 5′-CCAATN16GATC-3′ sequence , which is also present in the cccA promoter of A . terreus and—with one mismatch—in the terR promoter ( position −215 ) . However , further studies will be required to confirm this motif as a HapX binding site . In conclusion , we have shown that the terrein biosynthesis gene cluster is activated by at least three independent environmental signals: methionine , nitrogen limitation , and iron starvation . Additionally , the three global transcription factors AreA , AtfA , and HapX are essential to transfer these signals to the terrein biosynthesis gene cluster . Although one environmental signal is sufficient to induce the gene cluster , a combination of different signals provides an environmental image of higher resolution and allows an adequate adaptation to the ecological niche . The phytotoxic and antifungal potential of terrein combined with its iron reductive properties enables A . terreus to acquire nutrients that might otherwise be consumed by competitors . Notably , similar to terrein , the mycotoxin production in Fusarium species is generally associated with sensing of nitrogen availability . Although AreA plays an important regulatory function , involvement of AtfA as an additional regulator and methionine as a second inducer in these species needs to be tested . However , from our results on the terrein biosynthesis gene cluster , we propose that manifold regulation of SM gene clusters and the interplay of different global transcriptional activators might depict a general mechanism to regulate the flexibility of gene cluster activation in fungi . All genotypes of strains used in this study are summarised in ( Figure 4—source data 1 ) . A . terreus strain SBUG844 and its derivative SBUG844ΔakuB that show increased rates of homologous integration served as parental strains ( Gressler et al . , 2011 ) . If not indicated otherwise , all strains were cultivated in Aspergillus minimal medium ( AMM; http://www . fgsc . net/methods/anidmed . html ) supplemented with different carbon and nitrogen sources: AMM with 100 mM glucose ( AMM-G100 ) , with 100 mM glucose and 50 mM glutamine ( AMM-G100Gln50 ) , with 100 mM glucose , 50 mM glutamine and 10 mM methionine ( AMM-G100Gln50Met10 ) , with 1% casamino acids ( AMM-CA1% ) . Additionally , the following complex media were used: Sabouraud ( Sab , Sigma Aldrich , Taufkirchen , Germany ) , YPD ( 10 g/l yeast extract , 20 g/l peptone , 20 g/l dextrose; PDB , Sigma Aldrich ) , potato broth ( 2% potato extract , Sigma Aldrich ) , apple juice ( pH adjusted to 6 . 0 with NaOH ) , carrot juice with honey ( both Wiesgart , ALDI Nord , Jena , Germany ) , banana juice or peach juice ( both FruchtOase , Kiberg , Germany ) . When required , 140 µg/ml hygromycine B ( Carl Roth GmbH , Karlsruhe , Germany ) , 0 . 1 µg/ml pyrithiamine or 80 µg/ml phleomycin ( both Sigma Aldrich ) were added . For preparation of conidia suspensions , all strains were cultivated for 4 d at 37°C on solid 2% AMM-G50Gln10 agar plates . Conidia were harvested by overlaying colonies with water and filtering off the suspension over 40 μm cell strainers ( VWR , Darmstadt , Germany ) . Liquid media were generally inoculated at a final concentration of 1 × 106 conidia per ml . For the ΔareA strains , inoculation densities were generally doubled to compensate for reduced growth rates . For metabolite extraction and quantification liquid cultures in 50 ml scale were used and incubated at 30°C for 48 hr or 72 hr depending on the specific experiment . For nitrogen or iron shift experiments , strains were pre-cultivated for 40 hr in 150 ml AMM-G100Gln50 or AMM-G100Gln50 with 200 µM FeCl3 . Mycelia were harvested over sterile miracloth ( Merck , Darmstadt , Germany ) washed twice with AMM lacking a nitrogen ( AMM-N ) or iron source ( AMM-Fe ) and aliquots were transferred to fresh 50 ml media with or without 50 mM Gln or with or without 200 µM FeCl3 . Depending on the specific experiments , samples were analysed after 12 , 24 , or 48 hr of cultivation . Metabolites were extracted from culture broth as described previously ( Gressler et al . , 2011 ) . In brief , an equal volume of ethyl acetate was added and collected after defined shaking of the mixture . The procedure was repeated once . After evaporation of the solvent , residues were taken up in 1 ml methanol each and filtered . Standard extract analyses were performed on an Agilent 1100 series HPLC-DAD system coupled with a MSD trap ( Agilent Technologies , Waldbronn , Germany ) operating in alternating ionisation mode . Terrein quantification was carried out from 50 ml cultures as described elsewhere ( Zaehle et al . , 2014 ) . For quantification of the siderophore coprogen , the complete 50 ml culture supernatants were filtered and lyophilised to dryness . The remaining solids were extracted three times with 10 ml MeOH . The solvent from the combined organic extracts was removed under reduced pressure and residues were re-dissolved in 2 ml MeOH . The resulting slurries were filtered and the filtrates analysed by HPLC measurements . HPLC analyses were carried out on an Agilent 1260 device equipped with a quaternary pump and a UV/Vis detector ( Agilent Technologies; Column: Zorbax Eclipse XDB-C8 , 5 µm , 150 × 4 . 6 mm; flow rate 1 ml/min; eluent A: H2O/0 . 1% HCOOH , eluent B: MeOH ) . The gradient started with 10% B and reached 30% B after 4 min , increased to 55% B within 10 min and reached 100% B after 2 min , where it was retained for an additional 4 min . Quantification of coprogen was performed from a calibration curve of known coprogen concentrations . For correlation of coprogen to the fungal biomass , mycelia from the cultures were dried for 48 hr at 37°C and balanced and coprogen concentrations per gram dried mycelium were calculated . All quantifications were carried out in biological triplicates and technical duplicates . Isolation of coprogen for generation of the calibration curve was performed by semi-preparative HPLC from culture supernatants of the ΔakuB and ΔakuBΔterA strains and fractions were collected by automatic fraction collection . Separation was carried out on a Zorbax Eclipse XDB-C8 , 5 µm , 250 × 4 . 6 mm with a flow rate of 4 . 0 ml/min using H2O as eluent A and MeOH as eluent B . The gradient started with 10% B , reached 30% B after 6 . 5 min , increased to 55% B within 16 . 5 min , reached 100% B after 2 min , and was retained at 100% B for an additional 6 min . For isolation of 2- ( ( E ) -prop-1-en-1-yl ) maleic acid , the crude product from upscaled terrein reduction assays ( see below ) was subjected to semi-preparative HPLC using a Zorbax Eclipse XDB-C8 , 5 µm , 250 × 4 . 6 mm with a flow rate of 4 . 0 ml/min , eluent A: H2O/0 . 1% HCOOH , eluent B: acetonitrile . The gradient started with 5% B and was held for 14 min , increased to 10% B within 9 min , increased to 100% B within 2 min where it was retained for an additional 7 min . Fractions from the new metabolite formed from ferric iron reduction were collected and evaporated resulting in a white solid which revealed a m/z value of 150 . 0495 [M + H+] by HRESI-MS that perfectly matched a calculated molecular formula of C7H8O4 containing four double-bond equivalents . 13C-NMR measurements ( Figure 5—figure supplement 4 ) revealed the presence of two carbonyl groups , one terminal methyl group , and four carbons being part of a conjugated system . Two-dimensional NMR data ( Figure 5—figure supplement 5 ) and analysis of all proton coupling constants from the 1H-NMR spectrum ( Figure 5—figure supplement 4 ) finally confirmed the structure of 2- ( ( E ) -prop-1-en-1-yl ) maleic acid . NMR spectra were recorded on a Bruker Avance III 500 and a Bruker Avance III 600 spectrometer ( Bruker BioSpin GmbH , Rheinstetten , Germany ) equipped with a cryoprobe head using DMSO-d6 and methanol-d4 as solvents and internal standards . Fruit surfaces were wiped with a soft tissue saturated with 70% ethanol . Apples ( ALDI , type: Tenroy Royal/Gala; Germany ) , bananas ( type Bio , Fairverbindet; tegut , Jena , Germany ) , and nectarines ( tegut , type: Sweet Lady , class 1 , size A; Italy ) were cut lengthwise using a sterile scalpel . The resulting groove was infected with 200 µl of a conidia suspension containing 4 × 107 conidia and fruits were incubated for 7–10 d at room temperature in a humid chamber . Cut but uninfected fruits served as controls . For cultivation in the presence of high levels of nitrogen , 200 µl of a 3 . 5 M NH4Cl solution were applied prior to infection . For terrein quantification , fruits were homogenised and extracted twice with 100 ml ethyl acetate and the solid residues were collected and evaporated for dry weight determination . Terrein was quantified from extracts as described above and terrein production rates were calculated as mg terrein/g fruit dry weight . A detailed description on fruit infection , sample preparation and terrein quantification is described in Bio-protocols by Gressler and Brock ( 2016 ) . To determine the induction of lesion formation on banana fruit surfaces , organic bananas ( type Bio , Fairverbindet; tegut ) were softly cleaned with water and air dried . 5 µl of various sequential dilutions ( 1:2 to 1:512 ) of metabolite crude extracts ( dissolved in PBS ) were added as a single drop to the surface . Fruits were incubated in the dark at room temperature and photographs were taken after 40 hr and 60 hr . The determination of ammonia was performed as described by Weichselbaum et al . ( 1969 ) . From each culture 1 ml aliquots of broth were removed and centrifuged for 5 min at 16 , 000×g to remove residual mycelium . The supernatant was collected and adjusted to pH 7 . 0 by addition of NaOH . A 200 µl aliquot was sequentially diluted in a nitrogen-free medium and transferred to a transparent flat-bottom 96-well plate . After addition of 20 µl SC solution ( 6 . 5 g sodium salicylate ( C7H5O3Na ) , 6 . 5 g trisodium citrate ( C6H5O7Na3 × 2 H2O ) , 48 . 5 mg disodium pentacyanonitrosylferrate ( Na2Fe ( CN ) 5NO × 2 H2O ) in 50 ml water ) , the reaction was started with 20 µl DCIC solution ( 1 . 6 g NaOH , 100 mg sodium dichloroisocyanurate ( C3N3Cl2O3Na ) in 50 ml water ) . Readings were done after 4 . 5 hr of incubation at room temperature in a FLOUstar Omega microplate reader ( BMG Labtech , Ortenberg , Germany ) . Plates were shaken for 1 min in a double orbital direction and absorbance at 655 nm was measured ( 50 flashes/well , gain 2000 ) . Fresh growth medium served as positive and water as negative controls . A calibration curve was recorded for calculation of ammonia levels from culture broth . The ferric iron-reducing assay was performed virtually as described by Benzie and Strain ( 1996 ) . In brief , three different working solutions were prepared by mixing 20 parts of reagent A ( 0 . 4% sodium acetate , 1 . 6% acetic acid; pH 4 ) with 1 part of reagent B ( 0 . 4% 2 , 4 , 6-tripyridyl-S-triazine ( TPTZ; Sigma ) in 0 . 16% HCl and either 1 part of reagent C1 ( 88 mM–5 . 5 mM sequential dilutions of terrein ) , reagent C2 ( 2% ascorbic acid , serving as positive control ) or reagent C3 ( water , serving as negative control ) . From each working solution , 200 µl were transferred into wells of a 96-well plate . An iron stock solution ( 10 mg/ml FeCl3 in 0 . 2 M H2SO4 ) was diluted to 0 . 05 mg/ml and 100 µl were added to the working solution , which started the reaction . Reduction of ferric to ferrous iron was followed by determination of the change in absorbance at 590 nm for up to 10 hr using a FLOUstar Omega microplate reader ( BMG Labtech ) . For determination of the pH optimum of the terrein mediated reduction , reagent A was adjusted to pH values of 3 . 5–6 . 0 . Reductive activity was normalised against the activity determined from ascorbic acid , which was set as 100% . All experiments were carried out in triplicate . Upscaling of the iron reduction assays for metabolite extraction was performed by incubating 20 mg ( 125 µmol ) of terrein in a 1:2 molar ratio with either FeSO4 , FeCl3 , CH3COOCu , or ( CH3COO ) 2Cu in 10 ml of reagent A . Samples were incubated for 48 hr at room temperature under continuous vertical rotation ( 20 rpm ) . Metabolites were extracted three times with ethyl acetate ( 10 ml each ) . The solvent was evaporated and the residues dissolved in MeOH and analysed as described above . Chrome azol S agar plate assays were prepared as described by Milagres et al . ( 1999 ) . Holes of 8 mm diameter were punched into plates and inoculated with 100 µl methanolic solutions of coprogen , terrein , or propylene maleic acid in concentrations of 2 . 50 , 1 . 25 , 0 . 625 , and 0 . 3125 mg/ml . Plates were incubated for 48 hr at 30°C prior to photography . Mycelia from specified cultivations were briefly washed with water and ground under liquid nitrogen . RNA was isolated using the RiboPure RNA Purification Kit ( Ambion , Life Technologies , Darmstadt , Germany ) . Residual genomic DNA ( gDNA ) was removed by the DNA-free kit ( Ambion ) . cDNA was synthesised by Revert Aid Reverse transcriptase ( Thermo Scientific , Schwerte , Germany ) using anchored oligodT primers . qRT-PCR was carried out on a CFX384 Touch Real-Time PCR Detection System ( BioRad , Munich , Germany ) using the EvaGreen 5 × QPCR ( ROX ) Mix ( Bio & Sell , Feucht , Germany ) following the manufacturer's protocol and using 1:5 and 1:10 dilutions of cDNA samples serving as templates . The actin gene ( actA , ATEG_06973 ) and the enolase gene ( enoA , ATEG_02902 ) were used for normalisation of transcript levels yielding similar results . Normalised transcript levels were calculated as fold expression = 2Δ ( reference − target ) . Primers used for qRT-PCRs showed a primer efficiency of 1 . 89–2 . 0 and are listed in Figure 4—source data 2 ( Figure 4—source data 2 ) . In general , all PCR amplifications were performed using the Phusion DNA Polymerase ( Thermo Scientific ) . Plasmids were amplified in E . coli DH5α . A detailed description of all cloning procedures is given below . Oligonucleotides used for plasmid construction are listed in the Figure 4—source data 2 ( Figure 4—source data 2 ) . Fungal transformation was performed as described previously ( Gressler et al . , 2011 ) and transformants were checked by diagnostic PCR and Southern blot analyses ( see Figure 4—figure supplement 4 ) . The promoter of the terA gene ( PterA; 1220 bp ) was amplified with oligonucleotides P47/48 from gDNA of A . terreus SBUG844 and ligated into the NotI/BamHI digested plasmid lacZ:trpCT-pJET1 . 2 ( Gressler et al . , 2011 ) containing the E . coli lacZ gene and the trpC terminator sequence . The plasmid was linearised by NotI digestion and the NotI-excised ptrA cassette from plasmid ptrA-pJET ( Fleck and Brock , 2010 ) was inserted . The resulting plasmid was used for transformation of A . terreus SBUG844 wild-type resulting in SBUG844_PterA:lacZ . Constitutive strong expression of the atfA gene was performed by using the strong constitutive gpdA promoter from A . nidulans FGSC A4 ( AnPgpdA , 1387 bp; oligonucleotides P49/50 ) to control expression of A . terreus atfA . The atfA ORF together with its natural terminator was amplified from gDNA of A . terreus SBUG844 ( atfA + atfAT , 2466 bp ) using oligonucleotides P51/52 . Both fragments were fused by in vitro recombination with the SpeI- restricted hph-pCRIV vector using the InFusion HD cloning kit ( Clonetech Laboratories , Saint-Germain-en-Laye , France ) . The resulting plasmid AnPgpdA:atfA:atfAT_hph-pCRIV was used for transformation of A . terreus SBUG844 wild-type resulting in strain SBUG844_AnPgpdA:atfA . The upstream and downstream fragments of cpcA ( ATEG_03131 ) were amplified from gDNA of A . terreus SBUG844 using oligonucleotides P53/54 ( 943 bp ) and P55/56 ( 853 bp ) . Similarly , the upstream and downstream flanks of rhbA ( ATEG_09480 ) were amplified with oligonucleotides P57/58 ( 473 bp ) and P59/60 ( 418 bp ) . The respective fragments were fused by in vitro recombination with the ptrA resistance cassette ( 1950 bp ) from ptrA-pJET1 ( Fleck and Brock , 2010 ) into the KpnI-restricted pUC19 vector using the InFusion HD cloning kit ( Clonetech Laboratories ) , resulting in cpcAup-ptrA-cpcAdn-pUC19 , rhbAup-ptrA-rhbAdn-pUC19 , and dnmTup-ptrA-dnmTdn-pUC19 . The deletion cassettes were excised by KpnI restriction and used for transformation of SBUG844ΔakuB . The upstream and downstream flanks of areA ( ATEG_07264 ) were amplified from gDNA of A . terreus SBUG844 with oligonucleotides P61/62 ( 820 bp ) and P63/64 ( 731 bp ) . Fragments were fused by fusion PCR and ligated into the pJET1 . 2 cloning vector ( Thermo Scientific ) . The fragment was excised with SmaI and subcloned into pUC19 . After restriction with NotI , the ptrA resistance cassette ( 1950 bp ) from ptrA-pJET1 ( Fleck and Brock , 2010 ) was inserted resulting in areAup-ptrA-areAdn-pUC19 . The deletion cassette was excised by SmaI and used for transformation of SBUG844ΔakuB . For complementation , a fragment spanning the entire areA ORF including its promoter and terminator sequence was amplified from gDNA of the A . terreus type strain FGSC A1156 ( NIH2624 ) using oligonucleotides P65/64 ( 4181 bp ) and ligated into pJET1 . 2 . gDNA from the type strain was used because its areA promoter sequence contained an additional PstI restriction site that allowed discrimination of complemented strains from the parental SBUG844ΔakuB strain . The complementation fragment was excised by XhoI restriction from plasmid A1156 ( areAup:areA:areAdn ) -pJET1 . 2 and directly used for the transformation of SBUG844ΔakuBΔareA . Due to the inability of the ΔareA mutant to utilise various nitrogen sources , no additional resistance marker was required and transformants were regenerated on media containing nitrate as the sole nitrogen source . Complemented strains were checked by Southern blot analysis and an additional PstI control digest of the PCR-amplified areA upstream flank . The upstream and downstream fragments of atfA ( ATEG_04664 ) were amplified from gDNA of A . terreus SBUG844 with oligonucleotides P66/67 ( 1102 bp ) and P68/69 ( 890 bp ) and fused by in vitro recombination with the NotI-excised ptrA resistance cassette ( 1950 bp ) from ptrA-pJET1 ( Fleck and Brock , 2010 ) into the KpnI-excised pUC19 vector using the InFusion HD cloning kit ( Clonetech Laboratories ) resulting in atfAup-ptrA-atfAdn-pUC19 . The deletion cassette was excised by KpnI and used for transformation of SBUG844ΔakuB . For deletion of atfA in the ΔareA background , the ptrA resistance cassette was replaced by the phleomycin resistance ( ble ) cassette ( NotI digest from ble-pJET1 . 2 ) . The deletion cassette was excised by KpnI restriction and used for transformation of SBUG844ΔakuBΔareA . For complementation of the ΔatfA phenotype , a fragment spanning the entire atfA gene including its promoter and terminator region were amplified with oligonucleotides P66/70 ( 3319 bp ) and an additional downstream fragment was amplified with P71/69 ( 859 bp ) . The fragments were fused by in vitro recombination with the NotI-excised ble resistance cassette ( 2073 bp ) from ble-pJET1 . 2 into the KpnI-restricted pUC19 vector using the InFusion HD cloning kit ( Clonetech Laboratories ) resulting in atfAup:atfA:atfAT_ble_atfAdn-pUC19 . The complementation cassette was excised by KpnI and used for transformation of SBUG844ΔakuBΔatfA . The atfA flanking regions were used to integrate the terR gene under control of the A . nidulans gpdA promoter in the atfA locus , which results in an atfA deletion . The upstream and downstream flanks of atfA were amplified from gDNA of A . terreus SBUG844 using oligonucleotides P72/73 ( 1125 bp ) and P74/75 ( 913 bp ) . Subsequently , a fragment already containing the fusion of AnPgpdA with the terR ORF including its native terminator ( 4405 bp ) was amplified with oligonucleotides P76/77 from plasmid hph_AnPgpdA:terR-pUC19 ( Gressler et al . , 2015a ) . The plasmid hph-pCRIV ( Fleck and Brock , 2010 ) was restricted with EcoRI to remove the hph resistance cassette and all three fragments were fused by in vitro recombination into the EcoRI site using the InFusion HD cloning kit ( Clonetech Laboratories ) . The resulting plasmid atfAup_AnPgpdA:terR:terRT_atfAdn_pCRIV was linearised with NotI and the ble resistance cassette from ble-pJET1 . 2 was inserted between the terR terminator and the atfA downstream region . The final plasmid was restricted with EcoRI , and the fragment atfAup_AnPgpdA:terR:terRT_ble_atfAdn was used for transformation of A . terreus SBUG844ΔakuBΔareA to replace the atfA ORF with the terR overexpression construct . Because the complete sequence information of the hapX locus was lacking at the beginning of this study , only a partial deletion of the A . terreus hapX gene was performed . Upstream and downstream flanks inside the hapX coding region ( ATEG_08014 ) were amplified with oligonucleotides P82/83 ( 387 bp ) and P84/85 ( 345 bp ) from genomic DNA of SBUG844 and fused by in vitro recombination with the ptrA resistance cassette ( 1950 bp ) from ptrA-pJET1 ( Fleck and Brock , 2010 ) into the KpnI-restricted pUC19 vector using the InFusion HD cloning kit ( Clonetech Laboratories ) resulting in hapXup-ptrA-hapXdn-pUC19 . The deletion construct was excised by KpnI restriction and used for transformation of SBUG844ΔakuB . For complementation of the partial hapX deletion by the A . nidulans hapX , the upstream and downstream flanks of ATEG_08014 used for generation of the partial deletion construct were amplified with oligonucleotides P82/86 ( 381 bp ) and P87/85 ( 341 bp ) and the complete hapX ORF ( AN8251 ) including its native promoter and terminator sequence was amplified from gDNA of A . nidulans FGSC A4 using oligonucleotides P88/89 ( 2648 bp ) . All fragments were fused by in vitro recombination with the ble resistance cassette ( 2073 bp ) from ble-pJET1 . 2 ( 2073 bp ) into the KpnI-restricted pUC19 vector using the InFusion HD cloning kit ( Clonetech Laboratories ) resulting in hapXup_PAnhapX:AnhapX:AnhapXT_ble_hapXdn-pUC19 . The complementation cassette was excised by KpnI and used for transformation of SBUG844ΔakuBΔhapX . To identify the complete coding region of the A . terreus hapX gene , long run PCRs from within the hapX gene and the last nucleotides of the known 5′- and 3′- borders were performed with oligonucleotides P110/111 ( 800 bp 5′ fragment ) and P112/113 ( 2500 bp 3′ fragment ) . Bands were excised from agarose gels and cloned into the pJET1 . 2 cloning vector . Fragments were sequenced from both strands using the primer walking method ( oligonucleotides P114/115 for both fragments and additionally P116 , P117 and P78 for the 3′ fragment ) . Finally , RNA was isolated from iron starvation conditions , transcribed into cDNA , and the hapX ORF was amplified with gene-specific oligonucleotides 113/79 and sequenced with oligonucleotides P114/115 . The complete hapX locus information was submitted to EMBL and can be found under accession number KP233834 ( Gressler et al . , 2015b ) . The upstream and downstream flanking regions of sreA ( ATEG_07714 ) were amplified from gDNA of A . terreus SBUG844 with oligonucleotides P90/91 ( 763 bp ) and P92/93 ( 756 bp ) and fused by in vitro recombination with the ptrA resistance cassette ( 1950 bp ) from ptrA-pJET1 ( Fleck and Brock , 2010 ) into the KpnI-restricted pUC19 vector using the InFusion HD cloning kit ( Clonetech Laboratories ) resulting in sreAup-ptrA-sreAdn-pUC19 . The deletion cassette was excised by KpnI and used for transformation of SBUG844ΔakuB . For complementation , a fragment spanning the sreA coding region including its natural promoter and terminator sequence was amplified with oligonucleotides P94/95 ( 3018 bp ) and additional downstream fragment with P96/97 ( 970 bp ) . The fragments were fused by in vitro recombination with the ble resistance cassette ( 2073 bp ) from ble-pJET1 . 2 into the KpnI-restricted pUC19 vector using the InFusion HD cloning kit ( Clonetech Laboratories ) resulting in sreAup:sreA:sreAT_ble_sreAdn-pUC19 . The complementation cassette was excised by KpnI restriction and used for transformation of SBUG844ΔakuBΔsreA . The upstream and downstream fragments of the sidA gene ( ATEG_06879 ) were amplified from gDNA of A . terreus SBUG844 using oligonucleotides P98/99 ( 764 bp ) and P100/101 ( 939 bp ) and fused by in vitro recombination with the ptrA resistance cassette ( 1950 bp ) from ptrA-pJET1 ( Fleck and Brock , 2010 ) into the KpnI-restricted pUC19 vector using the InFusion HD cloning kit ( Clonetech Laboratories ) resulting in sidAup-ptrA-sidAdn-pUC19 . For transformation of the terA deletion mutant , the ptrA cassette was excised by NotI restriction and replaced by the ble cassette . The deletion cassette was excised by KpnI restriction and used for transformation of SBUG844ΔakuB and SBUG844ΔakuBΔterA . For complementation of strain SBUG844ΔakuBΔsidA , the sidA ORF including its promoter and terminator sequence was amplified with oligonucleotides P102/103 ( 2393 bp ) and an additional downstream fragment was amplified with P104/105 ( 919 bp ) . The fragments were fused by in vitro recombination with the ble resistance cassette ( 2073 bp ) from ble-pJET1 . 2 into the HindIII-restricted pUC19 vector using the InFusion HD cloning kit ( Clonetech Laboratories ) resulting in sidAup:sidA:sidAT_ble_sidAdn-pUC19 . The complementation cassette was excised by HindIII restriction and used for transformation of SBUG844ΔakuBΔsidA . The coding sequence for the A . nidulans AreA DNA-binding domain ( amino acids 663–797 ) was amplified from plasmid pGEX4T1-AreA-ZnF ( Muro-Pastor et al . , 2004 ) as BamHI-HindIII fragment using oligonucleotides P80/81 ( Figure 4—source data 2 ) . The amplified fragment was cloned into a modified pET-43 . 1a vector allowing the addition of a N-terminal ( His ) 6-tag and removable by a tobacco etch virus ( TEV ) protease ( Novagen ) . ( His ) 6-AreA663-797 was produced by E . coli BL21 ( DE3 ) cells grown at 30°C in 1 l Overnight Express Instant TB Medium ( Novagen , Darmstadt , Germany ) in the presence of 1 mM Zn ( OAc ) 2 . Cells ( 15–20 g wet weight ) were collected by centrifugation , resuspended in 200 ml lysis buffer ( 50 mM NaH2PO4 , 300 mM NaCl , 10% vol/vol glycerol , 40 mM imidazole , 5 mM β-mercaptoethanol , 1 mM AEBSF , pH 8 . 0 ) and disrupted using an Emulsiflex C5 high pressure homogeniser ( Avestin , Mannheim , Germany ) . Cleared cellular extract was loaded to a 25 ml Ni Sepharose FF ( GE Heathcare , Freiburg , Germany ) column and ( His ) 6-AreA663-797 was eluted with 200 mM imidazole . After removal of the ( His ) 6 tag by adding 4 μg TEV protease per mg peptide and overnight incubation at room temperature , samples were adjusted to 150 mM NaCl and applied on a 40 ml CellufineSulfate ( Millipore , Darmstatdt , Germany ) column that was equilibrated with 50 mM NaH2PO4 , 150 mM NaCl , 10% vol/vol glycerol , 5 mM β-mercaptoethanol , 10 µM Zn ( OAc ) 2 , pH 7 . 5 , followed by elution of AreA663–797 with a gradient up to 2 M NaCl . Peak fractions were concentrated with an Amicon Ultra-15 10K centrifugal filter device and purified to homogeneity by size exclusion chromatography on a Superdex 75 prep grade column ( GE Healthcare ) in 20 mM HEPES , 300 mM NaCl , 5 mM β-mercaptoethanol , 10 µM Zn ( OAc ) 2 , pH 7 . 5 as running buffer . The absolute molecular mass of AreA663–797 was determined in series on a miniDawn TREOS static light scattering monitor and an Optilab T-rEX differential refractometer ( Wyatt , Dernbach , Germany ) . The molar mass was calculated using ASTRA 6 software ( Wyatt ) . AreA663–797 was stored in 50% vol/vol glycerol at −20°C . Real-time analyses were performed on a Biacore 2000 system ( GE Healthcare ) at 25°C . DNA duplexes were produced by annealing complementary oligonucleotides using a fivefold molar excess of the non-biotinylated oligonucleotide . The dsDNA was injected on flow cells of a streptavidin ( Sigma ) -coated CM3 sensor chip at a flow rate of 10 µl/min until the calculated amount of DNA that gives a maximum AreA663–797 binding capacity of 100 RU were bound . AreA663–797 was injected in running buffer ( 10 mM HEPES pH 7 . 4 , 150 mM NaCl , 0 . 005% ( vol/vol ) surfactant P20 , 5 mM β-mercaptoethanol and 1 µM ZnCl2 ) at concentrations from 3 . 125 to 200 nM . Sample injection and dissociation times were set to 200 and 400 s at a flow rate of 30 µl/min . Regeneration was achieved with 10 mM Tris/HCl pH 7 . 5 , 0 . 5 M NaCl , 1 mM EDTA and 0 . 005% ( wt/vol ) SDS for 1 min . Refractive index errors due to bulk solvent effects were corrected with responses from DNA-free flow cell 1 as well as subtracting blank injections . Kinetic raw data were processed and globally fitted with Scrubber 2 . 0c ( BioLogic Software ) using a 1:1 interaction model including a mass transport term . All analyses in which statistical analyses were required were performed in biological triplicates with at least two technical replications . Significance was calculated by use of the Microsoft Excel 2007 software package using the Student's paired t-test with a two-tailed distribution; p values were marked as follows: *p<0 . 05 , **p<0 . 01; ***p<0 . 001 .

Organisms produce a wide variety of small molecules called metabolites through the break down of food and other chemical reactions . Some of these molecules—known as primary metabolites—are required for growth , reproduction and other vital processes . Other molecules called secondary metabolites are not strictly required by the organism , but generally have other roles that may improve the individual’s ability to survive and reproduce . Fungi and other microbes produce a large variety of secondary metabolites , many of which are used as medicines to treat diseases in humans and other animals . For example , a molecule called lovastatin—which is produced by a fungus known as Aspergillus terreus—can reduce a human patient's risk of heart disease . However , it is not known what role many secondary metabolites play in the microbe that produced them . A . terreus lives in the soil , but it can also infect plants and animals . In addition to lovastatin , it also makes another secondary metabolite called terrein . A recent study identified the genes responsible for making terrein , and discovered that this molecule is harmful to plant cells and may help the fungus to colonize and thrive in the area immediately around plant roots , which is known as the rhizosphere . Here , Gressler et al . studied how terrein may help the fungus to cope with competitors in this environment . The experiments show that terrein increases the availability of iron and inhibits the growth of competing microbes . A shortage of iron or nitrogen-containing nutrients can stimulate the fungus to produce terrein , and elevated levels of a molecule called methionine have the same effect . These conditions are commonly found in the rhizosphere and further experiments identified several proteins in the fungus that are required for sensing them . Gressler et al . 's findings suggest that terrein helps to ensure that the fungus has sufficient nitrogen and iron to thrive in the rhizosphere . Also , this study confirms that the production of secondary metabolites in microbes can happen in response to elaborate cues from the environment , which may explain why only a limited number of secondary metabolites are produced by microbes when they are grown in the laboratory . Future studies will analyze other ways to activate the production of secondary metabolites outside of the microbe's normal environment , which may lead to the discovery of new important drugs .

2015

Phytotoxin production in Aspergillus terreus is regulated by independent environmental signals

Learning to perform a complex motor task requires the optimization of specific behavioral features to cope with task constraints . We show that when mice learn a novel motor paradigm they differentially refine specific behavioral features . Animals trained to perform progressively faster sequences of lever presses to obtain reinforcement reduced variability in sequence frequency , but increased variability in an orthogonal feature ( sequence duration ) . Trial-to-trial variability of the activity of motor cortex and striatal projection neurons was higher early in training and subsequently decreased with learning , without changes in average firing rate . As training progressed , variability in corticostriatal activity became progressively more correlated with behavioral variability , but specifically with variability in frequency . Corticostriatal plasticity was required for the reduction in frequency variability , but not for variability in sequence duration . These data suggest that during motor learning corticostriatal dynamics encode the refinement of specific behavioral features that change the probability of obtaining outcomes . Animals have the ability to learn novel motor skills , allowing them to perform complex patterns of movement to improve the outcomes of their actions . Acquiring novel skills usually requires exploration of the behavioral space , which is critical for learning ( Skinner , 1981; Sutton and Barto , 1998; Grunow and Neuringer , 2002; Kao et al . , 2005; Olveczky et al . , 2005; Tumer and Brainard , 2007; Miller et al . , 2010; Wu et al . , 2014 ) . It also requires the selection of the appropriate behavioral features that lead to the desired outcomes ( Skinner , 1981 ) . It has been postulated that the motor system can learn complex movements by optimizing motor variability in task-relevant dimensions , correcting only deviations that interfere with the final output of the action ( Todorov and Jordan , 2002; Scott , 2004; Valero-Cuevas et al . , 2009; Diedrichsen et al . , 2010 ) . By optimizing the precision of an action endpoint , for example , humans can perform smooth movements even in the presence of noise ( Harris and Wolpert , 1998 ) . Selecting task-relevant features and decreasing task-relevant variability might therefore be a critical component of motor learning ( Franklin and Wolpert , 2008; Cohen and Sternad , 2009; Valero-Cuevas et al . , 2009; Costa , 2011; Shmuelof et al . , 2012 ) . The reduction of motor variability specifically in relevant domains suggests that the neural activity giving rise to the task-relevant output is selected during learning . However , it is still unclear how the differential refinement of behavioral variability is encoded at the neural level . It has been suggested that cortical and basal ganglia circuits are important for the selection of task-relevant features ( Costa et al . , 2004; Barnes et al . , 2005; Kao et al . , 2005; Olveczky et al . , 2005; Jin and Costa , 2010; Woolley et al . , 2014 ) . Consistently , it has been previously shown that the initial stages of learning have increased behavioral ( Tumer and Brainard , 2007; Jin and Costa , 2010; Miller et al . , 2010 ) and neuronal ( Costa et al . , 2004; Barnes et al . , 2005 ) variability , but as specific movements are consolidated , neural variability is reduced in these circuits ( Costa et al . , 2004; Kao et al . , 2005 ) . This suggests that after initial motor and neural exploration , specific patterns are selected and consolidated ( Costa , 2011 ) . In this study , we investigated if the dynamics of neural activity in cortical and striatal circuits reflect the changes of variability in specific behavioral domains , and if corticostriatal plasticity is critical for the refinement of particular behavior features . We trained mice to perform a fast lever-pressing task where they were required to press a lever at increasingly higher frequencies , in order to obtain a 20 mg food pellet . After introducing the animals ( N = 20 ) to the behavioral apparatus and 1 day of continuous reinforcement , where each lever-press was reinforced , animals were trained intensively with three daily sessions for 3 days to perform fast lever presses . In the fast press schedules we introduced a covert minimum frequency target , defined by the inverse of three consecutive inter-press intervals ( 3 IPIs , 4 presses ) , which increased across sessions from 0 Hz to a maximum of 4 . 5 Hz ( Figure 1A; see ‘Materials and methods’ ) . The total number of lever presses per minute increased throughout training ( F8 , 152 = 41 . 34 , p < 0 . 0001; Figure 1—figure supplement 1A ) and animals rapidly started to organize their behavior in self-paced bouts or sequences of lever presses , until there were almost no single presses ( Figure 1C , E and Video 1 ) . 10 . 7554/eLife . 09423 . 003Figure 1 . Mice learn a fast lever-pressing task , shaping their behavior to gradually approach the minimum frequency target . ( A ) Schematic of the training protocol , starting with magazine habituation and CRF training in the first 2 days , followed by 3 days of the fast press schedules ( S1–S9 ) where we introduce an increasingly higher covert target , defined as the inverse of the sum of three consecutive inter-press intervals ( IPIs ) . ( B ) Joint distribution of the frequency ( log scale ) for all individual IPIs , in the first , middle and last session of the fast press schedules , for all the 20 animals . Vertical dashed lines correspond to the IPI threshold used for sequence definition ( IPI = 2 s , 0 . 5 Hz ) and the final covert target ( IPI = 3/660 ms , 4 . 5 Hz ) . ( C ) Percentage of lever presses comprised within sequences . ( D ) Number of sequences performed per minute . ( E ) Left: Example of sequences performed by a representative animal , aligned at the time of sequence initiation . Individual lever presses are marked as black ticks , the full sequence duration is shaded in grey and the IPIs that meet the session minimum target are shaded in orange; Top right: Probability of a magazine check immediately after a successful covert target; Bottom right: Probability of a magazine check having occurred after a reinforced lever-press vs a non-reinforced lever-press . ( F ) Distance of all three consecutive IPIs ( summed ) from the final covert target ( ∑ ( 3 IPIs ) <660 ms , ∼4 . 5 Hz ) . ( G ) Spread of the distance between all three consecutive IPIs ( summed ) around the final minimum frequency target . ( H ) Percentage of sequences containing the minimum frequency target of the last session ( end-target: 3 IPIs <660 ms , ∼4 . 5 Hz ) . Shaded areas correspond to mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 09423 . 00310 . 7554/eLife . 09423 . 004Figure 1—figure supplement 1 . Lever-pressing rate increased and shifted towards higher speeds with training , and performance increased or plateaued when task difficulty did not change in consecutive sessions . ( A ) Lever presses per minute ( F8 , 152 = 41 . 34 , p < 0 . 0001 ) . ( B ) Percentage of reinforced sequences ( F8 , 152 = 57 . 31 , p < 0 . 0001 , Post hoc comparisons: Fisher's LSD test , ( 0 . 75 Hz ) Session 3 vs Session 4 t152 = 3 . 847 , p = 0 . 0002; ( 3 Hz ) Session 6 vs Session 7 t152 = 0 . 7681 , p = 0 . 4436; ( 4 . 5 Hz ) Session 8 vs Session 9 t152 = 2 . 639 , p = 0 . 0092 ) . Shaded areas correspond to mean ± SEM . ( C , D ) Histograms with distribution of instantaneous lever-press frequencies , using linear ( C ) and log scales ( D ) , defined as the inverse of all the individual IPIs from the 20 animals . Vertical dashed lines correspond to the IPI threshold used for sequence definition ( IPI = 2 s , 0 . 5 Hz ) and the final minimum frequency target ( IPI = 220 ms , 4 . 5 Hz ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09423 . 00410 . 7554/eLife . 09423 . 005Video 1 . Animal performing sequences of lever-presses , doing magazine checks and obtaining reinforcement during the last training session . A 20 mg food pellet was delivered in the magazine when the animal performed three consecutive presses within 660 ms ( covert target = 4 . 5 Hz ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09423 . 005 The distribution of the instantaneous lever press frequencies ( calculated as the inverse of the each IPI ) shows a clear shift from initial sessions , where animals did mostly slow frequency presses ( 0–0 . 5 Hz; but already some higher frequency presses of 0 . 5–4 . 5 Hz and >4 . 5 Hz ) , to latter sessions where the distribution was shifted towards faster pressing speeds ( Figure 1—figure supplement 1C ) . A clear multimodal distribution became evident in log scale , with long IPIs ( frequencies <0 . 5 Hz , Figure 1B and Figure 1—figure supplement 1D ) representing pauses in pressing or magazine checks . This allowed us to identify the sequences or bouts of pressing a posteriori , based on behavioral performance ( either by a pause in pressing higher than 2 s or by the occurrence of checking behavior , i . e . , magazine checks between presses; see ‘Materials and methods’ ) , independently of the requirements for a specific training session . Importantly , reinforcement delivery did not provide an external cue that could be used by the animals to anticipate a reward , as the probability of performing a magazine check immediately after a successful covert target ( instead of performing another press ) was not significantly different from 0 . 5 both on early ( t19 = 0 . 9232 , p = 0 . 3675 ) and late sessions ( t19 = 1 . 763 , p = 0 . 0940 ) , and did not change throughout learning ( F8 , 152 = 1 . 753 , p = 0 . 0907 , Figure 1E , top right ) . Because a large number of sequences did not contain covert patterns ( were not reinforced ) we have also calculated the probability of a magazine check having occurred after a reinforced lever-press vs a non-reinforced lever-press , and observed that this was rather low ( ∼0 . 25 ) and did not change from early to late sessions ( Post hoc comparison: t144 = 1 . 184 , p = 0 . 283 , Figure 1E , bottom right ) . The percentage of lever presses performed within a sequence increased significantly from 56 . 98 ± 3 . 98 in the first session of covert target introduction , to 98 . 26 ± 0 . 53 in the last training session ( F8 , 152 = 60 . 22 , p < 0 . 0001; Figure 1C ) , and the number of sequences performed per minute increased with training ( F8 , 152 = 32 . 23 , p < 0 . 0001; Figure 1D ) . The percentage of reinforced sequences tended to decrease , since the difficulty of the task increased across sessions , but tended to stabilize or increase when the same target difficulty was repeated in two consecutive sessions ( F8 , 152 = 57 . 31 , p < 0 . 0001; Figure 1—figure supplement 1B ) . Importantly , with training , the distance of consecutive IPIs ( summed in bins of 3 IPIs to mimic the online criteria ) to the final target frequency ( 3 IPIs <660 ms , ∼4 . 5 Hz ) decreased consistently ( F8 , 152 = 25 . 76 , p < 0 . 0001; Figure 1F ) , indicating that animals shaped their behavior gradually to approach the end target . Not only did the distance to the end target decrease , but the spread around the target also decreased ( F8 , 152 = 9 . 616 , p < 0 . 001; calculated as the standard deviation around the target frequency , Figure 1G ) . Consistently , animals gradually increased the percentage of press bouts that would achieve the minimum target frequency of the last session ( end-target: 3 IPIs <660 ms , ∼4 . 5 Hz; F8 , 152 = 14 . 15 , p < 0 . 0001; Figure 1H ) . These data indicate that animals learned to shape their behavior to get closer to the covert target . The mean frequency of each pressing bout ( sequence frequency ) decreased slightly ( F8 , 152 = 2 . 372 , p = 0 . 0195 , Figure 2A ) , while the duration of each pressing bout ( sequence duration ) increased with training ( F8 , 152 = 22 . 69 , p < 0 . 0001 , Figure 2B ) . Importantly , the sequence-to-sequence variability of the behavioral parameters ( measured both by the variance and by the Fano factor , Figure 2C–F ) was differentially modulated during training . While the variability of sequence frequency decreased significantly throughout training ( variance: F8 , 152 = 4 . 450 , p < 0 . 0001 , Figure 2C; Fano factor: F8 , 152 = 5 . 343 , p < 0 . 0001 , Figure 2E ) , the variability of sequence duration significantly increased ( variance: F8 , 152 = 11 . 15 , p < 0 . 0001 , Figure 2D; Fano factor: F8 , 152 = 16 . 86 , p < 0 . 0001 , Figure 2F ) . The sequence-to-sequence variability of these two behavioral features was independent as there was no correlation between the variability in sequence frequency and the variability in sequence duration ( variance: R2 = 0 . 0135; Fano factor: R2 = 0 . 0119 , Figure 2—figure supplement 1 ) . This is in contrast with a strong correlation observed between variability in sequence duration and the variability in sequence length—number of presses ( variance: R2 = 0 . 8710; Fano factor: R2 = 0 . 8839 , Figure 2—figure supplement 1 ) . The decrease in frequency variability cannot be explained by animals reaching a ceiling in pressing frequency , since the average frequency did not increase with training ( it actually decreased slightly ) . Furthermore , frequency variability started stabilizing after session 4 where the target constrains are still rather loose ( 3 IPIs in less than 4 s ) and this is a frequency that animals can reach in 78 . 91 ± 5 . 09% of the sequences at the end of training . 10 . 7554/eLife . 09423 . 006Figure 2 . Variability of behavioral dimensions evolves independently as animals learn a motor task . ( A , B ) Frequency and duration of lever-press sequences ( C–F ) Variability , measured as the variance and Fano factor , for sequence frequency and sequence duration . ( G–H ) Fano factor of both frequency and duration , normalized to the first session , for the frequency and control tasks . Shaded areas correspond to mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 09423 . 00610 . 7554/eLife . 09423 . 007Figure 2—figure supplement 1 . Significant correlation between variability of number of presses and duration , but not between variability of frequency and duration . Scatter plots of the paired values , variances and Fano factors , for frequency/duration and number of presses/duration . Each point corresponds to one session of one individual animal , with darker colors depicting later sessions . Line corresponds to the best linear fit of all the data , with the correspondent R2 value . DOI: http://dx . doi . org/10 . 7554/eLife . 09423 . 007 In order to test the specificity of these results , a different group of animals ( N = 8 ) was trained on a control task ( Figure 2H ) , where sequences of exactly four consecutive presses were reinforced but where the frequency at which these sequences were performed was not relevant . In contrast with the results observed for the frequency task , in which the sequence-to-sequence variability in frequency decreased ( F8 , 152 = 5 . 343 , p < 0 . 0001 ) and in duration increased ( F8 , 152 = 16 . 86 , p < 0 . 0001 ) ( Figure 2G ) , in this control task the variability of sequence frequency did not decrease with training ( F8 , 56 = 1 . 049 , p = 0 . 4113 ) , while variability in sequence duration did ( F8 , 56 = 4 . 589 , p = 0 . 0002 ) ( Figure 2H ) . These data indicate that the decrease in variability in sequence frequency was task-specific . To further investigate this , we analyzed if the variability of these two behavioral dimensions was different in reinforced vs non-reinforced sequences ( Figure 3 ) . We verified that sequences leading to reinforcement had indeed significantly lower variability in frequency compared to non-reinforced sequences ( main effect of reinforcement , F1 , 38 = 7 . 608 , p = 0 . 0089 , Figure 3C and F1 , 38 = 28 . 34 , p < 0 . 0001 , Figure 3E ) , but there were no significant differences in the variability of sequence duration between reinforced and non-reinforced sequences ( Figure 3D , F ) . These results suggest that mice selectively reduced variability in the behavioral domains where variability affected the probability of reinforcement ( sequence frequency ) , but not in domains where variability did not change this probability ( sequence duration ) . 10 . 7554/eLife . 09423 . 008Figure 3 . Behavior variability is differentially modulated during training . ( A , B ) Comparison of frequency and duration between reinforced ( RF ) and non-reinforced ( Non-RF ) sequences . ( C , D ) Variance and ( E , F ) variability , measured as the Fano factor , for reinforced and non-reinforced sequences . Black lines correspond to mean values for non-reinforced sequences . Red lines correspond to mean values for reinforced sequences . Shaded areas correspond to mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 09423 . 008 In order to investigate the dynamics of cortical and striatal circuits during the acquisition and performance of the fast lever pressing task , we continuously recorded extracellular neuronal activity simultaneously in layer 5 of the primary motor cortex ( M1 ) , and in the dorsal striatum ( DS ) of mice during the full duration of training ( 4 days , N = 7 animals , average of 18 M1 units and 10 DS units simultaneously recorded per animal , per session ) . Non-stop continuous electrophysiological recordings across 4 days encompassing all the sessions of training allowed us to track the activity of a subset of ‘stable’ cells throughout the whole period of training ( 49 M1 units , 21 DS Units ) . Putative single-units were isolated based on waveform characteristics , inter-spike intervals ( ISI ) and clustering statistics using principal component analysis ( PCA ) . Units were considered ‘stable’ if the statistics in PCA space and waveform proprieties did not change significantly across sessions ( see ‘Materials and methods’ and Figure 4—figure supplement 1C ) . We found a high sequence-to-sequence variability in the activity of individual neurons ( measured by the Fano factor of the firing rate ) in the first couple of sessions , that then decreased with training ( DS: F8 , 48 = 2 . 767 , p < 0 . 05; M1: F8 , 48 = 2 . 771 , p < 0 . 05; Figure 4A ) . These dynamics in neuronal variability were observed during the performance of lever-press sequences , but not during baseline periods ( measured from 5 to 2 s before the initiation of each sequence ) , when the animals were not actively engaged in lever pressing ( DS: F8 , 48 = 1 . 117 , p = 0 . 3324; M1: F8 , 48 = 1 . 459 , p = 0 . 1973; Figure 4B ) , or during periods flanking the sequence ( first press: DS F8 , 48 = 1 . 213 , p = 0 . 3121; M1 F8 , 48 = 0 . 1374 , p = 0 . 9971; last press: DS F8 , 48 = 0 . 5227 , p = 0 . 8335; M1 F8 , 48 = 0 . 8677 , p = 0 . 5499; Figure 4—figure supplement 2 ) . The decrease in neuronal variability was also observed when using exclusively ‘stable’ cells for this analysis ( DS: F8 , 160 = 5 . 223 , p < 0 . 0001; M1 F8 , 384 = 12 . 72 , p < 0 . 0001; Figure 4C ) , showing that the differences in variability throughout learning could be observed in individual cells , and did not represent a shift in the population of neurons recorded across days . Importantly , the average firing rate of individual cells did not change significantly , neither across sessions nor across days ( p > 0 . 05 for all conditions , Figure 4E–H ) , suggesting that the reduction in variability was not attributable to overall changes in firing rate , but instead to the selection/refinement of a particular firing patterns related to sequence execution . 10 . 7554/eLife . 09423 . 009Figure 4 . Trial-to-trial variability in corticostriatal circuits decreases throughout training . ( A–D ) Neuronal variability ( measured as the Fano factor of firing rates ) during sequence performance and baseline periods , for all the recorded neuronal units and exclusively for ‘stable units’ , for both M1 ( blue traces ) and dorsal striatum ( DS , red traces ) . ( E–H ) Firing rates during sequence performance and baseline , for all the recorded units and exclusively for stable units , for M1 ( blue traces ) and DS ( red traces ) . ( I , J ) Fano factor ( FF ) and firing rate ( FR ) modulation relative to baseline values , for individual units recorded across the training sessions ( stable units ) within DS ( top colorplots ) and M1 ( bottom colorplots ) . Right panels depict average modulation . Shaded areas correspond to mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 09423 . 00910 . 7554/eLife . 09423 . 010Figure 4—figure supplement 1 . Histological confirmation of electrode tip position and stable units criteria . ( A ) Depiction of electrode array tip localization for motor cortex ( top ) and DS ( bottom ) for each individual animal . ( B ) Example coronal brain slice magnification using cresyl violet staining for confirmation of electrodes position . Atlas adapted from Paxinos and Franklin ( 2004 ) . ( C ) Illustration of an example stable cell . Diagram illustrating the criteria for stability of cells across different recording sessions ( top left , c1: cluster centroid during one session , c2: centroid of the same cluster in the subsequent session , see ‘Materials and methods’ ) . Average waveform in each session ( bottom left ) . Cluster projection using principal component analysis ( PCA ) across the training sessions ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09423 . 01010 . 7554/eLife . 09423 . 011Figure 4—figure supplement 2 . Neuronal variability around the first and last press of a sequence does not change with training . Fano factor calculated for 1 s intervals around the first ( DS F8 , 48 = 1 . 213 , p = 0 . 3121; M1 F8 , 48 = 0 . 1374 , p = 0 . 9971 ) and last presses ( DS F8 , 48 = 0 . 5227 , p = 0 . 8335; M1 F8 , 48 = 0 . 8677 , p = 0 . 5499 ) of a sequence . Red lines correspond to mean value for DS , blue lines correspond to mean value for M1 . Shaded areas represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 09423 . 011 Further analysis of these dynamics for individual stable cells clearly showed higher variability relative to baseline during the initial sessions ( first session DS: W = 134 , p = 0 . 0107; first session M1: W = 1119 , p < 0 . 0001 ) , that decreased throughout training until it reached the same levels of baseline at the end of training ( last session DS: W = 73 , p = 0 . 2157; last session M1: W = 253 , p = 0 . 2121; Figure 4I ) . Again , average firing rates did not show any significant modulation in relation to baseline throughout the whole period of training ( DS: F8 , 160 = 1 . 031 , p = 0 . 4153; M1: F8 , 384 = 1 . 757 , p = 0 . 084; Figure 4J ) . This decrease in sequence-to-sequence variability of neural activity did not seem to result from the behavior becoming more stereotyped with training , as variability in behavior decreased for frequency but increased for duration ( Figure 2 ) . To further control that the decrease in neural variability was due to gross changes in behavior we restricted our analyses to sequences matched for frequency ( t48 = 1 . 800 , p = 0 . 0781 ) and duration ( t48 = 1 . 733 , p = 0 . 0895 ) between early and late sessions ( Figure 5A , B ) . We observed that neuronal variability was still elevated in early sessions and decreased as training progressed ( DS: F8 , 48 = 2 . 732 , p = 0 . 0144; M1: F8 , 48 = 2 . 491 , p = 0 . 0239; Figure 5C ) . Again , these dynamics were not observed during baseline periods ( DS: F8 , 48 = 1 . 483 , p = 0 . 1884; M1: F8 , 48 = 1 . 241 , p = 0 . 2965; Figure 5D ) and no changes in firing rates were evident in sequence ( DS: F8 , 48 = 0 . 4684 , p = 0 . 8723; M1: F8 , 48 = 0 . 4040 , p = 0 . 9128; Figure 5E ) or baseline periods ( DS: F8 , 48 = 0 . 2208 , p = 0 . 9855; M1: F8 , 48 = 0 . 3354 , p = 0 . 9479; Figure 5F ) . Single unit analysis also revealed a significant decrease in Fano factor modulation throughout training ( DS: F8 , 160 = 2 . 688 , p = 0 . 0084; M1:F8 , 384 = 9 . 705 , p < 0 . 0001; Figure 5G ) with no modulation in firing rates ( DS: F8 , 160 = 0 . 3008 , p = 0 . 9648; M1:F8 , 384 = 1 . 406 , p < 0 . 1923; Figure 5H ) . 10 . 7554/eLife . 09423 . 012Figure 5 . Neuronal variability dynamics are still evident when analysis is restricted to sequences with duration and frequency . ( A , B ) Frequency and duration of matched sequences . ( C , D ) Neuronal variability , measured as the Fano factor of the firing rate , for sequences of matched duration and frequency , for both recorded areas , during sequences and baseline . ( E , F ) Firing rates , for sequences of matched duration and frequency , during sequences and baseline . ( G , H ) Fano factor ( FF ) and firing rate ( FR ) modulation relative to baseline values , for individual units recorded across the training sessions ( stable units ) within DS ( top colorplots ) and M1 ( bottom colorplots ) , for sequences of matched duration and frequency . Right panels depict average modulation . Error bars correspond to mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 09423 . 012 The results above suggest that the decrease in corticostriatal variability is not related to a general decrease in behavioral variability . We therefore investigated if the changes in sequence-to-sequence variability in neural activity were related to the changes in sequence-to-sequence variability of specific behavioral dimensions . We re-calculated the Fano factor of the behavioral features and the neuronal activity using a moving average of a reduced number of trials ( 5 ) to provide a higher within session resolution of the variability dynamics and therefore permit the correlation of behavioral and neuronal dynamics across training for each animal ( Figure 6A , see ‘Materials and methods’ ) . Analyses of the relationship between the variability of the recorded units and the variability of each independent behavior feature revealed a significant increase in correlation between neuronal and behavior variability , specific for sequence frequency ( Figure 6C ) , but not for duration ( Figure 6D ) . These results were observed when using only task-relevant or non-task-relevant neurons ( data not shown ) . They were also observed using different number of trials for calculating the moving average of the Fano factor ( Figure 6—figure supplement 2 ) . 10 . 7554/eLife . 09423 . 013Figure 6 . Correlations between corticostriatal and behavioral variability emerge for specific behavioral features . ( A ) Example traces from a single animal representing variability , calculated as the Fano factor , using a moving window of five consecutive trials shifted by one for sequence frequency ( dark blue trace ) , sequence duration ( green trace ) , M1 units firing rate during sequences ( blue trace ) and baseline ( grey trace ) , and DS units firing rate during sequences ( red trace ) and baseline ( grey trace ) . Vertical dashed lines represent separation of different training sessions . Shaded areas correspond to mean ± SEM . ( B ) Correlation between the variability ( FF ) in M1 and DS . ( C , D ) Correlation between variability traces from neuronal firing rates in M1 ( blue bars ) or DS ( red bars ) , and variability of sequence frequency or duration . Error bars denote correlation coefficient ±standard error of the correlation . *p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 09423 . 01310 . 7554/eLife . 09423 . 014Figure 6—figure supplement 1 . No significant correlation was found between average firing rate and any of the behavior features . Correlation between the firing rate and ( A ) sequence frequency or ( B ) sequence duration . Error bars denote correlation coefficient ±standard error of the correlation . DOI: http://dx . doi . org/10 . 7554/eLife . 09423 . 01410 . 7554/eLife . 09423 . 015Figure 6—figure supplement 2 . Changing the number of trials used for Fano factor calculation did not affect the observed corticostriatal and neuronal/behavioral variability correlations . ( A , B ) Correlation between variability traces from neuronal firing rates of cortical and striatal units , and between neuronal variability and variability of sequence frequency and duration using 3 ( A ) or 7 ( B ) consecutive trials . Error bars denote correlation coefficient ±standard error of the correlation . *p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 09423 . 015 These results show that the decrease in variability in M1 and DS is not just a reflection of a more constrained performance of the movement as training progresses; variability of the movement decreased in a specific dimension but it increased in others were no significant correlation with neuronal variability was evident . Furthermore , no significant correlations were observed between the firing rate of neurons and the variability any of the behavior features ( Figure 6—figure supplement 1 ) , indicating again that the observed relationship between neuronal and behavior dynamics was not the reflex of a general increase in correlation between neuronal activity and behavior . The data presented above suggested that as training progressed variability in M1 and striatum became more correlated with variability in a specific domain of behavior that changed the probability of reinforcement . This suggests that neural variability in M1 and striatum could also become more coupled with training . We verified that at the onset of training the sequence-to-sequence variability of neural activity in DS and M1 in each animal was not correlated . However , a strong correlation between the variability in DS and M1 rapidly emerged during training ( p < 0 . 05 for all except the first training session , Figure 6B ) , suggesting that as behavioral variability is refined , neural variability in M1 and striatum becomes correlated . The results presented above show that a coupled reduction in corticostriatal variability accompanies the reduction in variability of sequence frequency , but not of sequence duration , suggesting that corticostriatal plasticity is necessary to select the appropriate motor features and hence reduce variability within specific domains . We decided to directly test if the observed reduction in sequence frequency variability is dependent on corticostriatal plasticity by using mutant mice with NMDA receptors deleted specifically at glutamatergic synapses of striatal projection neurons ( RGS9-LCre::Grin1tm1Yql; referred to in the figures as striatal projection neuron SPN NR1-KO ) , which have impaired corticostriatal plasticity ( Dang et al . , 2006 ) , and control littermates . Mutant animals had more difficulty learning the task , so we adapted the training protocol to one session per day for both mutant and littermate controls ( and repeated sessions when needed ) , in order to achieve comparable performance levels ( see ‘Materials and methods’ , Table 1 and Figure 7A ) . 10 . 7554/eLife . 09423 . 016Table 1 . Training protocol and respective number of animals reaching performance criteria for the SPN NR1-KO group and littermate controlsDOI: http://dx . doi . org/10 . 7554/eLife . 09423 . 016Training protocolFree0 . 375 Hz0 . 375/0 . 75 Hz ( 30 reinf ) 0 . 75 Hz1 . 5 Hz1 . 5/3 Hz ( 30 reinf ) 3/6 Hz ( 10 reinf ) 6/7 . 5 Hz ( 10 reinf ) # of subjects reaching criteriaNR1–KO77665411Controls5555555210 . 7554/eLife . 09423 . 017Figure 7 . Corticostriatal plasticity is necessary for the specific refinement of behavioral variability . ( A ) Schematic of the adapted training sessions for mutant animals and littermate controls . Animals would remain in the same training session until reaching a stable performance . ( B ) Distance of the sum of all three consecutive IPIs from the final covert target ( ∑ ( 3 IPIs ) <660 ms , ∼4 . 5 Hz ) in SPN NR1 mutants and littermate controls ( C ) Spread of the distance between three consecutive IPIs around the final covert target . ( D–G ) Behavior parameters and variability , measured as the Fano factor , during early and late training sessions in SPN NR1 mutants and littermate controls groups . Bars correspond to mean , with data from individual animals plotted on the background ( red: SPN NR1-KO; black: littermate controls ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09423 . 01710 . 7554/eLife . 09423 . 018Figure 7—figure supplement 1 . Bootstrapping statistics in the SPN NR1-KO data support the observations from the post hoc planned comparisons . Histograms depicting the sampled statistic ( difference between the means of two groups ) , after sampling with replacement the original data 100 . 000 times . Red vertical lines correspond to the 5% confidence intervals . Green vertical line corresponds to the mean of the sampled data . Blue vertical line corresponds to the difference between the original data groups ( H0 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09423 . 018 As expected , the distance to target ( Controls: p = 0 . 0450 , t5 = 2 . 657 , Figure 7B ) and spread around the target ( Controls: p = 0 . 0179 , t5 = 3 . 466 , Figure 7C ) decreased in littermate controls . However , neither of these measures changed with training in mutants ( Mutants: p = 0 . 3535 , t6 = 1 . 005; and p = 0 . 2817 , t6 = 1 . 183 , respectively; Figure 7B , C ) . In general , no significant difference was observed for any of the behavior features between the two groups of animals . However , planned comparisons did show that RGS9-LCre::Grin1tm1Yql mutants did not decrease sequence frequency variability during training , in contrast to littermate controls which did ( significant main effect of training time: F1 , 10 = 10 . 13 , p = 0 . 009; Posthocs: Mutant group: t10 = 1 . 38 , p = 0 . 1964; Control group: t10 = 3 . 00 , p = 0 . 0134 ) . Importantly , no differences in the modulation of sequence duration variability were observed between the two groups ( no significant main effect for genotype: Duration FF: F1 , 10 = 0 . 02 , p = 0 . 887 ) ( Figure 7D–G ) . These statistical results were robust as they were confirmed using bootstrapping statistics ( using 100 . 000 random samples of the data , with replacement ) ( Figure 7—figure supplement 1 ) . These data suggest that corticostriatal plasticity is required for the reduction in variability of specific behavioral features that change the probability of reinforcement . In this study we show that when mice are trained on a difficult operant paradigm they differentially refine specific behavioral features . When mice were asked to perform progressively faster covert patterns of lever presses to obtain a reinforcer , they reduced variability in sequence frequency , but increased variability in an orthogonal uncorrelated feature ( sequence duration ) . These results are interesting because both features would be classically considered task-relevant—a covert sequence of four presses , which is the minimum to produce a reinforcer in this task , has to have a minimal duration . However , although both features could be considered relevant for the task , only changes in frequency variability were differentially reinforced . Reinforced sequences had lower variability in frequency than non-reinforced sequences , but had equal variability in duration as non-reinforced sequences . Thus , our results indicate that animals reduced frequency variability because that was what was reinforced throughout training . Consistent with this interpretation , in a task where the exact number of presses ( correlated with duration ) was reinforced but the frequency at which the sequence was performed was not , variability in duration decreased and in frequency increased . This in line with data demonstrating differential modulation of the different components of task space during learning ( Todorov and Jordan , 2002; Müller and Sternad , 2004; Cohen and Sternad , 2009 ) . In previous studies from our group where animals performed operant tasks where the constrains were more relaxed ( Jin and Costa , 2010 ) , animals decreased variability in all behavioral domains ( i . e . , they became more stereotypical overall ) . However , when faced with a more challenging task as in the present study , they decreased variability in the domain that was critical for getting a reinforcer , but increased variability in orthogonal domains ( i . e . , they were more stereotypical in just a particular domain ) . It could be that the increase in variability in the orthogonal behavioral domains happens because in difficult tasks animals try to minimize the effort to obtain reinforcers , and hence do not attempt to reduce variability in more than one independent domain . Alternatively , it could also be that mice increased the duration of the sequence ( and the correlated number of presses ) as a strategy to try to increase the probability of getting a successful covert pattern in that sequence . However , this second possibility is less likely , given that the two behavioral features were not correlated , and that sequences of different durations were equally likely to get reinforced . These data suggest that in more challenging motor tasks it is difficult to reduce variability in all domains , and animals seem to differentially refine the motor patterns that led to reinforcement . Consistently , the number of sequences that comply with the minimum frequency required for the last session ( end-target ) increased with training and the distance to the end-target decreased with training , indicating that mice implicitly learned to shape their behavior to match the task requirements . At the neural level , we observed initial high sequence-to-sequence variability of neuronal activity in corticostriatal circuits that decreased with training . Variability in the spike patterns of individual neurons and populations of neurons may be the bases for a process of behavioral exploration ( or trial ) ( Olveczky et al . , 2005; Kao et al . , 2005; Mandelblat-Cerf et al . , 2009 ) , while a decrease in neural variability may reflect a process of selection of specific patterns of neural activity that lead to specific behavioral outputs ( Costa et al . , 2004; Kao et al . , 2005; Fee and Goldberg , 2011 ) . It has been suggested that a decrease in corticostriatal variability as a motor task is learned ( Costa et al . , 2004; Barnes et al . , 2005 ) could correspond to the process of selection and consolidation of specific motor patterns ( Costa , 2011 ) . Here , we show that this decrease in neural variability in corticostriatal circuits correlates specifically with the decrease in variability of a particular behavior domain . These data suggest that the neural patterns in motor cortex and sensorimotor striatum that give rise to the behavioral patterns that are reinforced are progressively selected . Provocatively , it also suggests that changes in motor variability that are not specifically reinforced but are part of a strategy or driven by effort reduction may be encoded somewhere else . Finally , we also show that corticostriatal plasticity is important for the refinement of specific behavior features . Our data therefore suggests an important role for corticostriatal plasticity in selecting the appropriate implicit neural and behavioral patterns that are reinforced ( Costa , 2011 ) . However corticostriatal plasticity did not seem to be necessary for the increase in behavioral variability in other domains ( Goldberg and Fee , 2011 ) . Although in this study we don't investigate the mechanisms underlying the generation of variability , several studies have suggested that the basal ganglia , dopaminergic system , specific cortical circuits , or cerebellar circuits could subserve this function ( Olveczky et al . , 2005; Costa et al . , 2006; Leblois et al . , 2010; Costa , 2011; Fee and Goldberg , 2011; Shmuelof and Krakauer , 2011; Woolley et al . , 2014 ) . Taken together , our findings suggest that corticostriatal plasticity is important to select the neural patterns that lead to the movement patterns that are reinforced . They highlight that corticostriatal plasticity is not only important for choosing which action to do , but also to shape how to do it to obtain a desired outcome . All experiments were carried in accordance to the ethics committee guidelines of the Champalimaud Foundation and Instituto Gulbenkian de Ciência , and with approval of the Portuguese DGAV ( Ref . 0421 ) . Experiments were carried out using 20 male , 3 to 5 month old C57BL6/J mice . From these , 13 animals were used exclusively for behavioral training while the remaining seven underwent microelectrode array implantation for neuronal data recordings . Animals were maintained on a light–dark cycle of 12 hr:12 hr starting at 7 AM . All experiments were done during the light cycle . Mice were housed in groups of four animals prior to surgery and individually after the electrodes were implanted . 3 to 6 months old RGS9-LCre::Grin1tm1Yql homozygous mice ( N = 7 ) and Cre negative littermate controls ( N = 5 ) were used for the mutant mouse behavioral experiments . Seven C57Bl6/J mice were implanted bilaterally with two micro-electrode arrays ( 2 × 8 ) , 35–50 µm tungsten electrodes with micro-polished tips . One array targeted the primary motor cortex ( M1 , layer 5 ) while the second was targeting the ( DS , sensorimotor area that receives projections from the same area in M1 ) . Craniotomies and electrode array positioning were done according to coordinates from the Mouse Brain Atlas ( Paxinos and Franklin , 2008 ) . M1 array was placed 1 mm rostral and 1 . 6 mm lateral from bregma , and lowered ∼1 mm from the surface of the brain . DS array was placed 0 . 5 mm rostral and 2 . 1 mm lateral from bregma , and lowered ∼2 . 3 mm from the surface of the brain . Electrodes were manually lowered at slow rates while constantly monitoring neural activity in all the channels in order to control for proper electrode function and correct positioning . Final verification of electrode position was done after all the experiments were finished , by perfusing animals with PFA and histological confirmation of Nissl stained 70 µm brain slices ( Figure 4—figure supplement 1A , B ) . After surgery animals were allowed to recover for at least 2 weeks before starting any other experimental procedure . Single and multi unit activity was recorded using Blackrock Microsystems Neural Signal Processor , allowing for online sorting of identified units . Further offline sorting of selected units was done using Plexon Offline Sorter v3 ( Plexon Inc , Dallas , TX , United States ) , based on waveform characteristics , ISI and PCA clustering . Units stability was assessed from waveforms and PCA cluster proprieties . For PCA cluster comparison data from all the training sessions was pooled together to calculate common eigen vectors . Data from individual sessions was then projected into this common PC space , allowing us to determine cluster centroids and dispersion for each session . Clusters were considered stable whenever the centroid in a given session was comprised within the interval of the centroid of the previous session ±1 . 96 * standard deviation of the cluster , in the first two principal components ( Figure 4—figure supplement 1C for a graphical representation of this criteria ) . Animals were trained using operant chambers ( MedAssociates Inc , St . Albans , VT , United States ) placed inside sound attenuating boxes . A retractable lever was extruded in the beginning of each session , simultaneous to the onset of a light . Animals were required to perform a sequence of presses at a minimum frequency in order to obtain a 20 mg food pellet ( Bio-Serv , Flemington , NJ , United States ) . 24 hr before the first training session animals were placed under a food restriction schedule . Body weight was constantly monitored in order to be kept above 85% of the initial weight . In order to facilitate learning , animals were initially exposed to one session of magazine training were food pellets would be available on a random time schedule , and to three sessions of continuous reinforcement schedule ( CRF ) 1 day before training , where single lever presses would be reinforced . On the following training sessions animals were reinforced if they performed a sequence of consecutive presses at a minimum frequency ( covert target ) , defined by the inverse of three consecutive inter-press intervals ( IPIs ) , which increased with training . On the first session there was no minimum frequency target , meaning that any consecutive 3 IPIs would lead to reinforcement . In consecutive sessions the minimum frequency that would lead to reinforcement was increased or maintained in the following order: 0 . 375 Hz , 0 . 75 Hz , 0 . 75 Hz , 1 . 5 Hz , 3 Hz , 3 Hz , 4 . 5 Hz and 4 . 5 Hz . This constant increase in the minimum frequency of the covert target forced the animals to systematically adapt to the task requirements and perform faster sequences of presses from session to session . The training protocol for mutant animals and littermate controls was adapted due to difficulties learning the task , to one daily session and using automatic progressive schedules once a minimum number of reinforcements ( 30 or 10 ) was achieved . ( Table 1 for performance summary . ) Sequences of presses were differentiated based on IPI and occurrence of a magazine head entry . An IPI >2 s ( determined based on the distribution of IPIs ) or a head-entry were used to define the bouts or sequences of presses . The 2 s cutoff was determined from the joint distribution of the instantaneous IPIs ( and the corresponding log distribution ) from all the animals , by determining the valley between the two main peaks of IPIs ( Figure 1—figure supplement 1C , D ) . Frequency of each sequence was defined as the inverse of the average IPI of each sequence . Duration of each sequence was defined as the time between the first and the last press event . Length of each sequence was defined as the number of press events in each sequence . For the matched sequences analysis , sequences with a duration of 0 . 2–2 s and a frequency higher than 2 Hz were selected . Neural activity was averaged in 20-ms bins , shifted by 1 ms , and averaged across trials to construct the peri-event histogram ( PETH ) . Data from the PETH from 5000 to 2000 ms before lever press were considered as baseline activity . A positive modulation in firing rate was defined if at least 20 consecutive bins had firing rate larger than a threshold of 99% above baseline activity , and a negative modulation of firing rate was defined if at least 20 consecutive bins had a firing rate smaller than a threshold of 95% below baseline activity ( Belova et al . , 2007 ) . Paired t-tests between baseline firing rate and sequence firing rate were used to classify individual neurons as sequence-related . The programs to run the tasks presented in this study can be found at http://tinyurl . com/or7ug72 . Analyses were done in Matlab ( MathWorks , Natick , MA , United States ) or GraphPad Prism ( GraphPad Software Inc , La Jolla , CA , United States ) . Normality was verified for all tests using the D'Agostino-Pearson omnibus normality test , or the Kolmogorov–Smirnov test when sample size was too small . Repeated measures ANOVA were used to evaluate changes in behavior and neuronal features . Probability of a magazine check after lever-press was evaluated using one-way ANOVA and post hoc comparisons using Fisher's LSD test , but one subject was excluded from these analysis due to a lack of recorded timestamps for magazine head-entries . Paired t-tests were used to evaluate differences in percentage of lever-presses . Increases in FF modulation were assessed by the Wilcoxon Rank Signed test . Repeated measures two-way ANOVA was used to verify the general effect of the RGS9-NR1 mutants experiment . Bootstrapping statistics were used on the data from the RGS9-NR1 mutants and littermate controls to validate the results from the post hoc tests . Histograms were built from 100000 randomized samples with replacement . Sample sizes were calculated based on α = 0 . 05 and power of 0 . 7 . Trial to trial variability of neuronal and behavior data was assessed using Fano factor . We calculate the Fano factor of individual units by dividing the variance of firing rates across all the trials of a session by the mean over those trials . Fano factor and firing rate modulations for individual stable cells were calculated as the ratio between the difference of values for sequence and baseline and the values during baseline ( Fano factor: [FFsequence − FFbaseline]/FFbaseline; firing rate: [FRsequence − FRbaseline]/FRbaseline ) . Fano factor of the behavioral features was calculated by dividing the variance in the individual features by the mean of the feature for all the trials . To establish correlations between the variability of the neuronal data and the variability of the behavior , Fano factors were calculated using three , five or seven consecutive trials , allowing us to increase the resolution of the variability measures . Correlations between neuronal and behavior data were evaluated using Pearson's linear correlations . To avoid correlations bias due to sample size , statistical significance of all the correlations was assessed using the significance criteria for the session with smaller size . Within animal correlations averaged using Fisher's z transformation ( Silver and Dunlap , 1987 ) returned similar results to grouped correlations for all the tested conditions ( data not shown ) .

Learning a new motor skill typically involves a degree of trial and error . Movements that achieve the desired outcome—from catching a ball to playing scales—are repeated and refined until they can be produced on demand . This process is made more difficult as the activity of individual neurons and muscle fibers can vary at random , and this reduces the ability to reproduce a given movement precisely and reliably . It has been suggested that the motor system overcomes this problem by identifying those parts of a task that are essential for achieving the end goal , and then focusing resources on reducing the variability in the performance of those parts alone . Santos et al . now provide direct evidence in support of this proposal by recording the activity of neurons in motor regions of the mouse brain as the animals learn a lever pressing task . By giving mice a food reward each time they pressed the lever four times in a row , Santos et al . trained the animals to press the lever in bouts . The experiment was then slightly modified , so that the mice had to perform the four lever presses more rapidly in order to earn their reward . Consistent with predictions , the average speed of lever pressing initially varied greatly , but this variability decreased as the animals learned the task . By contrast , the total duration of individual bouts of lever pressing—which depends largely on the number of times the mice press the lever—was just as variable after training as before . A similar pattern emerged for the activity of individual motor neurons in the mouse brain . Whereas their activity initially varied greatly , this variability decreased over training . Moreover , it became increasingly linked to the variability in the speed of lever pressing , but not with the variability in the duration of individual bouts . The work of Santos et al . has thus shown in real time how the motor system focuses its efforts on reducing variability in those specific parts of a task that are essential for achieving a goal . Without a process called corticostriatal plasticity , by which the motor system adapts , mice could not refine this variability .

2015

Corticostriatal dynamics encode the refinement of specific behavioral variability during skill learning

Animal behavior is directed by the integration of sensory information from internal states and the environment . Neuroendocrine regulation of diverse behaviors of Caenorhabditis elegans is under the control of the DAF-7/TGF-β ligand that is secreted from sensory neurons . Here , we show that C . elegans males exhibit an altered , male-specific expression pattern of daf-7 in the ASJ sensory neuron pair with the onset of reproductive maturity , which functions to promote male-specific mate-searching behavior . Molecular genetic analysis of the switch-like regulation of daf-7 expression in the ASJ neuron pair reveals a hierarchy of regulation among multiple inputs—sex , age , nutritional status , and microbial environment—which function in the modulation of behavior . Our results suggest that regulation of gene expression in sensory neurons can function in the integration of a wide array of sensory information and facilitate decision-making behaviors in C . elegans . In order to effectively adapt to an environment , organisms need to be able to analyze both their internal state and external surroundings to make behavioral decisions that will maximize their chances of survival . It has been hypothesized that decision making in animals is controlled primarily at the level of command interneurons or decision centers in the central nervous system , which receive input from an array of sensory neurons and integrate the information conveyed to inform behavioral decisions ( Kristan , 2008 ) . However , recent work in a number of animal species has demonstrated that modulation of gene expression and neuronal activity by both environmental cues and internal state can occur at the level of sensory neurons , suggesting that aspects of behavioral decision making may also occur in the peripheral nervous system ( Dey et al . , 2015; Farhadian et al . , 2012; Ryan et al . , 2014 ) . Two pervasive internally derived modifiers of decision making in animals are the nutritional state of the animal and biological sex . Feeding status and satiety levels have long been known to have dramatic effects on the behaviors and decisions of species across the animal kingdom ( Pool and Scott , 2014; Sengupta , 2013 ) . In many cases , the nutritional state of the animal will serve as a source of regulation directly on feeding and other food-related behaviors . For instance , in the medicinal leech , periods of food deprivation can lead to a heightened response to appetitive stimuli in its environment , a response that has been shown to be carried out via a correlation between serotonin production and time of last meal ( Gaudry and Kristan , 2012; Groome et al . , 1993; Lent et al . , 1991 ) . Starvation or nutritional deprivation can also have less direct effects on feeding behaviors as in Drosophila , where periods of starvation have been demonstrated to both enhance odor sensitivity and attraction as well as abrogate avoidance responses to normally aversive stimuli ( Bräcker et al . , 2013; Inagaki et al . , 2014; Ko et al . , 2015; Root et al . , 2011 ) . Similarly , the biological sex of an animal is a potent modifier of decision-making . Not only does the sex of the animal lead to differences in mating behaviors and pheromone attraction , but studies in several systems have suggested that biological sex can also alter olfactory and gustatory preferences , perhaps to guide decision-making in the context of mating and reproduction ( Dey et al . , 2015; Kimchi et al . , 2007; Kurtovic et al . , 2007; Lebreton et al . , 2015; Lee and Portman , 2007; Meunier et al . , 2000; Nakagawa et al . , 2005; Ribeiro and Dickson , 2010; Ryan et al . , 2014; Stowers and Logan , 2010; Stowers et al . , 2002; Vargas et al . , 2010 ) . The roundworm , Caenorhabditis elegans , and its well-defined nervous system provide a simple experimental system well suited to investigate how internal states and environmental cues can be integrated to modulate animal behavior . C . elegans exist as one of two sexes: hermaphrodite or male . Hermaphrodites produce a limited amount of sperm during larval development making them capable of self-fertilization in addition to cross-fertilization . In contrast , males are only cross-fertile and must find mating partners for reproduction . As such , the male displays a large behavioral repertoire primarily aimed at facilitating mate finding and successful mating encounters . These behaviors include complex mating rituals ( Liu and Sternberg , 1995 ) , differential sensitivity to hermaphrodite pheromone ( Srinivasan et al . , 2008 ) , and a mate-searching behavior in adult male animals that is driven by the prioritization of mating over feeding ( Lipton et al . , 2004 ) . Mate-searching behavior of C . elegans has thus emerged as an interesting case study of decision making in a simple animal , which is subject to regulation by multiple features of internal state—sex , age , and nutritional status—as well as by external sensory experiences such as the presence of food or hermaphrodites in the environment ( Barrios et al . , 2008; Lipton et al . , 2004; Ryan et al . , 2014 ) . Here , we focus on the DAF-7/TGF-β signaling pathway , which plays a pivotal role in the regulation of diverse aspects of C . elegans physiology and behavior including the dauer developmental decision , reproductive egg laying , feeding and foraging , fat storage , satiety quiescence , longevity , aggregation , aerotaxis and avoidance of pathogenic bacteria ( Chang et al . , 2006; Gallagher et al . , 2013; Greer et al . , 2008; Meisel et al . , 2014; Milward et al . , 2011; Ren et al . , 1996; Shaw et al . , 2007; White and Jorgensen , 2012; You et al . , 2008 ) . The expression pattern of daf-7 was originally identified as being restricted to a single bilaterally symmetric pair of neurons in the head of the worm , the ASI chemosensory neurons , but it is also expressed in the ADE and OLQ neurons under normal laboratory growth conditions ( Meisel et al . , 2014; Ren et al . , 1996; Schackwitz et al . , 1996 ) . We recently defined an altered expression pattern for daf-7 in hermaphrodites in response to exposure to bacterial metabolites produced by the pathogenic bacteria Pseudomonas aeruginosa ( Meisel et al . , 2014 ) . Within minutes of being exposed to P . aeruginosa , daf-7 expression is induced in an additional pair of chemosensory neurons , the ASJ neurons , which in turn promotes behavioral avoidance of the pathogenic bacterial lawn ( Meisel et al . , 2014 ) . Here , we show that in male C . elegans , daf-7 expression in the ASJ neuron pair is not only induced in response to environmental cues such as Pseudomonas aeruginosa , but can be switched ‘on’ or ‘off’ depending on multiple internal states such as sex , age , and experience . The differential effects of internal states and environmental experience on the DAF-7 transcriptional switch in the two ASJ sensory neurons reveal a hierarchy among different inputs that can function to modulate behavior . While expression of the DAF-7/TGF-β ligand was long thought to be confined to the ASI chemosensory neurons ( Ren et al . , 1996; Schackwitz et al . , 1996 ) , we recently reported that a change in the microbial environment of the worm—namely , the introduction of the pathogen P . aeruginosa—can alter the neuronal expression pattern of daf-7 , inducing expression in the ASJ neuron pair of hermaphrodite animals ( Meisel et al . , 2014 ) . In the current study , we observed that adult males exhibit daf-7 expression in both the ASI and the ASJ neuron pairs in the absence of P . aeruginosa ( Figure 1A and B ) . This expression pattern difference in adult C . elegans , can be observed both through the use of a pdaf-7::gfp transcriptional reporter ( Figure 1A and B ) as well as by fluorescence in situ hybridization ( FISH ) that directly probes for endogenous daf-7 mRNA ( Figure 1—figure supplement 1A–C ) . 10 . 7554/eLife . 21166 . 003Figure 1 . Sex-specific expression of daf-7 in the ASJ neurons of adult males . ( A ) pdaf-7::gfp expression pattern in hermaphrodites ( left ) and males ( right ) . Filled triangles indicate the ASI neurons; dashed triangles indicate the ASJ neurons . Scale bar indicates 50 µm . ( B ) Maximum fluorescence values of pdaf-7::gfp in the ASI ( left ) and ASJ ( right ) neurons of age-matched adult hermaphrodites ( grey ) and males ( black ) . *p<0 . 05 as determined by an unpaired t-test with Welch’s correction . Error bars indicate standard deviation ( SD ) . n . s . , not significant . ( C ) Maximum fluorescence values of pdaf-7::gfp in the ASJ neurons of males through larval development and early adulthood . The developmental timeline of C . elegans is shown to scale on top . Time values indicate hours after an egg is laid . Error bars indicate SD . ( D ) Maximum fluorescence values of pdaf-7::gfp in the ASJ neurons of both males and hermaphrodites on E . coli ( yellow ) or P . aeruginosa ( green ) . ***p<0 . 001 as determined by unpaired t-tests with Welch’s correction . Error bars indicate SD . DOI: http://dx . doi . org/10 . 7554/eLife . 21166 . 00310 . 7554/eLife . 21166 . 004Figure 1—figure supplement 1 . daf-7 expression in the ASJ neuron pair is specific to adult males . ( A–B ) FISH images of endogenous daf-7 mRNA in adult hermaphrodite ( A ) and adult male ( B ) . White dotted lines indicate outline of the ASJ neuron ( s ) . Both images are a single slice through the animal where ptrx-1::gfp was visible and in focus in one or both ASJ neurons . Other neurons visible anterior to the ASI neurons are likely the OLQ neurons , and the cells posterior to the ASJ neurons that express daf-7 mRNA are likely the ADE neurons . Scale bar represents 25 µm . Orientation provided in ( A ) applies to both images . ( C ) Fluorescence quantification for daf-7-Cy5 probe in the ASJ neurons ( as localized by ptrx-1::gfp ) in males and hermaphrodites . ***p<0 . 001 as determined by unpaired t-test with Welch’s correction . Error bars represent SD . ( D ) pdaf-7::gfp expression pattern in an L4 larval male . Filled triangles indicate the ASI neurons . Scale bar ( shown in C ) indicates 50 µm . ( E–F ) Maximum fluorescence values of pdaf-7::gfp in the ASI ( E ) and ASJ ( F ) neurons of developing males . **p<0 . 01 , ***p<0 . 001 , as determined by unpaired t-test with Welch’s correction . Error bars represent SD . n . s . , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 21166 . 004 Recent studies have suggested that , in addition to the generation of sex-specific neurons late in larval development , both male-specific and sex-shared neurons exhibit changes in gene expression as well as synaptic remodeling as the male undergoes the sexual maturation process ( Oren-Suissa et al . , 2016; Pereira et al . , 2015; Ryan et al . , 2014; Sammut et al . , 2015 ) . This predicts that a number of additional unidentified changes may occur in the anatomy , connectivity and gene expression profiles of the male nervous system as the animal transitions from larva to adult . Given this , we asked if daf-7 expression in males was correlated with the developmental stage of the animal . We observed that in populations of early larval animals carrying the pdaf-7::gfp transcriptional reporter , males and hermaphrodites were indistinguishable , with fluorescence observed only in the ASI neurons ( Figure 1—figure supplement 1D ) . To determine the onset of daf-7 expression in males , we imaged animals carrying the fluorescent reporter at multiple time points during development beginning in the L4 larval stage and extending into reproductive maturity . We found that daf-7 expression remains off in the ASJ of males through the end of the last ( L4 ) larval stage , but is switched on during young adulthood as the animals reach reproductive maturity ( Figure 1C ) . In addition , we examined expression of daf-7 in the ASI neuron pair at several matched timepoints during development and observed a significant drop in the expression of daf-7 in the ASI neurons from L4 to young adulthood ( Figure 1—figure supplements 1E and F , 52 hr and 64 . 5 hr timepoints ) . At subsequent timepoints during adulthood , however , expression levels in the ASI neurons remained constant as expression in the ASJ neurons increased ( Figure 1—figure supplements 1E , 3F 64 . 5 hr and 71 . 5 hr timepoints ) . We asked what the role of the genetic sex of the animal is in regulating daf-7 expression in the ASJ neurons . The C . elegans sex determination pathway terminates on the master regulator GLI-family transcription factor , TRA-1 ( Zarkower and Hodgkin , 1992; Zarkower , 2006 ) . TRA-1 activity functions to promote hermaphrodite cell fates and inhibit male development . We first looked at tra-1 ( e1099 ) mutants which are karyotypically XX , but develop as fertile pseudomales due to the lack of active TRA-1 ( Hodgkin and Brenner , 1977 ) . In these mutant XX males , daf-7 expression could be seen in both the ASI and ASJ neurons , suggesting that daf-7 is regulated by the genetic sex determination pathway either directly or indirectly ( Figure 2—figure supplement 1A ) . We next looked at the role of genetic sex specifically in the nervous system using transgenic sexual mosaics . TRA-1 is negatively regulated by the FEM proteins—FEM-1 , FEM-2 , and FEM-3—and it has been demonstrated by many groups that overexpression of FEM-3 cDNA is sufficient to lead to ‘masculinization’ in otherwise hermaphrodite animals ( Mehra et al . , 1999; Oren-Suissa et al . , 2016; Pereira et al . , 2015; Sammut et al . , 2015; White et al . , 2007 ) . We expressed the FEM-3 cDNA under the control of the pan-neural promoter , rab-3 , and found that daf-7 expression was up-regulated in the ASJ neurons of adult hermaphrodites , suggesting that a genetically ‘male’ nervous system is sufficient for daf-7 expression in ASJ ( Figure 2A , C and E ) . Although we observed an increase in daf-7 expression in these genetically masculinized animals , the overall expression of daf-7 in the ASJ of masculinized hermaphrodites was still notably lower than what we observe in wild-type males ( e . g . Figure 1B ) . This could possibly be explained by the lack of male-specific neurons in these transgenic animals . While genetic masculinization is effective in sexualizing neurons of the sex-shared nervous system , this technique fails to reliably retain the male-specific CEM neurons and produces hermaphrodites which also lack other male-specific neurons found in the tail ( White et al . , 2007 ) . To assess the importance of these male-specific neurons for daf-7 expression in ASJ , we examined pdaf-7::gfp fluorescence in ceh-30 ( n4289 ) , mab-3 ( e1240 ) , and pkd-2 ( sy606 ) ;lov-1 ( sy582 ) mutant males . While ceh-30 ( n4289 ) males lack the male-specific CEM neurons in the head ( Schwartz and Horvitz , 2007 ) , mab-3 ( e1240 ) mutants display marked defects in male-tail formation including ray differentiation as well as in male specific circuitry in the head of the animal ( Yi et al . , 2000 ) . Similarly , PKD-2 and LOV-1 constitute a TRP channel necessary for the function of the B-type ray neurons—another set of male-specific neurons in the tail of the animal ( Barr et al . , 2001; Barr and Sternberg , 1999; Barrios et al . , 2008 ) . Males of all mutant strains retained wild-type expression levels of daf-7 in ASJ , suggesting that the male-specific neural circuitry is not required for daf-7 expression and is unlikely to underlie the lower daf-7 expression we see in genetically masculinized hermaphrodites ( Figure 2—figure supplement 1B ) . 10 . 7554/eLife . 21166 . 005Figure 2 . Genetic sex regulates male-specific expression of daf-7 in the ASJ neuron pair . ( A–D ) Representative images of pdaf-7::gfp expression in: wild-type ( WT ) hermaphrodites ( A ) and males ( B ) , nervous-system-masculinized hermaphrodites ( C ) and nervous-system-feminized males ( D ) . Closed triangles indicate the ASI neurons; dashed triangles indicate the ASJ neurons . Scale bar indicates 50 µm . ( E ) Maximum fluorescence values of pdaf-7::gfp in the ASJ neurons of control ( black ) and partially masculinized hermaphrodites ( purple ) . Masculinization was effected by driving expression of fem-3 cDNA under pan-neural ( left ) and ASJ-specific ( right ) promoters . ***p<0 . 001 as determined by unpaired t-test with Welch’s correction . n . s . , not significant . Error bars represent SD . ( F ) Maximum fluorescence values of pdaf-7::gfp in ASJ of control ( black ) and partially feminized males ( orange ) . Feminization was effected by driving expression of tra-2IC under pan-neural ( left ) , ciliated neuron ( middle ) , and ASJ-specific ( right ) promoters . ***p<0 . 001 as determined by unpaired t-test with Welch’s correction . Error bars indicate SD . DOI: http://dx . doi . org/10 . 7554/eLife . 21166 . 00510 . 7554/eLife . 21166 . 006Figure 2—figure supplement 1 . daf-7 expression in the ASJ neuron pair is regulated by TRA-1 but not by male-specific neurons . ( A ) Representative image of pdaf-7::gfp expression pattern in tra-1 ( e1099 ) pseudomale . Filled triangles indicate the ASI neurons; dashed triangles indicate the ASJ neurons . Scale bar represents 50 µm . ( B ) Maximum fluorescence values of pdaf-7::gfp in the ASJ neurons of WT , ceh-30 ( n4289 ) , mab-3 ( e1240 ) , and pkd-2 ( sy606 ) ;lov-1 ( sy582 ) mutant animals . Significance determined by ordinary one-way ANOVA followed by Dunnett’s multiple comparisons test . Error bars represent SD . n . s . , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 21166 . 006 Complementary to masculinization , genetic ‘feminization’ of XO male animals can be achieved via the overexpression of the intracellular domain of the TRA-2 protein , which negatively interacts with FEM-3 to prevent TRA-1 degradation ( Mehra et al . , 1999 ) . Overexpression of TRA-2IC under the control of the rab-3 promoter led to complete loss of daf-7 expression in the ASJ of otherwise male animals ( Figure 2B , D and F ) . These results suggest that a genetically male nervous system is both necessary and sufficient for the sex-specific expression pattern of daf-7 . We next asked whether the sex determination pathway acts cell-autonomously in ASJ to regulate daf-7 expression . Using the trx-1 ASJ-specific promoter , we masculinized the ASJ neuron pair alone and saw no daf-7 expression in the resulting adult hermaphrodites ( Figure 2E ) . Similarly , no daf-7 expression could be observed when only the ciliated neurons ( of which ASJ is a member ) were masculinized ( data not shown ) . These data suggest the requirement that an additional neuron ( or neurons ) or other tissues have a male cell identity for daf-7 to be expressed in the ASJ neurons , although differences in the strength of the heterologous promoters used in the masculinization constructs may also account for these observations . In contrast , both ciliated neuron-specific and ASJ-specific feminization in males led to down-regulation of daf-7 expression , confirming that a male cell identity in ASJ is necessary , though not sufficient , for proper daf-7 expression ( Figure 2F ) . We next sought to determine the functional significance of the sex-specific changes in daf-7 expression that we observe in males . Mate-searching behavior is a behavioral program specific to adult male C . elegans , which features the prioritization of mating over feeding in these animals . While hermaphrodite animals prefer to remain in a patch of bacterial food indefinitely and exhibit limited exploratory behavior , males will wander away from a food source at a predictable and constant rate ( Lipton et al . , 2004 ) . This can be examined in an assay in which the movement of the animal is tracked over time to determine the leaving probability of any given genotype ( Figure 3A; Lipton et al . , 2004 ) . We observed that mutants in both the DAF-7/TGF-β ligand and its Type I receptor , DAF-1 , showed defects in the mate-searching assay ( Figure 3A and B , and Figure 3—figure supplement 1 ) . This behavioral defect in the daf-7 mutant can be fully suppressed by mutation of the downstream antagonistic co-SMAD , DAF-3 , suggesting that DAF-7 functions through its canonical TGF-β signaling pathway to regulate mate-searching behavior ( Figure 3B , and Figure 3—figure supplement 1B ) . Given the change in expression of daf-7 in the ASJ neurons of males at the transition to adulthood when this behavior is first displayed , we used genetic ablations of the ASJ neuron pair to ask what their contribution to this behavior might be . We observed that the ASJ-ablated animals were defective in mate-searching behavior , establishing a role for the ASJ neuron pair in promoting this behavior ( Figure 3C and Figure 3—figure supplement 1A ) . Importantly , the defects that we see in daf-7 mutant and ASJ-ablated male behavior are not accounted for by global defects in locomotion . When mutant animals are tested on leaving assay plates on which the bacterial food has been fully spread , these males reach the scoring distance at equivalent rates to WT controls ( Figure 3—figure supplement 2; Barrios et al . , 2012 ) . This suggests that the reduced leaving behavior we observe in these animals isspecific to the decision to leave the food source in search of a mate . 10 . 7554/eLife . 21166 . 007Figure 3 . DAF-7/TGF-β is required for male mate-searching behavior . ( A ) Schematic of mate-searching assay ( top ) . Animals are placed individually into the center of a lawn of bacteria . The tracks of the animal are followed over time and scored for movement beyond 3 cm away from the food source . A representative data curve is shown on the bottom depicting hermaphrodites ( solid black ) , males ( dashed black ) , and daf-7 ( e1372 ) mutant males ( blue ) . ( B ) Probability of leaving values for WT , daf-7 mutant , and daf-7;daf-3 double mutant males . Values plotted are the average + SEM for two independent experiments , n = 40 animals for all strains except daf-7 ( ok3125 ) where n = 29 . *p<0 . 05 , **p<0 . 01 as determined by ordinary one-way ANOVA followed by Dunnett’s multiple comparisons test . n . s . , not significant . ( C ) Probability of leaving values for two independent lines of male animals with genetic ablation of the ASJ neurons , compared with corresponding values for WT and daf-7 mutant males . Values plotted are the average + SEM for two independent experiments for daf-7 mutant animals and three independent experiments for WT control and ASJ ablation strains . n = 60 animals for all strains except daf-7 ( ok3125 ) where n = 29 animals . **p<0 . 01 as determined by ordinary one-way ANOVA followed by Dunnett’s multiple comparisons test . One replicate was performed with the daf-7;daf-3 double mutant strains in 1B and controls ( WT and daf-7 ) are identical , so probability values for that experiment were used in the averages shown in both B and C of this figure . ( D ) Probability of leaving values for daf-7 ( ok3125 ) mutant animals carrying transgenes expressing daf-7 cDNA specifically under the control of the indicated promoters . Values on graph are the average + SEM of two independent experiments with an n = 40 total animals for each strain . **p<0 . 01 as determined by ordinary one-way ANOVA followed by Dunnett’s multiple comparisons test . n . s . , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 21166 . 00710 . 7554/eLife . 21166 . 008Figure 3—figure supplement 1 . daf-7 regulates mate-searching behavior in males . ( A ) Representative mate-searching data for strains with genetic ablation of the ASJ neurons ( shown in shades of orange ) . WT hermaphrodites ( solid black ) , WT males ( dashed black ) , and daf-7 mutant ( blue ) animals are included in each assay as controls . ( B ) Representative data for daf-7;daf-3 double mutant ( green ) in the mate-searching assay . ( C ) Representative data for daf-7 rescue strains ( shown in shades of red ) . ( D ) Representative mate-searching curves for daf-1 mutants ( purple ) . ( E ) Probability of leaving values for WT , daf-7 , and daf-1 mutant animals . Values shown are the average + SEM of two independent experiments . n = 40 for WT and daf-7 mutants , n = 32 for daf-1 ( e1287 ) and n = 36 for daf-1 ( m402 ) . *p<0 . 05 as determined by ordinary one-way ANOVA followed by Dunnett’s multiple comparisons test . DOI: http://dx . doi . org/10 . 7554/eLife . 21166 . 00810 . 7554/eLife . 21166 . 009Figure 3—figure supplement 2 . daf-7 mutant animals and strains carrying genetic ablation of the ASJ neurons exhibit no locomotion defect . ( A ) Schematic of locomotion assay . E . coli food ( shown in yellow ) was spread to the edges of the plate and a standard mate searching assay performed to determine animal movement on a full lawn of bacteria . ( B–C ) Raw data curves for daf-7 ( blue ) and ASJ-ablation ( orange ) males in locomotion assay . WT males shown in black for reference . DOI: http://dx . doi . org/10 . 7554/eLife . 21166 . 009 Since DAF-7 is a secreted ligand , we anticipated that overexpression of DAF-7 from either the ASI or ASJ neurons would be sufficient to rescue daf-7-associated mutant phenotypes . As expected , we saw that transgenic rescue from the endogenous promoter as well as from ASI- and ASJ-specific promoters were sufficient to restore mate-searching behavior in the daf-7 ( ok3125 ) mutant background ( Figure 3D and Figure 3—figure supplement 1C ) . Taken together , these data imply that the dynamic expression of daf-7 in the ASJ neurons of adult males regulates mate-searching behavior . Mate-searching behavior has been demonstrated to be dependent on the feeding state of the animal—starved male animals will re-prioritize feeding over mate searching for a limited period of time until the animal has had sufficient opportunity to re-feed ( Figure 4C; Lipton et al . , 2004 ) . We asked whether daf-7 expression in the male ASJ neurons might be dependent on the nutritional state of the animal or the bacterial availability in its environment . We found that males subjected to starvation beginning early in the L4 larval stage failed to express daf-7 in the ASJ neurons as adults ( Figure 4A , conditions i and ii , and Figure 4B ) . This lack of daf-7 in the ASJ neuron pair was apparent both by GFP visualization as well as by FISH , with daf-7 mRNA entirely absent from the ASJ neurons ( Figure 4A and B ) . We noticed that daf-7 mRNA was still apparent in the ASI , OLQ , and ADE neurons of these starved animals , suggesting that the loss of daf-7 expression in the ASJ neurons is specific and not representative of a global down-regulation of transcription in these animals ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 21166 . 010Figure 4 . Nutritional state regulates daf-7 expression in the ASJ neurons of males . ( A ) Schematic of experiment design indicating timing and duration of starvation periods ( left ) . FISH images of endogenous daf-7 mRNA in the ASJ neurons for each experimental condition ( right ) . White dotted lines indicate outline of ASJ cell body as localized by ptrx-1::gfp . Scale bar represents 5 µm . ( B ) Maximum fluorescence values of pdaf-7::gfp in the ASJ neurons during re-feeding experiment . Re-fed animals were starved for a period of 24 hr before being reintroduced to E . coli for the indicated amount of time . ***p<0 . 001 as determined by ordinary one-way ANOVA followed by Dunnett’s multiple comparisons test . Error bars indicate SD . n . s . , not significant . ( C ) Representative curves of mate-searching data in fed and post-starvation WT males . Animals were starved for a period of 24 hr as in previous experiments before being assayed for mate-searching behavior . DOI: http://dx . doi . org/10 . 7554/eLife . 21166 . 01010 . 7554/eLife . 21166 . 011Figure 4—figure supplement 1 . daf-7 expression is down-regulated by starvation specifically in the ASJ neurons . ( A ) Representative FISH image of endogenous daf-7 mRNA in a starved adult male . White dotted line indicates outline of an ASJ neuron . Closed triangle indicates an ASI neuron . Other neurons visible anterior to the ASI neuron are likely the OLQ neurons , and posterior to the ASJ neuron are likely the ADE neurons . Scale bar represents 25 µm . ( B ) Fluorescence quantification for daf-7-Cy5 probe in the ASJ neurons ( as localized by ptrx-1::gfp ) of males fed and starved at various different points in their lives . ***p<0 . 001 as determined by ordinary one-way ANOVA followed by Dunnett’s multiple comparisons test . Error bars represent SD . n . s . , not significant . ( C ) Fluorescence quantification for daf-7-Cy5 probe in the ASI neurons of starved and fed males . Significance determined by unpaired t-test with Welch’s correction . Error bars represent SD . n . s . , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 21166 . 011 We next asked if daf-7 expression could be dynamically modulated by removal and re-introduction of food at various points during the life of the male . We first looked at daf-7 expression in animals that had experienced starvation during the L4-to-adulthood period before being returned to an E . coli food source as adults . In these animals , daf-7 expression could be detected both by the GFP reporter as well as by FISH ( Figure 4A , condition iii , and 4B ) . We asked whether fully mature adult animals would turn off daf-7 expression in response to starvation after having already induced expression of daf-7 in the ASJ neuron pair . For these experiments , we utilized FISH of the endogenous daf-7 mRNA because perdurance of the GFP reporter might complicate analysis . When males were allowed to develop into adulthood unperturbed and then subjected to a period of food deprivation , we observed a complete loss of daf-7 mRNA in the ASJ neurons ( Figure 4A , condition iv ) . Together this data demonstrates that daf-7 expression in the ASJ neurons is subject to dynamic regulation by periods of starvation and re-feeding . These data suggest two distinct possibilities for how bacterial food levels in the environment modulate daf-7 transcriptional regulation in the ASJ neuron pair of males . First , daf-7 transcription could be regulated by the internal nutritional state of the animal ( starved or fed ) . Alternatively , daf-7 transcription could be activated by the chemosensation ( smell or taste ) of bacterial food . To distinguish between these two possibilities , we measured the kinetics of daf-7 expression in response to re-feeding in starved animals . We observed that daf-7 expression in the ASJ neuron pair is not significantly increased compared to starved controls until 9 hr after the re-introduction of an E . coli food source ( Figure 4B ) . Although some animals could be observed expressing daf-7 weakly in the ASJ neuron pair at 4 hr of re-feeding , the majority of animals still exhibited no pdaf-7::gfp expression in the ASJ neuron pair until later time points . The extended period of time required to restore daf-7 expression in the ASJ neuron pair suggests that internal nutritional state , not the chemosensation of bacterial food , exerts control over daf-7 transcription in the ASJ neuron pair . Moreover , we observed that the onset of daf-7 expression in the ASJ neurons of males following the re-introduction of food are consistent with the previously noted kinetics of resumption of mate-searching behavior after males have been starved ( Figure 4C; Lipton et al . , 2004 ) . We recently showed that the microbial environment is an important regulator of daf-7 expression in the ASJ neurons in hermaphrodite animals ( Figure 1D; Meisel et al . , 2014 ) . Given that males induce expression of daf-7 in the same ASJ neuron pair independently of the bacterial food source , we sought to determine if they would be capable of responding to exposure to P . aeruginosa in the same way as their hermaphrodite siblings—namely , through additional up-regulation of daf-7 expression in the ASJ neurons . We observe that , when transferred to P . aeruginosa , males carrying the pdaf-7::gfp transcriptional reporter exhibited an increase in daf-7 expression in the ASJ neurons compared to E . coli-fed controls ( Figure 1D ) . The additive nature of DAF-7 levels in males suggests that the ASJ neuron pair of males retain the ability to respond to P . aeruginosa that we observe in hermaphrodites . Our data suggest that daf-7 expression in the ASJ neurons is subject to regulation by both sensory cues from the microbial environment as well as by internal state sensing mechanisms . Because daf-7 expression in the ASJ neuron pair and mate-searching behavior in males are dependent on nutritional state during feeding on the non-pathogenic bacterial food , E . coli , we sought to determine if expression of daf-7 in response to P . aeruginosa might also be dependent on the nutritional state in males . We tested this by challenging starved animals with E . coli or P . aeruginosa to see how daf-7 expression would be affected and to observe any behavioral differences . Looking at animals carrying the pdaf-7::gfp reporter during post-starvation re-feeding , we observed a rapid up-regulation of daf-7 expression in the ASJ neurons of starved males returned to P . aeruginosa as a food source ( Figure 5A and B , and Figure 5—figure supplement 1E ) . This increase in daf-7 expression in the ASJ neuron pair could be seen as early as 2 hr after first exposure to the pathogenic bacteria ( Figure 5—figure supplement 1E ) . In contrast , starved males that were returned to the non-pathogenic E . coli showed little to no daf-7 expression in the ASJ neurons until much later time points ( Figure 5A , B and 4B ) . The kinetics of the increase in fluorescence of the pdaf-7::gfp reporter in response to P . aeruginosa secondary metabolites , which lags behind the immediate response observed by FISH is consistent with what we observed previously for hermaphrodite animals ( Meisel et al . , 2014 ) , and notably faster than the slower increase in pdaf-7::gfp reporter fluorescence in response to repletion of the non-pathogenic E . coli bacterial food . 10 . 7554/eLife . 21166 . 012Figure 5 . Prioritization of multiple inputs through the regulation of daf-7 expression in the ASJ neurons of males . ( A ) Schematic of refeeding experimental design indicating timing and duration of each starving and feeding period ( left ) . On right , representative images of daf-7 expression following 4 hr of post-starvation refeeding on E . coli ( left ) or P . aeruginosa ( right ) . Closed triangles indicate the ASI neurons; dashed triangles indicate the ASJ neurons . Scale bar represents 50 µm . ( B ) Maximum fluorescence values of pdaf-7::gfp in the ASJ neurons of fed and starved animals and animals that were starved for a period of 24 hr and then reintroduced either to the normal E . coli food source or to pathogenic P . aeruginosa . Images were taken for quantification after 4 hr of re-feeding . ***p<0 . 001 as determined by ordinary one-way ANOVA followed by Dunnett’s multiple comparisons test . Error bars indicate SD . n . s . , not significant . ( C ) Representative curve of food-leaving data in starved animals on E . coli ( yellow ) and P . aeruginosa ( green ) . Animals were starved as in previous experiments before being run in the assay . DOI: http://dx . doi . org/10 . 7554/eLife . 21166 . 01210 . 7554/eLife . 21166 . 013Figure 5—figure supplement 1 . P . aeruginosa influences exploratory behavior and daf-7 expression in both males and hermaphrodites . ( A ) Representative leaving curves for males ( dashed ) and hermaphrodites ( solid ) on E . coli ( shades of grey ) and P . aeruginosa ( shades of green ) . All strains were fed in this experiment . ( B ) Probability of leaving values for males and hermaphrodites on E . coli and P . aeruginosa . All strains were fed in these experiments . Values plotted are the mean + SEM for two independent experiments , n = 40 animals for all conditions . *p<0 . 05 as determined by unpaired t-test with Welch’s correction . n . s . , not significant . ( C ) Representative raw data curves for starved hermaphrodites on E . coli and P . aeruginosa . Fed hermaphrodites shown as control . ( D ) Maximum fluorescence of pdaf-7::gfp in the ASJ neurons of starved hermaphrodites re-fed E . coli or P . aeruginosa for four hours ( compare to Figure 5B ) . Error bars indicate SD . ( E ) Maximum fluorescence values of pdaf-7::gfp in the ASJ neurons of fed and starved animals and animals that were starved for a period of 24 hr and then reintroduced to P . aeruginosa for the indicated amount of time . The experiment in Figure 4B was run in parallel , controls ( fed and starved ) were shared between conditions . These controls are shown in both figures for comparison purposes . *p<0 . 05 and ***p<0 . 001 , as determined by ordinary one-way ANOVA followed by Dunnett’s multiple comparisons test . Error bars indicate S . D . DOI: http://dx . doi . org/10 . 7554/eLife . 21166 . 013 We followed up by asking if this difference in expression after starvation is also reflected in behavioral differences of post-starvation animals experiencing different microbial environments . Normally , following a period of starvation , a male prefers feeding over leaving a lawn of bacteria in search of a mate ( Figure 4C; Lipton et al . , 2004 ) , but we reasoned that these priorities might be re-set if the only available food is pathogenic to the starved animal . Indeed , whereas starved males will remain on the E . coli lawn for several hours , starved males who are instead returned to a lawn of P . aeruginosa will readily leave the bacterial lawn at a rate similar to non-starved animals on the normal E . coli food source ( Figures 4C and 5C ) . This lawn-leaving behavior is not male-specific , as both fed and starved hermaphrodites will also readily leave a lawn of P . aeruginosa and starved hermaphrodites are also capable of responding to P . aeruginosa through an upregulation of daf-7 in the ASJ neurons ( Figure 5—figure supplement 1A–1D ) . In this regard , one might consider that lawn-leaving under these conditions is driven by the presence of pathogenic P . aeruginosa , and not mate-searching . Our data suggest that daf-7 expression in the ASJ neurons is one way through which animals are both able to interpret internal and external cues and prioritize their behaviors accordingly . We have presented a set of experiments that suggest that behavioral prioritization and decision-making may be effected through the transcriptional control of a key neuroendocrine regulator , the TGF-β ligand , DAF-7 , in a pair of chemosensory neurons , the ASJ neurons . Our data suggest that daf-7 expression in the ASJ neuron pair is utilized to promote exploratory behaviors in different physiological and ecological contexts . In hermaphrodites , expression of daf-7 is induced by the detection of secondary metabolites and functions to allow animals to discriminate between microbial species and promote avoidance of pathogenic bacteria ( Meisel et al . , 2014 ) . In essence , pathogen avoidance represents the behavioral choice an animal makes when confronted with two competing needs—the need to eat and remain well fed and the need to avoid danger and infection . Mate searching presents a paralagous but sex-specific behavior , where the decision instead is between food and mating . Our data suggest that modulation of daf-7 expression provides a striking example of how a neuroendocrine signaling pathway may be coopted to guide behavioral decisions in entirely different contexts . Our study suggests that internal states and environmental cues that converge on the transcriptional control of DAF-7 in the ASJ neuron pair are processed in a hierarchical manner that , in turn , serves to modulate the exploratory mate-searching behavior in the male ( Figure 6 ) . The work presented here suggests that the transcriptional regulation of daf-7 in the ASJ neurons functions as an ON/OFF switch , where each input holds the switch—and thus , daf-7 expression—in one of two states . However , the various regulatory inputs into this switch exert differential control , with environmental cues being prioritized over all others . This differential control of the transcriptional switch , in turn allows for re-adjustments of behavioral priorities in the worm based on its current and past experience . 10 . 7554/eLife . 21166 . 014Figure 6 . A hierarchy of inputs regulates daf-7 expression in the ASJ neurons to modulate exploratory behaviors . We show that the information from many different sources both internal and external is integrated hierarchically in the regulation of daf-7 expression in the ASJ neuron pair . This hierarchical regulation of daf-7 in turn leads to a distinct prioritization of exploratory behaviors in the male worm . DOI: http://dx . doi . org/10 . 7554/eLife . 21166 . 014 Interestingly , dynamic transcriptional regulation has previously been implicated in the regulation of sex-specific decision-making behavior in C . elegans . Expression of the chemoreceptor , ODR-10 , in the sex-shared sensory AWA neurons has also been shown to be dependent on both internal and external cues to regulate food-related decision making in male C . elegans ( Ryan et al . , 2014 ) . Given the similarities in the regulation of these two genes by both sex and satiety , it is possible that DAF-7 and ODR-10 function in a single pathway to regulate decision-making behaviors in the male . Previous studies of daf-7 have suggested that it is involved in regulating the expression of numerous chemoreceptors in many different neuron types ( Nolan et al . , 2002 ) . While no interaction has been identified between daf-7 and odr-10 in the hermaphrodite nervous system , there has been no examination as yet of the effects of daf-7 mutation on odr-10 in the relevant cell types , sex , or under different food conditions , which may be an interesting line of future research . DAF-7/TGF-β has previously been found to be involved in the regulation of other satiety-regulated behaviors such as quiescence and foraging ( Gallagher et al . , 2013; Milward et al . , 2011; You et al . , 2008 ) . daf-7 mutants show marked defects in the ability to respond to fasting through an increase in quiescence , suggesting that DAF-7 may serve as a critical signal of food availability in the worm ( You et al . , 2008 ) . Similarly , DAF-7 signaling has also been implicated in promoting food-leaving ( or foraging ) in the context of a rapidly depleting food source , further underscoring the importance of this pathway for communicating nutritional availability ( Milward et al . , 2011 ) . While DAF-7 from the ASI neurons , and not the ASJ neurons , contributes to these other possibly related phenotypes , similar mechanisms may be involved in regulating daf-7 expression and release in these three distinct food/satiety-dependent assay contexts . Taken together with our data , the important role of daf-7 in these different behavioral paradigms suggests that the daf-7 pathway may act as a more global food signal that is critical for allowing the animal to sense food and satiety and modulate its behaviors accordingly . Finally , our results support a model where integration of internal and external sensory information can occur at the level of the sensory neuron itself to regulate behavioral prioritization . Our data on the role of the ASJ sensory neurons in male mate-searching behavior are consistent with emerging evidence from studies in several species to suggest that the regulatory control of sensory neurons contributes to information processing and decision making behavior ( Dey et al . , 2015; Lebreton et al . , 2015; Peckol et al . , 2001; Ryan et al . , 2014 ) . C . elegans were grown on E . coli OP50 as a food source as previously described ( Brenner , 1974 ) . Strains were grown at 20°C except for assays containing dauer-constitutive mutants , in which case , all strains were grown at 16°C until they passed the dauer developmental decision . Unless otherwise indicated , all strains carry the him-5 ( e1490 ) mutation to increase the number of spontaneous males available . For a complete list of strains used in this study , please see Supplementary file 1 . For quantification of daf-7 expression in mutants and developing animals , animals were egg laid for 2 hr on OP50 to obtain synchronized populations of animals . The animals were then mounted and anesthetized in 50 mM sodium azide at the indicated hour of development or at 72 hr if not indicated . The animals were imaged at 40x using a Zeiss Axioimager Z1 microscope . 15–20 animals were imaged for each condition or strain . For quantification , maximum intensity values of GFP within the ASJ neurons was calculated using FIJI software ( Schindelin et al . , 2012 ) ; Fiji , RRID:SCR_002285 ) . For P . aeruginosa experiments , the P . aeruginosa PA14 strain was grown overnight at 37°C in 3 mL of LB as previously described and the next day 7 µL of culture was seeded onto SKA plates ( Tan et al . , 1999 ) . Plates were incubated at 37°C overnight and then allowed to grow for an additional two days at room temperature before being used for experiments . Animals for P . aeruginosa experiments were synchronized by treatment with bleach and then allowed to hatch and arrest as L1 larvae in M9 media overnight . This provided synchronized populations free from any possible bacterial contamination . Animals were transferred to PA14 plates or OP50 controls at the L4 larval stage . Imaging and quantification were performed as described above at 16 hr after transfer to PA14 plates . To generate the neural feminization and masculinization transgenes , the pGH8 plasmid containing the pan-neural rab-3 promoter along with mCherry::unc-54 3’UTR was linearized by PCR . The fem-3 and intracellular tra-2 cDNAs were amplified from cDNA generated with an Ambion RETROScript kit . The unc-54 3’UTR was amplified from the pPD95 . 75 vector and the SL2 trans-splicing sequence leader ( Huang et al . , 2001; Spieth et al . , 1993 ) was amplified from genomic DNA . The fem-3 or tra-2 cDNA along with the unc-54 3’UTR and SL2 sequence were cloned into the linearized pGH8 vector by Gibson assembly ( Gibson et al . , 2009 ) . For cell specific manipulations , the ciliated neuron bbs-1 promoter ( 1 . 9 kb ) and ASJ specific trx-1 promoter ( 1 . 1 kb; Fierro-González et al . , 2011 ) were amplified from genomic DNA and inserted in the place of the rab-3 promoter by Gibson assembly . All plasmids were sequenced to confirm identity and injected into ksIs2;him-5 ( e1490 ) animals at a concentration of 50 ng/µL , along with a plasmid carrying ofm-1::gfp ( 50 ng/µL ) . At least three independent transgenic lines were obtained and analyzed for each construct and one representative line is shown . For a complete list of primers used in this paper , please see Supplementary file 2 . Mate-searching behavior was assessed in animals through a food leaving assay as previously described ( See diagram in Figure 3A; Lipton et al . , 2004 ) . Briefly , animals were placed individually onto a 1 cm lawn ( ~20 µL of culture ) of E . coli OP50 in the center of a 10 cm plate . Movement of each animal was tracked at hourly time points . When an animal’s tracks extended a distance of 3 cm away from the starting lawn , the animal was scored as a ‘leaver’ . It has been previously demonstrated that male leaving behavior can be modeled by the single exponential decay function N ( t ) /N ( 0 ) =exp ( -λt ) , where N ( t ) refers to the number of non-leavers at a given time and the hazard value ( λ ) gives an estimate for the probability of leaving , which remains constant over the course of an assay ( Lipton et al . , 2004 ) . Hazard values were calculated by fitting data using the R survival package with an exponential parametric survival model using maximum likelihood ( R Project for Statistical Computing , RRID:SCR_001905 ) . These values were pooled across replicates . At least two independent experiments were performed for each genotype ( ~40 animals per genotype ) . For leaving assays with P . aeruginosa , plates were supplemented with additional peptone and were seeded with 15 µL of bacterial culture . In addition , all plates for PA14 assays were incubated overnight at 37°C after being seeded with bacteria and then for an additional 2 days at room temperature before being used for the assay . For locomotion assays , 10 cm mate-searching plates were seeded with ~400 µL of E . coli OP50 culture spread to the edge of the plate . Plates were allowed to dry and remained at room temperature for ~16 hr before being used in the assay . Males were placed individually on these plates and their movement tracked over time as in the standard mate-searching assay . The same scoring distance was used for locomotion assays as for the mate-searching assay . FISH was performed as described previously ( Raj et al . , 2008 ) . Animals were fixed in 4% formaldehyde for 45 min at 20°C . Following washing in PBS , animals were resuspended in 70% ethanol and incubated overnight at 4°C . Samples were then hybridized with probes to the endogenous daf-7 mRNA as previously published ( Meisel et al . , 2014 ) . Hybridization was carried out at 30°C overnight . Imaging was performed using a Nikon Eclipse Ti Inverted microscope with a Princeton Instruments PIXIS 1024 camera . All data were analyzed in FIJI . For quantifications of FISH data , fluorescence values were calculated within the ASJ ( or ASI ) neurons for the indicated conditions . All samples were hybridized using daf-7-Cy5 probes and imaged at an exposure time of 1500 ms . For all FISH experiments , animals carrying the ofEx4[ptrx-1::gfp] transgene were used in order to localize the ASJ neurons . For all starvation experiments , animals were synchronized by treatment with bleach and then were allowed to hatch and arrest as L1s in M9 media for 14 hr . After 14 hr , animals were dropped onto plates containing abundant OP50 E . coli and were allowed to develop for 34–36 hr at 20°C . At this time point , animals had already entered the L4 larval stage . All animals were washed from the plates and subjected to an additional 4–5 washes with M9 media to remove any remaining bacteria . Animals were placed either back onto plates seeded with E . coli or were dropped onto plates lacking peptone to inhibit any bacterial growth . For re-feeding experiments , animals were starved on peptone free plates for 24 hr at 20°C . After 24 hr of starvation , nearly all of the male animals had molted into adults as evident by tail and gonadal morphology . Only animals with adult morphology were used for experiments and imaging . They were subsequently collected and placed onto plates seeded with either E . coli or P . aeruginosa for the designated amount of time . For the FISH experiment shown in Figure 3 , animals were re-fed on E . coli OP50 for 16 hr at 20°C before being fixed for imaging . For the late starve experiments , animals dropped back onto E . coli after the washing step were allowed to grow for an additional 24 hr at 20°C before being collected and washed 4–5 times in M9 media . These animals were then placed onto peptone-free plates and starved for 16 hr at 20°C before being fixed for imaging . All statistical analysis was performed using the GraphPad Prism software ( Graphpad Prism , RRID:SCR_002798 ) . Statistical tests used are indicated in each figure legend .

For almost all species of animal , males and females will often behave differently in similar situations . Little is known about how these sex-specific differences are generated or , for example , how different the nervous system of a male is to that of a female . Moreover , it is also poorly understood how these underlying differences based on the biological sex of an animal are integrated with and influenced by its experiences and environment . The roundworm Caenorhabditis elegans has two sexes , hermaphrodites and males . The male worms behave differently to the hermaphrodites in a number of situations . This means that these animals offer the opportunity to explore and understand sex-specific differences in behavior . It is also possible to analyze the underlying factors that contribute to behavior in C . elegans , because it has a relatively simple and well-defined nervous system . Now , Hilbert and Kim show that a signal that influences how C . elegans explores in response to chemicals in its environment is expressed differently in male and hermaphrodite worms . The signal in question is molecule called DAF-7 , which is released by several sensory neurons—nerve cells that are used for detecting cues from the environment . The sensory neurons that release DAF-7 are found in both sexes of C . elegans but the specific way that the male worms express this signal encourages them to search for mates . Hermaphrodites , on the other hand , do not need to search for mates because they can fertilize their own eggs . Hilbert and Kim showed that the biological sex in combination with multiple other inputs – including the animal’s past diet and age – regulate how the DAF-7 signal is expressed in C . elegans . These inputs all converge onto a single pair of sensory neurons , which integrate the inputs and enable the worm to assess its current and past experiences and alter its behavior accordingly . Moving forward the next challenge is to understand how information about both external environment and internal states , such as hunger , are communicated to and integrated by these sensory neurons . Decoding the signals behind this process may illuminate how biological sex and internal states influence behavior in other species of animals .

2017

Sexually dimorphic control of gene expression in sensory neurons regulates decision-making behavior in C. elegans

SHARPIN regulates immune signaling and contributes to full transcriptional activity and prevention of cell death in response to TNF in vitro . The inactivating mouse Sharpin cpdm mutation causes TNF-dependent multi-organ inflammation , characterized by dermatitis , liver inflammation , splenomegaly , and loss of Peyer's patches . TNF-dependent cell death has been proposed to cause the inflammatory phenotype and consistent with this we show Tnfr1 , but not Tnfr2 , deficiency suppresses the phenotype ( and it does so more efficiently than Il1r1 loss ) . TNFR1-induced apoptosis can proceed through caspase-8 and BID , but reduction in or loss of these players generally did not suppress inflammation , although Casp8 heterozygosity significantly delayed dermatitis . Ripk3 or Mlkl deficiency partially ameliorated the multi-organ phenotype , and combined Ripk3 deletion and Casp8 heterozygosity almost completely suppressed it , even restoring Peyer's patches . Unexpectedly , Sharpin , Ripk3 and Casp8 triple deficiency caused perinatal lethality . These results provide unexpected insights into the developmental importance of SHARPIN . Chronic proliferative dermatitis mutation ( cpdm ) mice are deficient in SHARPIN ( Sharpincpdm/cpdm: henceforth referred to as Shpnm/m; protein: SHARPIN ) and develop dermatitis , multi-organ pathology and an immunological phenotype including disrupted lymphoid architecture , splenomegaly , liver inflammation and a loss of Peyer's patches in the gut ( HogenEsch et al . , 1993 , 1999; Seymour et al . , 2007 ) . SHARPIN is required for normal tumour necrosis factor ( TNF ) receptor 1 ( TNFR1 ) -mediated gene induction and prevention of TNF-mediated death of various cells , including epidermal keratinocytes , in vitro ( Gerlach et al . , 2011; Ikeda et al . , 2011; Tokunaga et al . , 2011 ) . The dermatitis is characterized by epidermal cell death marked by cleaved caspase-3- , -8- and -9-positive cells ( Ikeda et al . , 2011; Liang and Sundberg , 2011; Potter et al . , 2014 ) . Since the dermatitis and inflammatory phenotype were shown to be TNF dependent , and because the only TNF signaling output that was aberrantly increased in the absence of SHARPIN was cell death , we previously proposed TNF/TNFR1-mediated cell death to be causative of the cpdm phenotype ( Gerlach et al . , 2011 ) . The role of neither TNFR1 nor cell death has been confirmed in vivo , however . TNFR1 signaling typically involves the intracellular recruitment of TNFR1-associated death domain protein ( TRADD ) , TNF receptor-associated factor 2 ( TRAF2 ) , cellular inhibitor of apoptosis ( cIAPs ) , and receptor interacting protein kinase 1 ( RIPK1 ) ( Silke , 2011 ) . The heterotrimeric linear ubiquitin chain assembly complex ( LUBAC ) of SHARPIN ( also known as SIPL ) , HOIL-1 ( RBCK1/RNF54 ) and HOIL-1L-interacting protein ( HOIP; RNF31 ) ( Gerlach et al . , 2011; Ikeda et al . , 2011; Tokunaga et al . , 2011 ) is also recruited to the TNFR1 signaling complex . Here , it assembles a linear ubiquitin scaffold needed for full recruitment of the NF-κB essential modulator ( NEMO ) /NF-κB kinase subunit gamma ( IKKγ ) -containing IKK complex , which activates pro-survival NF-κB signaling . TNFR1-induced c-Jun N-terminal protein kinase ( JNK ) and p38 signaling is also regulated by LUBAC . SHARPIN deficiency blunts the TNFR1 pro-survival transcriptional signal and sensitizes cells to TNF-induced cell death . The E3 ligase activity of HOIP catalyzes the addition of linear ubiquitin to target proteins , and SHARPIN and HOIL-1 are key regulators of the stability and activity of HOIP ( Gerlach et al . , 2011 ) . In addition to TNFR1 , LUBAC has also been shown to regulate the transcriptional response from the interleukin-1 receptor ( IL-1R ) , CD40 , lymphotoxin beta receptor ( LTβR ) , toll-like-receptor 4 ( TLR4 ) , and nucleotide-binding oligomerization domain-containing protein 2 ( NOD2 ) receptor signaling complexes ( Schmukle and Walczak , 2012 ) . Deletion of Il1rap , needed for IL-1 signaling , has been reported to almost completely prevent the development of cpdm dermatitis ( Liang et al . , 2010 ) . This suggests that IL-1R signaling is a significant driver of disease , but the effect of Il1rap deficiency on the rest of the Shpnm/m phenotype was not reported . Cpdm mice have prominent eosinophil infiltration into the skin; however , deletion of Il5 , which dramatically reduces the number of cutaneous and circulating eosinophils , fails to ameliorate disease ( Renninger et al . , 2010 ) . Shpnm/mRag1−/− mice lacking functional lymphocytes develop dermatitis , indicating that T and B cell cells are not required for the skin phenotype ( Potter et al . , 2014 ) . Furthermore , hematopoietic cell transfer with bone marrow and spleen cells from cpdm mice to syngeneic wild-type C57BL/Ka mice failed to transfer disease in mice 2 months post reconstitution . Finally , cpdm skin transplanted onto nude mice retained the donor dermatitis phenotype 3 months post transplant , while syngeneic healthy skin transplanted onto cpdm mice did not acquire the disease over the same time ( HogenEsch et al . , 1993; Gijbels et al . , 1995 ) . Together these studies indicate that a skin-intrinsic defect in cpdm mice drives the inflammatory disease , however they do not rule out a role for the hematopoietic system in amplifying it . Impaired pro-survival TNFR1 signaling can induce both caspase-8-dependent apoptotic and RIPK3- and mixed lineage kinase domain-like protein ( MLKL ) -dependent necroptotic cell death via a cytosolic death platform ( Micheau and Tschopp , 2003; He et al . , 2009; Sun et al . , 2012; Zhao et al . , 2012; Murphy et al . , 2013 ) . Necroptosis involves the release of cellular contents including potential damage-associated molecular patterns ( DAMPs ) such as mitochondrial DNA , high mobility group box 1 protein ( HMGB1 ) , IL-33 , and IL-1α ( Kaczmarek et al . , 2013 ) . By contrast , apoptosis is considered to be immunologically silent , although this is clearly context dependent because excessive apoptosis resulting from conditional epidermal deletion of the caspase inhibitor cFLIP can cause severe skin inflammation ( Panayotova-Dimitrova et al . , 2013 ) . Caspase-8 can cleave both RIPK1 and RIPK3 and is needed to keep the necroptotic pathway in check ( Vandenabeele et al . , 2010; Kaiser et al . , 2011; Oberst et al . , 2011 ) . Regulation of necroptotic signaling is crucial for skin homeostasis because deletion of either caspase-8 , the caspase-8 adaptor protein FADD ( Fas-associated protein with death domain ) , or RIPK1 , leads to RIPK3- and MLKL-dependent epidermal hyperplasia and inflammation ( Kovalenko et al . , 2009; Lee et al . , 2009; Bonnet et al . , 2011; Kaiser et al . , 2011; Oberst et al . , 2011; Dannappel et al . , 2014; Dillon et al . , 2014; Rickard et al . , 2014 ) . Although the precise factors that determine whether TNFR1 mediates apoptosis or necroptosis are unclear , high levels of RIPK3 , loss of cIAPs , and CYLD-mediated deubiquitylation of RIPK1 appear conducive to necroptosis ( Silke and Vaux , 2014 ) . In addition to a crucial role in necroptosis , RIPK3 may also regulate inflammasome-induced IL-1ß production in the absence of IAPs or caspase-8 ( Vince et al . , 2012; Kang et al . , 2013 ) . Thus the effects of loss of RIPK3 on an inflammatory phenotype may not be due to loss of necroptotic cell death but to a less well-defined role in IL-1ß production . This complicates interpretation of the role of RIPK3 in a disease , particularly when IL-1 is pathogenic such as in cpdm dermatitis . MLKL is downstream of RIPK3 in necroptosis and appears not to be required for cytokine production in the same situations as RIPK3 ( Wong et al . , 2014 ) . Thus Mlkl−/− mice may provide an opportunity to disentangle the relative contribution of necroptosis and deregulated cytokine production in disease . Here we provide genetic evidence in support of our hypothesis that TNFR1-induced cell death is a driver of the inflammatory disease in cpdm mice ( Gerlach et al . , 2011 ) . We show that the cpdm phenotype is TNFR1 and cell-death dependent . Ripk3 or Mlkl deficiency largely prevented cpdm liver inflammation and ameliorated the splenic phenotype and leukocytosis . Remarkably , given the skin-inflammation phenotype of skin-specific Casp8 knock-out mice ( Kovalenko et al . , 2009; Lee et al . , 2009 ) , Casp8 heterozygosity potently suppressed the inflammatory skin phenotype while leaving the systemic inflammation unaffected . Strikingly , combined Ripk3 deficiency and Casp8 heterozygosity completely prevented the cpdm dermatitis in all but one of the mice analyzed at 42 to 45 weeks of age , prevented liver inflammation and grossly restored splenic architecture . Surprisingly , given the importance of LTßR ( also known as TNFRSF3 ) signaling in the formation of Peyer's patches ( De Togni et al . , 1994; Koni et al . , 1997 ) and the role of SHARPIN in this pathway ( Tokunaga et al . , 2011 ) , apparently normal Peyer's patches were also present in Shpnm/mCasp8+/-Ripk3−/− mice . The dermatitis in Shpnm/m mice has previously been shown to be driven by TNF ( Gerlach et al . , 2011 ) and IL-1 signaling ( Liang et al . , 2010 ) . Because the environment may influence the onset of the disease we wished to test the importance of TNF and IL-1 signaling in a head-to-head manner . Furthermore the relative contribution of TNFR1 and TNFR2 in cpdm dermatitis has not been determined . We therefore generated Shpnm/m mice deficient in Tnfr1 , Tnfr2 or Il1r1 ( Figure 1A ) . All the knock-out mouse strains used in this study have been backcrossed at least ten times onto C57BL/6 , or were generated on the C57BL/6 background ( Mlkl−/− mice ) . However , the cpdm mutation arose on a C57BL/Ka background ( HogenEsch et al . , 1993 ) . To control for background modifier effects , we backcrossed C57BL/Ka Shpnm/m mice once or twice onto the C57BL/6 background , equivalent to the strategy we used in generating all our Shpnm/m compound knock-out strains to generate Shpnm/m C57BL/6 control mice . These control mice developed the cpdm phenotype indistinguishably from the C57BL/Ka Shpnm/m mice , typically presenting with a visible skin phenotype by 5 to 6 weeks and invariably requiring euthanasia due to the dermatitis before 14 weeks of age . Tnfr2 deletion did not ameliorate the dermatitis but Shpnm/mTnfr1−/− mice showed no outward signs of disease even after 35 weeks ( Figure 1A–C ) . Il1r1 deletion significantly delayed the appearance of dermatitis , with markedly reduced epidermal hyperplasia in 13-week-old Shpnm/mIl1r1−/− mice compared with 12-week-old Shpnm/m mice ( Figure 1B , C ) . However , by 19–20 weeks , Shpnm/mIl1r1−/− mice typically developed disease and required euthanasia . Tnfr1 deletion reduced periportal liver inflammation and partially ameliorated the splenic phenotype , but did not restore Peyer's patches , whilst Tnfr2 and Il1r1 deletion did not prevent pathology in any of these organs ( Figure 1B , D ) . 10 . 7554/eLife . 03464 . 003Figure 1 . Cpdm dermatitis is mediated by TNFR1 , IL-1R to a lesser extent , and not TNFR2 . ( A ) Representative photos of mice of indicated genotypes and age . ( B ) Histological analysis of mice of genotype and age as indicated; representative of n ≥ 3 mice for each genotype or group . Black arrows in liver images point to areas of periportal inflammation . Black arrows in small intestine image point to Peyer's patches . Shpnm/m: Sharpincpdm/cpdm . Control mice are Shpn+/+ or +/m , Tnfr1−/− control mice are Shpn+/+ or +/mTnfr1−/− . Scale bars: Skin and liver 100 µm , spleen 500 µm , small intestine 1 mm . H&E: hematoxylin and eosin . ( C ) Epidermal thickness of mice of indicated age and genotypes . Each point represents the average of at least 14 measurements from multiple fields of view per mouse that were taken by a researcher who was blind to the genotype . Dotted lines are drawn at 30 µm and 14 weeks . Red numbers ( and black for middle graph ) correspond to proportion of Shpnm/m mice with epidermal thickness >30 µm at < 14 weeks of age ( upper left quadrant ) . Control mice are Shpn+/+ or +/m , Shpn+/+ or +/mTnfr1−/− and Shpn+/+ or +/mIl1r1−/− in upper , middle and lower graphs respectively . *** Significantly different to control group ( Fisher's exact test ) , p ≤ 0 . 005 . ( D ) Average spleen weights of mice of indicated genotypes . Spleen weights were taken from 12-week-old mice ( except Tnfr1−/− mice that were 15 or 35 weeks old ) , or younger mice if they required euthanasia due to their dermatitis . Data are represented as mean + SEM , *p ≤ 0 . 05 , ***p ≤ 0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 03464 . 003 To gain insight into the pathogenesis of cpdm dermatitis we assessed cytokine levels in skin lysates from 6-week-old cpdm and control mice using a BioPlex kit ( BioRad ) ( Figure 2A ) to determine which cytokines were elevated early in the disease process . TNF levels were slightly elevated ( log scale , Figure 2A ) and , consistent with reports of eosinophilia in cpdm mice ( HogenEsch et al . , 1993; Gijbels et al . , 1996 ) , IL-5 ( a key inducer of eosinophil maturation ) was also elevated . The monocyte and macrophage chemoattractant protein MCP-1 was significantly elevated in cpdm skin , as was the IL-12 p40 subunit . There was also a trend for macrophage inflammatory protein 1α ( MIP-1α ) levels to be elevated . Consistent with this , there was an increase in F4/80+ cells in the Shpnm/m dermis , and this was evident in patches at 3 weeks , before significant epidermal hyperplasia was present ( Figure 2B ) . Ikeda et al . reported cleaved caspase-3-positive keratinocytes in 10-week-old Shpnm/m mice ( Ikeda et al . , 2011 ) , and we found cleaved caspase-3-positive cells were already present in the epidermis of 3-week-old Shpnm/m mice , indicating that apoptosis is an early event in the dermatitis and occurs before significant hyperplasia ( Figure 2C ) . There were slightly more cleaved caspase-3-positive cells in Shpnm/m livers than controls ( almost exclusively in the infiltrating cells ) , however they were infrequent . An increased number of cleaved caspase-3-positive cells were detected in Shpnm/m spleens , but , again , were much less appreciable than in the epidermis ( Figure 2C ) . 10 . 7554/eLife . 03464 . 004Figure 2 . Keratinocyte cell death and dermal macrophage infiltration are early events in cpdm dermatitis . ( A ) BioPlex cytokine analysis of skin lysates from mice of indicated genotypes . Data are represented as mean +S . E . M . *p ≤ 0 . 05 . ( B ) F4/80 staining ( brown ) of skin sections counterstained with hematoxylin ( blue ) . Control: Shpn+/+ or +/m . ( C ) Cleaved caspase-3 staining ( brown ) of skin , liver and spleen sections counterstained with hematoxylin ( blue ) . Black arrows show examples of cleaved-caspase-3 positive cells . Control: Shpn+/+ or +/m . ( A–C ) Three mice were analyzed for each genotype or group . Scale bars: 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03464 . 004 Reconstitution of wild-type mice with Shpnm/m bone marrow and/or spleen cells fails to transfer the disease ( HogenEsch et al . , 1993 ) , suggesting a skin-intrinsic defect underlies cpdm dermatitis . These mice , however , were only followed for 8 weeks post reconstitution and this may not have allowed sufficient time for the dermatitis to develop . To test this possibility , we reconstituted wild-type mice with Shpnm/m bone marrow cells and followed them for 12 months . Reconstitution efficiency was high but the mice did not demonstrate any skin , liver , or spleen phenotype during this time ( Figure 3 ) . Collectively these results suggest that the hematopoietic system in isolation cannot cause Shpnm/m dermatitis , and that macrophages may play a role in the amplification of the disease , particularly given they can be a prominent source of TNF . 10 . 7554/eLife . 03464 . 005Figure 3 . No dermatitis in wild-type mice reconstituted with Shpnm/m bone marrow after 12 months . ( A ) No dermatitis was observed in wild-type mice 12 months after reconstitution with Shpnm/m bone marrow . ( B ) Flow cytometry analysis showing percentage contribution of Ly5 . 1 ( recipient ) vs Ly5 . 2 ( donor ) white blood cells in peripheral blood 12 months after reconstitution . ( C ) Histological and immunofluorescence analysis of dorsal skin from mice 12 months after reconstitution . ( D ) Histological analysis of the spleen and liver from mice 12 months after reconstitution . Scale bars: Skin 100 µm , liver 200 µm and spleen 500 µm . H&E: hematoxylin and eosin . DOI: http://dx . doi . org/10 . 7554/eLife . 03464 . 005 Tnf deletion prevents cpdm dermatitis and TNF kills Shpnm/m keratinocytes in vitro , suggesting that TNF-induced cell death drives the Shpnm/m skin phenotype ( Gerlach et al . , 2011 ) . Consistent with this , TNF-induced cell death in Shpnm/m keratinocytes that was partially blocked by Q-VD-OPh or Nec-1 treatment ( Figure 4A ) . Shpnm/m mouse dermal fibroblasts ( MDFs ) were also sensitive to TNF-induced cell death , and this death was almost completely blocked by the RIPK1 kinase inhibitor Necrostatin-1 ( Nec-1 ) , but not by the pan-caspase inhibitor Q-VD-OPh ( Figure 4B ) . To test whether TNF could induce caspase cleavage we treated primary Shpnm/m keratinocytes and MDFs with TNF for up to 4 hr ( Figure 4C , D , Figure 4—figure supplement 1 ) . TNF treatment led to caspase-3 and -8 cleavage at 4 hr in Shpnm/m , but not wild-type , keratinocytes . In Shpnm/m MDFs there was also caspase-3 and -8 cleavage , and , surprisingly , this was reduced by Nec-1 , but not Q-VD , treatment ( Figure 4D ) . In Shpnm/m keratinocytes and MDFs , the increased caspase-8 activity coincided with a marked processing of cFLIP , but other outcomes of the signaling pathway were less obviously affected . Together with the early presence of cleaved caspase-3 staining and the elevation of TNF in Shpnm/m skin , these results indicate that TNF-induced apoptosis may play a pathogenic role in the skin phenotype . 10 . 7554/eLife . 03464 . 006Figure 4 . TNF induces death in multiple primary cell types , and is marked by caspase-3 and -8 cleavage in primary keratinocytes and mouse dermal fibroblasts in vitro . ( A and B ) Primary keratinocytes ( A ) and mouse dermal fibroblasts ( MDFs ) ( B ) were treated with 100 ng/ml human Fc-TNF , 50 µM Nec-1 or 10 µM Q-VD-OPh for 24 hr as indicated , then viability was assessed by propidium iodide ( PI ) uptake and flow cytometry . Cells were generated from three different mice for each group . Control: Shpn+/+ or +/m . ( C and D ) Western blot analysis of primary keratinocytes ( C ) and MDFs ( D ) treated as indicated ( concentrations as in A and B ) then lysed and lysates separated on SDS/PAGE and western blotted with indicated antibodies . Shpn+/m: n = 1; Shpnm/m: n = 2 ( n = 1 for MDFs ) mice analyzed shown above . * Indicates non-specific band . Data from additional mice is shown in Figure 4—figure supplement 1 . ( E ) Neutrophils , monocytes and bone-marrow-derived macrophages were treated with 100 ng/ml human Fc-TNF , 50 µM Nec-1 , 20 µM Q-VD-OPh ( 10 µM for macrophages ) or 500 nM Compound A ( CpdA ) for 20 hr ( 24 hr for macrophages ) as indicated . Viability was assessed by PI uptake and flow cytometry . The Smac mimetic CpdA sensitizes cells to TNF-induced cell death and serves as a control . Cells were generated from three different mice for each group except for macrophages , where six to eight mice were analyzed . ( A , B , E ) Data are represented as mean + SEM , *p ≤ 0 . 05 , **p ≤ 0 . 01 , ***p ≤ 0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 03464 . 00610 . 7554/eLife . 03464 . 007Figure 4—figure supplement 1 . TNF induces caspase-3 and -8 cleavage in primary keratinocytes and mouse dermal fibroblasts in vitro . Western blot analysis of primary keratinocytes and mouse dermal fibroblasts ( MDFs ) treated as indicated ( reagent concentrations as per Figure 4A , B ) then lysed and lysates separated on SDS/PAGE and western blotted with indicated antibodies . * Indicates non-specific band . Lysates generated from different mice to those shown in Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 03464 . 007 Given the appearance of cleaved caspase-3-positive cells in the Shpnm/m dermis ( Figure 2C ) , we sought to determine whether some of these might be non-fibroblast cells such as immune cells . Since lymphoid cells are not required for cpdm dermatitis ( Potter et al . , 2014 ) and macrophage infiltration appears early , we tested whether myeloid cells were also sensitive to TNF-induced death . Shpnm/m neutrophils , monocytes and bone-marrow-derived macrophages ( BMDMs ) were all sensitive to killing by TNF , whereas the wild-type cells were not ( Figure 4E ) . Neutrophil and monocyte cell death was more efficiently blocked by a combination of Nec-1 and Q-VD , whereas in macrophages Nec-1 was sufficient to block cell death . These data suggest that myeloid cell death may contribute to the inflammatory skin phenotype . We did not detect a defect in p38 , JNK or NF-κB pro-survival signaling in SHARPIN-deficient keratinocytes and MDFs , as has been shown for other cell types ( Gerlach et al . , 2011; Ikeda et al . , 2011; Tokunaga et al . , 2011 ) . Consistent with earlier reports ( Gerlach et al . , 2011; Peltzer et al . , 2014 ) , however , using an antibody that specifically recognises linear ubiquitin we detected substantially reduced linear ubiquitylation in the native TNFR1 signaling complex obtained from Shpnm/m vs wild-type mouse embryonic fibroblasts ( MEFs ) ( Figure 5A ) . We also found that HaCaT keratinocytes ( a human immortalized cell line ) stably expressing catalytically inactive HOIPC885S were sensitive to TNF-induced cell death ( Figure 5B ) , indicating a requirement for LUBAC and its linear-ubiquitin-chain-forming activity in preventing TNF-induced keratinocyte death . 10 . 7554/eLife . 03464 . 008Figure 5 . Sharpin is required for normal linear ubiquitylation of the TNF-R1 signaling complex , and HOIP protects keratinocytes from TNF-induced cell death . ( A ) Anti Flag Immuno-Precipitation ( IP ) of the TNF receptor signaling complex in immortalised mouse embryonic fibroblasts ( MEFs ) treated with Flag-TNF for the times indicated . ( B ) HaCaT human keratinocytes stably expressing HOIP , the catalytically inactive HOIPC885S , or an empty vector ( control ) were treated with 100 ng/ml TNF for 24 hr . Viability was assessed by propidium iodide ( PI ) uptake and flow cytometry . Data are presented as mean + SEM , n = 3 , *p ≤ 0 . 05 , **p ≤ 0 . 01 , ***p ≤ 0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 03464 . 008 To investigate the importance of the caspase-8-mediated apoptotic pathway in vivo we generated Shpnm/mCasp8+/− mice . We did not attempt to generate Shpnm/mCasp8−/− mice because Casp8 deletion results in embryonic lethality at around E10 . 5 ( Varfolomeev et al . , 1998 ) . In contrast to Shpnm/m mice , 12-week-old Shpnm/mCasp8+/− mice had almost no epidermal hyperplasia and largely normal keratin 6 and 14 expression ( Figure 6A , B ) . By 15 weeks of age , however , significant hyperplasia was observed in one of these mice ( Figure 6B ) . Shpnm/mCasp8+/− mice retained other aspects of the phenotype: splenomegaly with disrupted splenic architecture , liver inflammation , and a lack of intestinal Peyer's patches ( Figure 6C ) . Some Shpnm/mCasp8+/− mice succumbed to a pulmonary infection with Pasteurella and required euthanasia while littermates were unaffected , suggesting that Shpnm/mCasp8+/− mice are partially immunocompromised . Mice with obvious signs of infection were not used for the spleen , blood , or liver analyses but were included in the skin analysis . Occult infection , however , in Shpnm/mCasp8+/− mice used for hematopoietic , liver , and spleen analyses cannot be completely excluded . 10 . 7554/eLife . 03464 . 009Figure 6 . Protection from cpdm dermatitis with Casp8 heterozygosity but not Bid deletion . ( A ) Histological and immunofluorescence skin analysis . ( B ) Epidermal thickness of mice of indicated age and genotypes determined as in Figure 1B , by an investigator blinded to genotype . Top panel is a repeat of data in 1B for reference purposes only . Dotted lines are drawn at 30 µm and 14 weeks . Red numbers correspond to proportion of Shpnm/m mice with epidermal thickness >30 µm at < 14 weeks of age ( upper left quadrant ) . Control mice are Shpn+/+ or +/m , Shpn+/+ or +/mBid−/− , and Shpn+/+ or +/mCasp8+/− in upper , middle and lower graphs , respectively . *** Significantly different to control group ( Fisher's exact test ) , p ≤ 0 . 005 . ( C ) Histological analysis of spleen , liver , and small intestine . Black arrows in liver images point to areas of periportal inflammation . Black arrow in small intestine image points to Peyer's patch . ( D ) Average spleen weights of mice of indicated genotypes . Spleen weights were taken from 12–14-week-old mice , or younger mice if they required euthanasia due to their dermatitis . Data are represented as mean + SEM , ***p ≤ 0 . 005 . ( A and C ) Control mice are Shpn+/+ or +/m , n ≥ 3 mice analyzed each genotype or group . Scale bars: skin and liver 100 µm , spleen 500 µm , small intestine 1 mm . H&E: hematoxylin and eosin . DOI: http://dx . doi . org/10 . 7554/eLife . 03464 . 009 In certain cell types such as hepatocytes , BID can be cleaved by caspase-8 to generate truncated BID ( tBID ) that , in turn , mediates caspase-9- and mitochondria-dependent apoptosis ( Czabotar et al . , 2014 ) . BID is a key regulator of UV-induced apoptosis in keratinocytes , indicating that this pathway is of importance in the skin ( Pradhan et al . , 2008 ) . Because cleaved caspase-9 is found in the Shpnm/m epidermis ( Ikeda et al . , 2011 ) and Shpnm/m keratinocyte extracts contained caspase-9-substrate-cleaving activity ( Liang and Sundberg , 2011 ) , we hypothesized that TNF-induced BID-dependent apoptosis may be a driver of the cpdm phenotype and , hence , generated Shpnm/mBid−/− mice . We observed no protection from any aspects of the cpdm phenotype ( Figure 6A–D ) , however , indicating that caspase-8 , but not BID , is an important driver of this disease . Caspase-8 heterozygosity significantly delayed epidermal hyperplasia , but other aspects of the cpdm phenotype ( e . g . splenomegaly and liver inflammation ) remained . We therefore sought to test the role of necroptosis in the inflammatory disease by generating Shpnm/mRipk3−/− mice . Whereas control mice invariably developed severe dermatitis by 12 weeks of age , roughly half of the 12-week-old Shpnm/mRipk3−/− mice had a less severe epidermal phenotype at this point ( Figure 7A , B ) . When aged over 12 weeks , all Shpnm/mRipk3−/− animals went on to develop severe disease and were euthanized due to their skin phenotype before 18 weeks of age . The remainder of Shpnm/mRipk3−/− mice developed skin disease at the same rate as control mice . 12 week-old Shpnm/mRipk3−/− mice had no signs of liver inflammation and significantly less splenomegaly , although they still lacked Peyer's patches ( Figure 7C , D ) . We also generated Shpnm/mMlkl−/− mice and found they had a similar epidermal phenotype to Shpnm/m mice ( Figure 7B ) . Like Shpnm/mRipk3−/− mice , however , 12-week-old Shpnm/mMlkl−/− animals had reduced splenomegaly and only one out of 12 mice showed signs of liver inflammation ( Figure 7C , D ) . Collectively these results indicate that RIPK3 and MLKL are important drivers of the liver and splenic cpdm phenotype , and that RIPK3 contributes to the epidermal phenotype , importantly in a non-MLKL-dependent ( hence , most likely , necroptosis-independent ) manner . 10 . 7554/eLife . 03464 . 010Figure 7 . Ripk3 deficiency slightly delays cpdm dermatitis onset , and Ripk3 and Mlkl deficiency partially protects against the cpdm splenic phenotype and markedly attenuates liver inflammation . ( A ) Histological and immunofluorescence skin analysis . ( B ) Epidermal thickness of mice of indicated age and genotypes determined as in Figure 1B , by an investigator blinded to genotype . Top panel is a repeat of data in 1B for reference purposes only . Dotted lines are drawn at 30 µm and 14 weeks . Red numbers correspond to proportion of Shpnm/m mice with epidermal thickness >30 µm at < 14 weeks of age ( upper left quadrant ) . Control mice are Shpn+/+ or +/m , Shpn+/+ or +/mRipk3−/− , and Shpn+/+ or +/mMlkl−/− in upper , middle and lower graphs , respectively . *** Significantly different to control group ( Fisher's exact test ) , p ≤ 0 . 005 , # significantly different to Shpnm/m mice ( Fisher's exact test ) , p ≤ 0 . 05 . ( C ) Histological analysis of spleen , liver , and small intestine . Black arrows in liver image points to areas of periportal inflammation . Black arrows in small intestine image points to Peyer's patches . ( D ) Average spleen weights of mice of indicated genotypes . Spleen weights were taken from 12-week-old mice , or younger mice if they required euthanasia due to their dermatitis . Data are represented as mean + SEM , ***p ≤ 0 . 005 . ( A and C ) Control mice are Shpn+/+ or +/m , n ≥ 3 mice analyzed each genotype or group . Scale bars: skin and liver 100 µm , spleen 500 µm , small intestine 1 mm . H&E: hematoxylin and eosin . DOI: http://dx . doi . org/10 . 7554/eLife . 03464 . 010 Since deletion neither of one allele of Casp8 nor of two alleles of Ripk3 was able to fully rescue all the multi-organ pathology in cpdm mice , we sought to generate Shpnm/mCasp8−/−Ripk3−/− mice , taking advantage of the fact that Ripk3 deletion prevents Casp8−/−-mediated embryonic lethality ( Kaiser et al . , 2011; Oberst et al . , 2011 ) . We generated these mice independently at two separate facilities . At one facility , most of these triple-deficient mice died perinatally , typically in a window between E17 and 1 to 2 days after birth , and no Shpnm/mCasp8-/-Ripk3−/− animals were obtained at weaning ( Figure 8A , B; numbers in A refer to mice analyzed at first facility only ) . Most Shpnm/mCasp8−/−Ripk3−/− embryos obtained by Caesarean section at E19 appeared edematous or were dead . Some were able to breathe; however , these died within the first 2 days after birth . Hematopoietic analysis at E19 did not reveal any consistent differences ( Figure 8C ) . At a separate facility , two viable Shpnm/mCasp8−/−Ripk3−/− mice were obtained from a limited number of matings . By roughly 3 months of age these mice appeared runted and were euthanized . Histologically these mice had no epidermal phenotype , but disrupted splenic architecture was apparent ( Figure 8D ) . 10 . 7554/eLife . 03464 . 011Figure 8 . Shpnm/mCasp8−/−Ripk3−/− mice are prone to perinatal lethality . ( A ) Table of segregation of expected and observed genotypes from Shpn+/mCasp8-/-Ripk3−/− or Shpnm/mCasp8+/-Ripk3−/− intercrosses at various developmental stages . E10 . 5 controls: Shpn+/+ or +/mCasp8-/-Ripk3−/−; E19 controls: Shpnm/mCasp8+/+ or +/−Ripk3−/−; weaning controls: Shpn+/+ or +/mCasp8−/−Ripk3−/− or Shpnm/mCasp8+/+ or +/−Ripk3−/− . *Significantly different to expected value , Fisher's exact test p < 0 . 0005 . ( B ) Photos of E19 embryos obtained by Caesarian section . The Shpnm/mCasp8−/−Ripk3−/− mouse on the right was recovered dead , all others were alive . Other embryos are control mice: Shpnm/mCasp8+/+ or +/−Ripk3−/− . ( C ) ADVIA blood analysis from E19 embryos . RBC: red blood cells; WBC: white blood cells; MCV: mean cell volume . Horizontal lines depict data means . Control mice: Shpnm/mCasp8+/+ or +/−Ripk3−/− . ( D ) Histological analysis of tissue from 12-week-old mice of indicated genotypes . Mice were from a separate facility to those in A–C and are not included in the table in A . Two mice were analyzed for each genotype . Scale bars: skin 100 µm , spleen 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03464 . 011 While the reason for the lethality of the triple-deficient animals is unknown , we were readily able to obtain viable Shpnm/mCasp8+/−Ripk3−/− mice . These mice were indistinguishable from control mice at 12 weeks of age and had no skin , liver , or spleen pathology ( Figure 9A–E ) . Remarkably , Peyer's patches were present ( Figure 9C , Figure 9—figure supplement 1 ) . One 67-week-old and three 45-week-old Shpnm/mCasp8+/−Ripk3−/− mice had no signs of any skin phenotype , however another mouse developed dermatitis at 42 weeks ( Figure 9A , B , D; note that epidermal thickness was not quantified for the 42-week-old mouse ) . 10 . 7554/eLife . 03464 . 012Figure 9 . Ripk3 deletion and Casp8 heterozygosity markedly delays emergence of cpdm dermatitis and liver inflammation and restores Peyer’s patches . ( A ) Representative photos of mice of indicated genotypes and age . Three Shpnm/mCasp8+/-Ripk3−/− mice were analyzed at 45 weeks and one at 67 weeks with no detectable dermatitis , however another mouse developed dermatitis by 42 weeks of age . ( B ) Histological and immunofluorescence skin analysis . ( C ) Histological analysis of spleen , liver , and small intestine . Black arrows in liver image point to areas of periportal inflammation . Black arrows in small intestine images point to Peyer's patches . ( D ) Epidermal thickness of mice of indicated age and genotypes measured as described in Figure 1B . Top panel is a repeat of data in 1B for reference purposes only . Dotted lines are drawn at 30 µm and 14 weeks . Red numbers correspond to proportion of Shpnm/m mice with epidermal thickness >30 µm at < 14 weeks of age ( upper left quadrant ) . Control mice are Shpn+/+ or +/m and Shpn+/+ or +/mCasp8+/−Ripk3−/− in upper and lower graphs , respectively . *** Significantly different to control group ( Fisher's exact test ) , p ≤ 0 . 005 . ( E ) Average spleen weights of mice of indicated genotypes . Spleen weights were taken from 12-week-old mice , or younger mice if they required euthanasia due to their dermatitis . Data are represented as mean + SEM , ***p ≤ 0 . 005 . ( B and C ) Control mice are Shpn+/+ or +/m , n ≥ 3 mice analysed each genotype or group . Scale bars: skin and liver 100 µm , spleen 500 µm , small intestine 1 mm . H&E: hematoxylin and eosin . DOI: http://dx . doi . org/10 . 7554/eLife . 03464 . 01210 . 7554/eLife . 03464 . 013Figure 9—figure supplement 1 . Restoration of Peyer's patches in 8-week-old Shpnm/mCasp8+/−Ripk3−/− mice . Numbers of Peyer's patches per small intestine counted from 8-week-old mice of indicated genotypes . Horizontal lines depict data means . DOI: http://dx . doi . org/10 . 7554/eLife . 03464 . 013 To investigate the effect of the various genetic crosses on Shpnm/m leukocytosis , we analyzed white blood cell levels in peripheral blood using an ADVIA hematological analyzer ( Figure 10 ) . Tnfr1 deletion was more effective at reducing leukocyte numbers than Tnfr2 or Il1r1 deletion , although all of these compound knock-out mice had leukocyte subsets that were elevated compared to controls . Caspase-8 heterozygosity or Bid deletion did not prevent the leukocytosis , whereas Ripk3 and Mlkl deletion markedly reduced it , suggesting the hematopoietic phenotype is driven predominantly by necroptosis . Shpnm/mRipk3−/− and Shpnm/mMlkl−/− mice still had elevated neutrophils , however Shpnm/mCasp8+/−Ripk3−/− mice had no white blood cell elevation . 10 . 7554/eLife . 03464 . 014Figure 10 . Peripheral blood counts from various crosses . Peripheral blood was collected from 11–14-week-old mice ( reconstituted mice: Shpnm/m → wild type were 12 months old and Tnfr1−/− mice were 35 weeks old ) , or younger if the mice were euthanized due to severe dermatitis . Shpn+/+ or +/m and Shpnm/m mice are highlighted in blue and red for reference purposes . Blood was analyzed using an ADVIA 2120 hematological analyzer . RBC: red blood cells . Horizontal lines depict data means . Control mice are Shpn+/+ or +/m , for example Tnfr1−/− control mice are Shpn+/+ or +/mTnfr1−/− . DOI: http://dx . doi . org/10 . 7554/eLife . 03464 . 014 LUBAC , composed of three proteins , HOIL-1 , HOIP and SHARPIN , has recently emerged as a regulator of a diverse set of signaling complexes ( Zak et al . , 2011; Schmukle and Walczak , 2012; Tokunaga , 2013 ) . Lack of HOIP causes mid-gestational embryonic lethality ( embryonic day 10 . 5 ) , and this is prevented by simultaneous loss of Tnfr1 ( Peltzer et al . , 2014 ) . Together , these data demonstrate that LUBAC is required for full-strength IL-1ß and TNF signaling and deficiency in these signaling pathways would be expected to impair an inflammatory response . The consequence , however , of Sharpin deficiency in mice is multi-organ inflammation ( HogenEsch et al . , 1993 ) . Likewise , loss-of-expression and loss-of-function mutations in HOIL-1 result in a fatal , human , inherited disorder characterized by chronic inflammation; consistent with loss of the major defensive inflammatory signaling pathways , affected patients suffer from invasive bacterial infections ( Boisson et al . , 2012 ) . In vitro experiments suggest that HOIL-1 is an essential component of LUBAC , which is present in many cell types . With this in mind it is surprising that HOIL-1-deficient mice were reported to be normal ( Tokunaga et al . , 2009 ) . If , however , we set this observation to one side , it seems that in humans and mice loss of LUBAC can precipitate an inflammatory disease . This could be because the loss of the homeostatic component of TNFR1 , IL-1R1 , and TLR signaling reduces the ability of an organism to resist infection , which , even in the weakened signaling environment of LUBAC deficiency , drives inflammatory signaling . Another possibility is that full-strength signaling from TNF , IL-1ß , or TLR ligands is required to upregulate essential negative feedback regulators such as IκBα and A20 . Signaling , therefore , although reduced , is constitutively active . This explanation seems less likely because IκBα and A20 transcripts were still upregulated in response to TNF or Pam3CSK4 in HOIL-1/HOIP-deficient and cpdm cells ( Haas et al . , 2009; Tokunaga et al . , 2009; Gerlach et al . , 2011; Zak et al . , 2011 ) . The fact that the phenotype from the cpdm mice could not be transferred by the hematopoietic compartment , and that the skin phenotype is maintained following skin transplantation onto nude mice , indicates that these animals suffer from an intrinsic skin defect ( HogenEsch et al . , 1993; Gijbels et al . , 1995 ) . This is not unprecedented; loss of Notch signaling in the skin causes a dermatitis disease in mice that is sufficient to drive systemic inflammation with similar features to the cpdm phenotype ( Dumortier et al . , 2010 ) . The earlier hematopoietic reconstitution experiments , however , may not have allowed sufficient time for the inflammatory phenotype to develop . The work described here excludes this caveat because reconstituted mice did not develop signs of the cpdm phenotype even a year after reconstitution; thus , the inflammatory skin phenotype in the Shpnm/m mutant mice is the result of an intrinsic skin defect . It is , however , noteworthy that loss of Mlkl did not affect the Shpnm/m skin phenotype but significantly reduced or prevented splenomegaly and liver inflammation . Conversely , reduction in caspase-8 markedly ameliorated the skin phenotype but did not prevent splenomegaly or liver inflammation . This shows that the skin and systemic phenotype are separable , although the exact mechanism is unclear . The fact that cpdm keratinocytes are sensitized to TNF-induced apoptosis and necroptosis led us to hypothesize that TNF/TNFR1-dependent cell death was causative for the skin phenotype ( Gerlach et al . , 2011 ) . Because of the systemic inflammation we suspected that necroptotic cell death and the release of DAMPs was the driver . Consistent with this hypothesis , we show here that Tnfr1 deficiency ( as with Tnf deficiency ) suppresses the cpdm phenotype , whereas Tnfr2 deficiency had no effect on the phenotype . Loss of IL-1 signaling has been shown to suppress the cpdm phenotype ( Liang et al . , 2010 ) , but because the environment of the mice likely plays a large part in the onset and severity of the inflammatory disease it was not possible to determine whether loss of IL-1 signaling is as potent as Tnf deficiency at preventing the phenotype . We therefore generated Shpnm/mIl1r1−/− mice , which succumbed to the inflammatory disease much later than Shpnm/m mice , but were far more inflammation prone than any of the Shpnm/mTnf−/+ , Shpnm/mTnf−/− , or Shpnm/mTnfr1−/− mice . These data are again consistent with our original hypothesis , suggesting that TNF/TNFR1 signaling is the main driver of inflammation and that IL-1 contributes to exacerbating the disease . Casp8 deficiency results in early embryonic lethality in mice ( Varfolomeev et al . , 1998 ) , and loss of Casp8 or Fadd in the skin leads to keratinocyte hyperplasia and inflammatory skin disease ( Kovalenko et al . , 2009; Bonnet et al . , 2011 ) . Therefore , we considered it unlikely that loss of caspase-8 would diminish the chronic proliferative dermatitis of the Shpnm/m mutant mice . To our surprise , however , loss of a single allele of Casp8 was strikingly effective at reducing the skin phenotype and appearance of cleaved caspase-3 in the skin . It has been proposed that the intrinsic mitochondrial apoptosis pathway mediated by caspase-9 and caspase-3 plays a role in the keratinocyte hyperplasia observed in Shpnm/m mutant mice based on the presence of disrupted mitochondria in cpdm skin and in vitro experiments ( Liang and Sundberg , 2011 ) . Given the prominent role that caspase-8 plays in the dermatitis , it would be expected that if the intrinsic apoptosis pathway is engaged it should be downstream of caspase-8 and require cleavage of the BH3 protein , BID ( Czabotar et al . , 2014 ) . Cells that require BID cleavage by caspase-8 in order to undergo apoptosis are known as type II cells ( Barnhart et al . , 2003 ) , but it is unclear whether keratinocytes are type I or type II ( Pradhan et al . , 2008; Geserick et al . , 2014 ) . Genetic deletion of Bid did not suppress the dermatitis or any other aspect of the cpdm phenotype , therefore we conclude that the intrinsic mitochondrial pathway is unlikely to play a significant role in the disease . Because necroptotic cell death can be inflammatory , by provoking the release of DAMPs , we expected that deficiency in RIPK3 or MLKL , essential effectors of the necroptotic cell death pathway , would reduce the severity of the inflammation in Shpnm/m mice . Whilst there was a modest delay in the appearance of the dermatitis in Shpnm/mRipk3−/− mice , all animals went on to develop severe skin disease . In contrast to RIPK3 deficiency , deletion of Mlkl did not even delay appearance of the dermatitis . Together , this suggests that RIPK3 may exacerbate the skin phenotype independently of necroptosis , possibly via a direct role in cytokine production . Combined with the markedly ameliorated dermatitis seen in Shpnm/mCasp8+/− mice , this indicates that apoptosis is the main driver of Shpnm/m dermatitis . Although apoptosis is generally regarded as being immunologically inert , when in excess it can result in severe inflammation . For example , conditional deletion of cFLIP in adult keratinocytes caused severe inflammation in the epidermis ( Panayotova-Dimitrova et al . , 2013 ) . Like the cpdm phenotype , this inflammation was TNF dependent . An excess of apoptotic cells may cause disease by overwhelming phagocytosis and clearance of apoptotic bodies , leading to secondary necrosis and DAMP release . It has also been proposed that in certain contexts , such as viral infections , apoptosis may be inflammatory ( Cullen et al . , 2013 ) . In other Shpnm/m organs such as the liver , spleen , and hematopoietic system , Ripk3 and Mlkl deletion ameliorated or protected against disease , indicating that Sharpin deficiency triggers apoptosis in some tissues but necroptosis in others . Because it has been shown that RIPK3 or MLKL can mediate cytokine production induced by the absence of caspase-8 ( Kang et al . , 2013 ) , and potentially other , as yet undescribed , pathways , we cannot completely exclude the possibility that Ripk3 or Mlkl deletion affords protection by blunting cytokine production . Absence of SHARPIN , however , leads to increased , not decreased , caspase-8 activity , so it is not obvious whether these observations apply in this case . Furthermore , whereas it is clear that RIPK3 plays a role in promoting cytokine production in response to a number of stimuli ( Vince et al . , 2012 ) , we have so far only observed a defect in necroptosis and not in inflammatory cytokine production in Mlkl−/− cells ( Murphy et al . , 2013; Allam et al . , 2014; Rickard et al . , 2014 ) . Whilst Shpnm/mCasp8+/− , Shpnm/mRipk3−/− , and Shpnm/mMlkl−/− mice all developed significant disease in either the liver , spleen , hematopoietic compartment , or skin , Shpnm/mCasp8+/−Ripk3−/− mice were almost completely protected from disease for approximately 10–11 months , and in one case 15 months . One of these mice developed the skin phenotype at 10 months of age , indicating that the single allele of Casp8 is enough to eventually cause dermatitis . This is supported by the development of epidermal phenotype in 14–15-week-old Shpnm/mCasp8+/− mice . SHARPIN has been shown to regulate LTß/LTßR signaling , and LTßR is required for the development of Peyer's patches ( De Togni et al . , 1994 ) , therefore we had assumed that the absence of Peyer's patches was due to defective LTßR signaling . A recent study , however , demonstrated that rudimentary Peyer's patches formed in Shpnm/m embryos , but that these regressed post-natally ( Seymour et al . , 2013 ) . Our data show that whilst neither Casp8 heterozygosity nor Ripk3 deletion in isolation restored intestinal Peyer's patches , the combination was able to do so . This suggests that the post-natal regression of these secondary lymphoid organs in Shpnm/m mice is due to deregulated cell death . In a wider sense , our results are particularly important when interpreting the recent finding that crossing cpdm mice with Ripk1K45A/K45A mice ( which lack RIPK1 kinase activity ) completely suppresses the cpdm phenotype ( Berger et al . , 2014 ) . RIPK1 has a promiscuous role in immune signaling , regulating pro-survival , necroptotic , and apoptotic pathways . However , while the RIPK1 kinase domain is well known for its ability to activate the necroptotic pathway , it is not believed to be required to cause apoptosis ( Vandenabeele et al . , 2010 ) . Yet we show here that the Shpnm/m skin phenotype is associated with the appearance of processed caspase-3 in the skin , and both the epidermal hyperplasia and appearance of processed caspase-3 is markedly reduced by loss of a single allele of Casp8 . Consistent with the hypothesis that Shpnm/m keratinocyte hyperplasia is due to apoptosis , it is not prevented by loss of the key necroptotic effector MLKL . Furthermore , purified Shpnm/m keratinocytes and dermal fibroblasts rapidly activate caspase-8 and caspase-3 in response to TNF . Taken together with the work of Berger et al . our work inescapably suggests that in the context of the Shpnm/m epidermis the dominant role of the RIPK1 kinase domain is to activate apoptosis . Unexpectedly , but supporting this conclusion , we showed that Nec-1 ( the RIPK1 kinase inhibitor ) blocked TNF-induced caspase-8 and caspase-3 activation in Shpnm/m dermal fibroblasts . This conclusion is particularly confronting because we and others have shown that loss of RIPK1 in the skin results in a RIPK3/MLKL-dependent hyperplasia that is presumably dependent on necroptosis ( Dannappel et al . , 2014; Rickard et al . , 2014 ) . Thus RIPK1 is able to both activate and inhibit either apoptosis or necroptosis in a highly context-dependent manner . Ripk3 deletion prevents the embryonic lethality seen with either Fadd or Casp8 deletion and the mice survive to adulthood ( Kaiser et al . , 2011; Oberst et al . , 2011; Dillon et al . , 2012 ) . Given this , the perinatal lethality we observed in Shpnm/mCasp8−/−Ripk3−/− mice was completely unexpected . At E19 some of these mice were recovered alive by Caesarean section and established regular breathing , some failed to establish normal breathing , some appeared edematous and were not recovered alive , whilst others were in the process of being resorbed from earlier embryonic lethality . To complicate the picture , two viable Shpnm/mCasp8−/−Ripk3−/− mice were obtained at a separate facility . At 3 months of age these mice had no epidermal phenotype but were runted . One potential explanation for the different penetrance of the phenotype may be a different genetic background because the mice described in Figure 8A were exon 3-deleted caspase-8 mice ( Beisner et al . , 2005 ) , whereas the mice in Figure 8D were exon 3- and 4-deleted caspase-8 ( Salmena et al . , 2003 ) . Exon 3- and 4-deleted caspase-8 mice in yet a third facility ( not shown ) , however , did not survive past weaning , indicating that environmental differences undoubtedly also contribute to the variable penetrance . Future efforts aimed at understanding the lethality of Shpnm/mCasp8−/−Ripk3−/− mice should yield important insights into not only the biology of SHARPIN and linear ubiquitin , but also that of caspase-8 and RIPK3 . In summary , we provide strong evidence that Sharpin deficiency sensitizes keratinocytes to TNF/TNFR1-induced , caspase-8-mediated apoptosis , and that this defect appears to drive the cpdm dermatitis . Ripk3 deletion provided only a modest and variable delay in the presentation of dermatitis . Unlike Casp8 heterozygosity , Ripk3 and Mlkl deletion ameliorated many aspects of the systemic cpdm phenotype . This indicates a tissue-specific role for Sharpin in regulating cell death pathways . Only combined Casp8 heterozygosity and Ripk3 deficiency was able to almost completely prevent all aspects of the cpdm systemic phenotype that we evaluated , including the early loss of Peyer's patches . Whilst inflammation is a known sequelae to necroptotic DAMP release , these findings provide further evidence that excessive apoptosis can also cause inflammatory disease . Furthermore , these results indicate that the suppression of cpdm dermatitis seen by crossing to mice lacking RIPK1 kinase activity ( Berger et al . , 2014 ) may , surprisingly , be due to RIPK1’s kinase activity being upstream of caspase-8 in mediating TNF-induced apoptosis . Mice were maintained at the Walter and Eliza Hall Institute of Medical Research ( WEHI ) and University College London ( UCL ) . C57BL/Ka Sharpincpdm/cpdm mice were obtained from Jax ( Bar Harbor , ME ) , and then either backcrossed one to two times onto C57BL/6 or crossed with C57BL/6 Ripk3−/− , Mlkl−/− , Casp8+/− , Casp8+/-Ripk3−/− , Bid−/− , Il1r1−/− , Tnfr1−/− , or Tnfr2−/− mice . For timed matings , mice were analyzed by Caesarean section . For E19 timed matings , pregnant females were injected at E17 and E18 with progesterone . Primary keratinocytes and MDFs were isolated and cultured as described previously ( Gerlach et al . , 2011; Etemadi et al . , 2013 ) . Cell lysates were prepared using DISC buffer ( 1% NP-40 , 10% glycerol , 150 mM NaCl , 20 mM Tris pH 7 . 5 , 2 mM EDTA , cOmplete protease inhibitor cocktail ( Roche; Penzberg , Germany ) , 2 mM sodium orthovanadate , 10 mM sodium fluoride , β-glycerophosphate , N2O2PO7 ) . Cell lysates were loaded in NuPAGE Bis-Tris gels ( Life Technologies/Thermo Fisher Scientific; Waltham , MA ) and transferred on to Immobilon-P PVDF membranes ( Millipore; Billerica , MA ) or Hybond-C Extra ( GE Healthcare; Little Chalfont , UK ) . Membranes were blocked and antibodies diluted in 5% skim milk powder or Bovine Serum Albumin ( BSA ) in 0 . 1% PBS or TBS-Tween20 . Antibodies used for western blot: cleaved caspase-3 ( 9661 ) and −8 ( 8592 ) , phospho-JNK1/2 ( 4668P ) , phospho-p38 ( 4511 ) , p38 ( 9212 ) , caspase-8 ( 4927 ) , JNK1/2 ( 9252 ) , IκBα ( CN: 9242 ) , and phospho-p65 ( 3033 ) from Cell Signaling Technology ( Danvers , MA ) , β-actin ( A-1978; Sigma Aldrich; St . Louis , MO ) , RIPK1 ( 610458; BD Biosciences; Franklin Lakes , NJ ) , cFLIP ( AG-20B-0005; Adipogen; Liestal , Switzerland ) , FADD ( generated in-house; gift from Lorraine O’Reilly ) and MLKL ( generated in house; Murphy et al . , 2013 ) . Signals were detected by chemoluminescence ( Millipore ) after incubation with secondary antibodies conjugated to horseradish peroxidase . For isolation of neutrophils and monocytes , red blood cells were lysed and bone marrow cells were stained with flurochrome-conjugated anti-mouse Ig antibodies ( CD11b [Mac-1] and Ly6G [1A8] ) and sorted using a FACS ARIA instrument ( BD Biosciences ) . Neutrophils ( CD11b+ Ly6G+ ) and monocytes ( CD11b+Ly6G− ) were cultured in 5% FCS RPMI at 1 × 105 and 0 . 5 × 105 cells per well , respectively , in a 96-well u-bottom tissue culture plate . BMDMs were isolated and cultured as described previously ( Wong et al . , 2014 ) . Keratinocytes , MDFs , neutrophils , monocytes , and BMDMs were stimulated with TNF ( 100 ng/ml ) , Nec-1 ( 50 μM ) , QVD-Oph ( 10 μM , 20 μM for neutrophils and monocytes ) and CpdA ( 911 , 500 nM ) . After 24 hr ( 20 hr for neutrophils and monocytes ) all cells except keratinocytes were stained with propidium iodide ( PI ) and cell death analyzed on a FACScalibur instrument ( BD Biosciences ) . For the keratinocyte MTS viability assay phenazine methosulfate ( PMS; 0 . 92 mg/ml in PBS; Sigma-Aldrich ) and 3- ( 4 , 5-dimethylthiazol-2-yl ) -5- ( 3-carboxymethoxyphenyl ) -2- ( 4-sulfophenyl ) -2H-tetrazolium ( MTS; 2 mg/ml in PBS; Promega; Fitchburg , WI ) were combined in a 1:20 ratio . The mixture was added to cell culture media in a 1:5 ratio and incubated for 1–4 hr at 37°C in a humidified 5% CO2 incubator . Media was transferred to a flat-bottom 96-well plate and absorbance was measured at 490 nm . Viability was calculated relative to the untreated sample . For HaCaT experiments , cells stably expressing HOIP , HOIPC885S ( catalytically inactive ) , or an empty vector were seeded and incubated the next day with 100 ng/ml histidine-tagged TNF for 24 hr . Supernatant and adherent cells were harvested and resuspended in PBS containing 5 mg/ml PI . PI-positive cells were measured by flow cytometry ( BD Accuri; BD Biosciences ) . Paraffin-embedded tissue was fixed in 10% neutral buffered formalin then processed and stained with hematoxylin and eosin ( H&E ) according to standard practices . Immunohistochemistry and immunofluorescence analysis was performed as described previously ( Rickard et al . , 2014 ) . Ly5 . 1 mice were irradiated with 2 × 550 rads spaced 3 hr apart . Following red blood cell lysis , ∼ 5 × 106 BM cells from Shpnm/m ( Ly5 . 2 ) mice were intravenously injected . Mice were maintained on 2 mg/ml neomycin in drinking water for 3 weeks post irradiation . Reconstitution efficiency was assessed 6 weeks and 12 months post reconstitution by staining for Ly5 . 1 and Ly5 . 2 in peripheral blood obtained from retro-orbital bleeding . Cytokines were analyzed using a BioPlex Pro mouse cytokine 23-plex kit ( Bio-Rad; Hercules , CA ) , or for analysis of TNF levels a mouse TNF ELISA kit ( eBioscience; San Diego , CA ) was used . Skin lysates were prepared by homogenizing skin in ice cold protein lysis buffer ( 20 mM Tris pH 7 . 5 , 150 mM NaCl , 2 mM EDTA , 1% Triton-X100 , 10% glycerol ) using a Tissue Lyser II ( QIAGEN; Hilden , Germany ) for 12 cycles of 30 s at 30 Hz . A BCA kit ( Thermo Fisher Scientific ) was used to normalise protein levels . Values below the reference range were assigned the value of the lowest standard . Peripheral blood was collected into EDTA tubes ( Sarstedt; Nümbrecht , Germany ) and analyzed using an ADVIA 2120 hematological analyzer . For TNF-RSC isolation , immortalised MEFs were stimulated with 3xFlag-2xStrep-TNF at 0 . 5 μg/ml for the indicated times , or left untreated . Cells were subsequently solubilized in lysis buffer ( 30 mM Tris–HCl ( pH 7 . 4 ) , 150 mM NaCl , 2 mM EDTA , 2 mM KCl , 10% Glycerol , 1% Triton X100 , EDTA-free proteinase inhibitor cocktail ( Roche ) and 1x phosphatase inhibitor cocktail 2 ( Sigma Aldrich ) ) at 4°C for 30 min . The lysates were cleared by centrifugation , and 3xFlag-2xStrep-TNF was added to the untreated samples . Next , lysates were subjected to anti-Flag immunoprecipitation using M2 beads ( Sigma Aldrich ) for 16 hr . The beads were washed three times with lysis buffer , proteins were eluted in 50 μl of LDS buffer ( Life Technologies/Thermo Fisher Scientific ) containing 50 mM DTT . Samples were analyzed by western blotting . Antibodies used were: HOIP ( custom-made , Thermo Fisher Scientific ) , SHARPIN ( 14626-1-AP; ProteinTech; Chicago , IL ) , TNFR1 ( ab19139; Abcam; Cambridge , UK ) , and linear ubiquitin ( Genentech; South San Francisco , CA ) . Pearson chi-square and Fisher's exact test were used to assess frequencies of observed vs expected genotypes during development , at birth , and at weaning . Fisher's exact test was used for epidermal thickness statistical calculations . Student's t test was used to calculate statistical significance shown for all other data .

In response to an injury or infection , areas of the body can become inflamed as the immune system attempts to repair the damage and/or destroy any microbes or toxins that have entered the body . At the level of individual cells inflammation can involve cells being programmed to die in one of two ways: apoptosis and necroptosis . Apoptosis is a highly controlled process during which the contents of the cell are safely destroyed in order to prevent damage to surrounding cells . Necroptosis , on the other hand , is not controlled: the cell bursts and releases its contents into the surroundings . Inflammation is activated by a protein called TNFR1 , which is controlled by a complex that includes a protein called SHARPIN . Mice that lack the SHARPIN protein develop inflammation on the skin and internal organs , even in the absence of injury or infection . However , it is not clear how SHARPIN controls TNFR1 to prevent inflammation . Rickard et al . and , independently Kumari et al . have now studied this process in detail . Rickard et al . cross bred mice that lack SHARPIN with mice lacking other proteins involved in inflammation and cell death . The experiments show that apoptosis is the main form of cell death in skin inflammation , but necroptosis has a bigger role in the inflammation of internal organs . Mice that lack both the apoptotic and necroptotic cell-death pathways can develop relatively normally , but they die shortly after birth if they also lack SHARPIN . Experiments on these mice could help us to understand how SHARPIN works .

2014

TNFR1-dependent cell death drives inflammation in Sharpin-deficient mice